Prev: [PATCH] timer_list: remove alignment padding on 64 bit when CONFIG_TIMER_STATS
Next: Why is next changing constantly?
From: Michal Simek on 26 Jul 2010 05:00
Andrew Morton wrote:
> On Wed, 26 May 2010 10:42:43 -0400 Steven Rostedt <rostedt(a)goodmis.org> wrote:
>
>> On Wed, 2010-05-26 at 10:48 +0200, monstr(a)monstr.eu wrote:
>>> From: Michal Simek <monstr(a)monstr.eu>
>>>
>>> start/stop_critical_timing function for preemptirqsoff, preemptoff
>>> and irqsoff tracers contains atomic_inc and atomic_dec operations.
>>>
>>> Atomic operations used local_irq_save/restore macros to ensure
>>> atomic access but they are traced by the same function which is causing
>>> recursion problem.
>>>
>>> The reason is when these tracers are turn ON then local_irq_save/restore
>>> macros are changed in include/linux/irqflags.h to call trace_hardirqs_on/off
>>> which call start/stop_critical_timing.
>>>
>>> Microblaze was affected because use generic atomic implementation.
>>>
>>> Signed-off-by: Michal Simek <monstr(a)monstr.eu>
>> Acked-by: Steven Rostedt <rostedt(a)goodmis.org>
>>
>
> Sighed-at-by: me.
>
>>> diff --git a/include/asm-generic/atomic.h b/include/asm-generic/atomic.h
>>> index 058129e..6c190fd 100644
>>> --- a/include/asm-generic/atomic.h
>>> +++ b/include/asm-generic/atomic.h
>>> @@ -57,11 +57,11 @@ static inline int atomic_add_return(int i, atomic_t *v)
>>> unsigned long flags;
>>> int temp;
>>>
>>> - local_irq_save(flags);
>>> + raw_local_irq_save(flags);
>>> temp = v->counter;
>>> temp += i;
>>> v->counter = temp;
>>> - local_irq_restore(flags);
>>> + raw_local_irq_restore(flags);
>>>
>>> return temp;
>>> }
>
> If a developer looks at atomic_add_return() and asks himself "why did
> this use raw_local_irq_save()", the only way of answering that question
> is to go groveling through the git logs, which is a right PITA if
> you're trying to get some coding work done.
>
> Guys, any time you add code which is non-obvious at the raw C level, it
> *needs* a comment!

Andrew: Can you please add this patch to your tree? I have it in
microblaze next branch but should go through different tree.

Thanks,
Michal

--
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian
From: Michal Simek on 28 Jul 2010 01:30
Andrew Morton wrote:
> On Mon, 26 Jul 2010 10:49:50 +0200
> Michal Simek <monstr(a)monstr.eu> wrote:
>
>> start/stop_critical_timing function for preemptirqsoff, preemptoff
>> and irqsoff tracers contains atomic_inc and atomic_dec operations.
>>
>> Atomic operations used local_irq_save/restore macros to ensure
>> atomic access but they are traced by the same function which is causing
>> recursion problem.
>>
>> The reason is when these tracers are turn ON then local_irq_save/restore
>> macros are changed in include/linux/irqflags.h to call trace_hardirqs_on/off
>> which call start/stop_critical_timing.
>>
>> Microblaze was affected because use generic atomic implementation.
>
> Seems that this will also affect blackfin, mn10300 and score. I guess
> they aren't supporting tracing yet?

If they uses asm-generic/atomic.h, then yes. It seems to me that
Blackfin uses own asm/atomic.h. nm10300 and score are the same case as
Microblaze.

In include/linux/irqflags.h is written this commentary.

/*
* The local_irq_*() APIs are equal to the raw_local_irq*()
* if !TRACE_IRQFLAGS.
*/

If architecture doesn't enable TRACE_IRQFLAGS then there is no
difference in behavior. If yes and use asm-generic/atomic.h code, then
IRQFLAG tracer freeze because it is traced part of irqsoff tracer
because of recursion as is describe in patch description.

>
>> Signed-off-by: Michal Simek <monstr(a)monstr.eu>
>> Acked-by: Steven Rostedt <rostedt(a)goodmis.org>
>> Signed-off-by: Andrew Morton <akpm(a)linux-foundation.org>
>
> hm, I wonder how my signoff got there. Doesn't matter.

http://lkml.org/lkml/2010/5/26/364

Michal


--
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Michal Simek on 28 Jul 2010 01:30
Mike Frysinger wrote:
> On Tue, Jul 27, 2010 at 19:18, Andrew Morton wrote:
>> On Mon, 26 Jul 2010 10:49:50 +0200 Michal Simek wrote:
>>> start/stop_critical_timing function for preemptirqsoff, preemptoff
>>> and irqsoff tracers contains atomic_inc and atomic_dec operations.
>>>
>>> Atomic operations used local_irq_save/restore macros to ensure
>>> atomic access but they are traced by the same function which is causing
>>> recursion problem.
>>>
>>> The reason is when these tracers are turn ON then local_irq_save/restore
>>> macros are changed in include/linux/irqflags.h to call trace_hardirqs_on/off
>>> which call start/stop_critical_timing.
>>>
>>> Microblaze was affected because use generic atomic implementation.
>> Seems that this will also affect blackfin, mn10300 and score. I guess
>> they aren't supporting tracing yet?
>
> do you mean TRACE_IRQFLAGS_SUPPORT ? Blackfin should ... my
> understanding was that arches just needed to implement asm/irqflags.h
> for it.

Blackin uses own include/asm/atomic.h implementation where shouldn't be
a problem with irqflags tracer. Look at my second post.

Michal


--
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [PATCH] timer_list: remove alignment padding on 64 bit when CONFIG_TIMER_STATS
Next: Why is next changing constantly?