Re: [PATCH v2] media: uvcvideo: Implement V4L2_EVENT_FRAME_SYNC

[Esker Wong]

[send again in text mode]
Hi Sakari,

Sequence number is important to us. We need it to measure the latency
from this event to the time we display the frame.

Regards,
Esker


On Mon, Nov 6, 2023 at 7:06 PM Sakari Ailus <sakari.ailus@xxxxxx> wrote:
>
> Hi Ricardo,
>
> On Mon, Nov 06, 2023 at 10:52:27AM +0000, Ricardo Ribalda wrote:
> > Add support for the frame_sync event, so user-space can become aware
> > earlier of new frames.
> >
> > Suggested-by: Esker Wong <esker@xxxxxxxxxxxx>
> > Tested-by: Esker Wong <esker@xxxxxxxxxxxx>
> > Signed-off-by: Ricardo Ribalda <ribalda@xxxxxxxxxxxx>
> > ---
> > We have measured a latency of around 30msecs between frame sync
> > and dqbuf.
> > ---
> > Changes in v2:
> > - Suggested by Laurent. Split sequence++ and event init.
> > - Link to v1: https://lore.kernel.org/r/20231020-uvc-event-v1-1-3baa0e9f6952@xxxxxxxxxxxx
> > ---
> >  drivers/media/usb/uvc/uvc_v4l2.c  | 2 ++
> >  drivers/media/usb/uvc/uvc_video.c | 7 +++++++
> >  2 files changed, 9 insertions(+)
> >
> > diff --git a/drivers/media/usb/uvc/uvc_v4l2.c b/drivers/media/usb/uvc/uvc_v4l2.c
> > index f4988f03640a..9f3fb5fd2375 100644
> > --- a/drivers/media/usb/uvc/uvc_v4l2.c
> > +++ b/drivers/media/usb/uvc/uvc_v4l2.c
> > @@ -1352,6 +1352,8 @@ static int uvc_ioctl_subscribe_event(struct v4l2_fh *fh,
> >       switch (sub->type) {
> >       case V4L2_EVENT_CTRL:
> >               return v4l2_event_subscribe(fh, sub, 0, &uvc_ctrl_sub_ev_ops);
> > +     case V4L2_EVENT_FRAME_SYNC:
> > +             return v4l2_event_subscribe(fh, sub, 0, NULL);
> >       default:
> >               return -EINVAL;
> >       }
> > diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
> > index 28dde08ec6c5..4f3a510ca4fe 100644
> > --- a/drivers/media/usb/uvc/uvc_video.c
> > +++ b/drivers/media/usb/uvc/uvc_video.c
> > @@ -1073,9 +1073,16 @@ static int uvc_video_decode_start(struct uvc_streaming *stream,
> >        * that discontinuous sequence numbers always indicate lost frames.
> >        */
> >       if (stream->last_fid != fid) {
> > +             struct v4l2_event event = {
> > +                     .type = V4L2_EVENT_FRAME_SYNC,
> > +             };
> > +
> >               stream->sequence++;
> >               if (stream->sequence)
> >                       uvc_video_stats_update(stream);
> > +
> > +             event.u.frame_sync.frame_sequence = stream->sequence,
> > +             v4l2_event_queue(&stream->vdev, &event);
>
> uvc_video_decode_start() is called when the reception of the entire frame
> has been completed. However, the documentation for V4L2_EVENT_FRAME_SYNC
> says that the event is "Triggered immediately when the reception of a frame
> has begun.". The functionality here doesn't seem to fit to this patch.
>
> Wouldn't V4L2_EVENT_VSYNC be a better fit, even if we don't really have a
> concept of vertical sync in the case of USB? That event doesn't have the
> sequence though but I guess it's not an issue at least if your case.
>
> Another technically correct option could be to create a new event for this
> but I'm not sure it's worth it.
>
> >       }
> >
> >       uvc_video_clock_decode(stream, buf, data, len);
> >
>
> --
> Regards,
>
> Sakari Ailus