Re: [PATCH] media: vb2: Fix regression on poll() for RW mode

From: Hans Verkuil
Date: Fri Apr 22 2016 - 10:31:39 EST


On 04/22/2016 04:21 PM, Mauro Carvalho Chehab wrote:
> Em Fri, 22 Apr 2016 14:37:07 +0200
> Hans Verkuil <hverkuil@xxxxxxxxx> escreveu:
>
>> On 04/22/2016 02:31 PM, Mauro Carvalho Chehab wrote:
>>> Em Fri, 22 Apr 2016 11:19:09 +0200
>>> Hans Verkuil <hverkuil@xxxxxxxxx> escreveu:
>>>
>>>> Hi Ricardo,
>>>>
>>>> On 04/21/2016 11:15 AM, Ricardo Ribalda Delgado wrote:
>>>>> When using a device is read/write mode, vb2 does not handle properly the
>>>>> first select/poll operation. It allways return POLLERR.
>>>>>
>>>>> The reason for this is that when this code has been refactored, some of
>>>>> the operations have changed their order, and now fileio emulator is not
>>>>> started by poll, due to a previous check.
>>>>>
>>>>> Reported-by: Dimitrios Katsaros <patcherwork@xxxxxxxxx>
>>>>> Cc: Junghak Sung <jh1009.sung@xxxxxxxxxxx>
>>>>> Cc: stable@xxxxxxxxxxxxxxx
>>>>> Fixes: 49d8ab9feaf2 ("media] media: videobuf2: Separate vb2_poll()")
>>>>> Signed-off-by: Ricardo Ribalda Delgado <ricardo.ribalda@xxxxxxxxx>
>>>>> ---
>>>>> drivers/media/v4l2-core/videobuf2-core.c | 8 ++++++++
>>>>> drivers/media/v4l2-core/videobuf2-v4l2.c | 8 --------
>>>>> 2 files changed, 8 insertions(+), 8 deletions(-)
>>>>>
>>>>> diff --git a/drivers/media/v4l2-core/videobuf2-core.c b/drivers/media/v4l2-core/videobuf2-core.c
>>>>> index 5d016f496e0e..199c65dbe330 100644
>>>>> --- a/drivers/media/v4l2-core/videobuf2-core.c
>>>>> +++ b/drivers/media/v4l2-core/videobuf2-core.c
>>>>> @@ -2298,6 +2298,14 @@ unsigned int vb2_core_poll(struct vb2_queue *q, struct file *file,
>>>>> return POLLERR;
>>>>>
>>>>> /*
>>>>> + * For compatibility with vb1: if QBUF hasn't been called yet, then
>>>>> + * return POLLERR as well. This only affects capture queues, output
>>>>> + * queues will always initialize waiting_for_buffers to false.
>>>>> + */
>>>>> + if (q->waiting_for_buffers && (req_events & (POLLIN | POLLRDNORM)))
>>>>> + return POLLERR;
>>>>
>>>> The problem I have with this is that this should be specific to V4L2. The only
>>>> reason we do this is that we had to stay backwards compatible with vb1.
>>>>
>>>> This is the reason this code was placed in videobuf2-v4l2.c. But you are correct
>>>> that this causes a regression, and I see no other choice but to put it in core.c.
>>>>
>>>> That said, I would still only honor this when called from v4l2, so I suggest that
>>>> a new flag 'check_waiting_for_buffers' is added that is only set in vb2_queue_init
>>>> in videobuf2-v4l2.c.
>>>>
>>>> So the test above becomes:
>>>>
>>>> if (q->check_waiting_for_buffers && q->waiting_for_buffers &&
>>>> (req_events & (POLLIN | POLLRDNORM)))
>>>>
>>>> It's not ideal, but at least this keeps this v4l2 specific.
>>>
>>> I don't like the above approach, for two reasons:
>>>
>>> 1) it is not obvious that this is V4L2 specific from the code;
>>
>> s/check_waiting_for_buffers/v4l2_needs_to_wait_for_buffers/
>
> Better, but still hell of a hack. Maybe we could add a quirks
> flag and add a flag like:
> VB2_FLAG_ENABLE_POLLERR_IF_WAITING_BUFFERS_AND_NO_QBUF
> (or some better naming, I'm not inspired today...)
>
> Of course, such quirk should be properly documented.

How about 'quirk_poll_must_check_waiting_for_buffers'? Something with 'quirk' in the
name is a good idea.

>
>>>
>>> 2) we should not mess the core due to some V4L2 mess.
>>
>> Well, the only other alternative I see is to split vb2_core_poll() into two
>> since the check has to happen in the middle. The v4l2 code would call core_poll1(),
>> then do the check and afterwards call core_poll2(). And that would really be ugly.
>
> Actually, the first callback would be better called as
> vb2_core_poll_check() - or something with similar name.
>
> IMHO, this is the cleaner solution, although it adds an extra cost.

I really don't like this. This has a much larger impact on vb2 core then adding
a simple quirk flag.

Regards,

Hans