[FFmpeg-devel] [PATCH v3] avcodec/v4l2_m2m: handle the v4l2 eos event

[Andriy Gelman]

On Tue, 17. Mar 19:39, Ming Qian wrote:
> when the last frame of capture is dequeueed,
> driver may send this V4L2_EVENT_EOS event,
> if this event is received, then we can set the capture port done

s/dequeueed/dequeued

Also I'd change the last line to:
"If this event is received, then the capture buffers have been flushed and   
avcodec_receive_packet()/avcodec_receive_frame() can return AVERROR_EOF.
Otherwise, the draining continues until all the capture buffers have been dequeued or
VIDIOC_DQBUF ioctl returns an EPIPE error.

> 
> if the v4l2 m2m driver don't support V4L2_EVENT_EOS,
> just output some error message, not make it error.

Some devices may not support V4L2_EVENT_EOS. This is logged as a warning
message and not treated as a fatal error during initialization.

> 
> Without this patch imx8qm often hangs at the end of encoding/decoding when
> flushing the capture buffers
> 
> Signed-off-by: Ming Qian <ming.qian at nxp.com>
> ---
>  libavcodec/v4l2_context.c |  5 +++++
>  libavcodec/v4l2_m2m_dec.c |  9 +++++++++
>  libavcodec/v4l2_m2m_enc.c | 20 ++++++++++++++++++++
>  3 files changed, 34 insertions(+)
> 
> diff --git a/libavcodec/v4l2_context.c b/libavcodec/v4l2_context.c
> index 8110bbb555..c10862aa12 100644
> --- a/libavcodec/v4l2_context.c
> +++ b/libavcodec/v4l2_context.c
> @@ -171,6 +171,11 @@ static int v4l2_handle_event(V4L2Context *ctx)
>          return 0;
>      }
>  
> +    if (evt.type == V4L2_EVENT_EOS) {
> +        ctx->done = 1;
> +        return 0;
> +    }
> +

The comments above the function definition should be updated 
because the function would also handle end of stream event. 

>      if (evt.type != V4L2_EVENT_SOURCE_CHANGE)
>          return 0;
>  
> diff --git a/libavcodec/v4l2_m2m_dec.c b/libavcodec/v4l2_m2m_dec.c
> index d666edffe4..7a4d2a43cb 100644
> --- a/libavcodec/v4l2_m2m_dec.c
> +++ b/libavcodec/v4l2_m2m_dec.c
> @@ -123,6 +123,15 @@ static int v4l2_prepare_decoder(V4L2m2mContext *s)
>          }
>      }
>  

> +    memset(&sub, 0, sizeof(sub));
> +    sub.type = V4L2_EVENT_EOS;
> +    ret = ioctl(s->fd, VIDIOC_SUBSCRIBE_EVENT, &sub);
> +    if (ret < 0) {
> +        av_log(s->avctx, AV_LOG_ERROR,
> +                "the v4l2 driver does not support VIDIOC_SUBSCRIBE_EVENT\n"
> +                "you must provide an eos event to finish encode\n");
> +    }
> +

Since some decoders may work fine without this event, I'd suggest to change to AV_LOG_WARNING,
and set the message as: 
"the v4l2 driver does not support end of stream VIDIOC_SUBSCRIBE_EVENT\n"

>      return 0;
>  }
>  
> diff --git a/libavcodec/v4l2_m2m_enc.c b/libavcodec/v4l2_m2m_enc.c
> index 84de63ec9d..13bd9d29f4 100644
> --- a/libavcodec/v4l2_m2m_enc.c
> +++ b/libavcodec/v4l2_m2m_enc.c
> @@ -155,6 +155,24 @@ static int v4l2_check_b_frame_support(V4L2m2mContext *s)
>      return AVERROR_PATCHWELCOME;
>  }
>  
> +static int v4l2_subscribe_eos_event(V4L2m2mContext *s)
> +{
> +    struct v4l2_event_subscription sub;
> +    int ret;
> +

> +    memset(&sub, 0, sizeof(sub));
> +    sub.type = V4L2_EVENT_EOS;
> +    ret = ioctl(s->fd, VIDIOC_SUBSCRIBE_EVENT, &sub);
> +    if (ret < 0) {
> +        av_log(s->avctx, AV_LOG_ERROR,
> +                "the v4l2 driver does not support VIDIOC_SUBSCRIBE_EVENT\n"
> +                "you must provide an eos event to finish encode\n");
> +        return ret;
> +    }

Same here for loglevel and message.

> +
> +    return 0;
> +}
> +
>  static int v4l2_prepare_encoder(V4L2m2mContext *s)
>  {
>      AVCodecContext *avctx = s->avctx;
> @@ -164,6 +182,8 @@ static int v4l2_prepare_encoder(V4L2m2mContext *s)
>      /**
>       * requirements
>       */
> +    v4l2_subscribe_eos_event(s);
> +
>      ret = v4l2_check_b_frame_support(s);
>      if (ret)
>          return ret;
> -- 
> 2.17.1
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".

Thanks
-- 
Andriy