Hi Niklas,
On Fri, May 10, 2024 at 12:36:56PM +0200, Niklas Söderlund wrote:
> On 2024-05-10 13:13:40 +0300, Laurent Pinchart wrote:
> > On Fri, May 10, 2024 at 12:03:47PM +0200, Niklas Söderlund wrote:
> > > Create and store the subdevice format in the subdevices active state.
> > > This change do not have a huge effect on the driver as it do not yet
> > > support changing the format.
> > >
> > > Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@xxxxxxxxxxxx>
> > > ---
> > > Hello,
> > >
> > > This change is mostly to align all driver used in the R-Car VIN pipeline
> > > to use the same set of API and to make it easier to extend them in
> > > tandem going forward.
> > >
> > > This should be seen as a compliment to Jacopo's larger work in "[PATCH
> > > v3 00/11] media: renesas: rcar-csi2: Use the subdev active state" which
> > > updates other drivers in the VIN pipeline to use the active state.
> > > ---
> > > drivers/staging/media/max96712/max96712.c | 36 +++++++++++++++++------
> > > 1 file changed, 27 insertions(+), 9 deletions(-)
> > >
> > > diff --git a/drivers/staging/media/max96712/max96712.c b/drivers/staging/media/max96712/max96712.c
> > > index c44145284aa1..2da65ccd2c57 100644
> > > --- a/drivers/staging/media/max96712/max96712.c
> > > +++ b/drivers/staging/media/max96712/max96712.c
> > > @@ -242,21 +242,34 @@ static const struct v4l2_subdev_video_ops max96712_video_ops = {
> > > .s_stream = max96712_s_stream,
> > > };
> > >
> > > -static int max96712_get_pad_format(struct v4l2_subdev *sd,
> > > - struct v4l2_subdev_state *sd_state,
> > > - struct v4l2_subdev_format *format)
> > > +static int max96712_init_state(struct v4l2_subdev *sd,
> > > + struct v4l2_subdev_state *state)
> > > {
> > > - format->format.width = 1920;
> > > - format->format.height = 1080;
> > > - format->format.code = MEDIA_BUS_FMT_RGB888_1X24;
> > > - format->format.field = V4L2_FIELD_NONE;
> > > + static const struct v4l2_mbus_framefmt default_fmt = {
> > > + .width = 1920,
> > > + .height = 1080,
> > > + .code = MEDIA_BUS_FMT_RGB888_1X24,
> > > + .colorspace = V4L2_COLORSPACE_SRGB,
> > > + .field = V4L2_FIELD_NONE,
> > > + .ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT,
> > > + .quantization = V4L2_QUANTIZATION_DEFAULT,
> > > + .xfer_func = V4L2_XFER_FUNC_DEFAULT,
> > > + };
> > > + struct v4l2_mbus_framefmt *fmt;
> > > +
> > > + fmt = v4l2_subdev_state_get_format(state, 0);
> > > + *fmt = default_fmt;
> > >
> > > return 0;
> > > }
> > >
> > > +static const struct v4l2_subdev_internal_ops max96712_internal_ops = {
> > > + .init_state = max96712_init_state,
> > > +};
> > > +
> > > static const struct v4l2_subdev_pad_ops max96712_pad_ops = {
> > > - .get_fmt = max96712_get_pad_format,
> > > - .set_fmt = max96712_get_pad_format,
> > > + .get_fmt = v4l2_subdev_get_fmt,
> > > + .set_fmt = v4l2_subdev_get_fmt,
> > > };
> > >
> > > static const struct v4l2_subdev_ops max96712_subdev_ops = {
> > > @@ -293,6 +306,7 @@ static int max96712_v4l2_register(struct max96712_priv *priv)
> > > long pixel_rate;
> > > int ret;
> > >
> > > + priv->sd.internal_ops = &max96712_internal_ops;
> > > v4l2_i2c_subdev_init(&priv->sd, priv->client, &max96712_subdev_ops);
> > > priv->sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> > > priv->sd.entity.function = MEDIA_ENT_F_VID_IF_BRIDGE;
> > > @@ -324,6 +338,10 @@ static int max96712_v4l2_register(struct max96712_priv *priv)
> > >
> > > v4l2_set_subdevdata(&priv->sd, priv);
> > >
> >
> > I think it would be nice to already use the control handler lock as the
> > active state lock. Apart from that, the patch looks good to me.
>
> I also worked on this to learn a bit more about the active state API. I
> assume what you are asking for here is ?
>
> priv->sd.state_lock = priv->ctrl_handler.lock;
That's right.
> What is the rational for using the control handler lock over the default
> state lock?
When there are interactions between controls and formats, using separate
locks can lead to AB/BA deadlocks if you need to access controls from a
context where the state is locked (e.g. calling
v4l2_ctrl_handler_setup() at stream enable time), and access the active
state from a context where the control handler is locked (e.g. accessing
the state in .s_ctrl()). Using the same lock helps there, you can then
call __v4l2_ctrl_handler_setup() at stream enable time, and get the
state without locking it in .s_ctrl().
At some point in the not too distant future I would like to move storage
of controls to the state. I don't know how painful that will be though,
the control framework has grown to be a very complex beast, with lots of
features that are not useful for MC-based drivers (such as control
handler inheritance). I'm worried that we'll need a full rewrite...
We'll see :-)
> I checked Jacopo's series when working on this and there he
> uses no lock for the R-Car CSI-2 driver and a private driver global
> state lock for ADV748x.
For the CSI-2 RX driver, that's because it doesn't have a control
handler. For the ADV748x driver, that's because the control handler is
configured to use the device-wide mutex.
> > > + ret = v4l2_subdev_init_finalize(&priv->sd);
> > > + if (ret)
> > > + goto error;
> > > +
> > > ret = v4l2_async_register_subdev(&priv->sd);
> > > if (ret < 0) {
> > > dev_err(&priv->client->dev, "Unable to register subdevice\n");
--
Regards,
Laurent Pinchart
