Re: [virtio-dev] Re: [RFC PATCH v6] virtio-video: Add virtio video device specification

[Cornelia Huck <cohuck@redhat.com>]

On Tue, Feb 07 2023, Alexandre Courbot <acourbot@chromium.org> wrote:

> On Mon, Feb 6, 2023 at 11:13 PM Cornelia Huck <cohuck@redhat.com> wrote:
>>
>> On Thu, Jan 19 2023, Alexander Gordeev <alexander.gordeev@opensynergy.com> wrote:
>>
>> > Hi Alexandre,
>> >
>> > On 12.01.23 07:39, Alexandre Courbot wrote:
>> >> On Thu, Jan 12, 2023 at 3:42 AM Alexander Gordeev
>> >> <alexander.gordeev@opensynergy.com> wrote:
>>
>> >>> Well, on the one hand mimicking v4l2 looks like an easy solution from
>> >>> virtio-video spec writing perspective. (But the implementers will have
>> >>> to read the V4L2 API instead AFAIU, which is probably longer...)
>> >> It should not necessarily be much longer as the parts we are
>> >> interested in have their own dedicated pages:
>> >>
>> >> https://docs.kernel.org/userspace-api/media/v4l/dev-decoder.html  <https://docs.kernel.org/userspace-api/media/v4l/dev-decoder.html>https://docs.kernel.org/userspace-api/media/v4l/dev-encoder.html  <https://docs.kernel.org/userspace-api/media/v4l/dev-encoder.html>
>> >>
>> >> Besides, the decoding and encoding processes are described with more
>> >> precision, not that we couldn't do that here but it would make the
>> >> spec grow longer than I am comfortable with...
>> >
>> > I read the references carefully, thanks. I am somewhat familiar with the
>> > stateful decoder API, but the stateless one still needs exploring.
>> >
>> > There is one serious issue with these references IMO: they represent
>> > guest user-space <-> v4l2 subsystem API, not v4l2 subsystem <->
>> > virtio-video driver API. Just to make sure we're on the same page:
>> >
>> > guest user-space <-> v4l2 kernel subsystem <-> virtio-video driver <-
>> > virtio-video protocol -> virtio-video device.
>> >
>> > I believe this is how it is supposed to work, right? So I thought, that
>> > your intention is to simplify virtio-video driver and virtio-video
>> > protocol by reusing the v4l2 subsystem <-> v4l2 driver API. But having
>> > these references I can assume, that you want to use user-space <-> v4l2
>> > subsystem API, right? Well, I think this cannot happen and therefore
>> > these references cannot be used directly unless:
>> >
>> > 1. You suggest that virtio-video driver should not use v4l2 subsystem,
>> > but should mimic its user-space API in every detail. Probably not a good
>> > idea.
>> >
>> > 2. There is already a way to bypass the subsystem completely. I'm not
>> > aware of that.
>> >
>> > 3. user-space <-> v4l2 subsystem API is already the same or very close
>> > to v4l2 subsystem <-> v4l2 driver API. I believe this is not the case
>> > even with stateful decoder/encoder. Even more with stateless decoders
>> > because I can see, that v4l2 subsystem actually stores some state in
>> > this case as well. Which is quite reasonable I think.
>> >
>> > So I think what we need to reference here is v4l2 subsystem <-> v4l2
>> > driver API. Do you have this reference? Well, I know there is some
>> > documentation, but still I doubt that. AFAIR kernel internal APIs are
>> > never fixed. Right?
>>
>> So, I'm not that familiar with v4l2, but if that's indeed the case,
>> depending on some kernel internal APIs is a no-go. First, because
>> in-kernel APIs are not stable, and second, because we want something
>> that's BSD-licenced (as opposed to GPLv2-licenced) to point to. The
>> kernel<->userspace API would work (BSD-licenced header and stable); I
>> had the impression that we wanted to reuse the various #defines in
>> there -- did I misunderstand?
>
> Sorry, I should have replied earlier to lift any misunderstanding. I
> am not suggesting to use any kernel internal API as reference. My
> suggestion is to stick strictly to the UAPI which is stable (as in,
> guaranteed to be backward-compatible) and well documented. Here is for
> instance the part documenting buffer queuing/dequeuing:
> https://www.kernel.org/doc/html/v4.9/media/uapi/v4l/vidioc-qbuf.html
>
> Changing anything in this documentation in a backward-incompatible way
> would break user-space, and everyone familiar with the kernel
> community knows what happens when someone breaks user-space. ;) So
> this can be a reliable source for virtio-video (and if you look
> closely, you will also notice many similarities between the two).

Yes, that sounds good to me :)

>
>>
>> (...)
>>
>> >> Let me try to summarize the case for using V4L2 over Virtio (I'll call
>> >> it virtio-v4l2 to differentiate it from the current spec).
>> >>
>> >> There is the argument that virtio-video turns out to be a recreation
>> >> of the stateful V4L2 decoder API, which itself works similarly to
>> >> other high-level decoder APIs. So it's not like we could or should
>> >> come with something very different. In parallel, virtio-camera is also
>> >> currently using V4L2 as its model. While this is subject to change, I
>> >> am starting to see a pattern here. :)
>> >>
>> >> Transporting V4L2 over virtio would considerably shorten the length of
>> >> this spec, as we would just need to care about the transport aspect
>> >> and minor amendments to the meaning of some V4L2 structure members,
>> >> and leave the rest to V4L2 which is properly documented and for which
>> >> there is a large collection of working examples.
>> >>
>> >> This would work very well for codec devices, but as a side-effect
>> >> would also enable other kinds of devices that may be useful to
>> >> virtualize, like image processors, DVB cards, and cameras. This
>> >> doesn't mean virtio-v4l2 should be the *only* way to support cameras
>> >> over virtio. It is a nice bonus of encapsulating V4L2, it may be
>> >> sufficient for simple (most?) use-cases, but also doesn't forbid more
>> >> specialized virtual devices for complex camera pipelines to be added
>> >> later. virtio-v4l2 would just be the generic virtual video device that
>> >> happens to be sufficient for our accelerated video needs - and if your
>> >> host camera is a USB UVC one, well feel free to use that too.
>> >>
>> >> In other words, I see an opportunity to enable a whole class of
>> >> devices instead of a single type for the same effort and think we
>> >> should seriously consider this.
>> >>
>> >> I have started to put down what a virtio-v4l2 transport might look
>> >> like, and am also planning on putting together a small
>> >> proof-of-concept. If I can get folks here to warm up to the idea, I
>> >> believe we should be able to share a spec and prototype in a month or
>> >> so.
>> >
>> > Thanks for the detailed explanation. Please check my comments above. I'd
>> > like to resolve the mentioned issue first.
>>
>> I hope we can sort this out soon -- I guess I'm not the only one who is
>> anxious about this spec moving forward :) Please let me know if I can
>> help in any way.
>
> I'll try to address Alexander's points in more detail, but I am not
> seeing any blocking issue with using the V4L2 UAPI as the basis for
> virtio-video (we are working on a small proof-of-concept and things
> are going smoothly so far).

Great to hear, looking forward to it!