Skip to contentSkip to conversation

ARM: make cpu_init() notrace

PATCH New Jon Medhurst (Tixy) 3 participants 6 messages Apr 18, 2013 – Apr 25, 2013 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

On resume from CPU power down any trace hooks enabled in cpu_init() will get called before that function has done set_my_cpu_offset(), so any use of per-cpu variables by trace hook code will cause bad things to happen. Prevent this by marking the function notrace.

This fixes lockups/crashes seen when enabling function tracer on TC2 with the not yet mainlined cpuidle driver.

---
 arch/arm/kernel/setup.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/setup.c +1 -1
index d343a6c..943cbf0 100644
@@ -390,7 +390,7 @@ static void __init feat_v6_fixup(void)
  *
  * cpu_init sets up the per-CPU stacks.
  */
-void cpu_init(void)
+void notrace cpu_init(void)
 {
 	unsigned int cpu = smp_processor_id();
 	struct stack *stk = &stacks[cpu];
Signature · 1 line
1.7.10.4
#1 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

Quoted 7 lines · Jon Medhurst (Tixy)

On resume from CPU power down any trace hooks enabled in cpu_init() will get called before that function has done set_my_cpu_offset(), so any use of per-cpu variables by trace hook code will cause bad things to happen. Prevent this by marking the function notrace.

This fixes lockups/crashes seen when enabling function tracer on TC2 with the not yet mainlined cpuidle driver.

This has potential to fix things on other platforms too.

Quoted 2 lines

I think you may send it to RMK's patch system.

Quoted 21 lines
---
 arch/arm/kernel/setup.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/kernel/setup.c +1 -1
index d343a6c..943cbf0 100644
@@ -390,7 +390,7 @@ static void __init feat_v6_fixup(void)
  *
  * cpu_init sets up the per-CPU stacks.
  */
-void cpu_init(void)
+void notrace cpu_init(void)
 {
 	unsigned int cpu = smp_processor_id();
 	struct stack *stk = &stacks[cpu];
Signature · 1 line
1.7.10.4
#2 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

In reply to Jon Medhurst (Tixy)
Quoted 7 lines · Jon Medhurst (Tixy)

On resume from CPU power down any trace hooks enabled in cpu_init() will get called before that function has done set_my_cpu_offset(), so any use of per-cpu variables by trace hook code will cause bad things to happen. Prevent this by marking the function notrace.

This fixes lockups/crashes seen when enabling function tracer on TC2 with the not yet mainlined cpuidle driver.

Looks sane. Needs to go to the patch system though. Not sure if it'll make 3.9 given its proximity (we're on -rc8).

#3 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

Quoted 11 lines · Russell King - ARM Linux
Quoted 7 lines · Jon Medhurst (Tixy)

On resume from CPU power down any trace hooks enabled in cpu_init() will get called before that function has done set_my_cpu_offset(), so any use of per-cpu variables by trace hook code will cause bad things to happen. Prevent this by marking the function notrace.

This fixes lockups/crashes seen when enabling function tracer on TC2 with the not yet mainlined cpuidle driver.

Looks sane. Needs to go to the patch system though. Not sure if it'll make 3.9 given its proximity (we're on -rc8).

Added as patch 7700/1. Don't think there is a big rush as the problem has probably been there since 3.8 when commit 14318efb (implement optimized percpu variable access) started using TPIDRPRW; which is presumably lost over CPU power down.

Signature · 1 line
Tixy
#4 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

Quoted 10 lines · Jon Medhurst (Tixy)
Quoted 9 lines · Russell King - ARM Linux
Quoted 7 lines · Jon Medhurst (Tixy)

On resume from CPU power down any trace hooks enabled in cpu_init() will get called before that function has done set_my_cpu_offset(), so any use of per-cpu variables by trace hook code will cause bad things to happen. Prevent this by marking the function notrace.

This fixes lockups/crashes seen when enabling function tracer on TC2 with the not yet mainlined cpuidle driver.

Quoted 7 lines
Quoted 2 lines

Looks sane. Needs to go to the patch system though. Not sure if it'll make 3.9 given its proximity (we're on -rc8).

Added as patch 7700/1. Don't think there is a big rush as the problem has probably been there since 3.8 when commit 14318efb (implement optimized percpu variable access) started using TPIDRPRW; which is presumably lost over CPU power down.

$ pdb getpatch 7700/1 |tr ' ' '_' |less
diff_--git_a/arch/arm/kernel/setup.c_b/arch/arm/kernel/setup.c
index_d343a6c..943cbf0_100644
---_a/arch/arm/kernel/setup.c
+++_b/arch/arm/kernel/setup.c
@@_-390,7_+390,7_@@_static_void___init_feat_v6_fixup(void)
__*
__*_cpu_init_sets_up_the_per-CPU_stacks.
__*/
-void_cpu_init(void)
+void_notrace_cpu_init(void)
_{
________unsigned_int_cpu_=_smp_processor_id();
________struct_stack_*stk_=_&stacks[cpu];

Can you guess why git has a problem appying the above... you seem to have sent the patch from your mailer, I guess your mailer thinks it knows better when it comes to what characters should be in your message... or maybe its a cut'n'paste issue?

#5 lore raw

PATCH 1/1 ARM: make cpu_init() notrace

Quoted 15 lines · Russell King - ARM Linux
$ pdb getpatch 7700/1 |tr ' ' '_' |less
diff_--git_a/arch/arm/kernel/setup.c_b/arch/arm/kernel/setup.c
index_d343a6c..943cbf0_100644
---_a/arch/arm/kernel/setup.c
+++_b/arch/arm/kernel/setup.c
@@_-390,7_+390,7_@@_static_void___init_feat_v6_fixup(void)
__*
__*_cpu_init_sets_up_the_per-CPU_stacks.
__*/
(quoted diff) +1 -1
-void_cpu_init(void)
+void_notrace_cpu_init(void)

_{
________unsigned_int_cpu_=_smp_processor_id();
________struct_stack_*stk_=_&stacks[cpu];

Quoted 1 line

Can you guess why git has a problem appying the above...

Sorry about that, the user forgot to tell their mailer it was pre-formatted text. I've sent a fixed version.

Note to self: bcc self on patch submissions and then run them though 'git am' to catch problems before others hit them.

Signature · 1 line
Tixy