linux_dsm_epyc7002/arch/avr32/kernel/entry-avr32b.S
Haavard Skinnemoen c0c3e81608 [AVR32] Optimize the TLB miss handler
Reorder some instructions and change the register usage to reduce
the number of pipeline stalls. Also use the bfextu and bfins
instructions for bitfield manipulations instead of shifting and
masking.

This makes gzipping a 80MB file approximately 2% faster.

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
2007-04-27 13:44:15 +02:00

727 lines
14 KiB
ArmAsm

/*
* Copyright (C) 2004-2006 Atmel Corporation
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
/*
* This file contains the low-level entry-points into the kernel, that is,
* exception handlers, debug trap handlers, interrupt handlers and the
* system call handler.
*/
#include <linux/errno.h>
#include <asm/asm.h>
#include <asm/hardirq.h>
#include <asm/irq.h>
#include <asm/ocd.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/ptrace.h>
#include <asm/sysreg.h>
#include <asm/thread_info.h>
#include <asm/unistd.h>
#ifdef CONFIG_PREEMPT
# define preempt_stop mask_interrupts
#else
# define preempt_stop
# define fault_resume_kernel fault_restore_all
#endif
#define __MASK(x) ((1 << (x)) - 1)
#define IRQ_MASK ((__MASK(SOFTIRQ_BITS) << SOFTIRQ_SHIFT) | \
(__MASK(HARDIRQ_BITS) << HARDIRQ_SHIFT))
.section .ex.text,"ax",@progbits
.align 2
exception_vectors:
bral handle_critical
.align 2
bral handle_critical
.align 2
bral do_bus_error_write
.align 2
bral do_bus_error_read
.align 2
bral do_nmi_ll
.align 2
bral handle_address_fault
.align 2
bral handle_protection_fault
.align 2
bral handle_debug
.align 2
bral do_illegal_opcode_ll
.align 2
bral do_illegal_opcode_ll
.align 2
bral do_illegal_opcode_ll
.align 2
bral do_fpe_ll
.align 2
bral do_illegal_opcode_ll
.align 2
bral handle_address_fault
.align 2
bral handle_address_fault
.align 2
bral handle_protection_fault
.align 2
bral handle_protection_fault
.align 2
bral do_dtlb_modified
/*
* r0 : PGD/PT/PTE
* r1 : Offending address
* r2 : Scratch register
* r3 : Cause (5, 12 or 13)
*/
#define tlbmiss_save pushm r0-r3
#define tlbmiss_restore popm r0-r3
.section .tlbx.ex.text,"ax",@progbits
.global itlb_miss
itlb_miss:
tlbmiss_save
rjmp tlb_miss_common
.section .tlbr.ex.text,"ax",@progbits
dtlb_miss_read:
tlbmiss_save
rjmp tlb_miss_common
.section .tlbw.ex.text,"ax",@progbits
dtlb_miss_write:
tlbmiss_save
.global tlb_miss_common
tlb_miss_common:
mfsr r0, SYSREG_TLBEAR
mfsr r1, SYSREG_PTBR
/* Is it the vmalloc space? */
bld r0, 31
brcs handle_vmalloc_miss
/* First level lookup */
pgtbl_lookup:
lsr r2, r0, PGDIR_SHIFT
ld.w r3, r1[r2 << 2]
bfextu r1, r0, PAGE_SHIFT, PGDIR_SHIFT - PAGE_SHIFT
bld r3, _PAGE_BIT_PRESENT
brcc page_table_not_present
/* Translate to virtual address in P1. */
andl r3, 0xf000
sbr r3, 31
/* Second level lookup */
ld.w r2, r3[r1 << 2]
mfsr r0, SYSREG_TLBARLO
bld r2, _PAGE_BIT_PRESENT
brcc page_not_present
/* Mark the page as accessed */
sbr r2, _PAGE_BIT_ACCESSED
st.w r3[r1 << 2], r2
/* Drop software flags */
andl r2, _PAGE_FLAGS_HARDWARE_MASK & 0xffff
mtsr SYSREG_TLBELO, r2
/* Figure out which entry we want to replace */
mfsr r1, SYSREG_MMUCR
clz r2, r0
brcc 1f
mov r3, -1 /* All entries have been accessed, */
mov r2, 0 /* so start at 0 */
mtsr SYSREG_TLBARLO, r3 /* and reset TLBAR */
1: bfins r1, r2, SYSREG_DRP_OFFSET, SYSREG_DRP_SIZE
mtsr SYSREG_MMUCR, r1
tlbw
tlbmiss_restore
rete
handle_vmalloc_miss:
/* Simply do the lookup in init's page table */
mov r1, lo(swapper_pg_dir)
orh r1, hi(swapper_pg_dir)
rjmp pgtbl_lookup
/* --- System Call --- */
.section .scall.text,"ax",@progbits
system_call:
pushm r12 /* r12_orig */
stmts --sp, r0-lr
zero_fp
mfsr r0, SYSREG_RAR_SUP
mfsr r1, SYSREG_RSR_SUP
stm --sp, r0-r1
/* check for syscall tracing */
get_thread_info r0
ld.w r1, r0[TI_flags]
bld r1, TIF_SYSCALL_TRACE
brcs syscall_trace_enter
syscall_trace_cont:
cp.w r8, NR_syscalls
brhs syscall_badsys
lddpc lr, syscall_table_addr
ld.w lr, lr[r8 << 2]
mov r8, r5 /* 5th argument (6th is pushed by stub) */
icall lr
.global syscall_return
syscall_return:
get_thread_info r0
mask_interrupts /* make sure we don't miss an interrupt
setting need_resched or sigpending
between sampling and the rets */
/* Store the return value so that the correct value is loaded below */
stdsp sp[REG_R12], r12
ld.w r1, r0[TI_flags]
andl r1, _TIF_ALLWORK_MASK, COH
brne syscall_exit_work
syscall_exit_cont:
popm r8-r9
mtsr SYSREG_RAR_SUP, r8
mtsr SYSREG_RSR_SUP, r9
ldmts sp++, r0-lr
sub sp, -4 /* r12_orig */
rets
.align 2
syscall_table_addr:
.long sys_call_table
syscall_badsys:
mov r12, -ENOSYS
rjmp syscall_return
.global ret_from_fork
ret_from_fork:
rcall schedule_tail
/* check for syscall tracing */
get_thread_info r0
ld.w r1, r0[TI_flags]
andl r1, _TIF_ALLWORK_MASK, COH
brne syscall_exit_work
rjmp syscall_exit_cont
syscall_trace_enter:
pushm r8-r12
rcall syscall_trace
popm r8-r12
rjmp syscall_trace_cont
syscall_exit_work:
bld r1, TIF_SYSCALL_TRACE
brcc 1f
unmask_interrupts
rcall syscall_trace
mask_interrupts
ld.w r1, r0[TI_flags]
1: bld r1, TIF_NEED_RESCHED
brcc 2f
unmask_interrupts
rcall schedule
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp 1b
2: mov r2, _TIF_SIGPENDING | _TIF_RESTORE_SIGMASK
tst r1, r2
breq 3f
unmask_interrupts
mov r12, sp
mov r11, r0
rcall do_notify_resume
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp 1b
3: bld r1, TIF_BREAKPOINT
brcc syscall_exit_cont
mfsr r3, SYSREG_TLBEHI
lddsp r2, sp[REG_PC]
andl r3, 0xff, COH
lsl r3, 1
sbr r3, 30
sbr r3, 0
mtdr DBGREG_BWA2A, r2
mtdr DBGREG_BWC2A, r3
rjmp syscall_exit_cont
/* The slow path of the TLB miss handler */
page_table_not_present:
page_not_present:
tlbmiss_restore
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_page_fault
rjmp ret_from_exception
/* This function expects to find offending PC in SYSREG_RAR_EX */
save_full_context_ex:
mfsr r8, SYSREG_RSR_EX
mov r12, r8
andh r8, (MODE_MASK >> 16), COH
mfsr r11, SYSREG_RAR_EX
brne 2f
1: pushm r11, r12 /* PC and SR */
unmask_exceptions
ret r12
2: sub r10, sp, -(FRAME_SIZE_FULL - REG_LR)
stdsp sp[4], r10 /* replace saved SP */
rjmp 1b
/* Low-level exception handlers */
handle_critical:
pushm r12
pushm r0-r12
rcall save_full_context_ex
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_critical_exception
/* We should never get here... */
bad_return:
sub r12, pc, (. - 1f)
bral panic
.align 2
1: .asciz "Return from critical exception!"
.align 1
do_bus_error_write:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mov r11, 1
rjmp 1f
do_bus_error_read:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mov r11, 0
1: mfsr r12, SYSREG_BEAR
mov r10, sp
rcall do_bus_error
rjmp ret_from_exception
.align 1
do_nmi_ll:
sub sp, 4
stmts --sp, r0-lr
mfsr r9, SYSREG_RSR_NMI
mfsr r8, SYSREG_RAR_NMI
bfextu r0, r9, MODE_SHIFT, 3
brne 2f
1: pushm r8, r9 /* PC and SR */
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_nmi
popm r8-r9
mtsr SYSREG_RAR_NMI, r8
tst r0, r0
mtsr SYSREG_RSR_NMI, r9
brne 3f
ldmts sp++, r0-lr
sub sp, -4 /* skip r12_orig */
rete
2: sub r10, sp, -(FRAME_SIZE_FULL - REG_LR)
stdsp sp[4], r10 /* replace saved SP */
rjmp 1b
3: popm lr
sub sp, -4 /* skip sp */
popm r0-r12
sub sp, -4 /* skip r12_orig */
rete
handle_address_fault:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_address_exception
rjmp ret_from_exception
handle_protection_fault:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_page_fault
rjmp ret_from_exception
.align 1
do_illegal_opcode_ll:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
mfsr r12, SYSREG_ECR
mov r11, sp
rcall do_illegal_opcode
rjmp ret_from_exception
do_dtlb_modified:
pushm r0-r3
mfsr r1, SYSREG_TLBEAR
mfsr r0, SYSREG_PTBR
lsr r2, r1, PGDIR_SHIFT
ld.w r0, r0[r2 << 2]
lsl r1, (32 - PGDIR_SHIFT)
lsr r1, (32 - PGDIR_SHIFT) + PAGE_SHIFT
/* Translate to virtual address in P1 */
andl r0, 0xf000
sbr r0, 31
add r2, r0, r1 << 2
ld.w r3, r2[0]
sbr r3, _PAGE_BIT_DIRTY
mov r0, r3
st.w r2[0], r3
/* The page table is up-to-date. Update the TLB entry as well */
andl r0, lo(_PAGE_FLAGS_HARDWARE_MASK)
mtsr SYSREG_TLBELO, r0
/* MMUCR[DRP] is updated automatically, so let's go... */
tlbw
popm r0-r3
rete
do_fpe_ll:
sub sp, 4
stmts --sp, r0-lr
rcall save_full_context_ex
unmask_interrupts
mov r12, 26
mov r11, sp
rcall do_fpe
rjmp ret_from_exception
ret_from_exception:
mask_interrupts
lddsp r4, sp[REG_SR]
andh r4, (MODE_MASK >> 16), COH
brne fault_resume_kernel
get_thread_info r0
ld.w r1, r0[TI_flags]
andl r1, _TIF_WORK_MASK, COH
brne fault_exit_work
fault_resume_user:
popm r8-r9
mask_exceptions
mtsr SYSREG_RAR_EX, r8
mtsr SYSREG_RSR_EX, r9
ldmts sp++, r0-lr
sub sp, -4
rete
fault_resume_kernel:
#ifdef CONFIG_PREEMPT
get_thread_info r0
ld.w r2, r0[TI_preempt_count]
cp.w r2, 0
brne 1f
ld.w r1, r0[TI_flags]
bld r1, TIF_NEED_RESCHED
brcc 1f
lddsp r4, sp[REG_SR]
bld r4, SYSREG_GM_OFFSET
brcs 1f
rcall preempt_schedule_irq
1:
#endif
popm r8-r9
mask_exceptions
mfsr r1, SYSREG_SR
mtsr SYSREG_RAR_EX, r8
mtsr SYSREG_RSR_EX, r9
popm lr
sub sp, -4 /* ignore SP */
popm r0-r12
sub sp, -4 /* ignore r12_orig */
rete
irq_exit_work:
/* Switch to exception mode so that we can share the same code. */
mfsr r8, SYSREG_SR
cbr r8, SYSREG_M0_OFFSET
orh r8, hi(SYSREG_BIT(M1) | SYSREG_BIT(M2))
mtsr SYSREG_SR, r8
sub pc, -2
get_thread_info r0
ld.w r1, r0[TI_flags]
fault_exit_work:
bld r1, TIF_NEED_RESCHED
brcc 1f
unmask_interrupts
rcall schedule
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp fault_exit_work
1: mov r2, _TIF_SIGPENDING | _TIF_RESTORE_SIGMASK
tst r1, r2
breq 2f
unmask_interrupts
mov r12, sp
mov r11, r0
rcall do_notify_resume
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp fault_exit_work
2: bld r1, TIF_BREAKPOINT
brcc fault_resume_user
mfsr r3, SYSREG_TLBEHI
lddsp r2, sp[REG_PC]
andl r3, 0xff, COH
lsl r3, 1
sbr r3, 30
sbr r3, 0
mtdr DBGREG_BWA2A, r2
mtdr DBGREG_BWC2A, r3
rjmp fault_resume_user
/* If we get a debug trap from privileged context we end up here */
handle_debug_priv:
/* Fix up LR and SP in regs. r11 contains the mode we came from */
mfsr r8, SYSREG_SR
mov r9, r8
andh r8, hi(~MODE_MASK)
or r8, r11
mtsr SYSREG_SR, r8
sub pc, -2
stdsp sp[REG_LR], lr
mtsr SYSREG_SR, r9
sub pc, -2
sub r10, sp, -FRAME_SIZE_FULL
stdsp sp[REG_SP], r10
mov r12, sp
rcall do_debug_priv
/* Now, put everything back */
ssrf SR_EM_BIT
popm r10, r11
mtsr SYSREG_RAR_DBG, r10
mtsr SYSREG_RSR_DBG, r11
mfsr r8, SYSREG_SR
mov r9, r8
andh r8, hi(~MODE_MASK)
andh r11, hi(MODE_MASK)
or r8, r11
mtsr SYSREG_SR, r8
sub pc, -2
popm lr
mtsr SYSREG_SR, r9
sub pc, -2
sub sp, -4 /* skip SP */
popm r0-r12
sub sp, -4
retd
/*
* At this point, everything is masked, that is, interrupts,
* exceptions and debugging traps. We might get called from
* interrupt or exception context in some rare cases, but this
* will be taken care of by do_debug(), so we're not going to
* do a 100% correct context save here.
*/
handle_debug:
sub sp, 4 /* r12_orig */
stmts --sp, r0-lr
mfsr r10, SYSREG_RAR_DBG
mfsr r11, SYSREG_RSR_DBG
unmask_exceptions
pushm r10,r11
andh r11, (MODE_MASK >> 16), COH
brne handle_debug_priv
mov r12, sp
rcall do_debug
lddsp r10, sp[REG_SR]
andh r10, (MODE_MASK >> 16), COH
breq debug_resume_user
debug_restore_all:
popm r10,r11
mask_exceptions
mtsr SYSREG_RSR_DBG, r11
mtsr SYSREG_RAR_DBG, r10
ldmts sp++, r0-lr
sub sp, -4
retd
debug_resume_user:
get_thread_info r0
mask_interrupts
ld.w r1, r0[TI_flags]
andl r1, _TIF_DBGWORK_MASK, COH
breq debug_restore_all
1: bld r1, TIF_NEED_RESCHED
brcc 2f
unmask_interrupts
rcall schedule
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp 1b
2: mov r2, _TIF_SIGPENDING | _TIF_RESTORE_SIGMASK
tst r1, r2
breq 3f
unmask_interrupts
mov r12, sp
mov r11, r0
rcall do_notify_resume
mask_interrupts
ld.w r1, r0[TI_flags]
rjmp 1b
3: bld r1, TIF_SINGLE_STEP
brcc debug_restore_all
mfdr r2, DBGREG_DC
sbr r2, DC_SS_BIT
mtdr DBGREG_DC, r2
rjmp debug_restore_all
.set rsr_int0, SYSREG_RSR_INT0
.set rsr_int1, SYSREG_RSR_INT1
.set rsr_int2, SYSREG_RSR_INT2
.set rsr_int3, SYSREG_RSR_INT3
.set rar_int0, SYSREG_RAR_INT0
.set rar_int1, SYSREG_RAR_INT1
.set rar_int2, SYSREG_RAR_INT2
.set rar_int3, SYSREG_RAR_INT3
.macro IRQ_LEVEL level
.type irq_level\level, @function
irq_level\level:
sub sp, 4 /* r12_orig */
stmts --sp,r0-lr
mfsr r8, rar_int\level
mfsr r9, rsr_int\level
pushm r8-r9
mov r11, sp
mov r12, \level
rcall do_IRQ
lddsp r4, sp[REG_SR]
bfextu r4, r4, SYSREG_M0_OFFSET, 3
cp.w r4, MODE_SUPERVISOR >> SYSREG_M0_OFFSET
breq 2f
cp.w r4, MODE_USER >> SYSREG_M0_OFFSET
#ifdef CONFIG_PREEMPT
brne 3f
#else
brne 1f
#endif
get_thread_info r0
ld.w r1, r0[TI_flags]
andl r1, _TIF_WORK_MASK, COH
brne irq_exit_work
1: popm r8-r9
mtsr rar_int\level, r8
mtsr rsr_int\level, r9
ldmts sp++,r0-lr
sub sp, -4 /* ignore r12_orig */
rete
2: get_thread_info r0
ld.w r1, r0[TI_flags]
bld r1, TIF_CPU_GOING_TO_SLEEP
#ifdef CONFIG_PREEMPT
brcc 3f
#else
brcc 1b
#endif
sub r1, pc, . - cpu_idle_skip_sleep
stdsp sp[REG_PC], r1
#ifdef CONFIG_PREEMPT
3: get_thread_info r0
ld.w r2, r0[TI_preempt_count]
cp.w r2, 0
brne 1b
ld.w r1, r0[TI_flags]
bld r1, TIF_NEED_RESCHED
brcc 1b
lddsp r4, sp[REG_SR]
bld r4, SYSREG_GM_OFFSET
brcs 1b
rcall preempt_schedule_irq
#endif
rjmp 1b
.endm
.section .irq.text,"ax",@progbits
.global cpu_idle_sleep
cpu_idle_sleep:
mask_interrupts
get_thread_info r8
ld.w r9, r8[TI_flags]
bld r9, TIF_NEED_RESCHED
brcs cpu_idle_enable_int_and_exit
sbr r9, TIF_CPU_GOING_TO_SLEEP
st.w r8[TI_flags], r9
unmask_interrupts
sleep 0
cpu_idle_skip_sleep:
mask_interrupts
ld.w r9, r8[TI_flags]
cbr r9, TIF_CPU_GOING_TO_SLEEP
st.w r8[TI_flags], r9
cpu_idle_enable_int_and_exit:
unmask_interrupts
retal r12
.global irq_level0
.global irq_level1
.global irq_level2
.global irq_level3
IRQ_LEVEL 0
IRQ_LEVEL 1
IRQ_LEVEL 2
IRQ_LEVEL 3