Re: [PATCH] Input: mpu3050: set IRQF_ONESHOT when requesting the interrupt

[Stephen Warren <swarren@wwwdotorg.org>]

On 05/11/2012 01:29 AM, Dmitry Torokhov wrote:
> On Wed, May 09, 2012 at 01:35:23PM -0600, Stephen Warren wrote:
>> On 05/08/2012 11:23 PM, Dmitry Torokhov wrote:
>>> Hi Stephen,
>>>
>>> On Tue, May 01, 2012 at 11:46:46AM -0600, Stephen Warren wrote:
>>>> From: Stephen Warren <swarren@nvidia.com>
>>>>
>>>> Commit 1c6c695 "genirq: Reject bogus threaded irq requests" requires
>>>> that request_threaded_irq() either be passed an explicit handler, or
>>>> that IRQF_ONESHOT be set. Set this flag.
>>>>
>>>> Signed-off-by: Stephen Warren <swarren@nvidia.com>
>>>
>>> Applied, thank you.
>>>
>>> BTW, since it appears you have the hardware any chance you could try the
>>> patch below? I had it in my queue for a while but Alan disappeared and
>>> I didn't have anyone to test it.
>>
>> I tested the patch with an interrupt supplied, and still see periodic
>> output from evbug that appears to correlate with when I move the board.
>>
>> I'm not sure how to test it in polled mode. What application would I use
>> to trigger the polling? evtest just seems to print out the initial
>> values, but not poll.
> 
> Then there is a problem with the patch; once someone (i.e. evtest) opens
> input device kernel should start polling at givein interval...
> 
> I think I know what might be wrong. We need to make sure we are calling
> runtime PM APIs instead of trying to wake up chip directly because
> parent might be sleeping as well.
> 
> Does the version below work any better?

Yes, but there's still one issue.

Case 1:
CONFIG_INPUT_MPU3050_POLLED_MODE=n
IRQ supplied to driver
Works

Case 2:
CONFIG_INPUT_MPU3050_POLLED_MODE=y
IRQ supplied to driver
Fails; no output from evtest

Case 2:
CONFIG_INPUT_MPU3050_POLLED_MODE=y
IRQ NOT supplied to driver
Works