Re: [PATCH RESEND v6 4/5] mfd: axp20x: Add support for AXP318W PMIC

[Jerome Brunet <jbrunet@baylibre.com>]

On jeu. 27 août 2026 at 14:42, Lee Jones <lee@kernel.org> wrote:

> On Fri, 14 Aug 2026, Jerome Brunet wrote:
>
>> On dim. 13 janv. 2008 at 11:56, Lee Jones <lee@kernel.org> wrote:
>> 
>> > On Thu, 13 Aug 2026, Jerome Brunet wrote:
>> >
>> >> On sam. 12 janv. 2008 at 23:40, sashiko-bot@kernel.org wrote:
>> >> 
>> >> > Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
>> >> > - [High] The corresponding regulator driver lacks support for
>> >> > `AXP318_ID`, causing the `axp20x-regulator` child device probe to fail.
>> >> 
>> >> Funny sashiko did not pick up on this when v6 was first sent. This is an
>> >> ordering problem. The regulator part should be applied before the mfd
>> >> one.
>> >> 
>> >> I could reorder here but there is not much point since the 2 changes will
>> >> land in different subsystems but it does bring a valid question:
>> >> 
>> >> Lee, Mark, are you OK with picking the patches indended for your
>> >> subsystem or would you prefer that I split the series, regulator on one
>> >> side, mfd on the other (with a dep on the regulator one, of course) ?
>> >
>> > Does this break existing devices or simply delay support for this new one?
>> 
>> Existing device are fine and untouched.
>> 
>> I've just checked and if someone tried to run without PATCH 5 (regulators),
>> the main axp driver loads fine but the regulator part gives this
>> 
>> axp20x-regulator axp20x-regulator.0: Unsupported AXP variant: 8
>
> Can you not enforce the deps with Kconfig?

This is not a build problem so I do not think Kconfig will change
anything here. There is no specific config per chip.

It is just a runtime ordering matter and not a problem. regulator and
mfd are split so at some point the support is incomplete.

* MFD first, it is fine it just cant instantiate the regulator until it
  lands. it is (mostly) useless without it, which is fine.
* Regulator first. regulator will never be instanciate until MFD support
  lands, just as useless.

IMO there is nothing to do here.

>
>> Seems sane enough if the regulator part is not present (and no board
>> using it has landed yet)
>> 
>> To completly avoid this, the ID would need to be in its own patch,
>> then the regulators, finally the MFD. It seems like an unecessary churn
>> but I can do that if you prefer.
>
> -- 
> Lee Jones

-- 
Jerome