On Fri, May 03, 2024 at 09:55:07AM +0200, Jacopo Mondi wrote:
> Hi Laurent
>
> On Thu, May 02, 2024 at 08:34:30PM GMT, Laurent Pinchart wrote:
> > Hi Jacopo,
> >
> > Thank you for the patch.
> >
> > On Tue, Apr 30, 2024 at 12:39:37PM +0200, Jacopo Mondi wrote:
> > > Initialize and use the subdev active state to store the subdevice
> > > format.
> > >
> > > This simplifies the implementation of the get_fmt and set_fmt pad
> > > operations.
> > >
> > > Signed-off-by: Jacopo Mondi <jacopo.mondi@xxxxxxxxxxxxxxxx>
> > > ---
> > > drivers/media/i2c/adv748x/adv748x-csi2.c | 69 ++++--------------------
> > > drivers/media/i2c/adv748x/adv748x.h | 1 -
> > > 2 files changed, 11 insertions(+), 59 deletions(-)
> > >
> > > diff --git a/drivers/media/i2c/adv748x/adv748x-csi2.c b/drivers/media/i2c/adv748x/adv748x-csi2.c
> > > index 5b265b722394..435b0909bbef 100644
> > > --- a/drivers/media/i2c/adv748x/adv748x-csi2.c
> > > +++ b/drivers/media/i2c/adv748x/adv748x-csi2.c
> > > @@ -139,78 +139,26 @@ static const struct v4l2_subdev_video_ops adv748x_csi2_video_ops = {
> > > * But we must support setting the pad formats for format propagation.
> > > */
> > >
> > > -static struct v4l2_mbus_framefmt *
> > > -adv748x_csi2_get_pad_format(struct v4l2_subdev *sd,
> > > - struct v4l2_subdev_state *sd_state,
> > > - unsigned int pad, u32 which)
> > > -{
> > > - struct adv748x_csi2 *tx = adv748x_sd_to_csi2(sd);
> > > -
> > > - if (which == V4L2_SUBDEV_FORMAT_TRY)
> > > - return v4l2_subdev_state_get_format(sd_state, pad);
> > > -
> > > - return &tx->format;
> > > -}
> > > -
> > > -static int adv748x_csi2_get_format(struct v4l2_subdev *sd,
> > > - struct v4l2_subdev_state *sd_state,
> > > - struct v4l2_subdev_format *sdformat)
> > > -{
> > > - struct adv748x_csi2 *tx = adv748x_sd_to_csi2(sd);
> > > - struct adv748x_state *state = tx->state;
> > > - struct v4l2_mbus_framefmt *mbusformat;
> > > -
> > > - mbusformat = adv748x_csi2_get_pad_format(sd, sd_state, sdformat->pad,
> > > - sdformat->which);
> > > - if (!mbusformat)
> > > - return -EINVAL;
> > > -
> > > - mutex_lock(&state->mutex);
> > > -
> > > - sdformat->format = *mbusformat;
> > > -
> > > - mutex_unlock(&state->mutex);
> > > -
> > > - return 0;
> > > -}
> > > -
> > > static int adv748x_csi2_set_format(struct v4l2_subdev *sd,
> > > struct v4l2_subdev_state *sd_state,
> > > struct v4l2_subdev_format *sdformat)
> > > {
> > > - struct adv748x_csi2 *tx = adv748x_sd_to_csi2(sd);
> > > - struct adv748x_state *state = tx->state;
> > > struct v4l2_mbus_framefmt *mbusformat;
> > > - int ret = 0;
> > > -
> > > - mbusformat = adv748x_csi2_get_pad_format(sd, sd_state, sdformat->pad,
> > > - sdformat->which);
> > > - if (!mbusformat)
> > > - return -EINVAL;
> > >
> > > - mutex_lock(&state->mutex);
> > > + mbusformat = v4l2_subdev_state_get_format(sd_state, sdformat->pad);
> > >
> > > + /* Format on the source pad is always copied from the sink one. */
> > > if (sdformat->pad == ADV748X_CSI2_SOURCE) {
> > > const struct v4l2_mbus_framefmt *sink_fmt;
> > >
> > > - sink_fmt = adv748x_csi2_get_pad_format(sd, sd_state,
> > > - ADV748X_CSI2_SINK,
> > > - sdformat->which);
> > > -
> > > - if (!sink_fmt) {
> > > - ret = -EINVAL;
> > > - goto unlock;
> > > - }
> > > -
> > > + sink_fmt = v4l2_subdev_state_get_format(sd_state,
> > > + ADV748X_CSI2_SINK);
> > > sdformat->format = *sink_fmt;
> >
> > That's not the right way to do it. You should propagate the format from
> > sink to source when pad == ADV748X_CSI2_SINK, and return
> > adv748x_csi2_get_format() when pad == ADV748X_CSI2_SOURCE. Otherwise
>
> I think it's done later
I didn't know when I reviewed this patch :-)
> and this patch doesn't change the currently implemented behaviour,
> doesn't it ?
I think it does.
> Anyway, I got that you would prefer to squash the first patches in a
> single one, so this will be solved
Agreed.
> > setting the format on the sink pad will not update the state of the
> > source pad, and a get format call on the source pad will return an
> > incorrect format.
> >
> > > }
> > >
> > > *mbusformat = sdformat->format;
> > >
> > > -unlock:
> > > - mutex_unlock(&state->mutex);
> > > -
> > > - return ret;
> > > + return 0;
> > > }
> > >
> > > static int adv748x_csi2_get_mbus_config(struct v4l2_subdev *sd, unsigned int pad,
> > > @@ -228,7 +176,7 @@ static int adv748x_csi2_get_mbus_config(struct v4l2_subdev *sd, unsigned int pad
> > > }
> > >
> > > static const struct v4l2_subdev_pad_ops adv748x_csi2_pad_ops = {
> > > - .get_fmt = adv748x_csi2_get_format,
> > > + .get_fmt = v4l2_subdev_get_fmt,
> > > .set_fmt = adv748x_csi2_set_format,
> > > .get_mbus_config = adv748x_csi2_get_mbus_config,
> > > };
> > > @@ -320,6 +268,11 @@ int adv748x_csi2_init(struct adv748x_state *state, struct adv748x_csi2 *tx)
> > > if (ret)
> > > goto err_cleanup_subdev;
> > >
> > > + tx->sd.state_lock = tx->ctrl_hdl.lock;
> >
> > Maybe that's addressed in subsequent patches, but do we need a
> > device-wide lock ? The code you replace above uses the
>
> device-wide as global to the CSI-2, the HDMI and the AFE subdevices ?
Yes. Multi-subdev drivers like CCS do so, because controls of one subdev
influence operations on other subdevs. I haven't checked if it's
actually required for adv748x.
> > adv748x_state.mutex lock, which covers all subdevs. I don't think this
> > patch introduces race conditions, so this could possibly be handled on
> > top.
> >
> > > + ret = v4l2_subdev_init_finalize(&tx->sd);
> > > + if (ret)
> > > + goto err_free_ctrl;
> > > +
> > > ret = v4l2_async_register_subdev(&tx->sd);
> > > if (ret)
> > > goto err_free_ctrl;
> > > diff --git a/drivers/media/i2c/adv748x/adv748x.h b/drivers/media/i2c/adv748x/adv748x.h
> > > index d2b5e722e997..9bc0121d0eff 100644
> > > --- a/drivers/media/i2c/adv748x/adv748x.h
> > > +++ b/drivers/media/i2c/adv748x/adv748x.h
> > > @@ -75,7 +75,6 @@ enum adv748x_csi2_pads {
> > >
> > > struct adv748x_csi2 {
> > > struct adv748x_state *state;
> > > - struct v4l2_mbus_framefmt format;
> > > unsigned int page;
> > > unsigned int port;
> > > unsigned int num_lanes;
--
Regards,
Laurent Pinchart
