On Wed, Mar 12, 2025 at 11:28:52AM +0000, Sudeep Holla wrote:

On Wed, Mar 12, 2025 at 10:52:23AM +0000, Sudeep Holla wrote:

quoted

On Tue, Mar 11, 2025 at 11:23:12AM +0000, Sudeep Holla wrote:

quoted

On Tue, Mar 11, 2025 at 11:12:45AM +0000, Peng Fan wrote:

quoted

So it is clear that wrong fw_devlink is created, it is because scmi cpufreq device is
created earlier and when device_add, the below logic makes the fwnode pointer points
to scmi cpufreq device.
        if (dev->fwnode && !dev->fwnode->dev) {
                dev->fwnode->dev = dev;
                fw_devlink_link_device(dev);
        }

Thanks, looks like simple way to reproduce the issue. I will give it a try.

I could reproduce but none of my solution solved the problem completely
or properly. And I don't like the DT proposal you came up with. I am
not inclined to just drop this fwnode setting in the scmi devices and
just use of_node.

Sorry for the typo that changes the meaning: s/not/now

I meant "I am now inclined ..", until we figure out a way to make this
work with devlinks properly.

when you have time, please give a look at
https://github.com/MrVan/linux/commit/b500c29cb7f6f32a38b1ed462e333db5a3e301e4

The upper patch was to follow Cristian's and Dan's suggestion in V2[1] to use
a flag SCMI_DEVICE_NO_FWNODE for scmi device.

I could post out the upper patch as V3 if it basically looks no design flaw.
I will drop the pinctrl patch in v3, considering we are first going
to resolve the fw_devlink issue for cpufreq/devfreq.

[1] https://lore.kernel.org/all/Z6SgFGb4Z88v783c@pluto/ (local)

Thanks,
Peng.

-- 
Regards,
Sudeep