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

[Andriy Gelman]

On Mon, 16. Mar 10:03, 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

Please add to your commit message (or something similar depending on what you
tested): 
"Without this patch the s5p-mfc often hangs at the end of encoding 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 | 10 ++++++++++
>  libavcodec/v4l2_m2m_enc.c | 22 ++++++++++++++++++++++
>  3 files changed, 37 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;
> +    }
> +
>      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..4862a4c0e5 100644
> --- a/libavcodec/v4l2_m2m_dec.c
> +++ b/libavcodec/v4l2_m2m_dec.c
> @@ -123,6 +123,16 @@ 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");
> +        return ret;
> +    }
> +
>      return 0;
>  }
>  
> diff --git a/libavcodec/v4l2_m2m_enc.c b/libavcodec/v4l2_m2m_enc.c
> index 5b954f4435..3abdd47a4a 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;
> +    }
> +
> +    return 0;
> +}
> +
>  static int v4l2_prepare_encoder(V4L2m2mContext *s)
>  {
>      AVCodecContext *avctx = s->avctx;
> @@ -164,6 +182,10 @@ static int v4l2_prepare_encoder(V4L2m2mContext *s)
>      /**
>       * requirements
>       */

> +    ret = v4l2_subscribe_eos_event(s);
> +    if (ret)
> +        return ret;
> +

I don't have a venus board to test, but it looks that the encoder doesn't support
V4L2_EVENT_EOS. It ends up calling v4l2_ctrl_subscribe_event() [1], which 
returns EINVAL. 

So I would log the error message, but not fail. Same in the decoder init. 
Ideally, it would be good to also test on a venus board too.

[1] https://github.com/torvalds/linux/blob/master/drivers/media/v4l2-core/v4l2-ctrls.c#L4565 

>      ret = v4l2_check_b_frame_support(s);
>      if (ret)
>          return ret;
> -- 
> 2.25.1
> 

Thanks,
-- 
Andriy