On Mon, Mar 15, 2021 at 1:50 PM Andy Shevchenko
[off-list ref] wrote:

On Mon, Mar 15, 2021 at 12:16:26PM +0200, Andy Shevchenko wrote:

quoted

On Mon, Mar 15, 2021 at 10:01:47AM +0100, Bartosz Golaszewski wrote:

quoted

On Fri, Mar 5, 2021 at 1:03 PM Andy Shevchenko
[off-list ref] wrote:

quoted

Unfortunately while this may fix the particular use-case on STM32, it
breaks all other users as the 'gpio-line-names' property doesn't live
on dev_fwnode(&gdev->dev) but on dev_fwnode(chip->parent).

How about we first look for this property on the latter and only if
it's not present descend down to the former fwnode?

Oops, I have tested on x86 and it worked the same way.

Lemme check this, but I think the issue rather in ordering when we apply fwnode
to the newly created device and when we actually retrieve gpio-line-names
property.

Hmm... I can't see how it's possible can be. Can you provide a platform name
and pointers to the DTS that has been broken by the change?

I noticed it with gpio-mockup (libgpiod tests failed on v5.12-rc3) and
the WiP gpio-sim - but it's the same on most DT platforms. The node
that contains the `gpio-line-names` is the one associated with the
platform device passed to the GPIO driver. The gpiolib then creates
another struct device that becomes the child of that node but it
doesn't copy the parent's properties to it (nor should it).

Every driver that reads device properties does it from the parent
device, not the one in gdev - whether it uses of_, fwnode_ or generic
device_ properties.

Bartosz