Re: [PATCH v1 1/1] media: imx: csi: Parse link configuration from fw_node

[Mathis Foerst <mathis.foerst@mt.com>]

On Mon, Mar 10, 2025 at 10:25:04AM +0000, Sakari Ailus wrote:
Hi Sakari,

> Hi Mathis,
> 
> On Fri, Mar 07, 2025 at 02:38:29PM +0100, Mathis Foerst wrote:
> > Hi Sakari, Hi Dan,
> > 
> > thanks a lot for your feedback.
> > 
> > On Thu, Mar 06, 2025 at 09:13:46PM +0000, Sakari Ailus wrote:
> > > Hi Dan,
> > > 
> > > On Thu, Mar 06, 2025 at 10:07:20PM +0300, Dan Carpenter wrote:
> > > > On Thu, Mar 06, 2025 at 04:33:18PM +0000, Sakari Ailus wrote:
> > > > > Hi Mathis,
> > > > > 
> > > > > Thanks for the patch.
> > > > > 
> > > > > On Wed, Mar 05, 2025 at 12:38:02PM +0100, Mathis Foerst wrote:
> > > > > > The imx-media-csi driver requires upstream camera drivers to implement
> > > > > > the subdev-pad-op "get_mbus_config" [0]. Camera drivers that don't
> > > > > > implement this function are not usable on the i.MX6.
> > > > > > 
> > > > > > The docs for get_mbus_config [1] say:
> > > > > > @get_mbus_config: get the media bus configuration of a remote sub-device.
> > > > > >             The media bus configuration is usually retrieved from the
> > > > > >             firmware interface at sub-device probe time, immediately
> > > > > >             applied to the hardware and eventually adjusted by the
> > > > > >             driver.
> > > > > > 
> > > > > > Currently, the imx-media-csi driver is not incorporating the information
> > > > > > from the firmware interface and therefore relies on the implementation of
> > > > > > get_mbus_config by the camera driver.
> > > > > > 
> > > > > > To be compatible with camera drivers not implementing get_mbus_config
> > > > > > (which is the usual case), use the bus information from the fw interface:
> > > > > > 
> > > > > > The camera does not necessarily has a direct media bus link to the CSI as
> > > > > > the video-mux and/or the MIPI CSI-2 receiver of the i.MX6 might be in
> > > > > > between them on the media pipeline.
> > > > > > The CSI driver already implements the functionality to find the connected
> > > > > > camera sub-device to call get_mbus_config on it.
> > > > > > 
> > > > > > At this point the driver is modified as follows:
> > > > > > In the case that get_mbus_config is not implemented by the upstream
> > > > > > camera, try to get its endpoint configuration from the firmware interface
> > > > > > usign v4l2_fwnode_endpoint_parse.
> > > > > > For the supported mbus_types (V4L2_MBUS_PARALLEL, V4L2_MBUS_BT656 and
> > > > > > V4L2_MBUS_CSI2_DPHY), extract the mbus_config from the endpoint
> > > > > > configuration.
> > > > > > For all other mbus_types, return an error.
> > > > > > 
> > > > > > Note that parsing the mbus_config from the fw interface is not done during
> > > > > > probing because the camera that's connected to the CSI can change based on
> > > > > > the selected input of the video-mux at runtime.
> > > > > > 
> > > > > > [0] drivers/staging/media/imx/imx-media-csi.c - line 211..216
> > > > > > [1] include/media/v4l2-subdev.h - line 814
> > > > > > 
> > > > > > Signed-off-by: Mathis Foerst <mathis.foerst@mt.com>
> > > > > > ---
> > > > > >  drivers/staging/media/imx/imx-media-csi.c | 36 ++++++++++++++++++++---
> > > > > >  1 file changed, 32 insertions(+), 4 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/staging/media/imx/imx-media-csi.c b/drivers/staging/media/imx/imx-media-csi.c
> > > > > > index 3edbc57be2ca..394a9321a10b 100644
> > > > > > --- a/drivers/staging/media/imx/imx-media-csi.c
> > > > > > +++ b/drivers/staging/media/imx/imx-media-csi.c
> > > > > > @@ -169,6 +169,8 @@ static int csi_get_upstream_mbus_config(struct csi_priv *priv,
> > > > > >  {
> > > > > >  	struct v4l2_subdev *sd, *remote_sd;
> > > > > >  	struct media_pad *remote_pad;
> > > > > > +	struct fwnode_handle *ep_node;
> > > > > > +	struct v4l2_fwnode_endpoint ep = { .bus_type = 0 };
> > > > > 
> > > > > Are there any defaults in DT bindings (other than 0's)? Also initialising a
> > > > > field to zero this way is redundant, just use {}.
> > > > > 
> > > > 
> > > > I was going to respond in much the same way.  This is equivalen to:
> > > > 
> > > > struct v4l2_fwnode_endpoint ep = { .bus_type = V4L2_MBUS_UNKNOWN };
> > > 
> > > Thinking about this in a context of parsing the endpoint, in fact the
> > > bus_type should be specified. Presumably the hardware is D-PHY, so the
> > > correct value would be V4L2_MBUS_CSI2_DPHY. This way
> > > v4l2_fwnode_endpoint_parse() doesn't need to guess.
> > 
> > I think we must use "bus_type = V4L2_MBUS_UNKNOWN" here:
> > 
> > The i.MX6 has two types of camera interfaces: Parallel and MIPI CSI-2.
> > They are connected either directly or via a video-mux to the CSIs
> > (See IMX6DQRM.pdf - Figure 9-3 for the connection diagram)
> > 
> > Pre-defining V4L2_MBUS_CSI2_DPHY here would let
> > v4l2_fwnode_endpoint_parse() fail if the camera uses the parallel bus.
> > 
> > We could distinguish between MIPI CSI-2 and Parallel input by checking
> > the grp_id of the upstream device like it's already done in
> > csi_get_upstream_mbus_config().
> > But for the Parallel case we still can't know if we should set bus_type
> > to V4L2_MBUS_PARALLEL or to V4L2_MBUS_BT656 - the i.MX6 supports both
> > formats on the parallel interface.
> > 
> > That's why I would argue that v4l2_fwnode_endpoint_parse() must figure
> > out the bus_type from the fw node.
> 
> Right, nowadays you can indeed do this -- it wasn't a long ago when you
> couldn't. I presume the bindings do specify the bus-type property is
> mandatory? Where are the bindings btw.?
> 

From my understanding, it's not even required to set the bus-type as 
v4l2_fwnode_endpoint_parse() will try to parse the endpoint first as a
CSI-2 bus and in case of failure as a Parallel bus if the bus-type is
unknown (see drivers/media/v4l2-core/v4l2-fwnode.c#L493).

About the bindings:

There are bindings for the MIPI CSI-2 receiver:
Documentation/devicetree/bindings/media/imx.txt
I think here it's not necessary to make the bus-type property mandatory
as the imx6-mipi-csi2 driver enforces V4L2_MBUS_CSI2_DPHY anyhow
(see drivers/staging/media/imx/imx6-mipi-csi2.c#L677).

For the case of a camera with parallel bus, the camera endpoint is
connected to a video-mux and not directly to the CSI. Therefore, we cannot
make the bus-type property mandatory on this endpoint as it this wouldn't
apply to other use-cases of video-mux.

> > 
> > > 
> > > > 
> > > > > >  	int ret;
> > > > > >  
> > > > > >  	if (!priv->src_sd)
> > > > > > @@ -210,11 +212,37 @@ static int csi_get_upstream_mbus_config(struct csi_priv *priv,
> > > > > >  
> > > > > >  	ret = v4l2_subdev_call(remote_sd, pad, get_mbus_config,
> > > > > >  			       remote_pad->index, mbus_cfg);
> > > > > > -	if (ret == -ENOIOCTLCMD)
> > > > > > -		v4l2_err(&priv->sd,
> > > > > > -			 "entity %s does not implement get_mbus_config()\n",
> > > > > > -			 remote_pad->entity->name);
> > > > > > +	if (ret == -ENOIOCTLCMD) {
> > > > > 
> > > > > 	if (!ret)
> > > > > 		return 0;
> > > > > 
> > > > > And you can unindent the rest.
> > > > 
> > > > I was going to say this too but then I thought actually this needs to
> > > > be:
> > > > 
> > > > 	if (ret != -ENOIOCTLCMD)
> > > > 		return ret;
> > > > 
> > > > Which is weird.  Better to break all the new code into a separate
> > > > helper function.
> > > > 
> > > > 	if (ret == -ENOIOCTLCMD)
> > > > 		ret = parse_fw_link_config_stuff();
> > > > 
> > > > 	return ret;
> > 
> > Good point. I factored out a helper function as suggested.
> > 
> > > 
> > > Indeed. get_mbus_config() presumably wouldn't return an error but
> > > correctness is usually a good idea.
> > > 
> 
> -- 
> Regards,
> 
> Sakari Ailus

Best regards,
Mathis Foerst