Hi Sylwester,


On 27-06-2017 21:34, Sylwester Nawrocki wrote:

Hi Jose,

On 06/27/2017 10:43 AM, Jose Abreu wrote:

quoted

On 25-06-2017 22:13, Sylwester Nawrocki wrote:

quoted

On 06/20/2017 07:26 PM, Jose Abreu wrote:

quoted

This is an initial submission for the Synopsys Designware HDMI RX
Controller Driver. This driver interacts with a phy driver so that
a communication between them is created and a video pipeline is
configured.
Signed-off-by: Jose Abreu <redacted>
+static int dw_hdmi_phy_init(struct dw_hdmi_dev *dw_dev)
+{
+	struct dw_phy_pdata *phy = &dw_dev->phy_config;
+	const struct of_device_id *of_id;
+	struct of_dev_auxdata lookup;

	struct of_dev_auxdata lookup = { };

You could initialize to 0 here and...

quoted

+	struct device_node *child;
+	const char *drvname;
+	int ret;
+
+	child = dw_hdmi_get_phy_of_node(dw_dev, &of_id);
+	if (!child || !of_id || !of_id->data) {
+		dev_err(dw_dev->dev, "no supported phy found in DT\n");
+		return -EINVAL;
+	}
+
+	drvname = of_id->data;
+	phy->funcs = &dw_hdmi_phy_funcs;
+	phy->funcs_arg = dw_dev;
+
+	lookup.compatible = (char *)of_id->compatible;
+	lookup.phys_addr = 0x0;
+	lookup.name = NULL;

...drop these two assignments.

Ok.

quoted

quoted

+	lookup.platform_data = phy;
+
+	request_module(drvname);

I'd say this request_module() is not needed when you use the v4l2-async
subnotifiers and the modules are properly installed in the file system.
I might be missing something though.

Hmm, well I didn't actually test without request_module but I
think its needed, otherwise I would have to do:

modprobe phy_module
modprobe controller_module

With request_module I just have to do:

modprobe controller_module

If you are sure you need it I'm not against that.  But assuming you have udev 
in your system it should also work like this, without request_module():

1. modprobe controller_module -> phy device is created in the kernel, uevent sent
2. udev receives uevent, finds matching module and does modprobe phy_module

Remaining part is as before: phy_module registers the driver which gets matched with 
phy device; probe() is called which registers v4l2 subdev which then is registered
to v4l2_device through the v4l2-async mechanism.

All this assumes udev is running and modules are installed in /lib/modules/$(uname -r).
E.g. there should be your module alias as shown by modinfo phy_module in
/lib/modules/$(uname -r)/modules.alias.

The modules are installed but I think I don't have udev :/ I'm
running this on an embedded platform called ARC AXS and I'm using
buildroot with minimal options.

quoted

quoted

quoted

+	ret = of_platform_populate(dw_dev->of_node, NULL, &lookup, dw_dev->dev);
+	if (ret) {
+		dev_err(dw_dev->dev, "failed to populate phy driver\n");
+		return ret;
+	}
+
+	return 0;
+}
+
+static void dw_hdmi_phy_exit(struct dw_hdmi_dev *dw_dev)
+{
+	of_platform_depopulate(dw_dev->dev);
+}
+static int dw_hdmi_v4l2_notify_complete(struct v4l2_async_notifier *notifier)
+{
+	struct dw_hdmi_dev *dw_dev = notifier_to_dw_dev(notifier);
+	int ret;
+
+	ret = v4l2_device_register_subdev_nodes(&dw_dev->v4l2_dev);

There shouldn't be multiple struct v4l2_device instances, instead we should
have only one created by the main driver. AFAIU, in your case it would be
driver associated with the dw-hdmi-soc DT node.  And normally such a top level
driver creates subdev device nodes when its all required sub-devices are
available.

I think this patch could be useful for you:
https://urldefense.proofpoint.com/v2/url?u=https-3A__patchwork.linuxtv.org_patch_41834&d=DwICaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=WHDsc6kcWAl4i96Vm5hJ_19IJiuxx_p_Rzo2g-uHDKw&m=uNHRQkbP_-z8v5d30KFx9pcPEUhlr4ciWY3ZDAVExTA&s=dB9wpgeP7AJg1eDRty0-RKhq3DY-7J5srIzyVoJey5I&e= 

With that the dw-hdmi-soc driver would have it's v4l2-async notifier's
notify_complete() callback called only when both the hdmi-rx and the
hdmi-phy subdevs are registered.

Yeah, I saw the patches. I just implemented this way because they
are not merged yet, right?

I think these patches will be merged in v4.14-rc1, so together with your driver.
You could apply them locally and indicate that your series depends on them in 
the cover letter.

Ok, will apply them locally and re-test.

quoted

quoted

quoted

+	if (ret) {
+		dev_err(dw_dev->dev, "failed to register subdev nodes\n");
+		return ret;
+	}
+
+	return 0;
+}
+static int dw_hdmi_rx_probe(struct platform_device *pdev)
+{
+	/* V4L2 initialization */
+	sd = &dw_dev->sd;
+	v4l2_subdev_init(sd, &dw_hdmi_sd_ops);
+	strlcpy(sd->name, dev_name(dev), sizeof(sd->name));
+	sd->dev = dev;
+	sd->internal_ops = &dw_hdmi_internal_ops;
+	sd->flags |= V4L2_SUBDEV_FL_HAS_EVENTS;

Don't you also need V4L2_SUBDEV_FL_HAS_DEVNODE flag set?

Ouch. Yes I need otherwise the subdev will not be associated with
the v4l2_device.

This flag indicates that the v4l2 subdev device node (/dev/v4l-subdev?)
should be created for this subdevice.

Ok, will add for controller driver only then: I think for phy
this should not be added because controller is responsible to
manage phy entirely so creating a /dev/ which can be seen by
userspace can lead to confusion, maybe?

Best regards,
Jose Miguel Abreu

---
Regards,
Sylwester