AuxXxilium/linux_dsm_epyc7002
1
0
Fork 0
mirror of https://github.com/AuxXxilium/linux_dsm_epyc7002.git synced 2024-12-19 01:37:22 +07:00
Code Issues Actions Packages Projects Releases Wiki Activity
350779a29f
linux_dsm_epyc7002/arch/powerpc/lib/sstep.c

2723 lines
59 KiB
C
Raw Normal View History

powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
/*
* Single-step support.
*
* Copyright (C) 2004 Paul Mackerras <paulus@au.ibm.com>, IBM
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/kernel.h>
[POWERPC] Disallow kprobes on emulate_step and branch_taken On powerpc, probing on emulate_step function will crash 2.6.18.1 when it is triggered. When kprobe is triggered, emulate_step() is on its kernel path and will cause recursive kprobe fault. And branch_taken() is called in emulate_step(). This disallows kprobes on both of them. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-11-01 09:50:15 +07:00
#include <linux/kprobes.h>
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
#include <linux/ptrace.h>
sanitize <linux/prefetch.h> usage Commit e66eed651fd1 ("list: remove prefetching from regular list iterators") removed the include of prefetch.h from list.h, which uncovered several cases that had apparently relied on that rather obscure header file dependency. So this fixes things up a bit, using grep -L linux/prefetch.h $(git grep -l '[^a-z_]prefetchw*(' -- '*.[ch]') grep -L 'prefetchw*(' $(git grep -l 'linux/prefetch.h' -- '*.[ch]') to guide us in finding files that either need <linux/prefetch.h> inclusion, or have it despite not needing it. There are more of them around (mostly network drivers), but this gets many core ones. Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-21 02:50:29 +07:00
#include <linux/prefetch.h>
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
#include <asm/sstep.h>
#include <asm/processor.h>
Replace <asm/uaccess.h> with <linux/uaccess.h> globally This was entirely automated, using the script by Al: PATT='^[[:blank:]]*#[[:blank:]]*include[[:blank:]]*<asm/uaccess.h>' sed -i -e "s!$PATT!#include <linux/uaccess.h>!" \ $(git grep -l "$PATT"|grep -v ^include/linux/uaccess.h) to do the replacement at the end of the merge window. Requested-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-25 02:46:01 +07:00
#include <linux/uaccess.h>
powerpc/lib: Fix randconfig build failure in sstep.c Under some configs we need to explicitly include cpu_has_feature.h, otherwise we fail with: arch/powerpc/lib/sstep.c:1992:7: error: implicit declaration of function 'cpu_has_feature' Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 07:51:14 +07:00
#include <asm/cpu_has_feature.h>
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#include <asm/cputable.h>
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
extern char system_call_common[];
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#ifdef CONFIG_PPC64
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
/* Bits in SRR1 that are copied from MSR */
[PATCH] powerpc: fix various sparse warnings Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-03-23 13:38:10 +07:00
#define MSR_MASK 0xffffffff87c0ffffUL
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#else
#define MSR_MASK 0x87c0ffff
#endif
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/* Bits in XER */
#define XER_SO 0x80000000U
#define XER_OV 0x40000000U
#define XER_CA 0x20000000U
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/*
* Functions in ldstfp.S
*/
extern int do_lfs(int rn, unsigned long ea);
extern int do_lfd(int rn, unsigned long ea);
extern int do_stfs(int rn, unsigned long ea);
extern int do_stfd(int rn, unsigned long ea);
extern int do_lvx(int rn, unsigned long ea);
extern int do_stvx(int rn, unsigned long ea);
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
extern void load_vsrn(int vsr, const void *p);
extern void store_vsrn(int vsr, void *p);
extern void conv_sp_to_dp(const float *sp, double *dp);
extern void conv_dp_to_sp(const double *dp, float *sp);
#endif
#ifdef __powerpc64__
/*
* Functions in quad.S
*/
extern int do_lq(unsigned long ea, unsigned long *regs);
extern int do_stq(unsigned long ea, unsigned long val0, unsigned long val1);
extern int do_lqarx(unsigned long ea, unsigned long *regs);
extern int do_stqcx(unsigned long ea, unsigned long val0, unsigned long val1,
unsigned int *crp);
#endif
#ifdef __LITTLE_ENDIAN__
#define IS_LE 1
#define IS_BE 0
#else
#define IS_LE 0
#define IS_BE 1
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
/*
* Emulate the truncation of 64 bit values in 32-bit mode.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long truncate_if_32bit(unsigned long msr,
unsigned long val)
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
{
#ifdef __powerpc64__
if ((msr & MSR_64BIT) == 0)
val &= 0xffffffffUL;
#endif
return val;
}
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
/*
* Determine whether a conditional branch instruction would branch.
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline int branch_taken(unsigned int instr,
const struct pt_regs *regs,
struct instruction_op *op)
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
{
unsigned int bo = (instr >> 21) & 0x1f;
unsigned int bi;
if ((bo & 4) == 0) {
/* decrement counter */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type |= DECCTR;
if (((bo >> 1) & 1) ^ (regs->ctr == 1))
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
return 0;
}
if ((bo & 0x10) == 0) {
/* check bit from CR */
bi = (instr >> 16) & 0x1f;
if (((regs->ccr >> (31 - bi)) & 1) != ((bo >> 3) & 1))
return 0;
}
return 1;
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline long address_ok(struct pt_regs *regs, unsigned long ea, int nb)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
if (!user_mode(regs))
return 1;
return __access_ok(ea, nb, USER_DS);
}
/*
* Calculate effective address for a D-form instruction
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline unsigned long dform_ea(unsigned int instr,
const struct pt_regs *regs)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int ra;
unsigned long ea;
ra = (instr >> 16) & 0x1f;
ea = (signed short) instr; /* sign-extend */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (ra)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
ea += regs->gpr[ra];
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
return truncate_if_32bit(regs->msr, ea);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
#ifdef __powerpc64__
/*
* Calculate effective address for a DS-form instruction
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline unsigned long dsform_ea(unsigned int instr,
const struct pt_regs *regs)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int ra;
unsigned long ea;
ra = (instr >> 16) & 0x1f;
ea = (signed short) (instr & ~3); /* sign-extend */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (ra)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
ea += regs->gpr[ra];
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
return truncate_if_32bit(regs->msr, ea);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
/*
* Calculate effective address for a DQ-form instruction
*/
static nokprobe_inline unsigned long dqform_ea(unsigned int instr,
const struct pt_regs *regs)
{
int ra;
unsigned long ea;
ra = (instr >> 16) & 0x1f;
ea = (signed short) (instr & ~0xf); /* sign-extend */
if (ra)
ea += regs->gpr[ra];
return truncate_if_32bit(regs->msr, ea);
}
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif /* __powerpc64 */
/*
* Calculate effective address for an X-form instruction
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long xform_ea(unsigned int instr,
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
const struct pt_regs *regs)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int ra, rb;
unsigned long ea;
ra = (instr >> 16) & 0x1f;
rb = (instr >> 11) & 0x1f;
ea = regs->gpr[rb];
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (ra)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
ea += regs->gpr[ra];
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
return truncate_if_32bit(regs->msr, ea);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
/*
* Return the largest power of 2, not greater than sizeof(unsigned long),
* such that x is a multiple of it.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long max_align(unsigned long x)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
x |= sizeof(unsigned long);
return x & -x; /* isolates rightmost bit */
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long byterev_2(unsigned long x)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
return ((x >> 8) & 0xff) | ((x & 0xff) << 8);
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long byterev_4(unsigned long x)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
return ((x >> 24) & 0xff) | ((x >> 8) & 0xff00) |
((x & 0xff00) << 8) | ((x & 0xff) << 24);
}
#ifdef __powerpc64__
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline unsigned long byterev_8(unsigned long x)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
return (byterev_4(x) << 32) | byterev_4(x >> 32);
}
#endif
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int read_mem_aligned(unsigned long *dest,
unsigned long ea, int nb)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err = 0;
unsigned long x = 0;
switch (nb) {
case 1:
err = __get_user(x, (unsigned char __user *) ea);
break;
case 2:
err = __get_user(x, (unsigned short __user *) ea);
break;
case 4:
err = __get_user(x, (unsigned int __user *) ea);
break;
#ifdef __powerpc64__
case 8:
err = __get_user(x, (unsigned long __user *) ea);
break;
#endif
}
if (!err)
*dest = x;
return err;
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int read_mem_unaligned(unsigned long *dest,
unsigned long ea, int nb, struct pt_regs *regs)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err;
unsigned long x, b, c;
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#ifdef __LITTLE_ENDIAN__
int len = nb; /* save a copy of the length for byte reversal */
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/* unaligned, do this in pieces */
x = 0;
for (; nb > 0; nb -= c) {
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#ifdef __LITTLE_ENDIAN__
c = 1;
#endif
#ifdef __BIG_ENDIAN__
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
c = max_align(ea);
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (c > nb)
c = max_align(nb);
err = read_mem_aligned(&b, ea, c);
if (err)
return err;
x = (x << (8 * c)) + b;
ea += c;
}
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#ifdef __LITTLE_ENDIAN__
switch (len) {
case 2:
*dest = byterev_2(x);
break;
case 4:
*dest = byterev_4(x);
break;
#ifdef __powerpc64__
case 8:
*dest = byterev_8(x);
break;
#endif
}
#endif
#ifdef __BIG_ENDIAN__
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
*dest = x;
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
return 0;
}
/*
* Read memory at address ea for nb bytes, return 0 for success
* or -EFAULT if an error occurred.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static int read_mem(unsigned long *dest, unsigned long ea, int nb,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
struct pt_regs *regs)
{
if (!address_ok(regs, ea, nb))
return -EFAULT;
if ((ea & (nb - 1)) == 0)
return read_mem_aligned(dest, ea, nb);
return read_mem_unaligned(dest, ea, nb, regs);
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(read_mem);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int write_mem_aligned(unsigned long val,
unsigned long ea, int nb)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err = 0;
switch (nb) {
case 1:
err = __put_user(val, (unsigned char __user *) ea);
break;
case 2:
err = __put_user(val, (unsigned short __user *) ea);
break;
case 4:
err = __put_user(val, (unsigned int __user *) ea);
break;
#ifdef __powerpc64__
case 8:
err = __put_user(val, (unsigned long __user *) ea);
break;
#endif
}
return err;
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int write_mem_unaligned(unsigned long val,
unsigned long ea, int nb, struct pt_regs *regs)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err;
unsigned long c;
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#ifdef __LITTLE_ENDIAN__
switch (nb) {
case 2:
val = byterev_2(val);
break;
case 4:
val = byterev_4(val);
break;
#ifdef __powerpc64__
case 8:
val = byterev_8(val);
break;
#endif
}
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/* unaligned or little-endian, do this in pieces */
for (; nb > 0; nb -= c) {
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#ifdef __LITTLE_ENDIAN__
c = 1;
#endif
#ifdef __BIG_ENDIAN__
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
c = max_align(ea);
powerpc: Fix Unaligned Loads and Stores This patch modifies the unaligned access routines of the sstep.c module so that it properly reverses the bytes of storage operands in the little endian kernel kernel. This is implemented by breaking an unaligned little endian access into a combination of single byte accesses plus an overal byte reversal operation. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:42:08 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (c > nb)
c = max_align(nb);
err = write_mem_aligned(val >> (nb - c) * 8, ea, c);
if (err)
return err;
powerpc: Unaligned stores and stmw are broken in emulation code The stmw instruction was incorrectly decoded as an update form instruction and thus the RA register was being clobbered. Also, the utility routine to write memory to unaligned addresses breaks the operation into smaller aligned accesses but was incorrectly incrementing the address by only one; it needs to increment the address by the size of the smaller aligned chunk. Signed-off-by: Tom Musta <tmusta@us.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-08-22 21:25:28 +07:00
ea += c;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
return 0;
}
/*
* Write memory at address ea for nb bytes, return 0 for success
* or -EFAULT if an error occurred.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static int write_mem(unsigned long val, unsigned long ea, int nb,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
struct pt_regs *regs)
{
if (!address_ok(regs, ea, nb))
return -EFAULT;
if ((ea & (nb - 1)) == 0)
return write_mem_aligned(val, ea, nb);
return write_mem_unaligned(val, ea, nb, regs);
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(write_mem);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
/*
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
* Check the address and alignment, and call func to do the actual
* load or store.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static int do_fp_load(int rn, int (*func)(int, unsigned long),
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ea, int nb,
struct pt_regs *regs)
{
int err;
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
union {
double dbl;
unsigned long ul[2];
struct {
#ifdef __BIG_ENDIAN__
unsigned _pad_;
unsigned word;
#endif
#ifdef __LITTLE_ENDIAN__
unsigned word;
unsigned _pad_;
#endif
} single;
} data;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ptr;
if (!address_ok(regs, ea, nb))
return -EFAULT;
if ((ea & 3) == 0)
return (*func)(rn, ea);
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
ptr = (unsigned long) &data.ul;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (sizeof(unsigned long) == 8 || nb == 4) {
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = read_mem_unaligned(&data.ul[0], ea, nb, regs);
if (nb == 4)
ptr = (unsigned long)&(data.single.word);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
} else {
/* reading a double on 32-bit */
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = read_mem_unaligned(&data.ul[0], ea, 4, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (!err)
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = read_mem_unaligned(&data.ul[1], ea + 4, 4, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
if (err)
return err;
return (*func)(rn, ptr);
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(do_fp_load);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static int do_fp_store(int rn, int (*func)(int, unsigned long),
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ea, int nb,
struct pt_regs *regs)
{
int err;
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
union {
double dbl;
unsigned long ul[2];
struct {
#ifdef __BIG_ENDIAN__
unsigned _pad_;
unsigned word;
#endif
#ifdef __LITTLE_ENDIAN__
unsigned word;
unsigned _pad_;
#endif
} single;
} data;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ptr;
if (!address_ok(regs, ea, nb))
return -EFAULT;
if ((ea & 3) == 0)
return (*func)(rn, ea);
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
ptr = (unsigned long) &data.ul[0];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (sizeof(unsigned long) == 8 || nb == 4) {
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
if (nb == 4)
ptr = (unsigned long)&(data.single.word);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
err = (*func)(rn, ptr);
if (err)
return err;
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = write_mem_unaligned(data.ul[0], ea, nb, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
} else {
/* writing a double on 32-bit */
err = (*func)(rn, ptr);
if (err)
return err;
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = write_mem_unaligned(data.ul[0], ea, 4, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (!err)
powerpc: Fix Unaligned LE Floating Point Loads and Stores This patch addresses unaligned single precision floating point loads and stores in the single-step code. The old implementation improperly treated an 8 byte structure as an array of two 4 byte words, which is a classic little endian bug. Signed-off-by: Tom Musta <tmusta@gmail.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-10-19 02:44:17 +07:00
err = write_mem_unaligned(data.ul[1], ea + 4, 4, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
return err;
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(do_fp_store);
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef CONFIG_ALTIVEC
/* For Altivec/VMX, no need to worry about alignment */
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int do_vec_load(int rn, int (*func)(int, unsigned long),
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ea, struct pt_regs *regs)
{
if (!address_ok(regs, ea & ~0xfUL, 16))
return -EFAULT;
return (*func)(rn, ea);
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int do_vec_store(int rn, int (*func)(int, unsigned long),
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long ea, struct pt_regs *regs)
{
if (!address_ok(regs, ea & ~0xfUL, 16))
return -EFAULT;
return (*func)(rn, ea);
}
#endif /* CONFIG_ALTIVEC */
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
static nokprobe_inline int emulate_lq(struct pt_regs *regs, unsigned long ea,
int reg)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err;
if (!address_ok(regs, ea, 16))
return -EFAULT;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
/* if aligned, should be atomic */
if ((ea & 0xf) == 0)
return do_lq(ea, &regs->gpr[reg]);
err = read_mem(&regs->gpr[reg + IS_LE], ea, 8, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (!err)
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
err = read_mem(&regs->gpr[reg + IS_BE], ea + 8, 8, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
return err;
}
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
static nokprobe_inline int emulate_stq(struct pt_regs *regs, unsigned long ea,
int reg)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
int err;
if (!address_ok(regs, ea, 16))
return -EFAULT;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
/* if aligned, should be atomic */
if ((ea & 0xf) == 0)
return do_stq(ea, regs->gpr[reg], regs->gpr[reg + 1]);
err = write_mem(regs->gpr[reg + IS_LE], ea, 8, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (!err)
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
err = write_mem(regs->gpr[reg + IS_BE], ea + 8, 8, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
return err;
}
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#endif /* __powerpc64 */
#ifdef CONFIG_VSX
void emulate_vsx_load(struct instruction_op *op, union vsx_reg *reg,
const void *mem)
{
int size, read_size;
int i, j;
const unsigned int *wp;
const unsigned short *hp;
const unsigned char *bp;
size = GETSIZE(op->type);
reg->d[0] = reg->d[1] = 0;
switch (op->element_size) {
case 16:
/* whole vector; lxv[x] or lxvl[l] */
if (size == 0)
break;
memcpy(reg, mem, size);
if (IS_LE && (op->vsx_flags & VSX_LDLEFT)) {
/* reverse 16 bytes */
unsigned long tmp;
tmp = byterev_8(reg->d[0]);
reg->d[0] = byterev_8(reg->d[1]);
reg->d[1] = tmp;
}
break;
case 8:
/* scalar loads, lxvd2x, lxvdsx */
read_size = (size >= 8) ? 8 : size;
i = IS_LE ? 8 : 8 - read_size;
memcpy(&reg->b[i], mem, read_size);
if (size < 8) {
if (op->type & SIGNEXT) {
/* size == 4 is the only case here */
reg->d[IS_LE] = (signed int) reg->d[IS_LE];
} else if (op->vsx_flags & VSX_FPCONV) {
preempt_disable();
conv_sp_to_dp(&reg->fp[1 + IS_LE],
&reg->dp[IS_LE]);
preempt_enable();
}
} else {
if (size == 16)
reg->d[IS_BE] = *(unsigned long *)(mem + 8);
else if (op->vsx_flags & VSX_SPLAT)
reg->d[IS_BE] = reg->d[IS_LE];
}
break;
case 4:
/* lxvw4x, lxvwsx */
wp = mem;
for (j = 0; j < size / 4; ++j) {
i = IS_LE ? 3 - j : j;
reg->w[i] = *wp++;
}
if (op->vsx_flags & VSX_SPLAT) {
u32 val = reg->w[IS_LE ? 3 : 0];
for (; j < 4; ++j) {
i = IS_LE ? 3 - j : j;
reg->w[i] = val;
}
}
break;
case 2:
/* lxvh8x */
hp = mem;
for (j = 0; j < size / 2; ++j) {
i = IS_LE ? 7 - j : j;
reg->h[i] = *hp++;
}
break;
case 1:
/* lxvb16x */
bp = mem;
for (j = 0; j < size; ++j) {
i = IS_LE ? 15 - j : j;
reg->b[i] = *bp++;
}
break;
}
}
EXPORT_SYMBOL_GPL(emulate_vsx_load);
NOKPROBE_SYMBOL(emulate_vsx_load);
void emulate_vsx_store(struct instruction_op *op, const union vsx_reg *reg,
void *mem)
{
int size, write_size;
int i, j;
union vsx_reg buf;
unsigned int *wp;
unsigned short *hp;
unsigned char *bp;
size = GETSIZE(op->type);
switch (op->element_size) {
case 16:
/* stxv, stxvx, stxvl, stxvll */
if (size == 0)
break;
if (IS_LE && (op->vsx_flags & VSX_LDLEFT)) {
/* reverse 16 bytes */
buf.d[0] = byterev_8(reg->d[1]);
buf.d[1] = byterev_8(reg->d[0]);
reg = &buf;
}
memcpy(mem, reg, size);
break;
case 8:
/* scalar stores, stxvd2x */
write_size = (size >= 8) ? 8 : size;
i = IS_LE ? 8 : 8 - write_size;
if (size < 8 && op->vsx_flags & VSX_FPCONV) {
buf.d[0] = buf.d[1] = 0;
preempt_disable();
conv_dp_to_sp(&reg->dp[IS_LE], &buf.fp[1 + IS_LE]);
preempt_enable();
reg = &buf;
}
memcpy(mem, &reg->b[i], write_size);
if (size == 16)
memcpy(mem + 8, &reg->d[IS_BE], 8);
break;
case 4:
/* stxvw4x */
wp = mem;
for (j = 0; j < size / 4; ++j) {
i = IS_LE ? 3 - j : j;
*wp++ = reg->w[i];
}
break;
case 2:
/* stxvh8x */
hp = mem;
for (j = 0; j < size / 2; ++j) {
i = IS_LE ? 7 - j : j;
*hp++ = reg->h[i];
}
break;
case 1:
/* stvxb16x */
bp = mem;
for (j = 0; j < size; ++j) {
i = IS_LE ? 15 - j : j;
*bp++ = reg->b[i];
}
break;
}
}
EXPORT_SYMBOL_GPL(emulate_vsx_store);
NOKPROBE_SYMBOL(emulate_vsx_store);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif /* CONFIG_VSX */
#define __put_user_asmx(x, addr, err, op, cr) \
__asm__ __volatile__( \
"1: " op " %2,0,%3\n" \
" mfcr %1\n" \
"2:\n" \
".section .fixup,\"ax\"\n" \
"3: li %0,%4\n" \
" b 2b\n" \
".previous\n" \
powerpc: EX_TABLE macro for exception tables This macro is taken from s390, and allows more flexibility in changing exception table format. mpe: Put it in ppc_asm.h and only define one version using stringinfy_in_c(). Add some empty definitions and headers to keep the selftests happy. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-10-13 12:42:53 +07:00
EX_TABLE(1b, 3b) \
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
: "=r" (err), "=r" (cr) \
: "r" (x), "r" (addr), "i" (-EFAULT), "0" (err))
#define __get_user_asmx(x, addr, err, op) \
__asm__ __volatile__( \
"1: "op" %1,0,%2\n" \
"2:\n" \
".section .fixup,\"ax\"\n" \
"3: li %0,%3\n" \
" b 2b\n" \
".previous\n" \
powerpc: EX_TABLE macro for exception tables This macro is taken from s390, and allows more flexibility in changing exception table format. mpe: Put it in ppc_asm.h and only define one version using stringinfy_in_c(). Add some empty definitions and headers to keep the selftests happy. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-10-13 12:42:53 +07:00
EX_TABLE(1b, 3b) \
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
: "=r" (err), "=r" (x) \
: "r" (addr), "i" (-EFAULT), "0" (err))
#define __cacheop_user_asmx(addr, err, op) \
__asm__ __volatile__( \
"1: "op" 0,%1\n" \
"2:\n" \
".section .fixup,\"ax\"\n" \
"3: li %0,%3\n" \
" b 2b\n" \
".previous\n" \
powerpc: EX_TABLE macro for exception tables This macro is taken from s390, and allows more flexibility in changing exception table format. mpe: Put it in ppc_asm.h and only define one version using stringinfy_in_c(). Add some empty definitions and headers to keep the selftests happy. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-10-13 12:42:53 +07:00
EX_TABLE(1b, 3b) \
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
: "=r" (err) \
: "r" (addr), "i" (-EFAULT), "0" (err))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void set_cr0(const struct pt_regs *regs,
struct instruction_op *op, int rd)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
long val = regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type |= SETCC;
op->ccval = (regs->ccr & 0x0fffffff) | ((regs->xer >> 3) & 0x10000000);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
if (!(regs->msr & MSR_64BIT))
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
val = (int) val;
#endif
if (val < 0)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval |= 0x80000000;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else if (val > 0)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval |= 0x40000000;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval |= 0x20000000;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void add_with_carry(const struct pt_regs *regs,
struct instruction_op *op, int rd,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long val1, unsigned long val2,
unsigned long carry_in)
{
unsigned long val = val1 + val2;
if (carry_in)
++val;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETREG + SETXER;
op->reg = rd;
op->val = val;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
if (!(regs->msr & MSR_64BIT)) {
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
val = (unsigned int) val;
val1 = (unsigned int) val1;
}
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval = regs->xer;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (val < val1 || (carry_in && val == val1))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval |= XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval &= ~XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_cmp_signed(const struct pt_regs *regs,
struct instruction_op *op,
long v1, long v2, int crfld)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
unsigned int crval, shift;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETCC;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
crval = (regs->xer >> 31) & 1; /* get SO bit */
if (v1 < v2)
crval |= 8;
else if (v1 > v2)
crval |= 4;
else
crval |= 2;
shift = (7 - crfld) * 4;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval = (regs->ccr & ~(0xf << shift)) | (crval << shift);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_cmp_unsigned(const struct pt_regs *regs,
struct instruction_op *op,
unsigned long v1,
unsigned long v2, int crfld)
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
{
unsigned int crval, shift;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETCC;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
crval = (regs->xer >> 31) & 1; /* get SO bit */
if (v1 < v2)
crval |= 8;
else if (v1 > v2)
crval |= 4;
else
crval |= 2;
shift = (7 - crfld) * 4;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval = (regs->ccr & ~(0xf << shift)) | (crval << shift);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_cmpb(const struct pt_regs *regs,
struct instruction_op *op,
unsigned long v1, unsigned long v2)
powerpc/lib/sstep: Add cmpb instruction emulation This patch adds emulation of the cmpb instruction, enabling xmon to emulate this instruction. Tested for correctness against the cmpb asm instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:22 +07:00
{
unsigned long long out_val, mask;
int i;
out_val = 0;
for (i = 0; i < 8; i++) {
mask = 0xffUL << (i * 8);
if ((v1 & mask) == (v2 & mask))
out_val |= mask;
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = out_val;
powerpc/lib/sstep: Add cmpb instruction emulation This patch adds emulation of the cmpb instruction, enabling xmon to emulate this instruction. Tested for correctness against the cmpb asm instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:22 +07:00
}
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
/*
* The size parameter is used to adjust the equivalent popcnt instruction.
* popcntb = 8, popcntw = 32, popcntd = 64
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_popcnt(const struct pt_regs *regs,
struct instruction_op *op,
unsigned long v1, int size)
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
{
unsigned long long out = v1;
out -= (out >> 1) & 0x5555555555555555;
out = (0x3333333333333333 & out) + (0x3333333333333333 & (out >> 2));
out = (out + (out >> 4)) & 0x0f0f0f0f0f0f0f0f;
if (size == 8) { /* popcntb */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = out;
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
return;
}
out += out >> 8;
out += out >> 16;
if (size == 32) { /* popcntw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = out & 0x0000003f0000003f;
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
return;
}
out = (out + (out >> 32)) & 0x7f;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = out; /* popcntd */
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
}
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
#ifdef CONFIG_PPC64
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_bpermd(const struct pt_regs *regs,
struct instruction_op *op,
unsigned long v1, unsigned long v2)
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
{
unsigned char perm, idx;
unsigned int i;
perm = 0;
for (i = 0; i < 8; i++) {
idx = (v1 >> (i * 8)) & 0xff;
if (idx < 64)
if (v2 & PPC_BIT(idx))
perm |= 1 << i;
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = perm;
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
}
#endif /* CONFIG_PPC64 */
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
/*
* The size parameter adjusts the equivalent prty instruction.
* prtyw = 32, prtyd = 64
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
static nokprobe_inline void do_prty(const struct pt_regs *regs,
struct instruction_op *op,
unsigned long v, int size)
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
{
unsigned long long res = v ^ (v >> 8);
res ^= res >> 16;
if (size == 32) { /* prtyw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = res & 0x0000000100000001;
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
return;
}
res ^= res >> 32;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = res & 1; /*prtyd */
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
}
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int trap_compare(long v1, long v2)
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
{
int ret = 0;
if (v1 < v2)
ret |= 0x10;
else if (v1 > v2)
ret |= 0x08;
else
ret |= 0x04;
if ((unsigned long)v1 < (unsigned long)v2)
ret |= 0x02;
else if ((unsigned long)v1 > (unsigned long)v2)
ret |= 0x01;
return ret;
}
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/*
* Elements of 32-bit rotate and mask instructions.
*/
#define MASK32(mb, me) ((0xffffffffUL >> (mb)) + \
((signed long)-0x80000000L >> (me)) + ((me) >= (mb)))
#ifdef __powerpc64__
#define MASK64_L(mb) (~0UL >> (mb))
#define MASK64_R(me) ((signed long)-0x8000000000000000L >> (me))
#define MASK64(mb, me) (MASK64_L(mb) + MASK64_R(me) + ((me) >= (mb)))
#define DATA32(x) (((x) & 0xffffffffUL) | (((x) & 0xffffffffUL) << 32))
#else
#define DATA32(x) (x)
#endif
#define ROTATE(x, n) ((n) ? (((x) << (n)) | ((x) >> (8 * sizeof(long) - (n)))) : (x))
/*
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
* Decode an instruction, and return information about it in *op
* without changing *regs.
* Integer arithmetic and logical instructions, branches, and barrier
* instructions can be emulated just using the information in *op.
*
* Return value is 1 if the instruction can be emulated just by
* updating *regs with the information in *op, -1 if we need the
* GPRs but *regs doesn't contain the full register set, or 0
* otherwise.
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
*/
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
int analyse_instr(struct instruction_op *op, const struct pt_regs *regs,
unsigned int instr)
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
{
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned int opcode, ra, rb, rd, spr, u;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
unsigned long int imm;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
unsigned long int val, val2;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
unsigned int mb, me, sh;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
long ival;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = COMPUTE;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
opcode = instr >> 26;
switch (opcode) {
case 16: /* bc */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = BRANCH;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
imm = (signed short)(instr & 0xfffc);
if ((instr & 2) == 0)
imm += regs->nip;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = truncate_if_32bit(regs->msr, imm);
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
if (instr & 1)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type |= SETLK;
if (branch_taken(instr, regs, op))
op->type |= BRTAKEN;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
return 1;
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#ifdef CONFIG_PPC64
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
case 17: /* sc */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if ((instr & 0xfe2) == 2)
op->type = SYSCALL;
else
op->type = UNKNOWN;
return 0;
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#endif
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
case 18: /* b */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BRANCH | BRTAKEN;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
imm = instr & 0x03fffffc;
if (imm & 0x02000000)
imm -= 0x04000000;
if ((instr & 2) == 0)
imm += regs->nip;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = truncate_if_32bit(regs->msr, imm);
powerpc: Use MSR_64BIT in sstep.c, fix kprobes on BOOK3E We check MSR_SF a lot in sstep.c, to decide if we need to emulate the truncation of values when running in 32-bit mode. Factor out that code into a helper, and convert it and the other uses to use MSR_64BIT. This fixes a bug on BOOK3E where kprobes would end up returning to a 32-bit address, because regs->nip was truncated, because (msr & MSR_SF) was false. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-04-08 04:56:04 +07:00
if (instr & 1)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type |= SETLK;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
return 1;
case 19:
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch ((instr >> 1) & 0x3ff) {
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
case 0: /* mcrf */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETCC;
powerpc: Fix emulation of mcrf in emulate_step() The mcrf emulation code was using the CR field number directly as the shift value, without taking into account that CR fields are numbered from 0-7 starting at the high bits. That meant it was looking at the CR fields in the reverse order. Fixes: cf87c3f6b647 ("powerpc: Emulate icbi, mcrf and conditional-trap instructions") Cc: stable@vger.kernel.org # v3.18+ Signed-off-by: Anton Blanchard <anton@samba.org> Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-06-15 06:46:38 +07:00
rd = 7 - ((instr >> 23) & 0x7);
ra = 7 - ((instr >> 18) & 0x7);
rd *= 4;
ra *= 4;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
val = (regs->ccr >> ra) & 0xf;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval = (regs->ccr & ~(0xfUL << rd)) | (val << rd);
return 1;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 16: /* bclr */
case 528: /* bcctr */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = BRANCH;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
imm = (instr & 0x400)? regs->ctr: regs->link;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = truncate_if_32bit(regs->msr, imm);
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
if (instr & 1)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type |= SETLK;
if (branch_taken(instr, regs, op))
op->type |= BRTAKEN;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 18: /* rfid, scary */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (regs->msr & MSR_PR)
goto priv;
op->type = RFI;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 150: /* isync */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BARRIER | BARRIER_ISYNC;
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 33: /* crnor */
case 129: /* crandc */
case 193: /* crxor */
case 225: /* crnand */
case 257: /* crand */
case 289: /* creqv */
case 417: /* crorc */
case 449: /* cror */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETCC;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
ra = (instr >> 16) & 0x1f;
rb = (instr >> 11) & 0x1f;
rd = (instr >> 21) & 0x1f;
ra = (regs->ccr >> (31 - ra)) & 1;
rb = (regs->ccr >> (31 - rb)) & 1;
val = (instr >> (6 + ra * 2 + rb)) & 1;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->ccval = (regs->ccr & ~(1UL << (31 - rd))) |
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(val << (31 - rd));
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
default:
op->type = UNKNOWN;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
break;
case 31:
switch ((instr >> 1) & 0x3ff) {
case 598: /* sync */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BARRIER + BARRIER_SYNC;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
switch ((instr >> 21) & 3) {
case 1: /* lwsync */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BARRIER + BARRIER_LWSYNC;
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 2: /* ptesync */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BARRIER + BARRIER_PTESYNC;
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 854: /* eieio */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = BARRIER + BARRIER_EIEIO;
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
break;
}
/* Following cases refer to regs->gpr[], so we need all regs */
if (!FULL_REGS(regs))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return -1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
rd = (instr >> 21) & 0x1f;
ra = (instr >> 16) & 0x1f;
rb = (instr >> 11) & 0x1f;
switch (opcode) {
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
#ifdef __powerpc64__
case 2: /* tdi */
if (rd & trap_compare(regs->gpr[ra], (short) instr))
goto trap;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
#endif
case 3: /* twi */
if (rd & trap_compare((int)regs->gpr[ra], (short) instr))
goto trap;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 7: /* mulli */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[ra] * (short) instr;
goto compute_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 8: /* subfic */
imm = (short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, ~regs->gpr[ra], imm, 1);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 10: /* cmpli */
imm = (unsigned short) instr;
val = regs->gpr[ra];
#ifdef __powerpc64__
if ((rd & 1) == 0)
val = (unsigned int) val;
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_cmp_unsigned(regs, op, val, imm, rd >> 2);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 11: /* cmpi */
imm = (short) instr;
val = regs->gpr[ra];
#ifdef __powerpc64__
if ((rd & 1) == 0)
val = (int) val;
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_cmp_signed(regs, op, val, imm, rd >> 2);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 12: /* addic */
imm = (short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra], imm, 0);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 13: /* addic. */
imm = (short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra], imm, 0);
set_cr0(regs, op, rd);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 14: /* addi */
imm = (short) instr;
if (ra)
imm += regs->gpr[ra];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = imm;
goto compute_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 15: /* addis */
imm = ((short) instr) << 16;
if (ra)
imm += regs->gpr[ra];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = imm;
goto compute_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 20: /* rlwimi */
mb = (instr >> 6) & 0x1f;
me = (instr >> 1) & 0x1f;
val = DATA32(regs->gpr[rd]);
imm = MASK32(mb, me);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (regs->gpr[ra] & ~imm) | (ROTATE(val, rb) & imm);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 21: /* rlwinm */
mb = (instr >> 6) & 0x1f;
me = (instr >> 1) & 0x1f;
val = DATA32(regs->gpr[rd]);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ROTATE(val, rb) & MASK32(mb, me);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 23: /* rlwnm */
mb = (instr >> 6) & 0x1f;
me = (instr >> 1) & 0x1f;
rb = regs->gpr[rb] & 0x1f;
val = DATA32(regs->gpr[rd]);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ROTATE(val, rb) & MASK32(mb, me);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 24: /* ori */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] | (unsigned short) instr;
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 25: /* oris */
imm = (unsigned short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] | (imm << 16);
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 26: /* xori */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] ^ (unsigned short) instr;
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 27: /* xoris */
imm = (unsigned short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] ^ (imm << 16);
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 28: /* andi. */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] & (unsigned short) instr;
set_cr0(regs, op, ra);
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 29: /* andis. */
imm = (unsigned short) instr;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] & (imm << 16);
set_cr0(regs, op, ra);
goto logical_done_nocc;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 30: /* rld* */
mb = ((instr >> 6) & 0x1f) | (instr & 0x20);
val = regs->gpr[rd];
if ((instr & 0x10) == 0) {
sh = rb | ((instr & 2) << 4);
val = ROTATE(val, sh);
switch ((instr >> 2) & 3) {
case 0: /* rldicl */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
val &= MASK64_L(mb);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 1: /* rldicr */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
val &= MASK64_R(mb);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 2: /* rldic */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
val &= MASK64(mb, 63 - sh);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 3: /* rldimi */
imm = MASK64(mb, 63 - sh);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
val = (regs->gpr[ra] & ~imm) |
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(val & imm);
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = val;
goto logical_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
} else {
sh = regs->gpr[rb] & 0x3f;
val = ROTATE(val, sh);
switch ((instr >> 1) & 7) {
case 0: /* rldcl */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = val & MASK64_L(mb);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 1: /* rldcr */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = val & MASK64_R(mb);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
}
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
}
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = UNKNOWN; /* illegal instruction */
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
case 31:
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch ((instr >> 1) & 0x3ff) {
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
case 4: /* tw */
if (rd == 0x1f ||
(rd & trap_compare((int)regs->gpr[ra],
(int)regs->gpr[rb])))
goto trap;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
#ifdef __powerpc64__
case 68: /* td */
if (rd & trap_compare(regs->gpr[ra], regs->gpr[rb]))
goto trap;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 83: /* mfmsr */
if (regs->msr & MSR_PR)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto priv;
op->type = MFMSR;
op->reg = rd;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 146: /* mtmsr */
if (regs->msr & MSR_PR)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto priv;
op->type = MTMSR;
op->reg = rd;
op->val = 0xffffffff & ~(MSR_ME | MSR_LE);
return 0;
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#ifdef CONFIG_PPC64
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 178: /* mtmsrd */
if (regs->msr & MSR_PR)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto priv;
op->type = MTMSR;
op->reg = rd;
/* only MSR_EE and MSR_RI get changed if bit 15 set */
/* mtmsrd doesn't change MSR_HV, MSR_ME or MSR_LE */
imm = (instr & 0x10000)? 0x8002: 0xefffffffffffeffeUL;
op->val = imm;
return 0;
powerpc: Make single-stepping emulation (mostly) usable on 32-bit The sc instruction emulation can't be done the same way on 32-bit as 64-bit yet, but this should work OK. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-10-28 19:48:08 +07:00
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 19: /* mfcr */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
imm = 0xffffffffUL;
powerpc: Fix emulation of mfocrf in emulate_step() From POWER4 onwards, mfocrf() only places the specified CR field into the destination GPR, and the rest of it is set to 0. The PowerPC AS from version 3.0 now requires this behaviour. The emulation code currently puts the entire CR into the destination GPR. Fix it. Fixes: 6888199f7fe5 ("[POWERPC] Emulate more instructions in software") Cc: stable@vger.kernel.org # v2.6.22+ Signed-off-by: Anton Blanchard <anton@samba.org> Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-06-15 06:46:39 +07:00
if ((instr >> 20) & 1) {
imm = 0xf0000000UL;
for (sh = 0; sh < 8; ++sh) {
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
if (instr & (0x80000 >> sh))
powerpc: Fix emulation of mfocrf in emulate_step() From POWER4 onwards, mfocrf() only places the specified CR field into the destination GPR, and the rest of it is set to 0. The PowerPC AS from version 3.0 now requires this behaviour. The emulation code currently puts the entire CR into the destination GPR. Fix it. Fixes: 6888199f7fe5 ("[POWERPC] Emulate more instructions in software") Cc: stable@vger.kernel.org # v2.6.22+ Signed-off-by: Anton Blanchard <anton@samba.org> Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-06-15 06:46:39 +07:00
break;
imm >>= 4;
}
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->ccr & imm;
goto compute_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 144: /* mtcrf */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETCC;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
imm = 0xf0000000UL;
val = regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->ccr;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
for (sh = 0; sh < 8; ++sh) {
if (instr & (0x80000 >> sh))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (op->val & ~imm) |
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(val & imm);
imm >>= 4;
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 339: /* mfspr */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
spr = ((instr >> 16) & 0x1f) | ((instr >> 6) & 0x3e0);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = MFSPR;
op->reg = rd;
op->spr = spr;
if (spr == SPRN_XER || spr == SPRN_LR ||
spr == SPRN_CTR)
return 1;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 467: /* mtspr */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
spr = ((instr >> 16) & 0x1f) | ((instr >> 6) & 0x3e0);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = MTSPR;
op->val = regs->gpr[rd];
op->spr = spr;
if (spr == SPRN_XER || spr == SPRN_LR ||
spr == SPRN_CTR)
return 1;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/*
* Compare instructions
*/
case 0: /* cmp */
val = regs->gpr[ra];
val2 = regs->gpr[rb];
#ifdef __powerpc64__
if ((rd & 1) == 0) {
/* word (32-bit) compare */
val = (int) val;
val2 = (int) val2;
}
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_cmp_signed(regs, op, val, val2, rd >> 2);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 32: /* cmpl */
val = regs->gpr[ra];
val2 = regs->gpr[rb];
#ifdef __powerpc64__
if ((rd & 1) == 0) {
/* word (32-bit) compare */
val = (unsigned int) val;
val2 = (unsigned int) val2;
}
#endif
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_cmp_unsigned(regs, op, val, val2, rd >> 2);
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc/lib/sstep: Add cmpb instruction emulation This patch adds emulation of the cmpb instruction, enabling xmon to emulate this instruction. Tested for correctness against the cmpb asm instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:22 +07:00
case 508: /* cmpb */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_cmpb(regs, op, regs->gpr[rd], regs->gpr[rb]);
goto logical_done_nocc;
powerpc/lib/sstep: Add cmpb instruction emulation This patch adds emulation of the cmpb instruction, enabling xmon to emulate this instruction. Tested for correctness against the cmpb asm instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:22 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
/*
* Arithmetic instructions
*/
case 8: /* subfc */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, ~regs->gpr[ra],
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->gpr[rb], 1);
goto arith_done;
#ifdef __powerpc64__
case 9: /* mulhdu */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
asm("mulhdu %0,%1,%2" : "=r" (op->val) :
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
"r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
goto arith_done;
#endif
case 10: /* addc */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra],
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->gpr[rb], 0);
goto arith_done;
case 11: /* mulhwu */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
asm("mulhwu %0,%1,%2" : "=r" (op->val) :
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
"r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
goto arith_done;
case 40: /* subf */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rb] - regs->gpr[ra];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
#ifdef __powerpc64__
case 73: /* mulhd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
asm("mulhd %0,%1,%2" : "=r" (op->val) :
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
"r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
goto arith_done;
#endif
case 75: /* mulhw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
asm("mulhw %0,%1,%2" : "=r" (op->val) :
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
"r" (regs->gpr[ra]), "r" (regs->gpr[rb]));
goto arith_done;
case 104: /* neg */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = -regs->gpr[ra];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
case 136: /* subfe */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, ~regs->gpr[ra],
regs->gpr[rb], regs->xer & XER_CA);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
case 138: /* adde */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra],
regs->gpr[rb], regs->xer & XER_CA);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
case 200: /* subfze */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, ~regs->gpr[ra], 0L,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->xer & XER_CA);
goto arith_done;
case 202: /* addze */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra], 0L,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->xer & XER_CA);
goto arith_done;
case 232: /* subfme */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, ~regs->gpr[ra], -1L,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->xer & XER_CA);
goto arith_done;
#ifdef __powerpc64__
case 233: /* mulld */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[ra] * regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
#endif
case 234: /* addme */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
add_with_carry(regs, op, rd, regs->gpr[ra], -1L,
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
regs->xer & XER_CA);
goto arith_done;
case 235: /* mullw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (unsigned int) regs->gpr[ra] *
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(unsigned int) regs->gpr[rb];
goto arith_done;
case 266: /* add */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[ra] + regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
#ifdef __powerpc64__
case 457: /* divdu */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[ra] / regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto arith_done;
#endif
case 459: /* divwu */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (unsigned int) regs->gpr[ra] /
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(unsigned int) regs->gpr[rb];
goto arith_done;
#ifdef __powerpc64__
case 489: /* divd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (long int) regs->gpr[ra] /
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(long int) regs->gpr[rb];
goto arith_done;
#endif
case 491: /* divw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (int) regs->gpr[ra] /
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
(int) regs->gpr[rb];
goto arith_done;
/*
* Logical instructions
*/
powerpc/lib/sstep: Add isel instruction emulation This adds emulation for the isel instruction. Tested for correctness against the isel instruction and its extended mnemonics (lt, gt, eq) on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:26 +07:00
case 15: /* isel */
mb = (instr >> 6) & 0x1f; /* bc */
val = (regs->ccr >> (31 - mb)) & 1;
val2 = (ra) ? regs->gpr[ra] : 0;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (val) ? val2 : regs->gpr[rb];
goto compute_done;
powerpc/lib/sstep: Add isel instruction emulation This adds emulation for the isel instruction. Tested for correctness against the isel instruction and its extended mnemonics (lt, gt, eq) on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:26 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 26: /* cntlzw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = __builtin_clz((unsigned int) regs->gpr[rd]);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#ifdef __powerpc64__
case 58: /* cntlzd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = __builtin_clzl(regs->gpr[rd]);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#endif
case 28: /* and */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] & regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 60: /* andc */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] & ~regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
case 122: /* popcntb */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_popcnt(regs, op, regs->gpr[rd], 8);
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
goto logical_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 124: /* nor */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ~(regs->gpr[rd] | regs->gpr[rb]);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
case 154: /* prtyw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_prty(regs, op, regs->gpr[rd], 32);
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
goto logical_done;
case 186: /* prtyd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_prty(regs, op, regs->gpr[rd], 64);
powerpc/lib/sstep: Add prty instruction emulation This adds emulation for the prtyw and prtyd instructions. Tested for logical correctness against the prtyw and prtyd instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:25 +07:00
goto logical_done;
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
#ifdef CONFIG_PPC64
case 252: /* bpermd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_bpermd(regs, op, regs->gpr[rd], regs->gpr[rb]);
powerpc/lib/sstep: Add bpermd instruction emulation This adds emulation for the bpermd instruction. Tested for correctness against the bpermd instruction on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:24 +07:00
goto logical_done;
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 284: /* xor */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ~(regs->gpr[rd] ^ regs->gpr[rb]);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 316: /* xor */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] ^ regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
case 378: /* popcntw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_popcnt(regs, op, regs->gpr[rd], 32);
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
goto logical_done;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 412: /* orc */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] | ~regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 444: /* or */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] | regs->gpr[rb];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 476: /* nand */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ~(regs->gpr[rd] & regs->gpr[rb]);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
#ifdef CONFIG_PPC64
case 506: /* popcntd */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
do_popcnt(regs, op, regs->gpr[rd], 64);
powerpc/lib/sstep: Add popcnt instruction emulation This adds emulations for the popcntb, popcntw, and popcntd instructions. Tested for correctness against the popcnt{b,w,d} instructions on ppc64le. Signed-off-by: Matt Brown <matthew.brown.dev@gmail.com> Reviewed-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-31 07:58:23 +07:00
goto logical_done;
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 922: /* extsh */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (signed short) regs->gpr[rd];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 954: /* extsb */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (signed char) regs->gpr[rd];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#ifdef __powerpc64__
case 986: /* extsw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (signed int) regs->gpr[rd];
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#endif
/*
* Shift instructions
*/
case 24: /* slw */
sh = regs->gpr[rb] & 0x3f;
if (sh < 32)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (regs->gpr[rd] << sh) & 0xffffffffUL;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 536: /* srw */
sh = regs->gpr[rb] & 0x3f;
if (sh < 32)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = (regs->gpr[rd] & 0xffffffffUL) >> sh;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 792: /* sraw */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETREG + SETXER;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
sh = regs->gpr[rb] & 0x3f;
ival = (signed int) regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ival >> (sh < 32 ? sh : 31);
op->xerval = regs->xer;
powerpc: Fix bugs in emulate_step() This fixes some bugs in emulate_step(). First, the setting of the carry bit for the arithmetic right-shift instructions was not correct on 64-bit machines because we were masking with a mask of type int rather than unsigned long. Secondly, the sld (shift left doubleword) instruction was using the wrong instruction field for the register containing the shift count. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-07-19 14:47:57 +07:00
if (ival < 0 && (sh >= 32 || (ival & ((1ul << sh) - 1)) != 0))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval |= XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval &= ~XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 824: /* srawi */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETREG + SETXER;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
sh = rb;
ival = (signed int) regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ival >> sh;
op->xerval = regs->xer;
powerpc: Fix bugs in emulate_step() This fixes some bugs in emulate_step(). First, the setting of the carry bit for the arithmetic right-shift instructions was not correct on 64-bit machines because we were masking with a mask of type int rather than unsigned long. Secondly, the sld (shift left doubleword) instruction was using the wrong instruction field for the register containing the shift count. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-07-19 14:47:57 +07:00
if (ival < 0 && (ival & ((1ul << sh) - 1)) != 0)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval |= XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval &= ~XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#ifdef __powerpc64__
case 27: /* sld */
powerpc: Fix bugs in emulate_step() This fixes some bugs in emulate_step(). First, the setting of the carry bit for the arithmetic right-shift instructions was not correct on 64-bit machines because we were masking with a mask of type int rather than unsigned long. Secondly, the sld (shift left doubleword) instruction was using the wrong instruction field for the register containing the shift count. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-07-19 14:47:57 +07:00
sh = regs->gpr[rb] & 0x7f;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
if (sh < 64)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] << sh;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 539: /* srd */
sh = regs->gpr[rb] & 0x7f;
if (sh < 64)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = regs->gpr[rd] >> sh;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 794: /* srad */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETREG + SETXER;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
sh = regs->gpr[rb] & 0x7f;
ival = (signed long int) regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ival >> (sh < 64 ? sh : 63);
op->xerval = regs->xer;
powerpc: Fix bugs in emulate_step() This fixes some bugs in emulate_step(). First, the setting of the carry bit for the arithmetic right-shift instructions was not correct on 64-bit machines because we were masking with a mask of type int rather than unsigned long. Secondly, the sld (shift left doubleword) instruction was using the wrong instruction field for the register containing the shift count. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-07-19 14:47:57 +07:00
if (ival < 0 && (sh >= 64 || (ival & ((1ul << sh) - 1)) != 0))
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval |= XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval &= ~XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
case 826: /* sradi with sh_5 = 0 */
case 827: /* sradi with sh_5 = 1 */
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->type = COMPUTE + SETREG + SETXER;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
sh = rb | ((instr & 2) << 4);
ival = (signed long int) regs->gpr[rd];
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->val = ival >> sh;
op->xerval = regs->xer;
powerpc: Fix bugs in emulate_step() This fixes some bugs in emulate_step(). First, the setting of the carry bit for the arithmetic right-shift instructions was not correct on 64-bit machines because we were masking with a mask of type int rather than unsigned long. Secondly, the sld (shift left doubleword) instruction was using the wrong instruction field for the register containing the shift count. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-07-19 14:47:57 +07:00
if (ival < 0 && (ival & ((1ul << sh) - 1)) != 0)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval |= XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
else
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
op->xerval &= ~XER_CA;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
goto logical_done;
#endif /* __powerpc64__ */
/*
* Cache instructions
*/
case 54: /* dcbst */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(CACHEOP, DCBST, 0);
op->ea = xform_ea(instr, regs);
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 86: /* dcbf */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(CACHEOP, DCBF, 0);
op->ea = xform_ea(instr, regs);
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 246: /* dcbtst */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(CACHEOP, DCBTST, 0);
op->ea = xform_ea(instr, regs);
op->reg = rd;
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 278: /* dcbt */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(CACHEOP, DCBTST, 0);
op->ea = xform_ea(instr, regs);
op->reg = rd;
return 0;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
case 982: /* icbi */
op->type = MKOP(CACHEOP, ICBI, 0);
op->ea = xform_ea(instr, regs);
return 0;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
}
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
break;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
}
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
/*
* Loads and stores.
*/
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = UNKNOWN;
op->update_reg = ra;
op->reg = rd;
op->val = regs->gpr[rd];
u = (instr >> 20) & UPDATE;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
op->vsx_flags = 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch (opcode) {
case 31:
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
u = instr & UPDATE;
op->ea = xform_ea(instr, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch ((instr >> 1) & 0x3ff) {
case 20: /* lwarx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LARX, 0, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 150: /* stwcx. */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STCX, 0, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 84: /* ldarx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LARX, 0, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 214: /* stdcx. */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STCX, 0, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 52: /* lbarx */
op->type = MKOP(LARX, 0, 1);
break;
case 694: /* stbcx. */
op->type = MKOP(STCX, 0, 1);
break;
case 116: /* lharx */
op->type = MKOP(LARX, 0, 2);
break;
case 726: /* sthcx. */
op->type = MKOP(STCX, 0, 2);
break;
case 276: /* lqarx */
if (!((rd & 1) || rd == ra || rd == rb))
op->type = MKOP(LARX, 0, 16);
break;
case 182: /* stqcx. */
if (!(rd & 1))
op->type = MKOP(STCX, 0, 16);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
case 23: /* lwzx */
case 55: /* lwzux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 87: /* lbzx */
case 119: /* lbzux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 1);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef CONFIG_ALTIVEC
case 103: /* lvx */
case 359: /* lvxl */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD_VMX, 0, 16);
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
op->element_size = 16;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 231: /* stvx */
case 487: /* stvxl */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE_VMX, 0, 16);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif /* CONFIG_ALTIVEC */
#ifdef __powerpc64__
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 21: /* ldx */
case 53: /* ldux */
op->type = MKOP(LOAD, u, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 149: /* stdx */
case 181: /* stdux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
case 151: /* stwx */
case 183: /* stwux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 215: /* stbx */
case 247: /* stbux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 1);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 279: /* lhzx */
case 311: /* lhzux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 2);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 341: /* lwax */
case 373: /* lwaux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, SIGNEXT | u, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
case 343: /* lhax */
case 375: /* lhaux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, SIGNEXT | u, 2);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 407: /* sthx */
case 439: /* sthux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 2);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 532: /* ldbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, BYTEREV, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
case 533: /* lswx */
op->type = MKOP(LOAD_MULTI, 0, regs->xer & 0x7f);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 534: /* lwbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, BYTEREV, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
case 597: /* lswi */
if (rb == 0)
rb = 32; /* # bytes to load */
op->type = MKOP(LOAD_MULTI, 0, rb);
op->ea = 0;
if (ra)
op->ea = truncate_if_32bit(regs->msr,
regs->gpr[ra]);
break;
powerpc: fix typo 'CONFIG_PPC_CPU' Commit cd64d1697cf0 ("powerpc: mtmsrd not defined") added a check for CONFIG_PPC_CPU were a check for CONFIG_PPC_FPU was clearly intended. Fixes: cd64d1697cf0 ("powerpc: mtmsrd not defined") Signed-off-by: Paul Bolle <pebolle@tiscali.nl> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-05-21 02:59:42 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 535: /* lfsx */
case 567: /* lfsux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD_FP, u, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 599: /* lfdx */
case 631: /* lfdux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD_FP, u, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 663: /* stfsx */
case 695: /* stfsux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE_FP, u, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 727: /* stfdx */
case 759: /* stfdux */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE_FP, u, 8);
break;
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 660: /* stdbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, BYTEREV, 8);
op->val = byterev_8(regs->gpr[rd]);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
case 661: /* stswx */
op->type = MKOP(STORE_MULTI, 0, regs->xer & 0x7f);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 662: /* stwbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, BYTEREV, 4);
op->val = byterev_4(regs->gpr[rd]);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
case 725:
if (rb == 0)
rb = 32; /* # bytes to store */
op->type = MKOP(STORE_MULTI, 0, rb);
op->ea = 0;
if (ra)
op->ea = truncate_if_32bit(regs->msr,
regs->gpr[ra]);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 790: /* lhbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, BYTEREV, 2);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 918: /* sthbrx */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, BYTEREV, 2);
op->val = byterev_2(regs->gpr[rd]);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef CONFIG_VSX
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 12: /* lxsiwzx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 4);
op->element_size = 8;
break;
case 76: /* lxsiwax */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, SIGNEXT, 4);
op->element_size = 8;
break;
case 140: /* stxsiwx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 4);
op->element_size = 8;
break;
case 268: /* lxvx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 16;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 269: /* lxvl */
case 301: { /* lxvll */
int nb;
op->reg = rd | ((instr & 1) << 5);
op->ea = ra ? regs->gpr[ra] : 0;
nb = regs->gpr[rb] & 0xff;
if (nb > 16)
nb = 16;
op->type = MKOP(LOAD_VSX, 0, nb);
op->element_size = 16;
op->vsx_flags = ((instr & 0x20) ? VSX_LDLEFT : 0) |
VSX_CHECK_VEC;
break;
}
case 332: /* lxvdsx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 8);
op->element_size = 8;
op->vsx_flags = VSX_SPLAT;
break;
case 364: /* lxvwsx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 4);
op->element_size = 4;
op->vsx_flags = VSX_SPLAT | VSX_CHECK_VEC;
break;
case 396: /* stxvx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 16;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 397: /* stxvl */
case 429: { /* stxvll */
int nb;
op->reg = rd | ((instr & 1) << 5);
op->ea = ra ? regs->gpr[ra] : 0;
nb = regs->gpr[rb] & 0xff;
if (nb > 16)
nb = 16;
op->type = MKOP(STORE_VSX, 0, nb);
op->element_size = 16;
op->vsx_flags = ((instr & 0x20) ? VSX_LDLEFT : 0) |
VSX_CHECK_VEC;
break;
}
case 524: /* lxsspx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 4);
op->element_size = 8;
op->vsx_flags = VSX_FPCONV;
break;
case 588: /* lxsdx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 8);
op->element_size = 8;
break;
case 652: /* stxsspx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 4);
op->element_size = 8;
op->vsx_flags = VSX_FPCONV;
break;
case 716: /* stxsdx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 8);
op->element_size = 8;
break;
case 780: /* lxvw4x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 4;
break;
case 781: /* lxsibzx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 1);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 812: /* lxvh8x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 2;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 813: /* lxsihzx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 2);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 844: /* lxvd2x */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->reg = rd | ((instr & 1) << 5);
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 8;
break;
case 876: /* lxvb16x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 1;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 908: /* stxvw4x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 4;
break;
case 909: /* stxsibx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 1);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 940: /* stxvh8x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 2;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 941: /* stxsihx */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 2);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 972: /* stxvd2x */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->reg = rd | ((instr & 1) << 5);
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 8;
break;
case 1004: /* stxvb16x */
op->reg = rd | ((instr & 1) << 5);
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 1;
op->vsx_flags = VSX_CHECK_VEC;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#endif /* CONFIG_VSX */
}
break;
case 32: /* lwz */
case 33: /* lwzu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 4);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 34: /* lbz */
case 35: /* lbzu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 1);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 36: /* stw */
powerpc/kprobe: Don't emulate store when kprobe stwu r1 We don't do the real store operation for kprobing 'stwu Rx,(y)R1' since this may corrupt the exception frame, now we will do this operation safely in exception return code after migrate current exception frame below the kprobed function stack. So we only update gpr[1] here and trigger a thread flag to mask this. Note we should make sure if we trigger kernel stack over flow. Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-09-17 06:54:31 +07:00
case 37: /* stwu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 4);
op->ea = dform_ea(instr, regs);
break;
powerpc/kprobe: Don't emulate store when kprobe stwu r1 We don't do the real store operation for kprobing 'stwu Rx,(y)R1' since this may corrupt the exception frame, now we will do this operation safely in exception return code after migrate current exception frame below the kprobed function stack. So we only update gpr[1] here and trigger a thread flag to mask this. Note we should make sure if we trigger kernel stack over flow. Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-09-17 06:54:31 +07:00
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 38: /* stb */
case 39: /* stbu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 1);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 40: /* lhz */
case 41: /* lhzu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, u, 2);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 42: /* lha */
case 43: /* lhau */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, SIGNEXT | u, 2);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 44: /* sth */
case 45: /* sthu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, u, 2);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 46: /* lmw */
if (ra >= rd)
break; /* invalid form, ra in range to load */
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
op->type = MKOP(LOAD_MULTI, 0, 4 * (32 - rd));
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 47: /* stmw */
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
op->type = MKOP(STORE_MULTI, 0, 4 * (32 - rd));
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 48: /* lfs */
case 49: /* lfsu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD_FP, u, 4);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 50: /* lfd */
case 51: /* lfdu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD_FP, u, 8);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 52: /* stfs */
case 53: /* stfsu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE_FP, u, 4);
op->ea = dform_ea(instr, regs);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 54: /* stfd */
case 55: /* stfdu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE_FP, u, 8);
op->ea = dform_ea(instr, regs);
break;
powerpc: mtmsrd not defined Replace the BOOK3S_64 specific mtmsrd with the generic MTMSRD macro. Only enable ldstfp when CONFIG_PPC_FPU is set. Signed-off-by: Sean MacLennan <smaclennan@pikatech.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-09-01 14:21:21 +07:00
#endif
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
case 56: /* lq */
if (!((rd & 1) || (rd == ra)))
op->type = MKOP(LOAD, 0, 16);
op->ea = dqform_ea(instr, regs);
break;
#endif
#ifdef CONFIG_VSX
case 57: /* lxsd, lxssp */
op->ea = dsform_ea(instr, regs);
switch (instr & 3) {
case 2: /* lxsd */
op->reg = rd + 32;
op->type = MKOP(LOAD_VSX, 0, 8);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 3: /* lxssp */
op->reg = rd + 32;
op->type = MKOP(LOAD_VSX, 0, 4);
op->element_size = 8;
op->vsx_flags = VSX_FPCONV | VSX_CHECK_VEC;
break;
}
break;
#endif /* CONFIG_VSX */
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
#ifdef __powerpc64__
case 58: /* ld[u], lwa */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->ea = dsform_ea(instr, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch (instr & 3) {
case 0: /* ld */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, 0, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 1: /* ldu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, UPDATE, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 2: /* lwa */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(LOAD, SIGNEXT, 4);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
break;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#endif
#ifdef CONFIG_VSX
case 61: /* lxv, stxsd, stxssp, stxv */
switch (instr & 7) {
case 1: /* lxv */
op->ea = dqform_ea(instr, regs);
if (instr & 8)
op->reg = rd + 32;
op->type = MKOP(LOAD_VSX, 0, 16);
op->element_size = 16;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 2: /* stxsd with LSB of DS field = 0 */
case 6: /* stxsd with LSB of DS field = 1 */
op->ea = dsform_ea(instr, regs);
op->reg = rd + 32;
op->type = MKOP(STORE_VSX, 0, 8);
op->element_size = 8;
op->vsx_flags = VSX_CHECK_VEC;
break;
case 3: /* stxssp with LSB of DS field = 0 */
case 7: /* stxssp with LSB of DS field = 1 */
op->ea = dsform_ea(instr, regs);
op->reg = rd + 32;
op->type = MKOP(STORE_VSX, 0, 4);
op->element_size = 8;
op->vsx_flags = VSX_FPCONV | VSX_CHECK_VEC;
break;
case 5: /* stxv */
op->ea = dqform_ea(instr, regs);
if (instr & 8)
op->reg = rd + 32;
op->type = MKOP(STORE_VSX, 0, 16);
op->element_size = 16;
op->vsx_flags = VSX_CHECK_VEC;
break;
}
break;
#endif /* CONFIG_VSX */
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 62: /* std[u] */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->ea = dsform_ea(instr, regs);
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
switch (instr & 3) {
case 0: /* std */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, 0, 8);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
case 1: /* stdu */
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op->type = MKOP(STORE, UPDATE, 8);
break;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 2: /* stq */
if (!(rd & 1))
op->type = MKOP(STORE, 0, 16);
break;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
}
break;
#endif /* __powerpc64__ */
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
return 0;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
logical_done:
if (instr & 1)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
set_cr0(regs, op, ra);
logical_done_nocc:
op->reg = ra;
op->type |= SETREG;
return 1;
powerpc: Emulate most Book I instructions in emulate_step() This extends the emulate_step() function to handle a large proportion of the Book I instructions implemented on current 64-bit server processors. The aim is to handle all the load and store instructions used in the kernel, plus all of the instructions that appear between l[wd]arx and st[wd]cx., so this handles the Altivec/VMX lvx and stvx and the VSX lxv2dx and stxv2dx instructions (implemented in POWER7). The new code can emulate user mode instructions, and checks the effective address for a load or store if the saved state is for user mode. It doesn't handle little-endian mode at present. For floating-point, Altivec/VMX and VSX instructions, it checks that the saved MSR has the enable bit for the relevant facility set, and if so, assumes that the FP/VMX/VSX registers contain valid state, and does loads or stores directly to/from the FP/VMX/VSX registers, using assembly helpers in ldstfp.S. Instructions supported now include: * Loads and stores, including some but not all VMX and VSX instructions, and lmw/stmw * Atomic loads and stores (l[dw]arx, st[dw]cx.) * Arithmetic instructions (add, subtract, multiply, divide, etc.) * Compare instructions * Rotate and mask instructions * Shift instructions * Logical instructions (and, or, xor, etc.) * Condition register logical instructions * mtcrf, cntlz[wd], exts[bhw] * isync, sync, lwsync, ptesync, eieio * Cache operations (dcbf, dcbst, dcbt, dcbtst) The overflow-checking arithmetic instructions are not included, but they appear not to be ever used in C code. This uses decimal values for the minor opcodes in the switch statements because that is what appears in the Power ISA specification, thus it is easier to check that they are correct if they are in decimal. If this is used to single-step an instruction where a data breakpoint interrupt occurred, then there is the possibility that the instruction is a lwarx or ldarx. In that case we have to be careful not to lose the reservation until we get to the matching st[wd]cx., or we'll never make forward progress. One alternative is to try to arrange that we can return from interrupts and handle data breakpoint interrupts without losing the reservation, which means not using any spinlocks, mutexes, or atomic ops (including bitops). That seems rather fragile. The other alternative is to emulate the larx/stcx and all the instructions in between. This is why this commit adds support for a wide range of integer instructions. Signed-off-by: Paul Mackerras <paulus@samba.org>
2010-06-15 11:48:58 +07:00
arith_done:
if (instr & 1)
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
set_cr0(regs, op, rd);
compute_done:
op->reg = rd;
op->type |= SETREG;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
return 1;
priv:
op->type = INTERRUPT | 0x700;
op->val = SRR1_PROGPRIV;
return 0;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
trap:
op->type = INTERRUPT | 0x700;
op->val = SRR1_PROGTRAP;
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
}
EXPORT_SYMBOL_GPL(analyse_instr);
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(analyse_instr);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
/*
* For PPC32 we always use stwu with r1 to change the stack pointer.
* So this emulated store may corrupt the exception frame, now we
* have to provide the exception frame trampoline, which is pushed
* below the kprobed function stack. So we only update gpr[1] but
* don't emulate the real store operation. We will do real store
* operation safely in exception return code by checking this flag.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline int handle_stack_update(unsigned long ea, struct pt_regs *regs)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
{
#ifdef CONFIG_PPC32
/*
* Check if we will touch kernel stack overflow
*/
if (ea - STACK_INT_FRAME_SIZE <= current->thread.ksp_limit) {
printk(KERN_CRIT "Can't kprobe this since kernel stack would overflow.\n");
return -EINVAL;
}
#endif /* CONFIG_PPC32 */
/*
* Check if we already set since that means we'll
* lose the previous value.
*/
WARN_ON(test_thread_flag(TIF_EMULATE_STACK_STORE));
set_thread_flag(TIF_EMULATE_STACK_STORE);
return 0;
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline void do_signext(unsigned long *valp, int size)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
{
switch (size) {
case 2:
*valp = (signed short) *valp;
break;
case 4:
*valp = (signed int) *valp;
break;
}
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
static nokprobe_inline void do_byterev(unsigned long *valp, int size)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
{
switch (size) {
case 2:
*valp = byterev_2(*valp);
break;
case 4:
*valp = byterev_4(*valp);
break;
#ifdef __powerpc64__
case 8:
*valp = byterev_8(*valp);
break;
#endif
}
}
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
/*
* Emulate an instruction that can be executed just by updating
* fields in *regs.
*/
void emulate_update_regs(struct pt_regs *regs, struct instruction_op *op)
{
unsigned long next_pc;
next_pc = truncate_if_32bit(regs->msr, regs->nip + 4);
switch (op->type & INSTR_TYPE_MASK) {
case COMPUTE:
if (op->type & SETREG)
regs->gpr[op->reg] = op->val;
if (op->type & SETCC)
regs->ccr = op->ccval;
if (op->type & SETXER)
regs->xer = op->xerval;
break;
case BRANCH:
if (op->type & SETLK)
regs->link = next_pc;
if (op->type & BRTAKEN)
next_pc = op->val;
if (op->type & DECCTR)
--regs->ctr;
break;
case BARRIER:
switch (op->type & BARRIER_MASK) {
case BARRIER_SYNC:
mb();
break;
case BARRIER_ISYNC:
isync();
break;
case BARRIER_EIEIO:
eieio();
break;
case BARRIER_LWSYNC:
asm volatile("lwsync" : : : "memory");
break;
case BARRIER_PTESYNC:
asm volatile("ptesync" : : : "memory");
break;
}
break;
case MFSPR:
switch (op->spr) {
case SPRN_XER:
regs->gpr[op->reg] = regs->xer & 0xffffffffUL;
break;
case SPRN_LR:
regs->gpr[op->reg] = regs->link;
break;
case SPRN_CTR:
regs->gpr[op->reg] = regs->ctr;
break;
default:
WARN_ON_ONCE(1);
}
break;
case MTSPR:
switch (op->spr) {
case SPRN_XER:
regs->xer = op->val & 0xffffffffUL;
break;
case SPRN_LR:
regs->link = op->val;
break;
case SPRN_CTR:
regs->ctr = op->val;
break;
default:
WARN_ON_ONCE(1);
}
break;
default:
WARN_ON_ONCE(1);
}
regs->nip = next_pc;
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
/*
* Emulate instructions that cause a transfer of control,
* loads and stores, and a few other instructions.
* Returns 1 if the step was emulated, 0 if not,
* or -1 if the instruction is one that should not be stepped,
* such as an rfid, or a mtmsrd that would clear MSR_RI.
*/
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
int emulate_step(struct pt_regs *regs, unsigned int instr)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
{
struct instruction_op op;
int r, err, size;
unsigned long val;
unsigned int cr;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
int i, rd, nb;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
r = analyse_instr(&op, regs, instr);
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
if (r < 0)
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
return r;
powerpc: Change analyse_instr so it doesn't modify *regs The analyse_instr function currently doesn't just work out what an instruction does, it also executes those instructions whose effect is only to update CPU registers that are stored in struct pt_regs. This is undesirable because optprobes uses analyse_instr to work out if an instruction could be successfully emulated in future. This changes analyse_instr so it doesn't modify *regs; instead it stores information in the instruction_op structure to indicate what registers (GPRs, CR, XER, LR) would be set and what value they would be set to. A companion function called emulate_update_regs() can then use that information to update a pt_regs struct appropriately. As a minor cleanup, this replaces inline asm using the cntlzw and cntlzd instructions with calls to __builtin_clz() and __builtin_clzl(). Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:25 +07:00
if (r > 0) {
emulate_update_regs(regs, &op);
return 1;
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
err = 0;
size = GETSIZE(op.type);
switch (op.type & INSTR_TYPE_MASK) {
case CACHEOP:
if (!address_ok(regs, op.ea, 8))
return 0;
switch (op.type & CACHEOP_MASK) {
case DCBST:
__cacheop_user_asmx(op.ea, err, "dcbst");
break;
case DCBF:
__cacheop_user_asmx(op.ea, err, "dcbf");
break;
case DCBTST:
if (op.reg == 0)
prefetchw((void *) op.ea);
break;
case DCBT:
if (op.reg == 0)
prefetch((void *) op.ea);
break;
powerpc: Emulate icbi, mcrf and conditional-trap instructions This extends the instruction emulation done by analyse_instr() and emulate_step() to handle a few more instructions that are found in the kernel. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:08 +07:00
case ICBI:
__cacheop_user_asmx(op.ea, err, "icbi");
break;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
}
if (err)
return 0;
goto instr_done;
case LARX:
if (op.ea & (size - 1))
break; /* can't handle misaligned */
if (!address_ok(regs, op.ea, size))
powerpc/sstep: Return directly after a failed address_ok() in emulate_step() Setting err and going to ldst_done just returns 0, without using err, so just return 0 directly. We already do that for other call sites in this function. Signed-off-by: Markus Elfring <elfring@users.sourceforge.net> [mpe: Rewrite change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-21 21:30:15 +07:00
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
err = 0;
switch (size) {
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
case 1:
__get_user_asmx(val, op.ea, err, "lbarx");
break;
case 2:
__get_user_asmx(val, op.ea, err, "lharx");
break;
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case 4:
__get_user_asmx(val, op.ea, err, "lwarx");
break;
powerpc/sstep: Fix sstep.c compile on powerpcspe Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") introduced ldarx and stdcx into the instructions in sstep.c, which are not accepted by the assembler on powerpcspe, but does seem to be accepted by the normal powerpc assembler even in 32 bit mode. Wrap these two instructions in a __powerpc64__ check like it is everywhere else in the file. Fixes: be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") Signed-off-by: Len Sorensen <lsorense@csclub.uwaterloo.ca> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-05-06 03:44:44 +07:00
#ifdef __powerpc64__
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case 8:
__get_user_asmx(val, op.ea, err, "ldarx");
break;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 16:
err = do_lqarx(op.ea, &regs->gpr[op.reg]);
goto ldst_done;
powerpc/sstep: Fix sstep.c compile on powerpcspe Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") introduced ldarx and stdcx into the instructions in sstep.c, which are not accepted by the assembler on powerpcspe, but does seem to be accepted by the normal powerpc assembler even in 32 bit mode. Wrap these two instructions in a __powerpc64__ check like it is everywhere else in the file. Fixes: be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") Signed-off-by: Len Sorensen <lsorense@csclub.uwaterloo.ca> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-05-06 03:44:44 +07:00
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
default:
return 0;
}
if (!err)
regs->gpr[op.reg] = val;
goto ldst_done;
case STCX:
if (op.ea & (size - 1))
break; /* can't handle misaligned */
if (!address_ok(regs, op.ea, size))
powerpc/sstep: Return directly after a failed address_ok() in emulate_step() Setting err and going to ldst_done just returns 0, without using err, so just return 0 directly. We already do that for other call sites in this function. Signed-off-by: Markus Elfring <elfring@users.sourceforge.net> [mpe: Rewrite change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-21 21:30:15 +07:00
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
err = 0;
switch (size) {
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
case 1:
__put_user_asmx(op.val, op.ea, err, "stbcx.", cr);
break;
case 2:
__put_user_asmx(op.val, op.ea, err, "stbcx.", cr);
break;
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case 4:
__put_user_asmx(op.val, op.ea, err, "stwcx.", cr);
break;
powerpc/sstep: Fix sstep.c compile on powerpcspe Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") introduced ldarx and stdcx into the instructions in sstep.c, which are not accepted by the assembler on powerpcspe, but does seem to be accepted by the normal powerpc assembler even in 32 bit mode. Wrap these two instructions in a __powerpc64__ check like it is everywhere else in the file. Fixes: be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") Signed-off-by: Len Sorensen <lsorense@csclub.uwaterloo.ca> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-05-06 03:44:44 +07:00
#ifdef __powerpc64__
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case 8:
__put_user_asmx(op.val, op.ea, err, "stdcx.", cr);
break;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case 16:
err = do_stqcx(op.ea, regs->gpr[op.reg],
regs->gpr[op.reg + 1], &cr);
break;
powerpc/sstep: Fix sstep.c compile on powerpcspe Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") introduced ldarx and stdcx into the instructions in sstep.c, which are not accepted by the assembler on powerpcspe, but does seem to be accepted by the normal powerpc assembler even in 32 bit mode. Wrap these two instructions in a __powerpc64__ check like it is everywhere else in the file. Fixes: be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") Signed-off-by: Len Sorensen <lsorense@csclub.uwaterloo.ca> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-05-06 03:44:44 +07:00
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
default:
return 0;
}
if (!err)
regs->ccr = (regs->ccr & 0x0fffffff) |
(cr & 0xe0000000) |
((regs->xer >> 3) & 0x10000000);
goto ldst_done;
case LOAD:
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
if (size == 16) {
err = emulate_lq(regs, op.ea, op.reg);
goto ldst_done;
}
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
err = read_mem(&regs->gpr[op.reg], op.ea, size, regs);
if (!err) {
if (op.type & SIGNEXT)
do_signext(&regs->gpr[op.reg], size);
if (op.type & BYTEREV)
do_byterev(&regs->gpr[op.reg], size);
}
goto ldst_done;
powerpc: Fix compilation of emulate_step() Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") added some calls to do_fp_load() and do_fp_store(), which fail to compile on configs with CONFIG_PPC_FPU=n and CONFIG_PPC_EMULATE_SSTEP=y. This fixes the compile by adding #ifdef CONFIG_PPC_FPU around the code that calls these functions. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-11-03 11:46:43 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case LOAD_FP:
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
if (!(regs->msr & MSR_FP))
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (size == 4)
err = do_fp_load(op.reg, do_lfs, op.ea, size, regs);
else
err = do_fp_load(op.reg, do_lfd, op.ea, size, regs);
goto ldst_done;
powerpc: Fix compilation of emulate_step() Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") added some calls to do_fp_load() and do_fp_store(), which fail to compile on configs with CONFIG_PPC_FPU=n and CONFIG_PPC_EMULATE_SSTEP=y. This fixes the compile by adding #ifdef CONFIG_PPC_FPU around the code that calls these functions. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-11-03 11:46:43 +07:00
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
#ifdef CONFIG_ALTIVEC
case LOAD_VMX:
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
if (!(regs->msr & MSR_VEC))
return 0;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
err = do_vec_load(op.reg, do_lvx, op.ea, regs);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto ldst_done;
#endif
#ifdef CONFIG_VSX
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case LOAD_VSX: {
char mem[16];
union vsx_reg buf;
unsigned long msrbit = MSR_VSX;
/*
* Some VSX instructions check the MSR_VEC bit rather than MSR_VSX
* when the target of the instruction is a vector register.
*/
if (op.reg >= 32 && (op.vsx_flags & VSX_CHECK_VEC))
msrbit = MSR_VEC;
if (!(regs->msr & msrbit))
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
return 0;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
if (!address_ok(regs, op.ea, size) ||
__copy_from_user(mem, (void __user *)op.ea, size))
return 0;
emulate_vsx_load(&op, &buf, mem);
load_vsrn(op.reg, &buf);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto ldst_done;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
#endif
case LOAD_MULTI:
if (regs->msr & MSR_LE)
return 0;
rd = op.reg;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
for (i = 0; i < size; i += 4) {
nb = size - i;
if (nb > 4)
nb = 4;
err = read_mem(&regs->gpr[rd], op.ea, nb, regs);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (err)
return 0;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
if (nb < 4) /* left-justify last bytes */
regs->gpr[rd] <<= 32 - 8 * nb;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
op.ea += 4;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
++rd;
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto instr_done;
case STORE:
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
#ifdef __powerpc64__
if (size == 16) {
err = emulate_stq(regs, op.ea, op.reg);
goto ldst_done;
}
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if ((op.type & UPDATE) && size == sizeof(long) &&
op.reg == 1 && op.update_reg == 1 &&
!(regs->msr & MSR_PR) &&
op.ea >= regs->gpr[1] - STACK_INT_FRAME_SIZE) {
err = handle_stack_update(op.ea, regs);
goto ldst_done;
}
err = write_mem(op.val, op.ea, size, regs);
goto ldst_done;
powerpc: Fix compilation of emulate_step() Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") added some calls to do_fp_load() and do_fp_store(), which fail to compile on configs with CONFIG_PPC_FPU=n and CONFIG_PPC_EMULATE_SSTEP=y. This fixes the compile by adding #ifdef CONFIG_PPC_FPU around the code that calls these functions. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-11-03 11:46:43 +07:00
#ifdef CONFIG_PPC_FPU
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
case STORE_FP:
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
if (!(regs->msr & MSR_FP))
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (size == 4)
err = do_fp_store(op.reg, do_stfs, op.ea, size, regs);
else
err = do_fp_store(op.reg, do_stfd, op.ea, size, regs);
goto ldst_done;
powerpc: Fix compilation of emulate_step() Commit be96f63375a1 ("powerpc: Split out instruction analysis part of emulate_step()") added some calls to do_fp_load() and do_fp_store(), which fail to compile on configs with CONFIG_PPC_FPU=n and CONFIG_PPC_EMULATE_SSTEP=y. This fixes the compile by adding #ifdef CONFIG_PPC_FPU around the code that calls these functions. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-11-03 11:46:43 +07:00
#endif
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
#ifdef CONFIG_ALTIVEC
case STORE_VMX:
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
if (!(regs->msr & MSR_VEC))
return 0;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
err = do_vec_store(op.reg, do_stvx, op.ea, regs);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto ldst_done;
#endif
#ifdef CONFIG_VSX
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
case STORE_VSX: {
char mem[16];
union vsx_reg buf;
unsigned long msrbit = MSR_VSX;
/*
* Some VSX instructions check the MSR_VEC bit rather than MSR_VSX
* when the target of the instruction is a vector register.
*/
if (op.reg >= 32 && (op.vsx_flags & VSX_CHECK_VEC))
msrbit = MSR_VEC;
if (!(regs->msr & msrbit))
return 0;
if (!address_ok(regs, op.ea, size))
return 0;
store_vsrn(op.reg, &buf);
emulate_vsx_store(&op, &buf, mem);
if (__copy_to_user((void __user *)op.ea, mem, size))
powerpc: Don't check MSR FP/VMX/VSX enable bits in analyse_instr() This removes the checks for the FP/VMX/VSX enable bits in the MSR from analyse_instr() and adds them to emulate_step() instead. The reason for this is that we may want to use analyse_instr() in a situation where the FP/VMX/VSX register values are stored in the current thread_struct and the FP/VMX/VSX enable bits in the MSR image in the pt_regs are zero. Since analyse_instr() doesn't make any changes to register state, it is reasonable for it to indicate what the effect of an instruction would be even though the relevant enable bit is off. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:26 +07:00
return 0;
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto ldst_done;
powerpc: Handle most loads and stores in instruction emulation code This extends the instruction emulation infrastructure in sstep.c to handle all the load and store instructions defined in the Power ISA v3.0, except for the atomic memory operations, ldmx (which was never implemented), lfdp/stfdp, and the vector element load/stores. The instructions added are: Integer loads and stores: lbarx, lharx, lqarx, stbcx., sthcx., stqcx., lq, stq. VSX loads and stores: lxsiwzx, lxsiwax, stxsiwx, lxvx, lxvl, lxvll, lxvdsx, lxvwsx, stxvx, stxvl, stxvll, lxsspx, lxsdx, stxsspx, stxsdx, lxvw4x, lxsibzx, lxvh8x, lxsihzx, lxvb16x, stxvw4x, stxsibx, stxvh8x, stxsihx, stxvb16x, lxsd, lxssp, lxv, stxsd, stxssp, stxv. These instructions are handled both in the analyse_instr phase and in the emulate_step phase. The code for lxvd2ux and stxvd2ux has been taken out, as those instructions were never implemented in any processor and have been taken out of the architecture, and their opcodes have been reused for other instructions in POWER9 (lxvb16x and stxvb16x). The emulation for the VSX loads and stores uses helper functions which don't access registers or memory directly, which can hopefully be reused by KVM later. Signed-off-by: Paul Mackerras <paulus@ozlabs.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-08-30 11:12:27 +07:00
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
#endif
case STORE_MULTI:
if (regs->msr & MSR_LE)
return 0;
rd = op.reg;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
for (i = 0; i < size; i += 4) {
val = regs->gpr[rd];
nb = size - i;
if (nb > 4)
nb = 4;
else
val >>= 32 - 8 * nb;
err = write_mem(val, op.ea, nb, regs);
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
if (err)
return 0;
op.ea += 4;
powerpc: Implement emulation of string loads and stores The size field of the op.type word is now the total number of bytes to be loaded or stored. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:09 +07:00
++rd;
}
powerpc: Split out instruction analysis part of emulate_step() This splits out the instruction analysis part of emulate_step() into a separate analyse_instr() function, which decodes the instruction, but doesn't execute any load or store instructions. It does execute integer instructions and branches which can be executed purely by updating register values in the pt_regs struct. For other instructions, it returns the instruction type and other details in a new instruction_op struct. emulate_step() then uses that information to execute loads, stores, cache operations, mfmsr, mtmsr[d], and (on 64-bit) sc instructions. The reason for doing this is so that the KVM code can use it instead of having its own separate instruction emulation code. Possibly the alignment interrupt handler could also use this. Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-09-02 11:35:07 +07:00
goto instr_done;
case MFMSR:
regs->gpr[op.reg] = regs->msr & MSR_MASK;
goto instr_done;
case MTMSR:
val = regs->gpr[op.reg];
if ((val & MSR_RI) == 0)
/* can't step mtmsr[d] that would clear MSR_RI */
return -1;
/* here op.val is the mask of bits to change */
regs->msr = (regs->msr & ~op.val) | (val & op.val);
goto instr_done;
#ifdef CONFIG_PPC64
case SYSCALL: /* sc */
/*
* N.B. this uses knowledge about how the syscall
* entry code works. If that is changed, this will
* need to be changed also.
*/
if (regs->gpr[0] == 0x1ebe &&
cpu_has_feature(CPU_FTR_REAL_LE)) {
regs->msr ^= MSR_LE;
goto instr_done;
}
regs->gpr[9] = regs->gpr[13];
regs->gpr[10] = MSR_KERNEL;
regs->gpr[11] = regs->nip + 4;
regs->gpr[12] = regs->msr & MSR_MASK;
regs->gpr[13] = (unsigned long) get_paca();
regs->nip = (unsigned long) &system_call_common;
regs->msr = MSR_KERNEL;
return 1;
case RFI:
return -1;
#endif
}
return 0;
ldst_done:
if (err)
return 0;
if (op.type & UPDATE)
regs->gpr[op.update_reg] = op.ea;
instr_done:
regs->nip = truncate_if_32bit(regs->msr, regs->nip + 4);
return 1;
powerpc: Merge enough to start building in arch/powerpc. This creates the directory structure under arch/powerpc and a bunch of Kconfig files. It does a first-cut merge of arch/powerpc/mm, arch/powerpc/lib and arch/powerpc/platforms/powermac. This is enough to build a 32-bit powermac kernel with ARCH=powerpc. For now we are getting some unmerged files from arch/ppc/kernel and arch/ppc/syslib, or arch/ppc64/kernel. This makes some minor changes to files in those directories and files outside arch/powerpc. The boot directory is still not merged. That's going to be interesting. Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-09-26 13:04:21 +07:00
}
powerpc/kprobes: Convert __kprobes to NOKPROBE_SYMBOL() Along similar lines as commit 9326638cbee2 ("kprobes, x86: Use NOKPROBE_SYMBOL() instead of __kprobes annotation"), convert __kprobes annotation to either NOKPROBE_SYMBOL() or nokprobe_inline. The latter forces inlining, in which case the caller needs to be added to NOKPROBE_SYMBOL(). Also: - blacklist arch_deref_entry_point(), and - convert a few regular inlines to nokprobe_inline in lib/sstep.c A key benefit is the ability to detect such symbols as being blacklisted. Before this patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem $ perf probe read_mem Failed to write event: Invalid argument Error: Failed to add events. $ dmesg | tail -1 [ 3736.112815] Could not insert probe at _text+10014968: -22 After patch: $ cat /sys/kernel/debug/kprobes/blacklist | grep read_mem 0xc000000000072b50-0xc000000000072d20 read_mem $ perf probe read_mem read_mem is blacklisted function, skip it. Added new events: (null):(null) (on read_mem) probe:read_mem (on read_mem) You can now use it in all perf tools, such as: perf record -e probe:read_mem -aR sleep 1 $ grep " read_mem" /proc/kallsyms c000000000072b50 t read_mem c0000000005f3b40 t read_mem $ cat /sys/kernel/debug/kprobes/list c0000000005f3b48 k read_mem+0x8 [DISABLED] Acked-by: Masami Hiramatsu <mhiramat@kernel.org> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> [mpe: Minor change log formatting, fix up some conflicts] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-04-12 18:18:51 +07:00
NOKPROBE_SYMBOL(emulate_step);
Reference in New Issue Copy Permalink