Re: [virtio-dev] Re: [PATCH v2 0/1] VirtIO video device specification

[Tomasz Figa <tfiga@chromium.org>]

On Sat, Jan 11, 2020 at 12:12 AM Dmitry Sepp
<dmitry.sepp@opensynergy.com> wrote:
>
> Hi Keiichi,
>
> On Freitag, 10. Januar 2020 14:53:01 CET Keiichi Watanabe wrote:
> > Hi,
> >
> > On Fri, Jan 10, 2020 at 7:16 PM Dmitry Sepp <dmitry.sepp@opensynergy.com>
> wrote:
> > > Hi,
> > >
> > > On Donnerstag, 9. Januar 2020 15:56:08 CET Dmitry Sepp wrote:
> > > > Hi,
> > > >
> > > > On Dienstag, 7. Januar 2020 11:25:56 CET Keiichi Watanabe wrote:
> > > > > Hi Dmitry,
> > > > >
> > > > > On Mon, Jan 6, 2020 at 8:28 PM Dmitry Sepp
> > > > > <dmitry.sepp@opensynergy.com>
> > > >
> > > > wrote:
> > > > > > Hi,
> > > > > >
> > > > > > On Montag, 6. Januar 2020 11:30:22 CET Keiichi Watanabe wrote:
> > > > > > > Hi Dmitry, Tomasz,
> > > > > > >
> > > > > > > On Fri, Jan 3, 2020 at 10:05 PM Dmitry Sepp
> > > > > > > <dmitry.sepp@opensynergy.com>
> > > > > >
> > > > > > wrote:
> > > > > > > > Hi Tomasz, Keiichi,
> > > > > > > >
> > > > > > > > On Samstag, 21. Dezember 2019 07:19:23 CET Tomasz Figa wrote:
> > > > > > > > > On Sat, Dec 21, 2019 at 3:18 PM Tomasz Figa
> > > > > > > > > <tfiga@chromium.org>
> > > >
> > > > wrote:
> > > > > > > > > > On Sat, Dec 21, 2019 at 1:36 PM Keiichi Watanabe
> > > > > > > > > > <keiichiw@chromium.org>
> > > > > > > >
> > > > > > > > wrote:
> > > > > > > > > > > Hi Dmitry,
> > > > > > > > > > >
> > > > > > > > > > > On Sat, Dec 21, 2019 at 12:59 AM Dmitry Sepp
> > > > > > > > > > >
> > > > > > > > > > > <dmitry.sepp@opensynergy.com> wrote:
> > > > > > > > > > > > Hi Keiichi,
> > > > > > > > > > > >
> > > > > > > > > > > > On Mittwoch, 18. Dezember 2019 14:02:13 CET Keiichi
> > > > > > > > > > > > Watanabe
> > > > > >
> > > > > > wrote:
> > > > > > > > > > > > > Hi,
> > > > > > > > > > > > > This is the 2nd version of virtio-video patch. The PDF
> > > > > > > > > > > > > is
> > > > > > > > > > > > > available
> > > > > > > > > > > > > in [1].
> > > > > > > > > > > > > The first version was sent at [2].
> > > > > > > > > > > > >
> > > > > > > > > > > > > Any feedback would be appreciated. Thank you.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Best,
> > > > > > > > > > > > > Keiichi
> > > > > > > > > > > > >
> > > > > > > > > > > > > [1]:
> > > > > > > > > > > > > https://drive.google.com/drive/folders/1eT5fEckBoor2iH
> > > > > > > > > > > > > ZR4f
> > > > > > > > > > > > > 4G
> > > > > > > > > > > > > LxYz
> > > > > > > > > > > > > FMVa
> > > > > > > > > > > > > pOFx?us
> > > > > > > > > > > > > p=sharing [2]:
> > > > > > > > > > > > > https://markmail.org/message/gc6h25acct22niut
> > > > > > > > > > > > >
> > > > > > > > > > > > > Change log:
> > > > > > > > > > > > >
> > > > > > > > > > > > > v2:
> > > > > > > > > > > > > * Removed functionalities except encoding and
> > > > > > > > > > > > > decoding.
> > > > > > > > > > > > > * Splited encoder and decoder into different devices
> > > > > > > > > > > > > that
> > > > > > > > > > > > > use
> > > > > > > > > > > > > the
> > > > > > > > > > > > > same
> > > > > > > > > > > > > protocol. * Replaced GET_FUNCS with GET_CAPABILITY.
> > > > > > > > > > > > > * Updated structs for capabilities.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - Defined new structs and enums such as image
> > > > > > > > > > > > >   formats,
> > > > > > > > > > > > >   profiles,
> > > > > > > > > > > > >   range
> > > > > > > > > > > > >
> > > > > > > > > > > > > (min, max, step), etc
> > > > > > > > > > > > >
> > > > > > > > > > > > >     * For virtio_video_pixel_format, chose a naming
> > > > > > > > > > > > >     convention
> > > > > > > > > > > > >     that
> > > > > > > > > > > > >     is used
> > > > > > > > > > > > >
> > > > > > > > > > > > >       in DRM. We removed XBGR, NV21 and I422, as they
> > > > > > > > > > > > >       are
> > > > > > > > > > > > >       not
> > > > > > > > > > > > >       used
> > > > > > > > > > > > >       in the
> > > > > > > > > > > > >       current draft implementation.
> > > > > > > > > > > > >       https://lwn.net/Articles/806416/
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - Removed virtio_video_control, whose usage was not
> > > > > > > > > > > > >   documented
> > > > > > > > > > > > >   yet
> > > > > > > > > > > > >   and
> > > > > > > > > > > > >
> > > > > > > > > > > > > which is not necessary for the simplest decoding
> > > > > > > > > > > > > scenario.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - Removed virtio_video_desc, as it is no longer
> > > > > > > > > > > > >   needed.
> > > > > > > > > > > > >
> > > > > > > > > > > > > * Updated struct virtio_video_config for changes
> > > > > > > > > > > > > around
> > > > > > > > > > > > > capabilities.
> > > > > > > > > > > > > * Added a way to represent supported combinations of
> > > > > > > > > > > > > formats.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - A field "mask" in virtio_video_format_desc plays
> > > > > > > > > > > > >   this
> > > > > > > > > > > > >   role.
> > > > > > > > > > > > >
> > > > > > > > > > > > > * Removed VIRTIO_VIDEO_T_STREAM_{START,STOP} because
> > > > > > > > > > > > > they
> > > > > > > > > > > > > don't
> > > > > > > > > > > > > play
> > > > > > > > > > > > > any
> > > > > > > > > > > > > meaningful roles. * Removed
> > > > > > > > > > > > > VIRTIO_VIDEO_T_STREAM_{ATTACH,
> > > > > > > > > > > > > DETACH}_BACKING
> > > > > > > > > > > > > and merged them into RESOURCE_{CREATE, DESTROY}. *
> > > > > > > > > > > > > Added a
> > > > > > > > > > > > > way
> > > > > > > > > > > > > to
> > > > > > > > > > > > > notify/specify resource creation method.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - Added a feature flag.
> > > > > > > > > > > > >   - Defined enum virtio_video_mem_type.
> > > > > > > > > > > > >   - Added new fields in video_stream_create.
> > > > > > > > > > > > >
> > > > > > > > > > > > > * Modified fields in virtio_video_params.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - Added crop information.
> > > > > > > > > > > > >
> > > > > > > > > > > > > * Removed enum virtio_video_channel_type because we
> > > > > > > > > > > > > can
> > > > > > > > > > > > > get
> > > > > > > > > > > > > this
> > > > > > > > > > > > > information by image format.
> > > > > > > > > > > >
> > > > > > > > > > > > Could you please explain this? How do you get the
> > > > > > > > > > > > information?
> > > > > > > > > > >
> > > > > > > > > > > It means that if image formats are well-defined, channel
> > > > > > > > > > > information
> > > > > > > > > > > (e.g. the order of channels) is uniquely determined.
> > > > > > > > > > >
> > > > > > > > > > > > Suppose you have some piece of HW on the host side that
> > > > > > > > > > > > wants
> > > > > > > > > > > > I420
> > > > > > > > > > > > as
> > > > > > > > > > > > one
> > > > > > > > > > > > contig buffer w/ some offsets. But on the driver side,
> > > > > > > > > > > > say,
> > > > > > > > > > > > gralloc
> > > > > > > > > > > > gives you three separate buffers, one per channel. How
> > > > > > > > > > > > do we
> > > > > > > > > > > > pass
> > > > > > > > > > > > those to the device then?
> > > > > > > > > > >
> > > > > > > > > > > You're talking about CrOS use case where buffers are
> > > > > > > > > > > allocated
> > > > > > > > > > > by
> > > > > > > > > > > virtio-gpu, right?
> > > > > > > > > > > In this case, virtio-gpu allocates one contiguous
> > > > > > > > > > > host-side
> > > > > > > > > > > buffer
> > > > > > > > > > > and
> > > > > > > > > > > the client regards a pair of (buffer FD, offset) as one
> > > > > > > > > > > channel.
> > > > > > > > > > > And, we can register this pair to the device when the
> > > > > > > > > > > buffer
> > > > > > > > > > > is
> > > > > > > > > > > imported.
> > > > > > > > > > > In the virtio-vdec spec draft, this pair corresponds to
> > > > > > > > > > > struct
> > > > > > > > > > > virtio_vdec_plane in struct virtio_vdec_plane.
> > > > > > > > > > >
> > > > > > > > > > > So, I suppose we will need similar structs when we add a
> > > > > > > > > > > control
> > > > > > > > > > > to
> > > > > > > > > > > import buffers. However, I don't think it's necessary when
> > > > > > > > > > > guest
> > > > > > > > > > > pages
> > > > > > > > > > > are used.
> > > > > > > > > >
> > > > > > > > > > I think we need some way for the guest to know whether it
> > > > > > > > > > can
> > > > > > > > > > allocate
> > > > > > > > > > the planes in separate buffers, even when guest pages are
> > > > > > > > > > used.
> > > > > > > > > > This
> > > > > > > > > > would be equivalent to V4L2 M and non-M formats, but mixing
> > > > > > > > > > this
> > > > > > > > > > into
> > > > > > > > > > FourCC in V4L2 is an acknowledged mistake, so we should add
> > > > > > > > > > a
> > > > > > > > > > query or
> > > > > > > > > > something.
> > > > > > > >
> > > > > > > > Yes, this is what I mean. In fact, we already do face the
> > > > > > > > situation
> > > > > > > > when
> > > > > > > > the device side is not happy with the sgt and wants contig. I
> > > > > > > > think
> > > > > > > > we'll
> > > > > > > > add a module parameter for now.
> > > > > > >
> > > > > > > Okay. So, I suppose we'll be able to update structs:
> > > > > > > * Add a flag in virtio_video_format_desc that indicates whether
> > > > > > > planes
> > > > > > > can be in separate buffers, and
> > > > > > > * Add a flag in virtio_video_format_desc that indicates that the
> > > > > > > device requires contiguous buffers for this format.
> > > > > > >
> > > > > > > Does it make sense?
> > > > > >
> > > > > > Sorry, I don't understand the difference between the two above:
> > > > > > isn't
> > > > > > the
> > > > > > first case is just when the flag is not set?
> > > > >
> > > > > Ah, I was confused and wrote something strange. Yeah,  these two are
> > > > > the
> > > > > same. Sorry for that.
> > > > >
> > > > > So, the suggestion is to add a field "planes_layout" in
> > > > > virtio_video_format_desc, which is one of the following enums:
> > > > >
> > > > > enum virtio_video_planes_layout {
> > > > >
> > > > >     VIRTIO_VIDEO_PLANES_LAYOUT_UNSPEC = 0,  /* no special requirement
> > > > >     */
> > > > >     VIRTIO_VIDEO_PLANES_LAYOUT_CONTIGUOUS,
> > > > >
> > > > > };
> > > > >
> > > > > If we have a better idea or naming, please let me know.
> > > >
> > > > The naming looks good for me, I might only change to CONTIG as we have
> > > > UNSPEC.
> > >
> > > So here we are talking about plane layout in memory, am I correct? But I
> > > think we also need a way to communicate memory requirements of the
> > > device: the device might require CMA buffers or it can be ok with SG
> > > lists. What about adding something like this to virtio_video_format_desc:
> > >
> > > enum virtio_video_mem_layout {
> > >
> > >         VIRTIO_VIDEO_MEM_LAYOUT_UNDEFINED = 0,
> > >
> > >         VIRTIO_VIDEO_MEM_LAYOUT_CONTIG = 0x100,
> > >         VIRTIO_VIDEO_MEM_LAYOUT_NON_CONTIG,
> > >
> > > };
> > >
> > > struct virtio_video_format_desc {
> > >
> > >     __le64 mask;
> > >     __le32 format; /* One of VIRTIO_VIDEO_FORMAT_* types */
> > >     __le32 planes_layout; /* One of VIRTIO_VIDEO_PLANES_LAYOUT_* types */
> > >     __le32 mem_layout; /* One of VIRTIO_VIDEO_MEM_LAYOUT_* types */
> > >     ...
> > >
> > > };
> >
> > Good.
> > I'd not like to call it NON_CONTIG, as it sounds like CMA buffers
> > aren't allowed.
> > Instead, how about this definition?
> >
> > enum virtio_video_mem_layout {
> >         VIRTIO_VIDEO_MEM_LAYOUT_UNSPEC = 0,  /* no special requirement */
> >         VIRTIO_VIDEO_MEM_LAYOUT_CONTIG = 1,
> > };
>
> Yes, I agree, that would be more correct. We just need an empty line to be
> aligned with other enums.
>
> >
> > With this enum,
> > * the device can simply ignore this field if it doesn't have any
> > requirement and the struct is zero-initialized, and
> > * if we need to add other types of memory layout requirements, we can
> > add them as 2, 4, 8, etc to represent combinations of requirements.
> >
> > Just to confirm, are the following combination of planes_layout and
> > mem_layout valid?
> > (1) (planes_layout, mem_layout) = (contig, not specified)
> > (2) (planes_layout, mem_layout) = (not specified, contig)
> >
> > In my understanding, (1) means that each plane must be a contiguous
> > buffer while different planes don't have to be in a contig memory, but
> > (2) is invalid.
> > Is it correct?
>
> Let me tell a bit more about my vision:
>
> mem_layout: device can handle SG lists (e.g. using iommu) or it cannot and
> needs CMA. If it can handle SG, CMA is also ok. So it is ether 'I don't care'
> or 'give me CMA'.
>
> planes_layout: some devices might want to see multiplanar data in one buffer.
> So if we allocate two sets of mem entries (one per each plane) and send them
> to the host, the device will fail to handle those.
>
> So from my perspective you example can be interpreted as follows:
> (1) means all your planes are hosted by one buffer instance (we can see it as
> one fd for all planes plus per-plane offsets inside the buffer), the buffer
> itself consists of arbitrary pages (or can consist, as it can of course also
> be CMA, because of UNSPEC).
> (2) valid, means each plane has its own buffer, each buffer is contiguous in
> memory (must be allocated from CMA), but planes are not necessarily adjacent
> to each other in memory.
>
> This also means that we cannot have unspec for planes layout. Device either
> expects planes in separate buffers or in one buffer with some offsets, there
> cannot be mixed cases.

I might be misunderstanding the above, but just to make sure we're on
the same page, here are the cases that we found to exist in practice
with V4L2 stateful decoders:

1) device expects planes in one buffer laid out one after another
without any padding; the device accepts only 1 pointer and no offsets
- this corresponds to V4L2 non-M formats, such as NV12, which exactly
specify the location of planes in the buffer,

2) device accepts separate pointers to all planes - the planes can be
located anywhere in memory, which could be separate buffers, 1 buffer
with planes laid out at arbitrary offsets or even exactly the same
layout as required for 1).

3) device _requires_ planes to be located in separate buffers
allocated from designated areas in memory, e.g. different physical
memory banks, for performance purposes. This is actually a very rare
case and observed only on old generations of Samsung Exynos SoCs.

Please let me know if that matches your expectation,

Best regards,
Tomasz

> So it should look like this:
>
> enum virtio_video_planes_layout {
>     VIRTIO_VIDEO_PLANES_LAYOUT_UNSPEC = 0,  /* default, invalid */
>
>     VIRTIO_VIDEO_PLANES_LAYOUT_CONTIG = 0x100,
>     VIRTIO_VIDEO_PLANES_LAYOUT_NON_CONTIG,
> };
>
> Best regards,
> Dmitry
>
> >
> > Best regards,
> > Keiichi
> >
> > > Best regards,
> > > Dmitry.
> > >
> > > > Best regards,
> > > > Dmitry.
> > > >
> > > > > Best regards,
> > > > > Keiichi
> > > > >
> > > > > > Regards,
> > > > > > Dmitry.
> > > > > >
> > > > > > > Best regards,
> > > > > > > Keiichi
> > > > > > >
> > > > > > > > Regards,
> > > > > > > > Dmitry.
> > > > > > > >
> > > > > > > > > > For future V4L2 development we came up with the idea of a
> > > > > > > > > > format
> > > > > > > > > > flag
> > > > > > > > > > which could mean that the hardware allows putting planes in
> > > > > > > > > > separate
> > > > > > > > > > buffers. We could have a similar per-format flag in the
> > > > > > > > > > capabilities,
> > > > > > > > > > as we already have a list of all the supported formats
> > > > > > > > > > there.
> > > > > > > > >
> > > > > > > > > Sorry, forgot to paste the link from future V4L2 work notes
> > > > > > > > > from
> > > > > > > > > this
> > > > > > > > > year
> > > > > > > > > ELCE: https://www.spinics.net/lists/linux-media/msg159789.html
> > > > > > > > >
> > > > > > > > > > Best regards,
> > > > > > > > > > Tomasz
> > > > > > > > > >
> > > > > > > > > > > Best regards,
> > > > > > > > > > > Keiichi
> > > > > > > > > > >
> > > > > > > > > > > > Best regards,
> > > > > > > > > > > > Dmitry.
> > > > > > > > > > > >
> > > > > > > > > > > > > * Renamed virtio_video_pin to virtio_video_buf_type.
> > > > > > > > > > > > >
> > > > > > > > > > > > >   - It's similar to V4L2_BUF_TYPE_VIDEO_{OUTPUT,
> > > > > > > > > > > > >   CAPTURE}.
> > > > > > > > > > > > >
> > > > > > > > > > > > > * Added an error event.
> > > > > > > > > > > > > * Reordered some subsections.
> > > > > > > > > > > > > * Changed styles to make it consistent with other
> > > > > > > > > > > > > devices.
> > > > > > > > > > > > >
> > > > > > > > > > > > > Dmitry Sepp (1):
> > > > > > > > > > > > >   virtio-video: Add virtio video device specification
> > > > > > > > > > > > >
> > > > > > > > > > > > >  content.tex      |   1 +
> > > > > > > > > > > > >  virtio-video.tex | 579
> > > > > > > > > > > > >  +++++++++++++++++++++++++++++++++++++++++++++++
> > > > > > > > > > > > >  2 files changed, 580 insertions(+)
> > > > > > > > > > > > >  create mode 100644 virtio-video.tex
> > > > > > > > > > > > >
> > > > > > > > > > > > > --
> > > > > > > > > > > > > 2.24.1.735.g03f4e72817-goog
> > > >
> > > > ---------------------------------------------------------------------
> > > > To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
> > > > For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org
>
>