ARM: kernel: irq: Simplify allocation of stack frame
RFC 1/1 ARM: kernel: irq: Simplify allocation of stack frame
On receiving IRQ exception in SVC mode, all the SVC mode registers are saved onto the stack very early on.
The stack frame allocation code for IRQ entry during SVC mode (svc_entry) is hard to read as 4-less is allocated initially only to be allocated later implicity using the mov r3, [sp, #-4]! instruction. We make code easier to read by allocating the 4 bytes on the stack frame in the beginning itself and remove all instances where 4 bytes is adjusted.
--- arch/arm/kernel/entry-armv.S | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
arch/arm/kernel/entry-armv.S +5 -5
@@ -149,7 +149,7 @@ ENDPROC(__und_invalid) .macro svc_entry, stack_hole=0 UNWIND(.fnstart ) UNWIND(.save {r0 - pc} ) - sub sp, sp, #(S_FRAME_SIZE + \stack_hole - 4) + sub sp, sp, #(S_FRAME_SIZE + \stack_hole) #ifdef CONFIG_THUMB2_KERNEL SPFIX( str r0, [sp] ) @ temporarily saved SPFIX( mov r0, sp )
@@ -159,14 +159,14 @@ ENDPROC(__und_invalid) SPFIX( tst sp, #4 ) #endif SPFIX( subeq sp, sp, #4 ) - stmia sp, {r1 - r12} + stmia sp, {r0 - r12} ldmia r0, {r3 - r5} - add r7, sp, #S_SP - 4 @ here for interlock avoidance + add r7, sp, #S_SP @ here for interlock avoidance mov r6, #-1 @ "" "" "" "" - add r2, sp, #(S_FRAME_SIZE + \stack_hole - 4) + add r2, sp, #(S_FRAME_SIZE + \stack_hole) SPFIX( addeq r2, r2, #4 ) - str r3, [sp, #-4]! @ save the "real" r0 copied + str r3, [sp] @ save the "real" r0 copied @ from the exception stack mov r3, lr
Signature · 1 line
1.8.1.2
Quoted 8 lines · Joel Fernandes
On receiving IRQ exception in SVC mode, all the SVC mode registers are saved onto the stack very early on.
The stack frame allocation code for IRQ entry during SVC mode (
svc_entry) is hard to read as 4-less is allocated initially only to be allocated later implicity using the mov r3, [sp, #-4]! instruction. We make code easier to read by allocating the 4 bytes on the stack frame in the beginning itself and remove all instances where 4 bytes is adjusted.
You omit to say that this results in saving one additional register unnecessarily in the stmia. We could use a stmib there instead which would avoid that issue while keeping the rest of the change.
Quoted 11 lines · Russell King - ARM Linux
Quoted 8 lines · Joel Fernandes
On receiving IRQ exception in SVC mode, all the SVC mode registers are saved onto the stack very early on.
The stack frame allocation code for IRQ entry during SVC mode (
svc_entry) is hard to read as 4-less is allocated initially only to be allocated later implicity using the mov r3, [sp, #-4]! instruction. We make code easier to read by allocating the 4 bytes on the stack frame in the beginning itself and remove all instances where 4 bytes is adjusted.You omit to say that this results in saving one additional register
Ok, I will add this detail to the commit message.
Quoted 2 lines
unnecessarily in the stmia. We could use a stmib there instead which would avoid that issue while keeping the rest of the change.
But stmib is not available in THUMB mode, so this will break the THUMB builds.
usr_entry does something similar:
ARM( stmib sp, {r1 - r12} )
THUMB( stmia sp, {r0 - r12} )Let me know your suggestions about using ARM/THUMB macros or stmia for both cases. Thanks.
Regards,
-Joel
Quoted 10 lines · Russell King - ARM Linux
Quoted 8 lines · Joel Fernandes
On receiving IRQ exception in SVC mode, all the SVC mode registers are saved onto the stack very early on.
The stack frame allocation code for IRQ entry during SVC mode (
svc_entry) is hard to read as 4-less is allocated initially only to be allocated later implicity using the mov r3, [sp, #-4]! instruction. We make code easier to read by allocating the 4 bytes on the stack frame in the beginning itself and remove all instances where 4 bytes is adjusted.
Quoted 4 lines
You omit to say that this results in saving one additional register unnecessarily in the stmia. We could use a stmib there instead which would avoid that issue while keeping the rest of the change.
Hi Russel,
BTW I used ETM to check the number of cycles used to store the extra register with and without this patch and with both cases it takes 7 cycles.
My platform uses Cortex-A8 (AM335x SoC).
Thanks,
-Joel