Re: [linux-next][PATCH] usb: dwc3: omap: remove IRQ_NOAUTOEN used with shared irq

From: Vignesh R
Date: Tue Jul 04 2017 - 08:50:45 EST


Hi,

On 7/3/2017 11:58 PM, Grygorii Strashko wrote:
>
>
> On 07/02/2017 11:27 PM, Vignesh R wrote:
>> Hi,
>>
>> On Saturday 01 July 2017 01:04 AM, Strashko, Grygorii wrote:
>>>
>>>
>>> On 06/29/2017 01:03 AM, Tony Lindgren wrote:
>>>> * Vignesh R <vigneshr@xxxxxx> [170628 21:21]:
>>>>>
>>>>>
>>>>> On Thursday 29 June 2017 05:01 AM, Strashko, Grygorii wrote:
>>>>>> IRQ_NOAUTOEN can't be used with shared IRQs and Kernel now will
>>>>>> triggers
>>>>>> warning if it happns, since commit 04c848d39879 ("genirq: Warn when
>>>>>> IRQ_NOAUTOEN is used with shared interrupts"). And this is the
>>>>>> case for
>>>>>> OMAP DWC 3 driver.
>>>>>>
>>>>>> Hence, remove IRQ_NOAUTOEN flag and instead call disable_irq() before
>>>>>> disabling PM runtime in probe error path handling.
>>>>>
>>>>> Or, how about requesting the irq at the end of probe after extcon
>>>>> registration?
>>>>
>>>> Hmm yeah, what prevents the issue that we tried to fix with commit
>>>> 12a7f17fac5b ("usb: dwc3: omap: fix race of pm runtime with irq handler
>>>> in probe")?
>>>
>>> No, I think. There should be the disable_irq() call any way
>>> before pm_runtime_put_sync(dev);| pm_runtime_disable(dev); -
>>> once IRQ is requested driver can't know if IRQ handler is running now
>>> or not
>>> unless it will call disable_irq()/free_irq().
>>>
>>> But, I've also thought about moving request_irq() down (seems
>>> it can be moved even after of_platform_populate()).
>>> looks like it can be separate patch.
>>>
>>
>> commit 12a7f17fac5b says PM runtime can race with IRQ handler when
>> deferred probing happening due to extcon, therefore IRQ_NOAUTOEN was set
>> and disable_irq() call was added. If IRQ handler registration is moved
>> to a place after of_platform_populate(), then we can be sure that extcon
>> is registered and there wont be any deferred probe or race with irq
>> handler and PM runtime. Would disable_irq() still be needed then?
>>
>
> You might be correct, so feel free to send a patch.

Ok, will send a patch moving irq registration after of_platform_populate().