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

[Tomi Valkeinen]

On 02/04/2024 15:05, Sakari Ailus wrote:
> On Wed, Mar 27, 2024 at 03:39:31PM +0200, Tomi Valkeinen wrote:
> > On 27/03/2024 15:32, Sakari Ailus wrote:
> > > Heissulivei,
> > >
> > > On Wed, Mar 27, 2024 at 01:06:42PM +0200, Tomi Valkeinen wrote:
> > > > On 27/03/2024 12:46, Sakari Ailus wrote:
> > > > > Heippa,
> > > > >
> > > > > On Mon, Mar 25, 2024 at 07:56:46PM +0200, Tomi Valkeinen wrote:
> > > > > > On 25/03/2024 19:52, Sakari Ailus wrote:
> > > > > > > Moi,
> > > > > > >
> > > > > > > On Mon, Mar 25, 2024 at 03:43:01PM +0200, Tomi Valkeinen wrote:
> > > > > > > > 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.
> > > > > > >
> > > > > > > I'd expect to do that but it'd take a long time. That being said, I think
> > > > > > > we need to consider devices without pads (VCMs) so it may well be this
> > > > > > > would remain after all.
> > > > > > >
> > > > > > > > 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?
> > > > > > >
> > > > > > > For new drivers I'd like to see an internal pad in fact.
> > > > > > > {enable,disable}_streams is still internal to the kernel.
> > > > > >
> > > > > > So, you think this patch should be dropped?
> > > > >
> > > > > No, no. Not all sub-device drivers with pads are camera sensor drivers. :-)
> > > >
> > > > Hmm, alright. So we want to support enable/disable_streams for sub-devices
> > > > with multiple source pads but no routing (so probably no sink pads)?
> > >
> > > That should be allowed indeed, in order to move from s_stream() to
> > > {enable,disable}_streams().
> > >
> > > > > > > > > > > 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.
> > > > > > >
> > > > > > > I think I'd actually prefer {enable,disable}_streams in fact.
> > > > > >
> > > > > > Hmm, sorry, now I'm confused =). What do you mean with that?
> > > > >
> > > > > I'd use V4L2_SUBDEV_FL_STREAMS flag instead of the number of pads. The
> > > > > number of pads is less related to routing.
> > > >
> > > > Well, with one pad you cannot have routing =).
> > > >
> > > > In this patch I used sd->enabled_streams to track the enabled streams, but
> > > > if we need to support multiple pads, I'll have to invent something new for
> > > > that.
> > >
> > > What exactly do you think needs to be changed? This is just about starting
> > > and stopping streaming using a different sent of callbacks, right?
> >
> > The helpers track which streams are enabled, so we need some place to store
> > the enabled streams.
> >
> > For V4L2_SUBDEV_FL_STREAMS we have that in state->stream_configs for each
> > stream. For the one-source-pad case we have a subdev wide
> > sd->enabled_streams to store that. But we don't have any place at the moment
> > to store if a pad is enabled.
>
> If there are is no support for routing, isn't streaming either enabled or
> disabled on all of them?

Hmm, no, I don't see why that would be the case. If a subdev has two 
source pads, and it gets an enable_streams() call on the first source 
pad, why would the second source pad be enabled too?

 Tomi