Re: [PATCH v4 08/11] media: ti: Add CSI2RX support for J721E

[Sakari Ailus]

On Thu, Oct 07, 2021 at 02:31:43AM +0530, Pratyush Yadav wrote:
> On 06/10/21 11:28PM, Sakari Ailus wrote:
> > Hi Pratyush,
> > 
> > On Wed, Sep 15, 2021 at 05:32:37PM +0530, Pratyush Yadav wrote:
> > ...
> > > +/*
> > > + * Find the input format. This is done by finding the first device in the
> > > + * pipeline which can tell us the current format. This could be the sensor, or
> > > + * this could be another device in the middle which is capable of format
> > > + * conversions.
> > > + */
> > > +static int ti_csi2rx_validate_pipeline(struct ti_csi2rx_dev *csi)
> > > +{
> > > +	struct media_pipeline *pipe = &csi->pipe;
> > > +	struct media_entity *entity;
> > > +	struct v4l2_subdev *sd;
> > > +	struct v4l2_subdev_format fmt;
> > > +	struct v4l2_pix_format *pix = &csi->v_fmt.fmt.pix;
> > > +	struct media_device *mdev = &csi->mdev;
> > > +	const struct ti_csi2rx_fmt *ti_fmt;
> > > +	int ret;
> > > +
> > > +	mutex_lock(&mdev->graph_mutex);
> > > +	ret = media_graph_walk_init(&pipe->graph, mdev);
> > > +	if (ret) {
> > > +		mutex_unlock(&mdev->graph_mutex);
> > > +		return ret;
> > > +	}
> > > +
> > > +	media_graph_walk_start(&pipe->graph, &csi->vdev.entity);
> > > +
> > > +	while ((entity = media_graph_walk_next(&pipe->graph))) {
> > > +		if (!is_media_entity_v4l2_subdev(entity))
> > > +			continue;
> > 
> > You shouldn't rely on media_graph_walk_next() to return entities in a
> > particular order.
> 
> Ah, right. Need to drop this.
> 
> > 
> > I'd suggest approach taken in isp_video_check_external_subdevs() (in
> > drivers/media/platform/omap3isp/ispvideo.c).
> > 
> > > +
> > > +		sd = media_entity_to_v4l2_subdev(entity);
> > > +
> > > +		fmt.which = V4L2_SUBDEV_FORMAT_ACTIVE;
> > > +		fmt.pad = media_get_pad_index(entity, 0, PAD_SIGNAL_DEFAULT);
> > > +
> > > +		ret = v4l2_subdev_call(sd, pad, get_fmt, NULL, &fmt);
> > > +		if (ret && ret != -ENOIOCTLCMD) {
> > > +			media_graph_walk_cleanup(&pipe->graph);
> > > +			mutex_unlock(&mdev->graph_mutex);
> > > +			return ret;
> > > +		}
> > > +
> > > +		if (!ret)
> > > +			break;
> > > +	}
> > > +
> > > +	media_graph_walk_cleanup(&pipe->graph);
> > > +	mutex_unlock(&mdev->graph_mutex);
> > > +
> > > +	/* Could not find input format. */
> > > +	if (!entity)
> > > +		return -EPIPE;
> > > +
> > > +	if (fmt.format.width != pix->width)
> > > +		return -EPIPE;
> > > +	if (fmt.format.height != pix->height)
> > > +		return -EPIPE;
> > 
> > Pipeline validation should take place during media_pipeline_start(). Why
> > are you doing it here?
> 
> How would be do that? Via the link_validate callback?

Yes, please. See other drivers for examples --- such as omap3isp.

-- 
Sakari Ailus