Re: virtio-v4l2 specification draft

[Albert Esteve <aesteve@redhat.com>]

Hi Alex,

SorryÂfor the late reply. I think this is a nice solution, specially for the host-guest

memory management part. I was intrigued with how would you solve this, as

the v4l2 struct memory fields can't be directly used.

In that sense, sending a field that is going to be ignored by both driver and device,

feels like kindof a waste. But I guess there is not a good solution to that.

Have you considered avoiding using struct v4l2_buffer for QBUF and DQBUF ioctls,

and having virtio-v4l2 specific struct for them? The device would have the burden to

copy the all the fields and leave the `m` field out, so it may have its own downsides.

In any case, the specs are short and clear toÂfollow. Great work!

I just have a couple comments/questions:

- The text in `0.1.6.3` and `0.1.6.3.1` is saying the same thing twice (?). Maybe would be

 clearer to unify?

- If I understood correctly, the `stream_id` is assigned by the device after receiving

 aÂVIRTIO_V4L2_CMD_OPEN? Is itÂsupposedÂto be a correlative natural number (1, 2, etc.)?

 I assume that after (successfully) closing the stream, its stream_id can be reused?

- For section `0.1.6.7` I think it would be good to refer (again) to the section `0.1.6.5` so that

 we can quickly navigate to the section that explains the memory fields in the presented events.

Best regards,

Albert

On Wed, May 31, 2023 at 8:08âAM Michael S. Tsirkin <mst@redhat.com> wrote:

On Wed, May 31, 2023 at 03:02:04PM +0900, Alexandre Courbot wrote:
> Hello everyone,
>
> With the virtio-video call taking place soon, I thought it would help
> everyone understand both proposals if I sent a more formal
> specification of what virtio-v4l2 looks like. Please find it attached
> to this email.
>
> I apologize for not finishing and sending this earlier, but hopefully
> at 7 and a half pages it should be rather quick to skim through. :)
>
> Despite its short size, this spec is capable of supporting camera and
> image processor devices (on top of decoder/encoders), and also allows
> for another kind of memory backing for buffers (host-managed) that is
> not supported by virtio-video.
>
> For convenience I have made use of clickable links to the relevant
> parts of the V4L2 documentation, since it is supposed to be used
> alongside this spec.
>
> Looking forward to a fruitful discussion tomorrow!
>
> Cheers,
> Alex.


Can you please post patch with tex source inline as opposed as
pdf as an attachment? Thanks!

--
MST