Hi Hans,

On Fri, Nov 03, 2023 at 10:05:18AM +0100, Hans Verkuil wrote:

Hi Maxime,

Thank you for posting v3, this time it runs fine on my RPi 4, thank you for
fixing that.

I'll start working on a conformity checker for this.

Awesome :)

quoted

+static int create_hdmi_infoframe_files(struct drm_connector *connector,
+				       struct dentry *parent)
+{
+	int ret;
+
+	ret = CREATE_HDMI_INFOFRAME_FILE(connector, parent, audio);
+	if (ret)
+		return ret;
+
+	ret = CREATE_HDMI_INFOFRAME_FILE(connector, parent, avi);
+	if (ret)
+		return ret;
+
+	ret = CREATE_HDMI_INFOFRAME_FILE(connector, parent, drm);

Hmm, I had to look into the code to figure out that 'drm' stands for
Dynamic Range and Mastering InfoFrame. While entirely correct, it is
also very confusing in the context of the 'drm' subsystem.

I am not quite certain what the best approach is here.

Internally in the drm code it is talking about 'hdr' or 'hdr metadata',
but that's a bit confusing as well since there is also an HDR Dynamic
Metadata Extended InfoFrame defined in CTA-861, even though support for
that is not (yet) implemented in drm.

At minimum there should be a comment in the code explaining what drm
stands for in this context.

One option to consider is renaming this file to hdr_drm, thus indicating
that this is HDR related.

I ended up doing both, thanks for the suggestion

quoted

+	if (ret)
+		return ret;
+
+	ret = CREATE_HDMI_INFOFRAME_FILE(connector, parent, spd);
+	if (ret)
+		return ret;
+
+	ret = CREATE_HDMI_INFOFRAME_FILE(connector, parent, vendor);

There may be multiple vendor specific InfoFrames in the future, so how
would that be handled? Perhaps add a comment here that currently only one
vendor specific InfoFrame is supported, but suggest how to handle multiple
VSIFs in the future.

What would actually be nice (although probably not that easy to fix) is if
the name of the file would be "vendor-XXXXXX' where 'XXXXXX' is the IEEE OUI
number.

I guess it's not entirely clear to me what that would look like. In
order for the framework to create the debugfs files, we would need some
enumeration mechanism (probably through a callback?), and then the
driver would generate the entire content of that file.

Which makes me question whether the framework should be initialised at
all? Maybe the simpler would be to just have drivers maintain their own
debugfs files and storing the content in their own, private, structure.

Maxime