linux_dsm_epyc7002/arch/powerpc/kernel/head_64.S

822 lines
21 KiB
ArmAsm
Raw Normal View History

/*
* PowerPC version
* Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
*
* Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
* Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
* Adapted for Power Macintosh by Paul Mackerras.
* Low-level exception handlers and MMU support
* rewritten by Paul Mackerras.
* Copyright (C) 1996 Paul Mackerras.
*
* Adapted for 64bit PowerPC by Dave Engebretsen, Peter Bergner, and
* Mike Corrigan {engebret|bergner|mikejc}@us.ibm.com
*
* This file contains the entry point for the 64-bit kernel along
* with some early initialization code common to all 64-bit powerpc
* variants.
*
* 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/threads.h>
#include <asm/reg.h>
#include <asm/page.h>
#include <asm/mmu.h>
#include <asm/ppc_asm.h>
#include <asm/asm-offsets.h>
#include <asm/bug.h>
#include <asm/cputable.h>
#include <asm/setup.h>
#include <asm/hvcall.h>
#include <asm/thread_info.h>
#include <asm/firmware.h>
#include <asm/page_64.h>
#include <asm/irqflags.h>
#include <asm/kvm_book3s_asm.h>
#include <asm/ptrace.h>
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
#include <asm/hw_irq.h>
/* The physical memory is laid out such that the secondary processor
* spin code sits at 0x0000...0x00ff. On server, the vectors follow
* using the layout described in exceptions-64s.S
*/
/*
* Entering into this code we make the following assumptions:
*
* For pSeries or server processors:
* 1. The MMU is off & open firmware is running in real mode.
* 2. The kernel is entered at __start
* -or- For OPAL entry:
* 1. The MMU is off, processor in HV mode, primary CPU enters at 0
* with device-tree in gpr3. We also get OPAL base in r8 and
* entry in r9 for debugging purposes
* 2. Secondary processors enter at 0x60 with PIR in gpr3
*
* For Book3E processors:
* 1. The MMU is on running in AS0 in a state defined in ePAPR
* 2. The kernel is entered at __start
*/
.text
.globl _stext
_stext:
_GLOBAL(__start)
/* NOP this out unconditionally */
BEGIN_FTR_SECTION
b .__start_initialization_multiplatform
END_FTR_SECTION(0, 1)
/* Catch branch to 0 in real mode */
trap
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
/* Secondary processors spin on this value until it becomes nonzero.
* When it does it contains the real address of the descriptor
* of the function that the cpu should jump to to continue
* initialization.
*/
.globl __secondary_hold_spinloop
__secondary_hold_spinloop:
.llong 0x0
/* Secondary processors write this value with their cpu # */
/* after they enter the spin loop immediately below. */
.globl __secondary_hold_acknowledge
__secondary_hold_acknowledge:
.llong 0x0
#ifdef CONFIG_RELOCATABLE
/* This flag is set to 1 by a loader if the kernel should run
* at the loaded address instead of the linked address. This
* is used by kexec-tools to keep the the kdump kernel in the
* crash_kernel region. The loader is responsible for
* observing the alignment requirement.
*/
/* Do not move this variable as kexec-tools knows about it. */
. = 0x5c
.globl __run_at_load
__run_at_load:
.long 0x72756e30 /* "run0" -- relocate to 0 by default */
#endif
. = 0x60
/*
* The following code is used to hold secondary processors
* in a spin loop after they have entered the kernel, but
* before the bulk of the kernel has been relocated. This code
* is relocated to physical address 0x60 before prom_init is run.
* All of it must fit below the first exception vector at 0x100.
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
* Use .globl here not _GLOBAL because we want __secondary_hold
* to be the actual text address, not a descriptor.
*/
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
.globl __secondary_hold
__secondary_hold:
#ifndef CONFIG_PPC_BOOK3E
mfmsr r24
ori r24,r24,MSR_RI
mtmsrd r24 /* RI on */
#endif
/* Grab our physical cpu number */
mr r24,r3
/* stash r4 for book3e */
mr r25,r4
/* Tell the master cpu we're here */
/* Relocation is off & we are located at an address less */
/* than 0x100, so only need to grab low order offset. */
std r24,__secondary_hold_acknowledge-_stext(0)
sync
li r26,0
#ifdef CONFIG_PPC_BOOK3E
tovirt(r26,r26)
#endif
/* All secondary cpus wait here until told to start. */
100: ld r4,__secondary_hold_spinloop-_stext(r26)
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
cmpdi 0,r4,0
beq 100b
#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
#ifdef CONFIG_PPC_BOOK3E
tovirt(r4,r4)
#endif
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
ld r4,0(r4) /* deref function descriptor */
mtctr r4
mr r3,r24
/*
* it may be the case that other platforms have r4 right to
* begin with, this gives us some safety in case it is not
*/
#ifdef CONFIG_PPC_BOOK3E
mr r4,r25
#else
li r4,0
#endif
/* Make sure that patched code is visible */
isync
bctr
#else
BUG_OPCODE
#endif
/* This value is used to mark exception frames on the stack. */
.section ".toc","aw"
exception_marker:
.tc ID_72656773_68657265[TC],0x7265677368657265
.text
/*
* On server, we include the exception vectors code here as it
* relies on absolute addressing which is only possible within
* this compilation unit
*/
#ifdef CONFIG_PPC_BOOK3S
#include "exceptions-64s.S"
powerpc: Make it possible to move the interrupt handlers away from the kernel This changes the way that the exception prologs transfer control to the handlers in 64-bit kernels with the aim of making it possible to have the prologs separate from the main body of the kernel. Now, instead of computing the address of the handler by taking the top 32 bits of the paca address (to get the 0xc0000000........ part) and ORing in something in the bottom 16 bits, we get the base address of the kernel by doing a load from the paca and add an offset. This also replaces an mfmsr and an ori to compute the MSR value for the handler with a load from the paca. That makes it unnecessary to have a separate version of EXCEPTION_PROLOG_PSERIES that forces 64-bit mode. We can no longer use a direct branches in the exception prolog code, which means that the SLB miss handlers can't branch directly to .slb_miss_realmode any more. Instead we have to compute the address and do an indirect branch. This is conditional on CONFIG_RELOCATABLE; for non-relocatable kernels we use a direct branch as before. (A later change will allow CONFIG_RELOCATABLE to be set on 64-bit powerpc.) Since the secondary CPUs on pSeries start execution in the first 0x100 bytes of real memory and then have to get to wherever the kernel is, we can't use a direct branch to get there. Instead this changes __secondary_hold_spinloop from a flag to a function pointer. When it is set to a non-NULL value, the secondary CPUs jump to the function pointed to by that value. Finally this eliminates one code difference between 32-bit and 64-bit by making __secondary_hold be the text address of the secondary CPU spinloop rather than a function descriptor for it. Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-08-30 08:40:24 +07:00
#endif
_GLOBAL(generic_secondary_thread_init)
mr r24,r3
/* turn on 64-bit mode */
bl .enable_64b_mode
/* get a valid TOC pointer, wherever we're mapped at */
bl .relative_toc
tovirt(r2,r2)
#ifdef CONFIG_PPC_BOOK3E
/* Book3E initialization */
mr r3,r24
bl .book3e_secondary_thread_init
#endif
b generic_secondary_common_init
/*
* On pSeries and most other platforms, secondary processors spin
* in the following code.
* At entry, r3 = this processor's number (physical cpu id)
*
* On Book3E, r4 = 1 to indicate that the initial TLB entry for
* this core already exists (setup via some other mechanism such
* as SCOM before entry).
*/
_GLOBAL(generic_secondary_smp_init)
mr r24,r3
mr r25,r4
/* turn on 64-bit mode */
bl .enable_64b_mode
/* get a valid TOC pointer, wherever we're mapped at */
bl .relative_toc
tovirt(r2,r2)
#ifdef CONFIG_PPC_BOOK3E
/* Book3E initialization */
mr r3,r24
mr r4,r25
bl .book3e_secondary_core_init
#endif
generic_secondary_common_init:
/* Set up a paca value for this processor. Since we have the
* physical cpu id in r24, we need to search the pacas to find
* which logical id maps to our physical one.
*/
LOAD_REG_ADDR(r13, paca) /* Load paca pointer */
ld r13,0(r13) /* Get base vaddr of paca array */
powerpc: Don't search for paca in freed memory Starting with 1426d5a3bd07589534286375998c0c8c6fdc5260 (powerpc: Dynamically allocate pacas) we free the memory for pacas beyond cpu_possible, but we failed to update the loop the secondary cpus use to find their paca. If the system has running cpu threads for which the kernel did not allocate a paca for they will search the memory that was freed. For instance this could happen when the device tree for a kdump kernel was not updated after a cpu hotplug, or the kernel is running with more cpus than the kernel was configured. Since c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) we set nr_cpu_ids before telling the cpus to advance, so use that to limit the search. We can't reference nr_cpu_ids without CONFIG_SMP because it is defined as 1 instead of a memory location, but any extra threads should be sent to kexec_wait in that case anyways, so make that explicit and remove the search loop for UP. Note to stable: The fix also requires c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) to function. Also 9d07bc841c9779b4d7902e417f4e509996ce805d (Properly handshake CPUs going out of boot spin loop) affects the second chunk, specifically the branch target was 3b before and is 4b after that patch, and there was a blank line before the #ifdef CONFIG_SMP that was removed Cc: <stable@kernel.org> # .34.x: c1854e0072 powerpc: Set nr_cpu_ids early Cc: <stable@kernel.org> # .34.x Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-11 02:28:37 +07:00
#ifndef CONFIG_SMP
addi r13,r13,PACA_SIZE /* know r13 if used accidentally */
b .kexec_wait /* wait for next kernel if !SMP */
#else
LOAD_REG_ADDR(r7, nr_cpu_ids) /* Load nr_cpu_ids address */
lwz r7,0(r7) /* also the max paca allocated */
li r5,0 /* logical cpu id */
1: lhz r6,PACAHWCPUID(r13) /* Load HW procid from paca */
cmpw r6,r24 /* Compare to our id */
beq 2f
addi r13,r13,PACA_SIZE /* Loop to next PACA on miss */
addi r5,r5,1
powerpc: Don't search for paca in freed memory Starting with 1426d5a3bd07589534286375998c0c8c6fdc5260 (powerpc: Dynamically allocate pacas) we free the memory for pacas beyond cpu_possible, but we failed to update the loop the secondary cpus use to find their paca. If the system has running cpu threads for which the kernel did not allocate a paca for they will search the memory that was freed. For instance this could happen when the device tree for a kdump kernel was not updated after a cpu hotplug, or the kernel is running with more cpus than the kernel was configured. Since c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) we set nr_cpu_ids before telling the cpus to advance, so use that to limit the search. We can't reference nr_cpu_ids without CONFIG_SMP because it is defined as 1 instead of a memory location, but any extra threads should be sent to kexec_wait in that case anyways, so make that explicit and remove the search loop for UP. Note to stable: The fix also requires c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) to function. Also 9d07bc841c9779b4d7902e417f4e509996ce805d (Properly handshake CPUs going out of boot spin loop) affects the second chunk, specifically the branch target was 3b before and is 4b after that patch, and there was a blank line before the #ifdef CONFIG_SMP that was removed Cc: <stable@kernel.org> # .34.x: c1854e0072 powerpc: Set nr_cpu_ids early Cc: <stable@kernel.org> # .34.x Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-11 02:28:37 +07:00
cmpw r5,r7 /* Check if more pacas exist */
blt 1b
mr r3,r24 /* not found, copy phys to r3 */
b .kexec_wait /* next kernel might do better */
2: SET_PACA(r13)
#ifdef CONFIG_PPC_BOOK3E
addi r12,r13,PACA_EXTLB /* and TLB exc frame in another */
mtspr SPRN_SPRG_TLB_EXFRAME,r12
#endif
/* From now on, r24 is expected to be logical cpuid */
mr r24,r5
/* See if we need to call a cpu state restore handler */
LOAD_REG_ADDR(r23, cur_cpu_spec)
ld r23,0(r23)
ld r23,CPU_SPEC_RESTORE(r23)
cmpdi 0,r23,0
beq 3f
ld r23,0(r23)
mtctr r23
bctrl
3: LOAD_REG_ADDR(r3, spinning_secondaries) /* Decrement spinning_secondaries */
lwarx r4,0,r3
subi r4,r4,1
stwcx. r4,0,r3
bne 3b
isync
4: HMT_LOW
lbz r23,PACAPROCSTART(r13) /* Test if this processor should */
/* start. */
cmpwi 0,r23,0
beq 4b /* Loop until told to go */
sync /* order paca.run and cur_cpu_spec */
isync /* In case code patching happened */
/* Create a temp kernel stack for use before relocation is on. */
ld r1,PACAEMERGSP(r13)
subi r1,r1,STACK_FRAME_OVERHEAD
b __secondary_start
powerpc: Don't search for paca in freed memory Starting with 1426d5a3bd07589534286375998c0c8c6fdc5260 (powerpc: Dynamically allocate pacas) we free the memory for pacas beyond cpu_possible, but we failed to update the loop the secondary cpus use to find their paca. If the system has running cpu threads for which the kernel did not allocate a paca for they will search the memory that was freed. For instance this could happen when the device tree for a kdump kernel was not updated after a cpu hotplug, or the kernel is running with more cpus than the kernel was configured. Since c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) we set nr_cpu_ids before telling the cpus to advance, so use that to limit the search. We can't reference nr_cpu_ids without CONFIG_SMP because it is defined as 1 instead of a memory location, but any extra threads should be sent to kexec_wait in that case anyways, so make that explicit and remove the search loop for UP. Note to stable: The fix also requires c1854e00727f50f7ac99e98d26ece04c087ef785 (powerpc: Set nr_cpu_ids early and use it to free PACAs) to function. Also 9d07bc841c9779b4d7902e417f4e509996ce805d (Properly handshake CPUs going out of boot spin loop) affects the second chunk, specifically the branch target was 3b before and is 4b after that patch, and there was a blank line before the #ifdef CONFIG_SMP that was removed Cc: <stable@kernel.org> # .34.x: c1854e0072 powerpc: Set nr_cpu_ids early Cc: <stable@kernel.org> # .34.x Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-11 02:28:37 +07:00
#endif /* SMP */
/*
* Turn the MMU off.
* Assumes we're mapped EA == RA if the MMU is on.
*/
#ifdef CONFIG_PPC_BOOK3S
_STATIC(__mmu_off)
mfmsr r3
andi. r0,r3,MSR_IR|MSR_DR
beqlr
mflr r4
andc r3,r3,r0
mtspr SPRN_SRR0,r4
mtspr SPRN_SRR1,r3
sync
rfid
b . /* prevent speculative execution */
#endif
/*
* Here is our main kernel entry point. We support currently 2 kind of entries
* depending on the value of r5.
*
* r5 != NULL -> OF entry, we go to prom_init, "legacy" parameter content
* in r3...r7
*
* r5 == NULL -> kexec style entry. r3 is a physical pointer to the
* DT block, r4 is a physical pointer to the kernel itself
*
*/
_GLOBAL(__start_initialization_multiplatform)
/* Make sure we are running in 64 bits mode */
bl .enable_64b_mode
/* Get TOC pointer (current runtime address) */
bl .relative_toc
/* find out where we are now */
bcl 20,31,$+4
0: mflr r26 /* r26 = runtime addr here */
addis r26,r26,(_stext - 0b)@ha
addi r26,r26,(_stext - 0b)@l /* current runtime base addr */
/*
* Are we booted from a PROM Of-type client-interface ?
*/
cmpldi cr0,r5,0
beq 1f
b .__boot_from_prom /* yes -> prom */
1:
/* Save parameters */
mr r31,r3
mr r30,r4
#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
/* Save OPAL entry */
mr r28,r8
mr r29,r9
#endif
#ifdef CONFIG_PPC_BOOK3E
bl .start_initialization_book3e
b .__after_prom_start
#else
/* Setup some critical 970 SPRs before switching MMU off */
mfspr r0,SPRN_PVR
srwi r0,r0,16
cmpwi r0,0x39 /* 970 */
beq 1f
cmpwi r0,0x3c /* 970FX */
beq 1f
cmpwi r0,0x44 /* 970MP */
beq 1f
cmpwi r0,0x45 /* 970GX */
bne 2f
1: bl .__cpu_preinit_ppc970
2:
/* Switch off MMU if not already off */
bl .__mmu_off
b .__after_prom_start
#endif /* CONFIG_PPC_BOOK3E */
_INIT_STATIC(__boot_from_prom)
#ifdef CONFIG_PPC_OF_BOOT_TRAMPOLINE
/* Save parameters */
mr r31,r3
mr r30,r4
mr r29,r5
mr r28,r6
mr r27,r7
/*
* Align the stack to 16-byte boundary
* Depending on the size and layout of the ELF sections in the initial
* boot binary, the stack pointer may be unaligned on PowerMac
*/
rldicr r1,r1,0,59
2008-08-30 08:43:47 +07:00
#ifdef CONFIG_RELOCATABLE
/* Relocate code for where we are now */
mr r3,r26
bl .relocate
#endif
/* Restore parameters */
mr r3,r31
mr r4,r30
mr r5,r29
mr r6,r28
mr r7,r27
/* Do all of the interaction with OF client interface */
2008-08-30 08:43:47 +07:00
mr r8,r26
bl .prom_init
#endif /* #CONFIG_PPC_OF_BOOT_TRAMPOLINE */
/* We never return. We also hit that trap if trying to boot
* from OF while CONFIG_PPC_OF_BOOT_TRAMPOLINE isn't selected */
trap
_STATIC(__after_prom_start)
2008-08-30 08:43:47 +07:00
#ifdef CONFIG_RELOCATABLE
/* process relocations for the final address of the kernel */
lis r25,PAGE_OFFSET@highest /* compute virtual base of kernel */
sldi r25,r25,32
lwz r7,__run_at_load-_stext(r26)
cmplwi cr0,r7,1 /* flagged to stay where we are ? */
bne 1f
add r25,r25,r26
1: mr r3,r25
2008-08-30 08:43:47 +07:00
bl .relocate
#endif
/*
* We need to run with _stext at physical address PHYSICAL_START.
* This will leave some code in the first 256B of
* real memory, which are reserved for software use.
*
* Note: This process overwrites the OF exception vectors.
*/
2008-08-30 08:43:47 +07:00
li r3,0 /* target addr */
#ifdef CONFIG_PPC_BOOK3E
tovirt(r3,r3) /* on booke, we already run at PAGE_OFFSET */
#endif
2008-08-30 08:43:47 +07:00
mr. r4,r26 /* In some cases the loader may */
beq 9f /* have already put us at zero */
li r6,0x100 /* Start offset, the first 0x100 */
/* bytes were copied earlier. */
#ifdef CONFIG_PPC_BOOK3E
tovirt(r6,r6) /* on booke, we already run at PAGE_OFFSET */
#endif
#ifdef CONFIG_RELOCATABLE
/*
* Check if the kernel has to be running as relocatable kernel based on the
* variable __run_at_load, if it is set the kernel is treated as relocatable
* kernel, otherwise it will be moved to PHYSICAL_START
*/
lwz r7,__run_at_load-_stext(r26)
cmplwi cr0,r7,1
bne 3f
/* just copy interrupts */
LOAD_REG_IMMEDIATE(r5, __end_interrupts - _stext)
b 5f
3:
#endif
lis r5,(copy_to_here - _stext)@ha
addi r5,r5,(copy_to_here - _stext)@l /* # bytes of memory to copy */
bl .copy_and_flush /* copy the first n bytes */
/* this includes the code being */
/* executed here. */
addis r8,r3,(4f - _stext)@ha /* Jump to the copy of this code */
addi r8,r8,(4f - _stext)@l /* that we just made */
mtctr r8
bctr
p_end: .llong _end - _stext
4: /* Now copy the rest of the kernel up to _end */
addis r5,r26,(p_end - _stext)@ha
ld r5,(p_end - _stext)@l(r5) /* get _end */
5: bl .copy_and_flush /* copy the rest */
9: b .start_here_multiplatform
/*
* Copy routine used to copy the kernel to start at physical address 0
* and flush and invalidate the caches as needed.
* r3 = dest addr, r4 = source addr, r5 = copy limit, r6 = start offset
* on exit, r3, r4, r5 are unchanged, r6 is updated to be >= r5.
*
* Note: this routine *only* clobbers r0, r6 and lr
*/
_GLOBAL(copy_and_flush)
addi r5,r5,-8
addi r6,r6,-8
4: li r0,8 /* Use the smallest common */
/* denominator cache line */
/* size. This results in */
/* extra cache line flushes */
/* but operation is correct. */
/* Can't get cache line size */
/* from NACA as it is being */
/* moved too. */
mtctr r0 /* put # words/line in ctr */
3: addi r6,r6,8 /* copy a cache line */
ldx r0,r6,r4
stdx r0,r6,r3
bdnz 3b
dcbst r6,r3 /* write it to memory */
sync
icbi r6,r3 /* flush the icache line */
cmpld 0,r6,r5
blt 4b
sync
addi r5,r5,8
addi r6,r6,8
isync
blr
.align 8
copy_to_here:
#ifdef CONFIG_SMP
#ifdef CONFIG_PPC_PMAC
/*
* On PowerMac, secondary processors starts from the reset vector, which
* is temporarily turned into a call to one of the functions below.
*/
.section ".text";
.align 2 ;
.globl __secondary_start_pmac_0
__secondary_start_pmac_0:
/* NB the entries for cpus 0, 1, 2 must each occupy 8 bytes. */
li r24,0
b 1f
li r24,1
b 1f
li r24,2
b 1f
li r24,3
1:
_GLOBAL(pmac_secondary_start)
/* turn on 64-bit mode */
bl .enable_64b_mode
li r0,0
mfspr r3,SPRN_HID4
rldimi r3,r0,40,23 /* clear bit 23 (rm_ci) */
sync
mtspr SPRN_HID4,r3
isync
sync
slbia
/* get TOC pointer (real address) */
bl .relative_toc
tovirt(r2,r2)
/* Copy some CPU settings from CPU 0 */
bl .__restore_cpu_ppc970
/* pSeries do that early though I don't think we really need it */
mfmsr r3
ori r3,r3,MSR_RI
mtmsrd r3 /* RI on */
/* Set up a paca value for this processor. */
LOAD_REG_ADDR(r4,paca) /* Load paca pointer */
ld r4,0(r4) /* Get base vaddr of paca array */
mulli r13,r24,PACA_SIZE /* Calculate vaddr of right paca */
add r13,r13,r4 /* for this processor. */
SET_PACA(r13) /* Save vaddr of paca in an SPRG*/
/* Mark interrupts soft and hard disabled (they might be enabled
* in the PACA when doing hotplug)
*/
li r0,0
stb r0,PACASOFTIRQEN(r13)
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
li r0,PACA_IRQ_HARD_DIS
stb r0,PACAIRQHAPPENED(r13)
/* Create a temp kernel stack for use before relocation is on. */
ld r1,PACAEMERGSP(r13)
subi r1,r1,STACK_FRAME_OVERHEAD
b __secondary_start
#endif /* CONFIG_PPC_PMAC */
/*
* This function is called after the master CPU has released the
* secondary processors. The execution environment is relocation off.
* The paca for this processor has the following fields initialized at
* this point:
* 1. Processor number
* 2. Segment table pointer (virtual address)
* On entry the following are set:
* r1 = stack pointer (real addr of temp stack)
* r24 = cpu# (in Linux terms)
* r13 = paca virtual address
* SPRG_PACA = paca virtual address
*/
.section ".text";
.align 2 ;
.globl __secondary_start
__secondary_start:
/* Set thread priority to MEDIUM */
HMT_MEDIUM
/* Initialize the kernel stack */
LOAD_REG_ADDR(r3, current_set)
sldi r28,r24,3 /* get current_set[cpu#] */
ldx r14,r3,r28
addi r14,r14,THREAD_SIZE-STACK_FRAME_OVERHEAD
std r14,PACAKSAVE(r13)
/* Do early setup for that CPU (stab, slb, hash table pointer) */
bl .early_setup_secondary
/*
* setup the new stack pointer, but *don't* use this until
* translation is on.
*/
mr r1, r14
/* Clear backchain so we get nice backtraces */
li r7,0
mtlr r7
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
/* Mark interrupts soft and hard disabled (they might be enabled
* in the PACA when doing hotplug)
*/
stb r7,PACASOFTIRQEN(r13)
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
li r0,PACA_IRQ_HARD_DIS
stb r0,PACAIRQHAPPENED(r13)
/* enable MMU and jump to start_secondary */
LOAD_REG_ADDR(r3, .start_secondary_prolog)
LOAD_REG_IMMEDIATE(r4, MSR_KERNEL)
mtspr SPRN_SRR0,r3
mtspr SPRN_SRR1,r4
RFI
b . /* prevent speculative execution */
/*
* Running with relocation on at this point. All we want to do is
* zero the stack back-chain pointer and get the TOC virtual address
* before going into C code.
*/
_GLOBAL(start_secondary_prolog)
ld r2,PACATOC(r13)
li r3,0
std r3,0(r1) /* Zero the stack frame pointer */
bl .start_secondary
b .
/*
* Reset stack pointer and call start_secondary
* to continue with online operation when woken up
* from cede in cpu offline.
*/
_GLOBAL(start_secondary_resume)
ld r1,PACAKSAVE(r13) /* Reload kernel stack pointer */
li r3,0
std r3,0(r1) /* Zero the stack frame pointer */
bl .start_secondary
b .
#endif
/*
* This subroutine clobbers r11 and r12
*/
_GLOBAL(enable_64b_mode)
mfmsr r11 /* grab the current MSR */
#ifdef CONFIG_PPC_BOOK3E
oris r11,r11,0x8000 /* CM bit set, we'll set ICM later */
mtmsr r11
#else /* CONFIG_PPC_BOOK3E */
li r12,(MSR_64BIT | MSR_ISF)@highest
sldi r12,r12,48
or r11,r11,r12
mtmsrd r11
isync
#endif
blr
/*
* This puts the TOC pointer into r2, offset by 0x8000 (as expected
* by the toolchain). It computes the correct value for wherever we
* are running at the moment, using position-independent code.
*
* Note: The compiler constructs pointers using offsets from the
* TOC in -mcmodel=medium mode. After we relocate to 0 but before
* the MMU is on we need our TOC to be a virtual address otherwise
* these pointers will be real addresses which may get stored and
* accessed later with the MMU on. We use tovirt() at the call
* sites to handle this.
*/
_GLOBAL(relative_toc)
mflr r0
bcl 20,31,$+4
0: mflr r11
ld r2,(p_toc - 0b)(r11)
add r2,r2,r11
mtlr r0
blr
.balign 8
p_toc: .llong __toc_start + 0x8000 - 0b
/*
* This is where the main kernel code starts.
*/
_INIT_STATIC(start_here_multiplatform)
/* set up the TOC */
bl .relative_toc
tovirt(r2,r2)
/* Clear out the BSS. It may have been done in prom_init,
* already but that's irrelevant since prom_init will soon
* be detached from the kernel completely. Besides, we need
* to clear it now for kexec-style entry.
*/
LOAD_REG_ADDR(r11,__bss_stop)
LOAD_REG_ADDR(r8,__bss_start)
sub r11,r11,r8 /* bss size */
addi r11,r11,7 /* round up to an even double word */
srdi. r11,r11,3 /* shift right by 3 */
beq 4f
addi r8,r8,-8
li r0,0
mtctr r11 /* zero this many doublewords */
3: stdu r0,8(r8)
bdnz 3b
4:
#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
/* Setup OPAL entry */
LOAD_REG_ADDR(r11, opal)
std r28,0(r11);
std r29,8(r11);
#endif
#ifndef CONFIG_PPC_BOOK3E
mfmsr r6
ori r6,r6,MSR_RI
mtmsrd r6 /* RI on */
#endif
2008-08-30 08:43:47 +07:00
#ifdef CONFIG_RELOCATABLE
/* Save the physical address we're running at in kernstart_addr */
LOAD_REG_ADDR(r4, kernstart_addr)
clrldi r0,r25,2
std r0,0(r4)
#endif
/* The following gets the stack set up with the regs */
/* pointing to the real addr of the kernel stack. This is */
/* all done to support the C function call below which sets */
/* up the htab. This is done because we have relocated the */
/* kernel but are still running in real mode. */
LOAD_REG_ADDR(r3,init_thread_union)
/* set up a stack pointer */
addi r1,r3,THREAD_SIZE
li r0,0
stdu r0,-STACK_FRAME_OVERHEAD(r1)
/* Do very early kernel initializations, including initial hash table,
* stab and slb setup before we turn on relocation. */
/* Restore parameters passed from prom_init/kexec */
mr r3,r31
bl .early_setup /* also sets r13 and SPRG_PACA */
LOAD_REG_ADDR(r3, .start_here_common)
ld r4,PACAKMSR(r13)
mtspr SPRN_SRR0,r3
mtspr SPRN_SRR1,r4
RFI
b . /* prevent speculative execution */
/* This is where all platforms converge execution */
_INIT_GLOBAL(start_here_common)
/* relocation is on at this point */
std r1,PACAKSAVE(r13)
/* Load the TOC (virtual address) */
ld r2,PACATOC(r13)
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
/* Do more system initializations in virtual mode */
bl .setup_system
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
/* Mark interrupts soft and hard disabled (they might be enabled
* in the PACA when doing hotplug)
*/
li r0,0
stb r0,PACASOFTIRQEN(r13)
li r0,PACA_IRQ_HARD_DIS
stb r0,PACAIRQHAPPENED(r13)
powerpc: Rework lazy-interrupt handling The current implementation of lazy interrupts handling has some issues that this tries to address. We don't do the various workarounds we need to do when re-enabling interrupts in some cases such as when returning from an interrupt and thus we may still lose or get delayed decrementer or doorbell interrupts. The current scheme also makes it much harder to handle the external "edge" interrupts provided by some BookE processors when using the EPR facility (External Proxy) and the Freescale Hypervisor. Additionally, we tend to keep interrupts hard disabled in a number of cases, such as decrementer interrupts, external interrupts, or when a masked decrementer interrupt is pending. This is sub-optimal. This is an attempt at fixing it all in one go by reworking the way we do the lazy interrupt disabling from the ground up. The base idea is to replace the "hard_enabled" field with a "irq_happened" field in which we store a bit mask of what interrupt occurred while soft-disabled. When re-enabling, either via arch_local_irq_restore() or when returning from an interrupt, we can now decide what to do by testing bits in that field. We then implement replaying of the missed interrupts either by re-using the existing exception frame (in exception exit case) or via the creation of a new one from an assembly trampoline (in the arch_local_irq_enable case). This removes the need to play with the decrementer to try to create fake interrupts, among others. In addition, this adds a few refinements: - We no longer hard disable decrementer interrupts that occur while soft-disabled. We now simply bump the decrementer back to max (on BookS) or leave it stopped (on BookE) and continue with hard interrupts enabled, which means that we'll potentially get better sample quality from performance monitor interrupts. - Timer, decrementer and doorbell interrupts now hard-enable shortly after removing the source of the interrupt, which means they no longer run entirely hard disabled. Again, this will improve perf sample quality. - On Book3E 64-bit, we now make the performance monitor interrupt act as an NMI like Book3S (the necessary C code for that to work appear to already be present in the FSL perf code, notably calling nmi_enter instead of irq_enter). (This also fixes a bug where BookE perfmon interrupts could clobber r14 ... oops) - We could make "masked" decrementer interrupts act as NMIs when doing timer-based perf sampling to improve the sample quality. Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org> --- v2: - Add hard-enable to decrementer, timer and doorbells - Fix CR clobber in masked irq handling on BookE - Make embedded perf interrupt act as an NMI - Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want to retrigger an interrupt without preventing hard-enable v3: - Fix or vs. ori bug on Book3E - Fix enabling of interrupts for some exceptions on Book3E v4: - Fix resend of doorbells on return from interrupt on Book3E v5: - Rebased on top of my latest series, which involves some significant rework of some aspects of the patch. v6: - 32-bit compile fix - more compile fixes with various .config combos - factor out the asm code to soft-disable interrupts - remove the C wrapper around preempt_schedule_irq v7: - Fix a bug with hard irq state tracking on native power7
2012-03-06 14:27:59 +07:00
/* Generic kernel entry */
[POWERPC] Fix iSeries hard irq enabling regression A subtle bug sneaked into iSeries recently. On this platform, we must not normally clear MSR:EE (the hardware external interrupt enable) except for short periods of time. Taking an interrupt while soft-disabled doesn't cause us to clear it for example. The iSeries kernel expects to mostly run with MSR:EE enabled at all times except in a few exception entry/exit code paths. Thus local_irq_enable() doesn't check if it needs to hard-enable as it expects this to be unnecessary on iSeries. However, hard_irq_disable() _does_ cause MSR:EE to be cleared, including on iSeries. A call to it was recently added to the context switch code, thus causing interrupts to become disabled for a long periods of time, causing the iSeries watchdog to kick in under some circumstances and other nasty things. This patch fixes it by making local_irq_enable() properly re-enable MSR:EE on iSeries. It basically removes a return statement here to make iSeries use the same code path as everybody else. That does mean that we might occasionally get spurious decrementer interrupts but I don't think that matters. Another option would have been to make hard_irq_disable() a nop on iSeries but I didn't like it much, in case we have good reasons to hard-disable. Part of the patch is fixes to make sure the hard_enabled PACA field is properly set on iSeries as it used not to be before, since it was mostly unused. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
2008-04-02 11:58:40 +07:00
bl .start_kernel
/* Not reached */
BUG_OPCODE
/*
* We put a few things here that have to be page-aligned.
* This stuff goes at the beginning of the bss, which is page-aligned.
*/
.section ".bss"
.align PAGE_SHIFT
.globl empty_zero_page
empty_zero_page:
.space PAGE_SIZE
.globl swapper_pg_dir
swapper_pg_dir:
.space PGD_TABLE_SIZE