Re: [PATCH v13 5/8] media: chips-media: wave5: Add the v4l2 layer

[Sebastian Fricke]

Hey Hans,

On 16.10.2023 13:57, Hans Verkuil wrote:
> Hi Sebastian,
>
> On 12/10/2023 13:01, Sebastian Fricke wrote:
> > Add the decoder and encoder implementing the v4l2
> > API. This patch also adds the Makefile and the VIDEO_WAVE_VPU config
> >
> > Signed-off-by: Sebastian Fricke <sebastian.fricke@xxxxxxxxxxxxx>
> > Signed-off-by: Nicolas Dufresne <nicolas.dufresne@xxxxxxxxxxxxx>
> > Signed-off-by: Deborah Brouwer <deborah.brouwer@xxxxxxxxxxxxx>
> > Signed-off-by: Robert Beckett <bob.beckett@xxxxxxxxxxxxx>
> > Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@xxxxxxxxxxxxx>
> > Signed-off-by: Nas Chung <nas.chung@xxxxxxxxxxxxxxx>
> > ---
> >  drivers/media/platform/chips-media/Kconfig         |    1 +
> >  drivers/media/platform/chips-media/Makefile        |    1 +
> >  drivers/media/platform/chips-media/wave5/Kconfig   |   12 +
> >  drivers/media/platform/chips-media/wave5/Makefile  |   10 +
> >  .../platform/chips-media/wave5/wave5-helper.c      |  213 +++
> >  .../platform/chips-media/wave5/wave5-helper.h      |   31 +
> >  .../platform/chips-media/wave5/wave5-vpu-dec.c     | 1953 ++++++++++++++++++++
> >  .../platform/chips-media/wave5/wave5-vpu-enc.c     | 1794 ++++++++++++++++++
> >  .../media/platform/chips-media/wave5/wave5-vpu.c   |  291 +++
> >  .../media/platform/chips-media/wave5/wave5-vpu.h   |   83 +
> >  .../platform/chips-media/wave5/wave5-vpuapi.h      |    2 -
> >  11 files changed, 4389 insertions(+), 2 deletions(-)
>
> <snip>
>
> > +static int wave5_vpu_dec_create_bufs(struct file *file, void *priv,
> > +				     struct v4l2_create_buffers *create)
> > +{
> > +	struct vpu_instance *inst = wave5_to_vpu_inst(priv);
> > +	struct v4l2_format *f = &create->format;
> > +
> > +	/* Firmware does not support CREATE_BUFS for CAPTURE queues. */
> > +	if (f->type == V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE) {
> > +		dev_dbg(inst->dev->dev,
> > +			"%s: VIDIOC_CREATE_BUFS not supported on CAPTURE queues.\n",
> > +			__func__);
> > +		return -EOPNOTSUPP;
> > +	}
> > +
> > +	return v4l2_m2m_ioctl_create_bufs(file, priv, create);
> > +}
>
> Regarding the EOPNOTSUPP discussion: I discussed this some more with
> Nicolas on irc, and we wonder if it isn't better to just drop create_bufs
> support for the wave5 decoder altogether. Is there any point in supporting
> it for OUTPUT but not CAPTURE?
>
> <snip>
>
> > +static const struct v4l2_ioctl_ops wave5_vpu_dec_ioctl_ops = {
> > +	.vidioc_querycap = wave5_vpu_dec_querycap,
> > +	.vidioc_enum_framesizes = wave5_vpu_dec_enum_framesizes,
> > +
> > +	.vidioc_enum_fmt_vid_cap	= wave5_vpu_dec_enum_fmt_cap,
> > +	.vidioc_s_fmt_vid_cap_mplane = wave5_vpu_dec_s_fmt_cap,
> > +	.vidioc_g_fmt_vid_cap_mplane = wave5_vpu_dec_g_fmt_cap,
> > +	.vidioc_try_fmt_vid_cap_mplane = wave5_vpu_dec_try_fmt_cap,
> > +
> > +	.vidioc_enum_fmt_vid_out	= wave5_vpu_dec_enum_fmt_out,
> > +	.vidioc_s_fmt_vid_out_mplane = wave5_vpu_dec_s_fmt_out,
> > +	.vidioc_g_fmt_vid_out_mplane = wave5_vpu_g_fmt_out,
> > +	.vidioc_try_fmt_vid_out_mplane = wave5_vpu_dec_try_fmt_out,
> > +
> > +	.vidioc_g_selection = wave5_vpu_dec_g_selection,
> > +	.vidioc_s_selection = wave5_vpu_dec_s_selection,
> > +
> > +	.vidioc_reqbufs = v4l2_m2m_ioctl_reqbufs,
> > +	.vidioc_querybuf = v4l2_m2m_ioctl_querybuf,
> > +	.vidioc_create_bufs = wave5_vpu_dec_create_bufs,
>
> So this would just be dropped.
>
> > +	.vidioc_prepare_buf = v4l2_m2m_ioctl_prepare_buf,
> > +	.vidioc_qbuf = v4l2_m2m_ioctl_qbuf,
> > +	.vidioc_expbuf = v4l2_m2m_ioctl_expbuf,
> > +	.vidioc_dqbuf = v4l2_m2m_ioctl_dqbuf,
> > +	.vidioc_streamon = v4l2_m2m_ioctl_streamon,
> > +	.vidioc_streamoff = v4l2_m2m_ioctl_streamoff,
> > +
> > +	.vidioc_try_decoder_cmd = v4l2_m2m_ioctl_try_decoder_cmd,
> > +	.vidioc_decoder_cmd = wave5_vpu_dec_decoder_cmd,
> > +
> > +	.vidioc_subscribe_event = wave5_vpu_subscribe_event,
> > +	.vidioc_unsubscribe_event = v4l2_event_unsubscribe,
> > +};
>
> This also means there is no need to document the new EOPNOTSUPP error
> code in VIDIOC_CREATE_BUFS, or to modify v4l2-compliance.
>
> You *do* need to add a comment somewhere explaining why you don't
> support this ioctl. I think it would be best to do that right after
> '.vidioc_reqbufs = v4l2_m2m_ioctl_reqbufs,'.

So, besides this issue would you judge the v4l2 layer of the driver to
be ready? Do you want a reviewed by tag for it or would you take it like
this as well?

> Regards,
>
> 	Hans

Sincerly,
Sebastian
> _______________________________________________
> Kernel mailing list -- kernel@xxxxxxxxxxxxxxxxxxxxx
> To unsubscribe send an email to kernel-leave@xxxxxxxxxxxxxxxxxxxxx