[PATCH v4 00/10] gpio: improve support for shared GPIOs

[Bartosz Golaszewski]

On Fri, Jan 9, 2026 at 3:41 PM Michael Walle <mwalle at kernel.org> wrote:
>
> > I am aware of the issue where
> > false-positive shared GPIOs are detected, I posted a fix for that too.
> > Without logs, I can't really tell if that's the case for you though.
> > :(
>
> What I mean is, GPIO_SHARED is automatically selected by ARCH_QCOM (and
> probably others in the future). But GPIO_SHARED_PROXY is
> selectable by the user and it can be deselected. But you cannot
> specifically deactivate GPIO_SHARED. So there is no way to go back
> to the former hacky shared gpio handling; which might be intended :)
>
> If so, shouldn't be GPIO_SHARED_PROXY be either y or m if GPIO_SHARED=y ?
> I.e. don't allow it to be deselected.
>

Fair point. I'll send a patch.

> >>
> >> This broke my board (using the arm64 defconfig, works without
> >> GPIO_SHARED of course). I'm seeing two issues here with my board
> >> (arch/arm64/boot/dts/ti/k3-am67a-kontron-sa67*):
> >>
> >>  (1) It's GPIO controller (gpio-davinci) doesn't support
> >>      .get_direction so I'm getting ENOTSUPP during probing of the
> >>      (some?) shared GPIOs.
> >>
> >
> > Unless this board really shares GPIOs, it may be due to the
> > false-positives that will be fixed by this patch[1]. If you enable
>
> Yeah this board shares an enable GPIO for two regulators
> (regulator-5 and regulator-6,
> arch/arm64/boot/dts/ti/k3-am67a-kontron-sa67-base.dts).
>
> > CONFIG_DEBUG_GPIO and post the kernel log, I'll be able to tell for
> > sure.
>
> See end of this mail. I've applied the mentioned patch.
>

Yes, I see there are indeed shared pins.

> > Though thanks for bringing this to my attention as I now see there is
> > indeed an issue when the proxied chip doesn't support .get_direction()
> > as well as a duplicated check in GPIO core. I'll fix it too.
> >
> >>  (2) GPIO_SHARED_PROXY is default m in the defconfig, but I need the
> >>      pins for the root filesystem medium, i.e. the SD card regulators.
> >>
> >
> > I'll take care of this is you confirm, the issue persists even with patch[1].
>
> Not sure this is still valid. Because I've just learned that
> apparently, the arm64 Image shall be made smaller and thus "need
> a driver for rootfs" isn't a valid reason for =y anymore.
>

Is a switch to disable shared GPIO management entirely via Kconfig
(depending on CONFIG_EXPERT) good enough for you?

Bart