x86/CPU: Adapt assembly for PIE support
Review
Quoted 6 lines · Thomas Garnier
(quoted diff) +2
-1- "pushq $1f\n\t" + "movabsq $1f, %q0\n\t" + "pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"
Fake PIE. True PIE looks like this:
ffffffff81022d70 <do_sync_core>:
ffffffff81022d70: 8c d0 mov eax,ss ffffffff81022d72: 50 push rax ffffffff81022d73: 54 push rsp ffffffff81022d74: 48 83 04 24 08 add QWORD PTR [rsp],0x8 ffffffff81022d79: 9c pushf ffffffff81022d7a: 8c c8 mov eax,cs ffffffff81022d7c: 50 push rax ffffffff81022d7d: ===> 48 8d 05 03 00 00 00 lea rax,[rip+0x3] # ffffffff81022d87 <do_sync_core+0x17> ffffffff81022d84: 50 push rax ffffffff81022d85: 48 cf iretq ffffffff81022d87: c3 ret
arch/x86/include/asm/processor.h +2 -1
@@ -710,7 +710,8 @@ static inline void sync_core(void)"pushfq\n\t" "mov %%cs, %0\n\t" "pushq %q0\n\t" -"pushq $1f\n\t" +"leaq 1f(%%rip), %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"
Quoted 10 lines · Alexey Dobriyan
arch/x86/include/asm/processor.h static inline void sync_core(void) · line 717
Quoted 6 lines · Thomas Garnier
arch/x86/include/asm/processor.h +2
-1-"pushq $1f\n\t" +"movabsq $1f, %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:" Fake PIE. True PIE looks like this:
I used movabsq in couple assembly changes where the memory context is unclear and relative reference might lead to issues. It happened on early boot and hibernation save/restore paths. Do you think a relative reference in this function will always be accurate?
Quoted 28 lines
ffffffff81022d70<do_sync_core>:ffffffff81022d70: 8c d0 mov eax,ss ffffffff81022d72: 50 push rax ffffffff81022d73: 54 push rsp ffffffff81022d74: 48 83 04 24 08 add QWORD PTR [rsp],0x8 ffffffff81022d79: 9c pushf ffffffff81022d7a: 8c c8 mov eax,cs ffffffff81022d7c: 50 push rax ffffffff81022d7d: ===> 48 8d 05 03 00 00 00 lea rax,[rip+0x3] # ffffffff81022d87 <do_sync_core+0x17> ffffffff81022d84: 50 push rax ffffffff81022d85: 48 cf iretq ffffffff81022d87: c3 retSigned-off-by Alexey Dobriyan <[email protected]>arch/x86/include/asm/processor.h +2
-1@@ -710,7 +710,8 @@ static inline void sync_core(void)"pushfq\n\t" "mov %%cs, %0\n\t" "pushq %q0\n\t" -"pushq $1f\n\t" +"leaq 1f(%%rip), %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"
Quoted 10 lines · Thomas Garnier
Quoted 9 lines · Alexey Dobriyan
Quoted 6 lines · Thomas Garnier
arch/x86/include/asm/processor.h +2
-1-"pushq $1f\n\t" +"movabsq $1f, %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"
Quoted 6 lines
Quoted 1 line
Fake PIE. True PIE looks like this:
I used movabsq in couple assembly changes where the memory context is unclear and relative reference might lead to issues. It happened on early boot and hibernation save/restore paths. Do you think a relative reference in this function will always be accurate?
As long as iretq target is not too far it should be OK.
I'm not really sure which issues can pop up.
IRETQ is 64-bit only, RIP-relative addressing is 64-bit only. Assembler (hopefully) will error compilation if target is too far.
And it is shorter than movabsq.
Quoted 27 lines
Quoted 27 lines
ffffffff81022d70<do_sync_core>:ffffffff81022d70: 8c d0 mov eax,ss ffffffff81022d72: 50 push rax ffffffff81022d73: 54 push rsp ffffffff81022d74: 48 83 04 24 08 add QWORD PTR [rsp],0x8 ffffffff81022d79: 9c pushf ffffffff81022d7a: 8c c8 mov eax,cs ffffffff81022d7c: 50 push rax ffffffff81022d7d: ===> 48 8d 05 03 00 00 00 lea rax,[rip+0x3] # ffffffff81022d87 <do_sync_core+0x17> ffffffff81022d84: 50 push rax ffffffff81022d85: 48 cf iretq ffffffff81022d87: c3 retSigned-off-by Alexey Dobriyan <[email protected]>arch/x86/include/asm/processor.h +2
-1@@ -710,7 +710,8 @@ static inline void sync_core(void)"pushfq\n\t" "mov %%cs, %0\n\t" "pushq %q0\n\t" -"pushq $1f\n\t" +"leaq 1f(%%rip), %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"
Quoted 19 lines · Alexey Dobriyan
Quoted 10 lines · Thomas Garnier
Quoted 9 lines · Alexey Dobriyan
Quoted 6 lines · Thomas Garnier
arch/x86/include/asm/processor.h +2
-1-"pushq $1f\n\t" +"movabsq $1f, %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:" Quoted 6 lines
Quoted 1 line
Fake PIE. True PIE looks like this:
I used movabsq in couple assembly changes where the memory context is unclear and relative reference might lead to issues. It happened on early boot and hibernation save/restore paths. Do you think a relative reference in this function will always be accurate?
Quoted 8 lines
As long as iretq target is not too far it should be OK.
I'm not really sure which issues can pop up.
IRETQ is 64-bit only, RIP-relative addressing is 64-bit only. Assembler (hopefully) will error compilation if target is too far.
And it is shorter than movabsq.
Agree, I will change it and run some tests for the next iteration.
Quoted 28 lines
Quoted 28 lines
Quoted 28 lines
ffffffff81022d70<do_sync_core>:ffffffff81022d70: 8c d0 mov eax,ss ffffffff81022d72: 50 push rax ffffffff81022d73: 54 push rsp ffffffff81022d74: 48 83 04 24 08 add QWORD PTR [rsp],0x8 ffffffff81022d79: 9c pushf ffffffff81022d7a: 8c c8 mov eax,cs ffffffff81022d7c: 50 push rax ffffffff81022d7d: ===> 48 8d 05 03 00 00 00 lea rax,[rip+0x3] # ffffffff81022d87 <do_sync_core+0x17> ffffffff81022d84: 50 push rax ffffffff81022d85: 48 cf iretq ffffffff81022d87: c3 retSigned-off-by Alexey Dobriyan <[email protected]>arch/x86/include/asm/processor.h +2
-1@@ -710,7 +710,8 @@ static inline void sync_core(void)"pushfq\n\t" "mov %%cs, %0\n\t" "pushq %q0\n\t" -"pushq $1f\n\t" +"leaq 1f(%%rip), %q0\n\t" +"pushq %q0\n\t" "iretq\n\t" UNWIND_HINT_RESTORE "1:"