Re: [PATCH v2] pinctrl: bcm63268: add gpio function

[Jonas Gorski <jonas.gorski@xxxxxxxxx>]

On Sat, Jan 4, 2025 at 6:02 PM Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
>
> El sáb, 4 ene 2025 a las 16:44, Álvaro Fernández Rojas
> (<noltari@xxxxxxxxx>) escribió:
> >
> > El vie, 3 ene 2025 a las 4:04, Kyle Hendry (<kylehendrydev@xxxxxxxxx>) escribió:
> > >
> > > On 2024-12-30 08:42, Jonas Gorski wrote:
> > > > Hi,
> > > >
> > > > On Tue, Dec 24, 2024 at 11:41 AM Álvaro Fernández Rojas
> > > > <noltari@xxxxxxxxx> wrote:
> > > >> From: Kyle Hendry <kylehendrydev@xxxxxxxxx>
> > > >>
> > > >> There is no guarantee that the bootloader will leave the pin configuration
> > > >> in a known default state, so pinctrl needs to be explicitly set in some
> > > >> cases. This patch adds a gpio function for drivers that need it, i.e.
> > > >> gpio-leds.
> > > >>
> > > >> Signed-off-by: Kyle Hendry <kylehendrydev@xxxxxxxxx>
> > > >> Signed-off-by: Álvaro Fernández Rojas <noltari@xxxxxxxxx>
> > > >
> > > > bcm63268-pinctrl implements pinmux_ops::gpio_request_enable(), which
> > > > should automatically set any requested GPIO pin to the GPIO function,
> > > > so explicitly requesting the gpio function for a pin should not be
> > > > needed. Is this not enough?
> > > >
> > > > Best Regards,
> > > > Jonas
> > > >
> > > >
> > > I assumed that as well, but nothing I tried worked with gpio-leds.
> > > gpiochip_generic_request() does call gpio_request_enable(), but gpio-leds
> > > uses devm_fwnode_gpiod_get() which looks like a different code path. The
> > > only place it seems to be configuring the gpio is in create_gpio_led()
> > > where it needs a pinctl node in the device tree. That's just my reading
> > > of the code, though. I haven't set up a ftrace to verify it.
> > >
> > > Best Regards,
> > > Kyle
> >
> > As Kyle pointed out it's not enough and gpio_request_enable() doesn't
> > get called from gpio-leds.
> > I will try to investigate this and report back.
> >
> > Best regards,
> > Álvaro.
>
> I've performed some tests and so far I haven't been able to get
> bcm63268_gpio_request_enable() called.
>
> Even exporting a gpio from userspace doesn't trigger that function,
> which seems strange because in that case we won't have a "gpio"
> function on the pinctrl for that gpio, even with this patch...

On a first glance it looks like that regmap-gpio does not populate
gpio_chip::request and gpio_chip:free, so the gpio subsystem is not
aware that it needs to request anything. Not sure if just using
gpiolib_generic_request() / gpiolib_generic_free() is fine. Currently
on vacation, so no hardware to test anything at hand ;-)

Best Regards,
Jonas