[PATCH/RFC 22/32] libcamera: Add CameraSensor implementation for raw V4L2 sensors

[Naushir Patuck]

On Wed, 13 Mar 2024 at 12:46, Tomi Valkeinen
<tomi.valkeinen at ideasonboard.com> wrote:
>
> On 13/03/2024 14:20, Naushir Patuck wrote:
>
> > ?> +               case MediaBusFormatInfo::Type::Metadata:
> >> +                       /*
> >> +                        * Skip metadata streams that are not sensor embedded
> >> +                        * data. The source stream reports a generic metadata
> >> +                        * format, check the sink stream for the exact format.
> >> +                        */
> >> +                       formats = subdev_->formats(route.sink);
> >> +                       if (formats.size() != 1)
> >> +                               continue;
> >
> > Should this test be if (!formats.size()) insead?  It might be possible
> > to have multiple metadata types.
>
> The driver in my branch is old and hacky. I should see what Laurent has
> done with the imx219 in his branch, and possibly just take that one.
>
> I think advertising only a single format makes sense here, as the
> embedded format is defined by the video format.
>
> >> +
> >> +                       if (MediaBusFormatInfo::info(formats.cbegin()->first).type !=
> >> +                           MediaBusFormatInfo::Type::EmbeddedData)
> >> +                               continue;
> >
> > The IMX219 driver (from Tomi's kernel tree) advertises
> > MEDIA_BUS_FMT_META_8 / MEDIA_BUS_FMT_META_10 formats for the embedded
> > data stream, which translates to a type of
> > MediaBusFormatInfo::Type::Metadata.  Does the driver need updating, or
> > should this check include MediaBusFormatInfo::Type::Metadata?
>
> Laurent's version should also report those same mbus formats. Hmm, oh,
> but it uses MEDIA_BUS_FMT_CCS_EMBEDDED for the internal pad...

Can you point to Laurent's tree?  I'll take the driver from there and
see if these comments are still valid.

>
>   Tomi
>