Re: [PATCH 10/10] pinctrl: Add AXP192 pin control driver

[Andy Shevchenko]

On Fri, Jun 17, 2022 at 2:14 PM Aidan MacDonald
<aidanmacdonald.0x0@xxxxxxxxx> wrote:
> Andy Shevchenko <andy.shevchenko@xxxxxxxxx> writes:
> > On Fri, Jun 3, 2022 at 6:29 PM Aidan MacDonald
> > <aidanmacdonald.0x0@xxxxxxxxx> wrote:

...

> >> +       if ((val & reginfo->mask) == (input_muxvals[offset] << (ffs(reginfo->mask) - 1)))
> >> +               return GPIO_LINE_DIRECTION_IN;
> >
> >> +       else
> >
> > Redundant.
> > Also applies for the other similar cases in your code. Note, this is
> > also redundant for 'continue' and 'break' in case of loops.
>
> Sorry, I'm not sure what you're referring to here. The "else"?

Yes.

> I'm missing the generalization.
>
> >> +               return GPIO_LINE_DIRECTION_OUT;

...

> >> +       pctl->chip.to_irq               = axp192_gpio_to_irq;
> >
> > Why a custom method?
>
> The irq chip is part of the mfd device, not the gpio chip. There does
> not seem to be any default implementation for this case so I have to
> provide one. A similar example is gpio-wm8994.
>
> I did notice I'm doing something wrong by calling regmap_irq_get_virq()
> in the probe function, which creates an irq mapping; I think I should be
> doing that in the to_irq() callback like the other drivers do.

It may be done using different approaches, but this part should be
carefully reviewed by GPIO / pin control maintainers.

...

> Ah, sorry, I see that function is deprecated. The documentation points
> to doing this in the device tree instead. So if I understand correctly
> I should follow the example of pinctrl-thunderbay and add gpio-ranges:
>
>     pinctrl0: gpio@0 {
>         compatible = "x-powers,axp192-gpio";
>         gpio-controller;
>         #gpio-cells = <2>;
>         gpio-ranges = <&pinctrl0 0 0 6>;
>     };
>
> which means I'll have to update the gpio DT bindings. I'm guessing the
> callback you mentioned is add_pin_ranges() or of_gpio_ranges_fallback()
> but neither of those seem appropriate in this case. The DT node should
> be good enough.

Sounds good.

-- 
With Best Regards,
Andy Shevchenko