[igt-dev] [PATCH i-g-t v4] tests/kms_async_flips: Add test to validate asynchronous flips

[Karthik B S]

On 8/7/2020 6:29 PM, Michel Dänzer wrote:
> On 2020-08-07 11:31 a.m., Karthik B S wrote:
>> On 8/6/2020 8:19 PM, Michel Dänzer wrote:
>>> On 2020-08-06 3:29 p.m., Karthik B S wrote:
>>>> Asynchronous flips are issued using the page flip IOCTL.
>>>> The test consists of two subtests. The first subtest waits for
>>>> the page flip event to be received before giving the next flip,
>>>> and the second subtest doesn't wait for page flip events.
>>>
>>> The latter sounds sketchy. It might get lucky most of the time, but if
>>> the previous flip is still pending, the ioctl will fail with EBUSY.
>>>
>>
>> Thanks for the review.
>> I will update the test to retry on getting EBUSY until 1 vblank time and
>> then fail if we're still getting EBUSY.
> 
> That could still fail with AMD GPUs, where the vblank interrupt can fire
> earlier than the flip completion interrupt.
> 
> I don't get the point of such a sub-test TBH. Userspace should always
> use the flip completion event to know when a flip has completed and
> another one can be submitted.
> 
This sub-test was added as the API could be called without requesting 
for events as well.

Do you suggest we document somewhere that this API cannot be used like 
that or is there a possibility of this API still being called from 
userspace without requesting for events?

Thanks,
Karthik.B.S
>