Hi Maxime,

On Fri, 25 Aug 2017 15:41:14 +0200
Maxime Ripard [off-list ref] wrote:

Hi Yong,

On Wed, Aug 23, 2017 at 10:32:16AM +0800, Yong wrote:

quoted

quoted

quoted

quoted

quoted

+static int sun6i_graph_notify_complete(struct v4l2_async_notifier *notifier)
+{
+	struct sun6i_csi *csi =
+			container_of(notifier, struct sun6i_csi, notifier);
+	struct sun6i_graph_entity *entity;
+	int ret;
+
+	dev_dbg(csi->dev, "notify complete, all subdevs registered\n");
+
+	/* Create links for every entity. */
+	list_for_each_entry(entity, &csi->entities, list) {
+		ret = sun6i_graph_build_one(csi, entity);
+		if (ret < 0)
+			return ret;
+	}
+
+	/* Create links for video node. */
+	ret = sun6i_graph_build_video(csi);
+	if (ret < 0)
+		return ret;

Can you elaborate a bit on the difference between a node parsed with
_graph_build_one and _graph_build_video? Can't you just store the
remote sensor when you build the notifier, and reuse it here?

There maybe many usercases:
1. CSI->Sensor.
2. CSI->MIPI->Sensor.
3. CSI->FPGA->Sensor1
            ->Sensor2.
FPGA maybe some other video processor. FPGA, MIPI, Sensor can be
registered as v4l2 subdevs. We do not care about the driver code
of them. But they should be linked together here.

So, the _graph_build_one is used to link CSI port and subdevs. 
_graph_build_video is used to link CSI port and video node.

So the graph_build_one is for the two first cases, and the
_build_video for the latter case?

No. 
The _graph_build_one is used to link the subdevs found in the device 
tree. _build_video is used to link the closest subdev to video node.
Video node is created in the driver, so the method to get it's pad is
diffrent to the subdevs.

Sorry for being slow here, I'm still not sure I get it.

In summary, both the sun6i_graph_build_one and sun6i_graph_build_video
will iterate over each endpoint, will retrieve the remote entity, and
will create the media link between the CSI pad and the remote pad.

As far as I can see, there's basically two things that
sun6i_graph_build_one does that sun6i_graph_build_video doesn't:
  - It skips all the links that would connect to one of the CSI sinks
  - It skips all the links that would connect to a remote node that is
    equal to the CSI node.

I assume the latter is because you want to avoid going in an infinite
loop when you would follow one of the CSI endpoint (going to the
sensor), and then follow back the same link in the opposite
direction. Right?

Not exactly. But any way, some code is true redundant here. I will 
make some improve.

I'm confused about the first one though. All the pads you create in
your driver are sink pads, so wouldn't that skip all the pads of the
CSI nodes?

Also, why do you iterate on all the CSI endpoints, when there's only
of them? You want to anticipate the future binding for devices with
multiple channels?

quoted

quoted

If so, you should take a look at the last iteration of the
subnotifiers rework by Nikas S?derlund (v4l2-async: add subnotifier
registration for subdevices).

It allows subdevs to register notifiers, and you don't have to build
the graph from the video device, each device and subdev can only care
about what's next in the pipeline, but not really what's behind it.

That would mean in your case that you can only deal with your single
CSI pad, and whatever subdev driver will use it care about its own.

Do you mean the subdevs create pad link in the notifier registered by
themself ?

Yes.

Thanks!
Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

Thanks,
Yong