Re: [PATCH RFC v1 1/3] ARM: Add irq disabled version of soft_restart.

[Russ Dill <Russ.Dill@ti.com>]

On 02/25/2014 02:27 AM, Thomas Gleixner wrote:
> On Mon, 24 Feb 2014, Russ Dill wrote:
>> On 02/24/2014 03:13 PM, Sebastian Capella wrote:
>>> Quoting Russell King - ARM Linux (2014-02-22 02:26:17)
>>>> On Tue, Feb 18, 2014 at 05:52:07PM -0800, Sebastian Capella
>>>> wrote:
>>>>> From: Russ Dill <Russ.Dill@ti.com>
>>>>> 
>>>>> This adds the ability to run soft_restart with
>>>>> local_irq/fiq_disable already called. This is helpful for
>>>>> the hibernation code paths.
>>>> 
>>>> I'd rather keep this simple.  There's no problem with calling
>>>> soft_restart with interrupts already disabled.
>>>> local_irq_disable()/local_fiq_disable() there should be
>>>> harmless.
>>> 
>>> Hi Russell,
>>> 
>>> I'm observing a data abort loop when I replace this call:
>>> 
>>> In the local_irq_disable, it ends up calling
>>> trace_hardirqs_off (CONFIG_TRACE_IRQFLAGS_SUPPORT is enabled),
>>> which calls trace_hardirqs_off_caller which checks
>>> lockdep_recursion in the current task, but we've switched to a
>>> temporary stack with the call_with_stack, and get_current is
>>> returning NULL.  This triggers a data abort, which calls
>>> trace_hardirqs_off again and so on.
>>> 
>>> Do you have any suggestions here?
>>> 
>>> Thanks,
>>> 
>>> Sebastian
>>> 
>> 
>> So the alternative is to have a version of the call that calls a
>> special no trace version of
>> local_irq_disable()/local_fiq_disable(). Which would be
>> preferable? Having a noirq version of soft_restart seems much
>> simpler to me.
> 
> If you want escape the tracer and in that case you really want it 
> being on a different stack, use raw_local_irq_* which are not
> traced.

So it might make sense to change soft_restart to use the
raw_local_irq_* calls.