Re: [PATCH 2/4] media: subdev: Fix use of sd->enabled_streams in call_s_stream()

[Tomi Valkeinen]

On 04/04/2024 16:00, Laurent Pinchart wrote:
> Hi Tomi,
>
> Thank you for the patch.
>
> On Thu, Apr 04, 2024 at 01:50:01PM +0300, Tomi Valkeinen wrote:
> > call_s_stream() uses sd->enabled_streams to track whether streaming has
> > already been enabled. However,
> > v4l2_subdev_enable/disable_streams_fallback(), which was the original
> > user of this field, already uses it, and
> > v4l2_subdev_enable/disable_streams_fallback() will call call_s_stream().
> >
> > This leads to a conflict as both functions set the field. Afaics, both
> > functions set the field to the same value, so it won't cause a runtime
> > bug, but it's still wrong and if we, e.g., change how
> > v4l2_subdev_enable/disable_streams_fallback() operates we might easily
> > cause bugs.
> >
> > Fix this by adding a new field, 'streaming_enabled', for
> > call_s_stream().
> >
> > Signed-off-by: Tomi Valkeinen <tomi.valkeinen@xxxxxxxxxxxxxxxx>
> > ---
> >   drivers/media/v4l2-core/v4l2-subdev.c | 8 ++------
> >   include/media/v4l2-subdev.h           | 2 ++
> >   2 files changed, 4 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/media/v4l2-core/v4l2-subdev.c b/drivers/media/v4l2-core/v4l2-subdev.c
> > index b90b5185e87f..3b3310bce5d4 100644
> > --- a/drivers/media/v4l2-core/v4l2-subdev.c
> > +++ b/drivers/media/v4l2-core/v4l2-subdev.c
> > @@ -404,12 +404,8 @@ static int call_s_stream(struct v4l2_subdev *sd, int enable)
> >   	 * The .s_stream() operation must never be called to start or stop an
> >   	 * already started or stopped subdev. Catch offenders but don't return
> >   	 * an error yet to avoid regressions.
> > -	 *
> > -	 * As .s_stream() is mutually exclusive with the .enable_streams() and
> > -	 * .disable_streams() operation, we can use the enabled_streams field
> > -	 * to store the subdev streaming state.
> >   	 */
> > -	if (WARN_ON(!!sd->enabled_streams == !!enable))
> > +	if (WARN_ON(!!sd->streaming_enabled == !!enable))
> >   		return 0;
> >   
> >   #if IS_REACHABLE(CONFIG_LEDS_CLASS)
> > @@ -429,7 +425,7 @@ static int call_s_stream(struct v4l2_subdev *sd, int enable)
> >   	}
> >   
> >   	if (!ret)
> > -		sd->enabled_streams = enable ? BIT(0) : 0;
> > +		sd->streaming_enabled = !!enable;
> >   
> >   	return ret;
> >   }
> > diff --git a/include/media/v4l2-subdev.h b/include/media/v4l2-subdev.h
> > index a9e6b8146279..8bd1e3c96d2b 100644
> > --- a/include/media/v4l2-subdev.h
> > +++ b/include/media/v4l2-subdev.h
> > @@ -1043,6 +1043,7 @@ struct v4l2_subdev_platform_data {
> >    *		     v4l2_subdev_enable_streams() and
> >    *		     v4l2_subdev_disable_streams() helper functions for fallback
> >    *		     cases.
> > + * @streaming_enabled: Tracks whether streaming has been enabled with s_stream.
>
> Could you extend this to indicate this field can't be used by anything
> else than call_s_stream() ?

Yes, I'll add that.

> I'm also getting a bit concerned about having multiple fields with
> similar purposes. Another option would be to call the .s_stream()

I agree, it's getting slightly complex. The reason is that we support 
many ways for the subdev to implement these things.

> operation directly from the fallback handlers, instead of going through
> call_s_stream(). That may be considered as a bit of a hack though.

I thought about the same thing, but I felt it might just make things 
more complex. However, with my new unsent series, I add the privacy_led 
code also to the v4l2_subdev_enable_streams() function, so it's even 
more similar to the call_s_stream.

Maybe I'll try it out and see how it looks like if I call the op directly.

 Tomi

> >    *
> >    * Each instance of a subdev driver should create this struct, either
> >    * stand-alone or embedded in a larger struct.
> > @@ -1091,6 +1092,7 @@ struct v4l2_subdev {
> >   	 */
> >   	struct v4l2_subdev_state *active_state;
> >   	u64 enabled_streams;
> > +	bool streaming_enabled;
> >   };
> >   
> >