Re: [PATCH 7/9] reset: make the provider of reset-gpios the parent of the reset device

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Wed, Oct 22, 2025 at 02:17:53PM +0200, Bartosz Golaszewski wrote:
> On Wed, Oct 22, 2025 at 10:39 AM Philipp Zabel <p.zabel@pengutronix.de> wrote:
> > On Di, 2025-10-21 at 18:47 +0300, Andy Shevchenko wrote:
> > > On Tue, Oct 21, 2025 at 05:23:33PM +0200, Bartosz Golaszewski wrote:
> > > > On Tue, Oct 21, 2025 at 5:03 PM Andy Shevchenko
> > > > <andriy.shevchenko@linux.intel.com> wrote:
> > > > > On Tue, Oct 21, 2025 at 05:55:02PM +0300, Andy Shevchenko wrote:
> > > > > > On Tue, Oct 21, 2025 at 11:39:41AM +0200, Bartosz Golaszewski wrote:
> > > > > > > On Tue, Oct 21, 2025 at 11:31 AM Philipp Zabel <p.zabel@pengutronix.de> wrote:
> > > > > > > > On Di, 2025-10-21 at 11:27 +0200, Bartosz Golaszewski wrote:

[...]

> > > > > > > > No need to convert all existing drivers right away, but I'd like to see
> > > > > > > > a user that benefits from the conversion.
> > > > > > > >
> > > > > > >
> > > > > > > The first obvious user will be the reset-gpio driver which will see
> > > > > > > its core code simplified as we won't need to cast between OF and
> > > > > > > fwnodes.
> > > > > >
> > > > > > +1 to Bart's work. reset-gpio in current form is useless in all my cases
> > > > > > (it's OF-centric in 2025! We should not do that in a new code).
> > > > > >
> > > > > > More over, conversion to reset-gpio from open coded GPIO APIs is a clear
> > > > > > regression and I want to NAK all those changes (if any already done) for
> > > > > > the discrete components that may be used outside of certainly OF-only niche of
> > > > > > the platforms.
> > > > >
> > > > > To be clear, the conversion that's done while reset-gpio is kept OF-centric.
> > > > > I'm in favour of using it, but we need to make it agnostic.
> > > >
> > > > As of now, the whole reset framework is completely OF-centric, I don't
> > > > know what good blocking any such conversions would bring? I intend to
> > > > convert the reset core but not individual drivers.
> > >
> > > Blocking making new regressions?
> > >
> > > Otherwise as long as reset framework and reset-gpio are agnostic, I'm pretty
> > > much fine with the idea and conversion.
> >
> > I think we might be talking about different "conversions" and different
> > "blocking" here?
> >
> > 1) Conversion of the reset core from of_node to fwnode.
> > 2) Conversion of reset controller drivers from of_node to fwnode.
> > 3) Conversion of consumer drivers from gpiod to reset_control API.
> >
> > My understanding is:
> >
> > Bartosz would like to convert the reset core to fwnode (1) but not
> > convert all the individual reset controller drivers (2). He doesn't
> > like blocking (1) - this statement was partially in reaction to me
> > bringing up a previous attempt that didn't go through.
> >
> > Andy would like to block consumer driver conversions from gpiod to
> > reset_control API (3) while the reset-gpio driver only works on OF
> > platforms.
> >
> > Please correct me if and where I misunderstood.
> 
> I think Andy is afraid that people will convert drivers that are used
> in the fwnode world to reset-gpio which only works with OF. I don't
> think that anyone's trying to do it though.

You are both right about my worries and there is of course the case.
https://patch.msgid.link/1720009575-11677-1-git-send-email-shengjiu.wang@nxp.com

The mentioned change should be reverted.

And this was just found by a couple of minutes of `git log --grep`. I am pretty
sure there are handful of a such wrong patches.

Compare to https://patch.msgid.link/20250815172353.2430981-3-mohammad.rafi.shaik@oss.qualcomm.com
which is done correctly (it doesn't  break old functionality on non-OF platforms).

> > I think fwnode conversion of the reset controller framework core is a
> > good idea, I'd just like to see a use case accompanying the conversion.
> > It seems like enabling the reset-gpio driver to be used on non-OF
> > platforms could be that. Andy, do you have an actual case in mind?
> 
> I'd say converting the reset core to fwnode has merits on its own. We
> should typically use the highest available abstraction layer (which is
> fwnode in this case) unless we absolutely have no choice (for
> instance: using some very OF-specific APIs).
> 
> That being said: the reset-gpio driver will be able to work with any
> firmware node once we do the conversion which is a good first
> use-case.

+1, as I already mentioned I am in favour of this change.

-- 
With Best Regards,
Andy Shevchenko