PM: x86-64 suspend-resume fix
Subsystems (3)
PATCH 1/2 x86_64: acpi/wakeup_64 cleanup
In reply to Rafael J. WysockiFrom: Jiri Slaby <[email protected]>
- remove %ds re-set, it's already set in wakeup_long64
- remove double labels and alignment (ENTRY already adds both)
- use meaningful resume point labelname
- skip alignment while jumping from wakeup_long64 to the resume point
- remove .size, .type and unused labels
[v2]
- added ENDPROCs
--- arch/x86/kernel/acpi/wakeup_64.S | 26 +++++++------------------- 1 file changed, 7 insertions(+), 19 deletions(-)
arch/x86/kernel/acpi/wakeup_64.S +7 -19
@@ -13,7 +13,6 @@ * Hooray, we are in Long 64-bit mode (but still running in low memory) */ ENTRY(wakeup_long64) -wakeup_long64: movq saved_magic, %rax movq $0x123456789abcdef0, %rdx cmpq %rdx, %rax
@@ -34,16 +33,12 @@ wakeup_long64: movq saved_rip, %rax jmp *%rax +ENDPROC(wakeup_long64) bogus_64_magic: jmp bogus_64_magic - .align 2 - .p2align 4,,15 -.globl do_suspend_lowlevel - .type do_suspend_lowlevel,@function -do_suspend_lowlevel: -.LFB5: +ENTRY(do_suspend_lowlevel) subq $8, %rsp xorl %eax, %eax call save_processor_state
@@ -67,7 +62,7 @@ do_suspend_lowlevel: pushfq popq pt_regs_flags(%rax) - movq $.L97, saved_rip(%rip) + movq $resume_point, saved_rip(%rip) movq %rsp, saved_rsp movq %rbp, saved_rbp
@@ -79,13 +74,9 @@ do_suspend_lowlevel: movl $3, %edi xorl %eax, %eax jmp acpi_enter_sleep_state -.L97: - .p2align 4,,7 -.L99: - .align 4 - movl $24, %eax - movw %ax, %ds + .align 4 +resume_point: /* We don't restore %rax, it must be 0 anyway */ movq $saved_context, %rax movq saved_context_cr4(%rax), %rbx
@@ -117,12 +108,9 @@ do_suspend_lowlevel: xorl %eax, %eax addq $8, %rsp jmp restore_processor_state -.LFE5: -.Lfe5: - .size do_suspend_lowlevel, .Lfe5-do_suspend_lowlevel - +ENDPROC(do_suspend_lowlevel) + .data -ALIGN ENTRY(saved_rbp) .quad 0 ENTRY(saved_rsi) .quad 0 ENTRY(saved_rdi) .quad 0
PATCH 2/2 x86_64: Fix S3 fail path
In reply to Rafael J. WysockiFrom: Jiri Slaby <[email protected]>
As acpi_enter_sleep_state can fail, take this into account in do_suspend_lowlevel and don't return to the do_suspend_lowlevel's caller. This would break (currently) fpu status and preempt count.
Technically, this means use call' instead of jmp' and jmp' to the resume_point' after the call' (i.e. if acpi_enter_sleep_state returns=fails). resume_point' will handle the restore of fpu and preempt count gracefully.
--- arch/x86/kernel/acpi/wakeup_64.S | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/kernel/acpi/wakeup_64.S +3 -1
@@ -73,7 +73,9 @@ ENTRY(do_suspend_lowlevel) addq $8, %rsp movl $3, %edi xorl %eax, %eax - jmp acpi_enter_sleep_state + call acpi_enter_sleep_state + /* in case something went wrong, restore the machine status and go on */ + jmp resume_point .align 4 resume_point:
PATCH 2/2 x86_64: Fix S3 fail path
In reply to Rafael J. WysockiFrom: Jiri Slaby <[email protected]>
As acpi_enter_sleep_state can fail, take this into account in do_suspend_lowlevel and don't return to the do_suspend_lowlevel's caller. This would break (currently) fpu status and preempt count.
Technically, this means use call' instead of jmp' and jmp' to the resume_point' after the call' (i.e. if acpi_enter_sleep_state returns=fails). resume_point' will handle the restore of fpu and preempt count gracefully.
--- arch/x86/kernel/acpi/wakeup_64.S | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/kernel/acpi/wakeup_64.S +3 -1
@@ -73,7 +73,9 @@ ENTRY(do_suspend_lowlevel) addq $8, %rsp movl $3, %edi xorl %eax, %eax - jmp acpi_enter_sleep_state + call acpi_enter_sleep_state + /* in case something went wrong, restore the machine status and go on */ + jmp resume_point .align 4 resume_point:
cover PM: x86-64 suspend-resume fix
Hi,
The following two patches clean-up and fix the x86-64 resume code.
1/2 is a cleanup, but it is simple enough to be applied now IMO.
2/2 is a bug fix based on 1/2, which is 2.6.29 material.
Please apply them to the suspend branch and push to Linus for 2.6.29 (Ingo says he doesn't mind if they go through your tree).
Thanks,
Rafael
PATCH 1/2 x86_64: acpi/wakeup_64 cleanup
From: Jiri Slaby <[email protected]>
- remove %ds re-set, it's already set in wakeup_long64
- remove double labels and alignment (ENTRY already adds both)
- use meaningful resume point labelname
- skip alignment while jumping from wakeup_long64 to the resume point
- remove .size, .type and unused labels
[v2]
- added ENDPROCs
--- arch/x86/kernel/acpi/wakeup_64.S | 26 +++++++------------------- 1 file changed, 7 insertions(+), 19 deletions(-)
arch/x86/kernel/acpi/wakeup_64.S +7 -19
@@ -13,7 +13,6 @@ * Hooray, we are in Long 64-bit mode (but still running in low memory) */ ENTRY(wakeup_long64) -wakeup_long64: movq saved_magic, %rax movq $0x123456789abcdef0, %rdx cmpq %rdx, %rax
@@ -34,16 +33,12 @@ wakeup_long64: movq saved_rip, %rax jmp *%rax +ENDPROC(wakeup_long64) bogus_64_magic: jmp bogus_64_magic - .align 2 - .p2align 4,,15 -.globl do_suspend_lowlevel - .type do_suspend_lowlevel,@function -do_suspend_lowlevel: -.LFB5: +ENTRY(do_suspend_lowlevel) subq $8, %rsp xorl %eax, %eax call save_processor_state
@@ -67,7 +62,7 @@ do_suspend_lowlevel: pushfq popq pt_regs_flags(%rax) - movq $.L97, saved_rip(%rip) + movq $resume_point, saved_rip(%rip) movq %rsp, saved_rsp movq %rbp, saved_rbp
@@ -79,13 +74,9 @@ do_suspend_lowlevel: movl $3, %edi xorl %eax, %eax jmp acpi_enter_sleep_state -.L97: - .p2align 4,,7 -.L99: - .align 4 - movl $24, %eax - movw %ax, %ds + .align 4 +resume_point: /* We don't restore %rax, it must be 0 anyway */ movq $saved_context, %rax movq saved_context_cr4(%rax), %rbx
@@ -117,12 +108,9 @@ do_suspend_lowlevel: xorl %eax, %eax addq $8, %rsp jmp restore_processor_state -.LFE5: -.Lfe5: - .size do_suspend_lowlevel, .Lfe5-do_suspend_lowlevel - +ENDPROC(do_suspend_lowlevel) + .data -ALIGN ENTRY(saved_rbp) .quad 0 ENTRY(saved_rsi) .quad 0 ENTRY(saved_rdi) .quad 0
Len Brown
reviewerQuoted 10 lines · Rafael J. Wysocki
Hi,
The following two patches clean-up and fix the x86-64 resume code.
1/2 is a cleanup, but it is simple enough to be applied now IMO.
2/2 is a bug fix based on 1/2, which is 2.6.29 material.
Please apply them to the suspend branch and push to Linus for 2.6.29 (Ingo says he doesn't mind if they go through your tree).
1 & 2 applied to suspend branch.
thanks,
Len Brown, Intel Open Source Technology Center
Len Brown
reviewerQuoted 10 lines · Rafael J. Wysocki
Hi,
The following two patches clean-up and fix the x86-64 resume code.
1/2 is a cleanup, but it is simple enough to be applied now IMO.
2/2 is a bug fix based on 1/2, which is 2.6.29 material.
Please apply them to the suspend branch and push to Linus for 2.6.29 (Ingo says he doesn't mind if they go through your tree).
1 & 2 applied to suspend branch.
thanks,
Len Brown, Intel Open Source Technology Center