Re: "gpio-line-names" property - issue with commit 9427ecbed46cc

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Thu, Dec 14, 2017 at 02:37:07PM +0100, Linus Walleij wrote:
> On Thu, Dec 14, 2017 at 12:54 PM, Christophe LEROY
> <christophe.leroy@c-s.fr> wrote:
> 
> > Why is it necessary now to set a parent to the GPIO chip whereas it was not
> > necessary before ?
> (...)
> > Is there an easy modification that can be done to your new function
> > devprop_gpiochip_set_names() in order to have the GPIO drivers work as
> > before ?
> 
> I am also worried about this.
> 
> I have felt that the device property paradigm is too ambitious and assume
> too much about the subtle semantic differences between DT and ACPI
> DSDT. But maybe I'm just whiney.
> 
> Looking forward to good ideas on how to solve this!

I think we can fix this by passing struct fwnode_handle to
devprop_gpiochip_set_names(). Then the existing drivers don't need to be
changed and we update of_gpiochip_add() to call it like:

  devprop_gpiochip_set_names(chip, of_fwnode_handle(chip->of_node));

or so.

> Mika: is is possible to revert this without breaking something else, if we
> need to?

It breaks ACPI "gpio-line-names" users but I think we don't need to
revert if if do what I'm proposing above :)