[FFmpeg-devel] [EXT] [PATCH 2/2] avcodec/v4l2_context: Log warning when all capture buffers are in userspace

[Andriy Gelman]

On Mon, 27. Apr 16:28, Andriy Gelman wrote:
> Hi Ming,
> 
> Thanks for looking over the patch.
> 
> On Thu, 09. Apr 07:13, Ming Qian wrote:
> > Lgtm,
> > 
> > But if don't prevent enqueue frame buffer of capture port, unlikely to happen this case.
> 
> The problem happens when all the capture packets/frames are buffered internally
> by ffmpeg (i.e. the buffer state is V4L2BUF_RET_USER). 
> 
> I see it on the s5p-mfc even with your patch: 
> https://patchwork.ffmpeg.org/project/ffmpeg/patch/20200316020208.27547-1-ming.qian@nxp.com/
> 
> > And I think we can get a proper num_frame_buffers by g_ctrl V4L2_CID_MIN_BUFFERS_FOR_CAPTURE and V4L2_CID_MIN_BUFFERS_FOR_OUTPUT.
> > 
> 
> The values for V4L2_CID_MIN_BUFFERS_FOR_CAPTURE and
> V4L2_CID_MIN_BUFFERS_FOR_OUTPUT are the minimum values for the hardware device
> and are lower than our defaults. I don't think we should use the minimum values
> because the frames/packets may be buffered.
> 
> There is another option to request additional buffers via 
> CREATE_BUFS ioctl, but it's not supported by all hardware (thanks to ndufresne
> on #v4l for suggesting). I feel adding this ioctl should be a separate patch and we
> still need a warning if it's not supported.
> 
> I'll apply this patch on Wednesday if no one objects. 
> 

Applied.

Thanks,
-- 
Andriy