Re: [PATCH] x86_fixup_irqs: Fix possible missing interrupt handlewhen disabling CPU

[Srivatsa S. Bhat]

[ Adding a few more CCs ]

On 03/28/2012 11:08 AM, Yanmin Zhang wrote:

> On Mon, 2012-03-26 at 07:11 +0000, Liu, Chuansheng wrote:
>> From: liu chuansheng <chuansheng.liu@xxxxxxxxx>
>> Subject: [PATCH] x86_fixup_irqs: Fix possible missing interrupt handle when disabling CPU
>>
>> When preparing to unmask the irq in fixup_irqs(), using irqd_irq_masked()
>> as the condition to determine if do real unmasking action or not.
>>
>> Because in some chips, the .irq_disable is NULL, so calling disable_irq()
>> does not mean the irq is masked immediately, and before enable_irq() if
>> the irq is coming, it can be pending state. Using irqd_irq_disabled() will
>> lose this chance.
>>
>> Signed-off-by: liu chuansheng <chuansheng.liu@xxxxxxxxx>
>> ---
>>  arch/x86/kernel/irq.c |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c
>> index 6c0802e..b9dd37f 100644
>> --- a/arch/x86/kernel/irq.c
>> +++ b/arch/x86/kernel/irq.c
>> @@ -277,7 +277,7 @@ void fixup_irqs(void)
>>                         set_affinity = 0;
>>  
>>                 if (!irqd_can_move_in_process_context(data) &&
>> -                   !irqd_irq_disabled(data) && chip->irq_unmask)
>> +                   !irqd_irq_masked(data) && chip->irq_unmask)
>>                         chip->irq_unmask(data);
>>  
>>                 raw_spin_unlock(&desc->lock);
> Thomas,
> 
> Originally, we found an issue when enabling Android (Gingerbread) with kernel
> 2.6.35 on Medfield.
> 
> When system goes to suspend-2-ram, some device drivers call disable_irq to
> disable some device irq at device driver suspend callbacks. In addition,
> some specific irq_chips just provide mask/unmask callbacks and have no
> enable/disable callbacks. When drivers call
> disable_irq=>...=>irq_disable=>desc->irq_data.chip->irq_disable, they don't
> really mask the irq at the irq_chip. With such specific irq_chip, we could
> keep the irq as wake up sources. When the irq arrives after driver disables them,
> IRQ handling framework would mark it PENDING and calls the irq handler when
> drivers call enable_irq.
> 
> After device driver's suspend/suspend_noirq callbacks are called, suspend-2-ram
> would disable nonboot cpus, which calls fixup_irqs. fixup_irqs really masks the irq
> at irq_chip as the drivers calls disable_irq before. So one of the side effects of
> calling fixup_irq is to really mask the irq and we lose the wakeup capability of
> the corresponding irq.
> 
> You might ask why drivers call disable_irq at suspend callback. That's because
> another specific reason. suspend callback calls disable_irq and resume callback calls
> enable_irq. Between suspend and resume callbacks, we don't want kernel to deliver
> the irq to the driver irq handler as the device is not FULLY resumed yet.
> 
> Chuangsheng's patch tries to fix above issue, to keep the difference between
> enable/disable and mask/unmask callbacks of irq_chip.
> 
> Hope my explanation could be a little clearer.
> 
> Sorry for taking you too much time.
> 
> Yanmin
> 
> 



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/