Hi Laurent,
On 4/14/20 4:47 PM, Steve Longerbeam wrote:
Hi Laurent,
On 4/14/20 4:16 PM, Laurent Pinchart wrote:
Hi Steve,
Thank you for the patch.
On Tue, Mar 03, 2020 at 03:42:48PM -0800, Steve Longerbeam wrote:
Implement a notifier bound op to register media links from the remote
sub-device's source pad(s) to the video-mux sink pad(s).
Signed-off-by: Steve Longerbeam <slongerbeam@xxxxxxxxx>
---
Changes in v3:
- this version does the work inline. The previous version called
a media_create_fwnode_links() which is removed in v3.
---
drivers/media/platform/video-mux.c | 92
++++++++++++++++++++++++++++++
1 file changed, 92 insertions(+)
diff --git a/drivers/media/platform/video-mux.c
b/drivers/media/platform/video-mux.c
index f446ada82176..3991b1ea671c 100644
--- a/drivers/media/platform/video-mux.c
+++ b/drivers/media/platform/video-mux.c
@@ -36,6 +36,12 @@ static const struct v4l2_mbus_framefmt
video_mux_format_mbus_default = {
.field = V4L2_FIELD_NONE,
};
+static inline struct video_mux *
+notifier_to_video_mux(struct v4l2_async_notifier *n)
+{
+ return container_of(n, struct video_mux, notifier);
+}
+
static inline struct video_mux *v4l2_subdev_to_video_mux(struct
v4l2_subdev *sd)
{
return container_of(sd, struct video_mux, subdev);
@@ -360,6 +366,90 @@ static const struct v4l2_subdev_ops
video_mux_subdev_ops = {
.video = &video_mux_subdev_video_ops,
};
+static int video_mux_notify_bound(struct v4l2_async_notifier
*notifier,
+ struct v4l2_subdev *sd,
+ struct v4l2_async_subdev *asd)
+{
+ struct video_mux *vmux = notifier_to_video_mux(notifier);
+ struct fwnode_handle *vmux_fwnode = dev_fwnode(vmux->subdev.dev);
+ struct fwnode_handle *sd_fwnode = dev_fwnode(sd->dev);
+ struct fwnode_handle *vmux_ep;
There doesn't seem to be anything in this function that is specific to
the video_mux driver. I think it would make sense to turn it into a
generic helper that creates links between two subdevs based on their
fwnode connections.
Agreed, in fact I wrote imx_media_create_fwnode_pad_links(src_sd,
sink_sd) (patch 8 in this series), and it is completely generic, it
could simply be renamed v4l2_create_fwnode_pad_links() and moved to core.
I also wonder if it wouldn't be more efficient to create links at
complete() time instead of bound() time, in which case the helper would
create all links for a given subdevice, not links between two specific
subdevices (or maybe all links for a given pad direction, to be able to
remove the existing link check below).
It looks like that should be possible. The bound sub-devices at
complete() time are available in the notifier->done list. I'll start
looking at that for v5.
Actually for sub-devices that won't work. The .complete() callback is
only called on the root notifier, not on subdev notifiers. I don't think
it's a big deal, it works fine to create links from one source subdev at
a time in .bound(), and I don't see it as much of a efficiency hit.
Steve
+
+ fwnode_graph_for_each_endpoint(vmux_fwnode, vmux_ep) {
+ struct fwnode_handle *remote_ep, *sd_ep;
+ struct media_pad *src_pad, *sink_pad;
+ struct fwnode_endpoint fwep;
+ int src_idx, sink_idx, ret;
+ bool remote_ep_belongs;
+
+ ret = fwnode_graph_parse_endpoint(vmux_ep, &fwep);
+ if (ret)
+ continue;
+
+ /* only create links to the vmux sink pads */
+ if (fwep.port >= vmux->subdev.entity.num_pads - 1)
+ continue;
+
+ sink_idx = fwep.port;
+ sink_pad = &vmux->subdev.entity.pads[sink_idx];
+
+ remote_ep = fwnode_graph_get_remote_endpoint(vmux_ep);
+ if (!remote_ep)
+ continue;
+
+ /*
+ * verify that this remote endpoint is owned by the
+ * sd, in case the sd does not check for that in its
+ * .get_fwnode_pad operation or does not implement it.
+ */
+ remote_ep_belongs = false;
+ fwnode_graph_for_each_endpoint(sd_fwnode, sd_ep) {
+ if (sd_ep == remote_ep) {
+ remote_ep_belongs = true;
+ fwnode_handle_put(sd_ep);
+ break;
+ }
+ }
+ if (!remote_ep_belongs)
+ continue;
+
+ src_idx = media_entity_get_fwnode_pad(&sd->entity, remote_ep,
+ MEDIA_PAD_FL_SOURCE);
+ fwnode_handle_put(remote_ep);
+
+ if (src_idx < 0)
+ continue;
+
+ src_pad = &sd->entity.pads[src_idx];
+
+ /* skip this link if it already exists */
+ if (media_entity_find_link(src_pad, sink_pad))
+ continue;
Have you encountered this in practice ? If we only create links for sink
pads this shouldn't happen.
+
+ ret = media_create_pad_link(&sd->entity, src_idx,
+ &vmux->subdev.entity,
+ sink_idx, 0);
+ if (ret) {
+ dev_err(vmux->subdev.dev,
+ "%s:%d -> %s:%d failed with %d\n",
+ sd->entity.name, src_idx,
+ vmux->subdev.entity.name, sink_idx, ret);
+ fwnode_handle_put(vmux_ep);
+ return ret;
+ }
+
+ dev_dbg(vmux->subdev.dev, "%s:%d -> %s:%d\n",
+ sd->entity.name, src_idx,
+ vmux->subdev.entity.name, sink_idx);
+ }
+
+ return 0;
+}
+
+static const struct v4l2_async_notifier_operations
video_mux_notify_ops = {
+ .bound = video_mux_notify_bound,
+};
+
static int video_mux_async_register(struct video_mux *vmux,
unsigned int num_input_pads)
{
@@ -397,6 +487,8 @@ static int video_mux_async_register(struct
video_mux *vmux,
}
}
+ vmux->notifier.ops = &video_mux_notify_ops;
+
ret = v4l2_async_subdev_notifier_register(&vmux->subdev,
&vmux->notifier);
if (ret)
