2020-03-15 10:28:11

by Lad, Prabhakar

[permalink] [raw]
Subject: [PATCH 2/2] media: rcar-csi2: Let the driver handle fwnode matching using match_custom callback

The rcar-csi2 driver uses the v4l2-async framework to do endpoint matching
instead of node matching. This is needed as it needs to work with the
adv748x driver which register it self in v4l2-async using endpoints
instead of nodes. The reason for this is that from a single DT node it
creates multiple subdevices, one for each endpoint.

But when using subdevs which register itself in v4l2-async using nodes,
the rcar-csi2 driver failed to find the matching endpoint because the
match.fwnode was pointing to remote endpoint instead of remote parent
port.

This commit adds support in rcar-csi2 driver to handle both the cases
where subdev registers in v4l2-async using endpoints/nodes, by using
match_type as V4L2_ASYNC_MATCH_CUSTOM and implementing the match()
callback to compare the fwnode of either remote/parent.

Signed-off-by: Lad Prabhakar <[email protected]>
---
drivers/media/platform/rcar-vin/rcar-csi2.c | 46 +++++++++++++++++++--
1 file changed, 43 insertions(+), 3 deletions(-)

diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c b/drivers/media/platform/rcar-vin/rcar-csi2.c
index faa9fb23a2e9..1bbf05e9f025 100644
--- a/drivers/media/platform/rcar-vin/rcar-csi2.c
+++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
@@ -808,6 +808,46 @@ static int rcsi2_parse_v4l2(struct rcar_csi2 *priv,
return 0;
}

+static bool rcsi2_asd_match(struct v4l2_subdev *sd,
+ struct v4l2_async_subdev *asd)
+{
+ struct rcar_csi2 *priv = (struct rcar_csi2 *)asd->match.custom.priv;
+ struct fwnode_handle *remote_endpoint;
+ struct fwnode_handle *subdev_endpoint;
+ struct device_node *np;
+ bool matched = false;
+
+ np = of_graph_get_endpoint_by_regs(priv->dev->of_node, 0, 0);
+ if (!np) {
+ dev_err(priv->dev, "Not connected to subdevice\n");
+ return matched;
+ }
+
+ remote_endpoint =
+ fwnode_graph_get_remote_endpoint(of_fwnode_handle(np));
+ if (!remote_endpoint) {
+ dev_err(priv->dev, "Failed to get remote endpoint\n");
+ of_node_put(np);
+ return matched;
+ }
+ of_node_put(np);
+
+ if (sd->fwnode != dev_fwnode(sd->dev)) {
+ if (remote_endpoint == sd->fwnode)
+ matched = true;
+ } else {
+ subdev_endpoint =
+ fwnode_graph_get_next_endpoint(dev_fwnode(sd->dev), NULL);
+ if (remote_endpoint == subdev_endpoint)
+ matched = true;
+ fwnode_handle_put(subdev_endpoint);
+ }
+
+ fwnode_handle_put(remote_endpoint);
+
+ return matched;
+}
+
static int rcsi2_parse_dt(struct rcar_csi2 *priv)
{
struct device_node *ep;
@@ -833,9 +873,9 @@ static int rcsi2_parse_dt(struct rcar_csi2 *priv)
return ret;
}

- priv->asd.match.fwnode =
- fwnode_graph_get_remote_endpoint(of_fwnode_handle(ep));
- priv->asd.match_type = V4L2_ASYNC_MATCH_FWNODE;
+ priv->asd.match.custom.match = &rcsi2_asd_match;
+ priv->asd.match.custom.priv = priv;
+ priv->asd.match_type = V4L2_ASYNC_MATCH_CUSTOM;

of_node_put(ep);

--
2.20.1


2020-03-15 10:31:47

by Laurent Pinchart

[permalink] [raw]
Subject: Re: [PATCH 2/2] media: rcar-csi2: Let the driver handle fwnode matching using match_custom callback

Hi Prabhakar,

Thank you for the patch.

On Sun, Mar 15, 2020 at 10:27:24AM +0000, Lad Prabhakar wrote:
> The rcar-csi2 driver uses the v4l2-async framework to do endpoint matching
> instead of node matching. This is needed as it needs to work with the
> adv748x driver which register it self in v4l2-async using endpoints
> instead of nodes. The reason for this is that from a single DT node it
> creates multiple subdevices, one for each endpoint.
>
> But when using subdevs which register itself in v4l2-async using nodes,
> the rcar-csi2 driver failed to find the matching endpoint because the
> match.fwnode was pointing to remote endpoint instead of remote parent
> port.
>
> This commit adds support in rcar-csi2 driver to handle both the cases
> where subdev registers in v4l2-async using endpoints/nodes, by using
> match_type as V4L2_ASYNC_MATCH_CUSTOM and implementing the match()
> callback to compare the fwnode of either remote/parent.

This is not the way to go. The v4l2-async framework needs to be fixed
instead, so that fwnode match will do the right thing automatically
regardless of whether the node is a device node or and endpoint node.

> Signed-off-by: Lad Prabhakar <[email protected]>
> ---
> drivers/media/platform/rcar-vin/rcar-csi2.c | 46 +++++++++++++++++++--
> 1 file changed, 43 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c b/drivers/media/platform/rcar-vin/rcar-csi2.c
> index faa9fb23a2e9..1bbf05e9f025 100644
> --- a/drivers/media/platform/rcar-vin/rcar-csi2.c
> +++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
> @@ -808,6 +808,46 @@ static int rcsi2_parse_v4l2(struct rcar_csi2 *priv,
> return 0;
> }
>
> +static bool rcsi2_asd_match(struct v4l2_subdev *sd,
> + struct v4l2_async_subdev *asd)
> +{
> + struct rcar_csi2 *priv = (struct rcar_csi2 *)asd->match.custom.priv;
> + struct fwnode_handle *remote_endpoint;
> + struct fwnode_handle *subdev_endpoint;
> + struct device_node *np;
> + bool matched = false;
> +
> + np = of_graph_get_endpoint_by_regs(priv->dev->of_node, 0, 0);
> + if (!np) {
> + dev_err(priv->dev, "Not connected to subdevice\n");
> + return matched;
> + }
> +
> + remote_endpoint =
> + fwnode_graph_get_remote_endpoint(of_fwnode_handle(np));
> + if (!remote_endpoint) {
> + dev_err(priv->dev, "Failed to get remote endpoint\n");
> + of_node_put(np);
> + return matched;
> + }
> + of_node_put(np);
> +
> + if (sd->fwnode != dev_fwnode(sd->dev)) {
> + if (remote_endpoint == sd->fwnode)
> + matched = true;
> + } else {
> + subdev_endpoint =
> + fwnode_graph_get_next_endpoint(dev_fwnode(sd->dev), NULL);
> + if (remote_endpoint == subdev_endpoint)
> + matched = true;
> + fwnode_handle_put(subdev_endpoint);
> + }
> +
> + fwnode_handle_put(remote_endpoint);
> +
> + return matched;
> +}
> +
> static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> {
> struct device_node *ep;
> @@ -833,9 +873,9 @@ static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> return ret;
> }
>
> - priv->asd.match.fwnode =
> - fwnode_graph_get_remote_endpoint(of_fwnode_handle(ep));
> - priv->asd.match_type = V4L2_ASYNC_MATCH_FWNODE;
> + priv->asd.match.custom.match = &rcsi2_asd_match;
> + priv->asd.match.custom.priv = priv;
> + priv->asd.match_type = V4L2_ASYNC_MATCH_CUSTOM;
>
> of_node_put(ep);
>

--
Regards,

Laurent Pinchart

2020-03-15 12:13:08

by Lad, Prabhakar

[permalink] [raw]
Subject: Re: [PATCH 2/2] media: rcar-csi2: Let the driver handle fwnode matching using match_custom callback

Hi Laurent,

Thank you for the quick review.

On Sun, Mar 15, 2020 at 10:30 AM Laurent Pinchart
<[email protected]> wrote:
>
> Hi Prabhakar,
>
> Thank you for the patch.
>
> On Sun, Mar 15, 2020 at 10:27:24AM +0000, Lad Prabhakar wrote:
> > The rcar-csi2 driver uses the v4l2-async framework to do endpoint matching
> > instead of node matching. This is needed as it needs to work with the
> > adv748x driver which register it self in v4l2-async using endpoints
> > instead of nodes. The reason for this is that from a single DT node it
> > creates multiple subdevices, one for each endpoint.
> >
> > But when using subdevs which register itself in v4l2-async using nodes,
> > the rcar-csi2 driver failed to find the matching endpoint because the
> > match.fwnode was pointing to remote endpoint instead of remote parent
> > port.
> >
> > This commit adds support in rcar-csi2 driver to handle both the cases
> > where subdev registers in v4l2-async using endpoints/nodes, by using
> > match_type as V4L2_ASYNC_MATCH_CUSTOM and implementing the match()
> > callback to compare the fwnode of either remote/parent.
>
> This is not the way to go. The v4l2-async framework needs to be fixed
> instead, so that fwnode match will do the right thing automatically
> regardless of whether the node is a device node or and endpoint node.
>
OK, so moving forward should the v4l2-async do strictly endpoint
matching only or
both nodes/endpoints. fwnode in all the bridge drivers be replaced to
remote endpoints ?

Looking at the adv7604 its registered as node to v4l2-async which can
have upto 3 endpoints,
adv748x is the single driver which registers itself as endpoint to
v4l2-async, and rest of the
other subdevices have single endpoint and are registered as node to
v4l2-async. How would you
suggest to handle these cases.

Cheers,
--Prabhakar Lad

> > Signed-off-by: Lad Prabhakar <[email protected]>
> > ---
> > drivers/media/platform/rcar-vin/rcar-csi2.c | 46 +++++++++++++++++++--
> > 1 file changed, 43 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > index faa9fb23a2e9..1bbf05e9f025 100644
> > --- a/drivers/media/platform/rcar-vin/rcar-csi2.c
> > +++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > @@ -808,6 +808,46 @@ static int rcsi2_parse_v4l2(struct rcar_csi2 *priv,
> > return 0;
> > }
> >
> > +static bool rcsi2_asd_match(struct v4l2_subdev *sd,
> > + struct v4l2_async_subdev *asd)
> > +{
> > + struct rcar_csi2 *priv = (struct rcar_csi2 *)asd->match.custom.priv;
> > + struct fwnode_handle *remote_endpoint;
> > + struct fwnode_handle *subdev_endpoint;
> > + struct device_node *np;
> > + bool matched = false;
> > +
> > + np = of_graph_get_endpoint_by_regs(priv->dev->of_node, 0, 0);
> > + if (!np) {
> > + dev_err(priv->dev, "Not connected to subdevice\n");
> > + return matched;
> > + }
> > +
> > + remote_endpoint =
> > + fwnode_graph_get_remote_endpoint(of_fwnode_handle(np));
> > + if (!remote_endpoint) {
> > + dev_err(priv->dev, "Failed to get remote endpoint\n");
> > + of_node_put(np);
> > + return matched;
> > + }
> > + of_node_put(np);
> > +
> > + if (sd->fwnode != dev_fwnode(sd->dev)) {
> > + if (remote_endpoint == sd->fwnode)
> > + matched = true;
> > + } else {
> > + subdev_endpoint =
> > + fwnode_graph_get_next_endpoint(dev_fwnode(sd->dev), NULL);
> > + if (remote_endpoint == subdev_endpoint)
> > + matched = true;
> > + fwnode_handle_put(subdev_endpoint);
> > + }
> > +
> > + fwnode_handle_put(remote_endpoint);
> > +
> > + return matched;
> > +}
> > +
> > static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> > {
> > struct device_node *ep;
> > @@ -833,9 +873,9 @@ static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> > return ret;
> > }
> >
> > - priv->asd.match.fwnode =
> > - fwnode_graph_get_remote_endpoint(of_fwnode_handle(ep));
> > - priv->asd.match_type = V4L2_ASYNC_MATCH_FWNODE;
> > + priv->asd.match.custom.match = &rcsi2_asd_match;
> > + priv->asd.match.custom.priv = priv;
> > + priv->asd.match_type = V4L2_ASYNC_MATCH_CUSTOM;
> >
> > of_node_put(ep);
> >
>
> --
> Regards,
>
> Laurent Pinchart

2020-03-15 12:58:24

by Laurent Pinchart

[permalink] [raw]
Subject: Re: [PATCH 2/2] media: rcar-csi2: Let the driver handle fwnode matching using match_custom callback

Hi Prabhakar,

On Sun, Mar 15, 2020 at 12:10:14PM +0000, Lad, Prabhakar wrote:
> On Sun, Mar 15, 2020 at 10:30 AM Laurent Pinchart wrote:
> > On Sun, Mar 15, 2020 at 10:27:24AM +0000, Lad Prabhakar wrote:
> > > The rcar-csi2 driver uses the v4l2-async framework to do endpoint matching
> > > instead of node matching. This is needed as it needs to work with the
> > > adv748x driver which register it self in v4l2-async using endpoints
> > > instead of nodes. The reason for this is that from a single DT node it
> > > creates multiple subdevices, one for each endpoint.
> > >
> > > But when using subdevs which register itself in v4l2-async using nodes,
> > > the rcar-csi2 driver failed to find the matching endpoint because the
> > > match.fwnode was pointing to remote endpoint instead of remote parent
> > > port.
> > >
> > > This commit adds support in rcar-csi2 driver to handle both the cases
> > > where subdev registers in v4l2-async using endpoints/nodes, by using
> > > match_type as V4L2_ASYNC_MATCH_CUSTOM and implementing the match()
> > > callback to compare the fwnode of either remote/parent.
> >
> > This is not the way to go. The v4l2-async framework needs to be fixed
> > instead, so that fwnode match will do the right thing automatically
> > regardless of whether the node is a device node or and endpoint node.
>
> OK, so moving forward should the v4l2-async do strictly endpoint
> matching only or both nodes/endpoints. fwnode in all the bridge
> drivers be replaced to remote endpoints ?

Long term I think everything should use endpoint matching, but to get
there we shouldn't transition all drivers in one go. I've submitted a
patch to v4l2-async that I believe will fix your problem and allow for a
smooth transition. Could you give it a try ?

> Looking at the adv7604 its registered as node to v4l2-async which can
> have upto 3 endpoints, adv748x is the single driver which registers
> itself as endpoint to v4l2-async, and rest of the other subdevices
> have single endpoint and are registered as node to v4l2-async. How
> would you suggest to handle these cases.
>
> > > Signed-off-by: Lad Prabhakar <[email protected]>
> > > ---
> > > drivers/media/platform/rcar-vin/rcar-csi2.c | 46 +++++++++++++++++++--
> > > 1 file changed, 43 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/media/platform/rcar-vin/rcar-csi2.c b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > > index faa9fb23a2e9..1bbf05e9f025 100644
> > > --- a/drivers/media/platform/rcar-vin/rcar-csi2.c
> > > +++ b/drivers/media/platform/rcar-vin/rcar-csi2.c
> > > @@ -808,6 +808,46 @@ static int rcsi2_parse_v4l2(struct rcar_csi2 *priv,
> > > return 0;
> > > }
> > >
> > > +static bool rcsi2_asd_match(struct v4l2_subdev *sd,
> > > + struct v4l2_async_subdev *asd)
> > > +{
> > > + struct rcar_csi2 *priv = (struct rcar_csi2 *)asd->match.custom.priv;
> > > + struct fwnode_handle *remote_endpoint;
> > > + struct fwnode_handle *subdev_endpoint;
> > > + struct device_node *np;
> > > + bool matched = false;
> > > +
> > > + np = of_graph_get_endpoint_by_regs(priv->dev->of_node, 0, 0);
> > > + if (!np) {
> > > + dev_err(priv->dev, "Not connected to subdevice\n");
> > > + return matched;
> > > + }
> > > +
> > > + remote_endpoint =
> > > + fwnode_graph_get_remote_endpoint(of_fwnode_handle(np));
> > > + if (!remote_endpoint) {
> > > + dev_err(priv->dev, "Failed to get remote endpoint\n");
> > > + of_node_put(np);
> > > + return matched;
> > > + }
> > > + of_node_put(np);
> > > +
> > > + if (sd->fwnode != dev_fwnode(sd->dev)) {
> > > + if (remote_endpoint == sd->fwnode)
> > > + matched = true;
> > > + } else {
> > > + subdev_endpoint =
> > > + fwnode_graph_get_next_endpoint(dev_fwnode(sd->dev), NULL);
> > > + if (remote_endpoint == subdev_endpoint)
> > > + matched = true;
> > > + fwnode_handle_put(subdev_endpoint);
> > > + }
> > > +
> > > + fwnode_handle_put(remote_endpoint);
> > > +
> > > + return matched;
> > > +}
> > > +
> > > static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> > > {
> > > struct device_node *ep;
> > > @@ -833,9 +873,9 @@ static int rcsi2_parse_dt(struct rcar_csi2 *priv)
> > > return ret;
> > > }
> > >
> > > - priv->asd.match.fwnode =
> > > - fwnode_graph_get_remote_endpoint(of_fwnode_handle(ep));
> > > - priv->asd.match_type = V4L2_ASYNC_MATCH_FWNODE;
> > > + priv->asd.match.custom.match = &rcsi2_asd_match;
> > > + priv->asd.match.custom.priv = priv;
> > > + priv->asd.match_type = V4L2_ASYNC_MATCH_CUSTOM;
> > >
> > > of_node_put(ep);
> > >

--
Regards,

Laurent Pinchart