Re: [PATCH v3] Add tw5864 driver

[Hans Verkuil <hverkuil@xs4all.nl>]

On 07/11/2016 01:48 PM, Andrey Utkin wrote:
> Thanks for review Hans!
> 
> On Mon, Jul 11, 2016 at 07:58:38AM +0200, Hans Verkuil wrote:
>>> +" v4l2-ctl --device $dev --set-ctrl=video_gop_size=1; done\n"
>>
>> Replace $dev by /dev/videoX
>>
>> Wouldn't it make more sense to default to this? And show the warning only if
>> P-frames are enabled?
> 
> I believe it's better to leave P-frames on by default. All-I-frames
> stream has huge bitrate. And the pixels artifacts is not very strong,
> it's 0 - 10 bad pixels on picture at same time in our dev environment,
> and probably up to 50 bad pixels max in other environments I know of.
> 
>>> +	dma_sync_single_for_cpu(&dev->pci->dev, cur_frame->vlc.dma_addr,
>>> +				H264_VLC_BUF_SIZE, DMA_FROM_DEVICE);
>>> +	dma_sync_single_for_cpu(&dev->pci->dev, cur_frame->mv.dma_addr,
>>> +				H264_MV_BUF_SIZE, DMA_FROM_DEVICE);
>>
>> This is almost certainly the wrong place. This should probably happen in the
>> tasklet. The tasklet runs after the isr, so by the time the tasklet runs
>> you've already called dma_sync_single_for_device.
> 
> Thanks, moved to tasklet subroutine tw5864_handle_frame().
> 
> I didn't seem to me like dma_sync_single_for_* can take long time or be
> otherwise bad to be done from interrupt context.
> 
>>> +static int tw5864_querycap(struct file *file, void *priv,
>>> +			   struct v4l2_capability *cap)
>>> +{
>>> +	struct tw5864_input *input = video_drvdata(file);
>>> +
>>> +	strcpy(cap->driver, "tw5864");
>>> +	snprintf(cap->card, sizeof(cap->card), "TW5864 Encoder %d",
>>> +		 input->nr);
>>> +	sprintf(cap->bus_info, "PCI:%s", pci_name(input->root->pci));
>>> +	cap->device_caps = V4L2_CAP_VIDEO_CAPTURE | V4L2_CAP_READWRITE |
>>> +		V4L2_CAP_STREAMING;
>>> +	cap->capabilities = cap->device_caps | V4L2_CAP_DEVICE_CAPS;
>>
>> This line can be dropped, the core will fill in the capabilities field for you.
> 
> No, removing this line causes v4l2-compliance failures and also ffmpeg fails to
> play the device.

My fault. I mixed things up. The struct video_device has a new device_caps field. That
should be filled in with the caps and then in the function above you can drop both
the device_caps and capabitlities assignments.

Sorry about that.

Regards,

	Hans

> 
> Required ioctls:
>                 fail: v4l2-compliance.cpp(550): dcaps & ~caps
>         test VIDIOC_QUERYCAP: FAIL
> 
> Allow for multiple opens:
>         test second video open: OK
>                 fail: v4l2-compliance.cpp(550): dcaps & ~caps
>         test VIDIOC_QUERYCAP: FAIL
>