[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[janboe]

__backtrace is empty when ARM_UNWIND enable, so instead dump_stack

Signed-off-by: janboe <janboe.ye@gmail.com>
---
 arch/arm/kernel/process.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
index acf5e6f..e7ad878 100644
--- a/arch/arm/kernel/process.c
+++ b/arch/arm/kernel/process.c
@@ -271,7 +271,11 @@ void show_regs(struct pt_regs * regs)
 	printk("\n");
 	printk("Pid: %d, comm: %20s\n", task_pid_nr(current), current->comm);
 	__show_regs(regs);
+#ifdef CONFIG_ARM_UNWIND
+	dump_stack();
+#else
 	__backtrace();
+#endif
 }
 
 ATOMIC_NOTIFIER_HEAD(thread_notify_head);
-- 
1.7.1

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[Catalin Marinas]

On Tue, 2010-07-20 at 23:24 +0800, janboe wrote:
> __backtrace is empty when ARM_UNWIND enable, so instead dump_stack
> 
> Signed-off-by: janboe <janboe.ye@gmail.com>
> ---
>  arch/arm/kernel/process.c |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
> index acf5e6f..e7ad878 100644
> --- a/arch/arm/kernel/process.c
> +++ b/arch/arm/kernel/process.c
> @@ -271,7 +271,11 @@ void show_regs(struct pt_regs * regs)
>  	printk("\n");
>  	printk("Pid: %d, comm: %20s\n", task_pid_nr(current), current->comm);
>  	__show_regs(regs);
> +#ifdef CONFIG_ARM_UNWIND
> +	dump_stack();
> +#else
>  	__backtrace();
> +#endif
>  }

It looks like both dump_stack() and __backtrace() (with FRAME_POINTER)
have pretty much the same semantics. Maybe we could define __backtrace()
to just call dump_stack() when ARM_UNWIND.

-- 
Catalin

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[ye janboe]

Yes. both are same.

Is it ok to remove backtrace.S if ARM_UNWIND is enable?

Catalin

Is it ok to merge this patch first and then try replace __bactrace in
next patch?

Janboe

2010/7/20 Catalin Marinas <catalin.marinas@arm.com>:
> On Tue, 2010-07-20 at 23:24 +0800, janboe wrote:
>> __backtrace is empty when ARM_UNWIND enable, so instead dump_stack
>>
>> Signed-off-by: janboe <janboe.ye@gmail.com>
>> ---
>> ?arch/arm/kernel/process.c | ? ?4 ++++
>> ?1 files changed, 4 insertions(+), 0 deletions(-)
>>
>> diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
>> index acf5e6f..e7ad878 100644
>> --- a/arch/arm/kernel/process.c
>> +++ b/arch/arm/kernel/process.c
>> @@ -271,7 +271,11 @@ void show_regs(struct pt_regs * regs)
>> ? ? ? printk("\n");
>> ? ? ? printk("Pid: %d, comm: %20s\n", task_pid_nr(current), current->comm);
>> ? ? ? __show_regs(regs);
>> +#ifdef CONFIG_ARM_UNWIND
>> + ? ? dump_stack();
>> +#else
>> ? ? ? __backtrace();
>> +#endif
>> ?}
>
> It looks like both dump_stack() and __backtrace() (with FRAME_POINTER)
> have pretty much the same semantics. Maybe we could define __backtrace()
> to just call dump_stack() when ARM_UNWIND.
>
> --
> Catalin
>
>

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[Catalin Marinas]

On Tue, 2010-07-20 at 16:54 +0100, ye janboe wrote:
> Yes. both are same.
> 
> Is it ok to remove backtrace.S if ARM_UNWIND is enable?

I think so, as long as you provide some macro or static inline function
for __backtrace and c_backtrace (if they are called at all).

-- 
Catalin

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[Janboe Ye]

Is there any way for arm unwind to do user space backtrace like c_backtrace in arm_syscall?

Thanks

On Wed, Jul 21, 2010 at 10:21:05AM +0100, Catalin Marinas wrote:
> On Tue, 2010-07-20 at 16:54 +0100, ye janboe wrote:
> > Yes. both are same.
> > 
> > Is it ok to remove backtrace.S if ARM_UNWIND is enable?
> 
> I think so, as long as you provide some macro or static inline function
> for __backtrace and c_backtrace (if they are called at all).
> 
> -- 
> Catalin
> 

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[Russell King - ARM Linux]

On Thu, Jul 22, 2010 at 01:24:28AM +0800, Janboe Ye wrote:
> Is there any way for arm unwind to do user space backtrace like
> c_backtrace in arm_syscall?

That's a job for userspace, which would have access to the program
debug information, which is necessary for unwinding EABI programs.
The debug information is not loaded into the programs address space
when you run it, and therefore would not be available to the kernel.

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[ye janboe]

hi, Russell

I agree with this.

But I am confused with c_bactrace in arm_syscall. This function is
used to dump stack when arm syscall failed.
regs->ARM_fp should be an userspace address and c_backtrace should
fail to dump the backtrace.

Do you know what the purpose c_backtrace here?

Thanks

Janboe


2010/7/22 Russell King - ARM Linux <linux@arm.linux.org.uk>:
> On Thu, Jul 22, 2010 at 01:24:28AM +0800, Janboe Ye wrote:
>> Is there any way for arm unwind to do user space backtrace like
>> c_backtrace in arm_syscall?
>
> That's a job for userspace, which would have access to the program
> debug information, which is necessary for unwinding EABI programs.
> The debug information is not loaded into the programs address space
> when you run it, and therefore would not be available to the kernel.
>

[PATCH] arm: show_regs dump stack when ARM_UNWIND enable

[Russell King - ARM Linux]

On Tue, Aug 10, 2010 at 11:27:54AM +0800, ye janboe wrote:
> hi, Russell
> 
> I agree with this.
> 
> But I am confused with c_bactrace in arm_syscall. This function is
> used to dump stack when arm syscall failed.
> regs->ARM_fp should be an userspace address and c_backtrace should
> fail to dump the backtrace.
> 
> Do you know what the purpose c_backtrace here?

Debugging kernel failures in earlier kernels when there wasn't such
stuff as EABI to worry about - from times when programs were built
with frame pointers.

That call should probably be removed now.