Re: [PATCH] media: v4l2-subdev: Support enable/disable_streams for single-pad subdevs

[Tomi Valkeinen]

On 25/03/2024 15:02, Sakari Ailus wrote:
> Moi,
>
> Thanks for the patch.
>
> On Mon, Mar 25, 2024 at 02:50:55PM +0200, Laurent Pinchart wrote:
> > Hi Tomi,
> >
> > On Mon, Mar 25, 2024 at 02:43:23PM +0200, Tomi Valkeinen wrote:
> > > Currently a subdevice with a single pad, e.g. a sensor subdevice, must
> > > use the v4l2_subdev_video_ops.s_stream op, instead of
> > > v4l2_subdev_pad_ops.enable/disable_streams. This is because the
> > > enable/disable_streams machinery requires a routing table which a subdev
> > > cannot have with a single pad.
> > >
> > > Implement enable/disable_streams support for these single-pad subdevices
> > > by assuming an implicit stream 0 when the subdevice has only one pad.
> > >
> > > Signed-off-by: Tomi Valkeinen <tomi.valkeinen@xxxxxxxxxxxxxxxx>
> > > ---
> > > Even though I did send this patch, I'm not sure if this is necessary.
> > > s_stream works fine for the subdevs with a single pad. With the upcoming
> > > internal pads, adding an internal pad to the subdev will create a
> > > routing table, and enable/disable_streams would get "fixed" that way.
>
> I'd like to get rid of a redundant way to control streaming.

We can't get rid of it anyway, can we? We're not going to convert old 
drivers to streams.

For new drivers, yes, we shouldn't use s_stream. But is the answer for 
new sensor drivers this patch, or requiring an internal pad?

> > > So perhaps the question is, do we want to support single-pad subdevs in
> > > the future, in which case something like this patch is necessary, or
> > > will all modern source subdev drivers have internal pads, in which
> > > case this is not needed...
> >
> > I think the latter would be best. I however can't guarantee we won't
> > have valid use cases for (enable|disable)_streams on single-pad subdevs
> > though, so you patch could still be interesting.
>
> Instead of the number of pads, could we use instead the
> V4L2_SUBDEV_FL_STREAMS flag or whether g_routing op is supported to
> determine the need for this?

Maybe, but are they better? Do you see some issue with checking for the 
number of pads? I considered a few options, but then thought that the 
most safest test for this case is 1) one pad 2) enable/disable_streams 
implemented.

 Tomi