Re: [PATCH v5 09/12] gpio: aggregator: handle runtime registration of gpio_desc in gpiochip_fwd

[Bartosz Golaszewski]

On Wed, May 7, 2025 at 5:24 PM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
>
> On Wed, May 7, 2025 at 4:54 PM Thomas Richard
> <thomas.richard@xxxxxxxxxxx> wrote:
> > On 5/7/25 15:24, Andy Shevchenko wrote:
> > > On Wed, May 7, 2025 at 1:10 PM Thomas Richard
> > > <thomas.richard@xxxxxxxxxxx> wrote:
> > >> On 5/7/25 08:34, Andy Shevchenko wrote:
> > >>> On Tue, May 6, 2025 at 6:21 PM Thomas Richard
> > >>> <thomas.richard@xxxxxxxxxxx> wrote:
>
> ...
>
> > >>>> +       /*
> > >>>> +        * get_direction() is called during gpiochip registration, return input
> > >>>> +        * direction if there is no descriptor for the line.
> > >>>> +        */
> > >>>> +       if (!test_bit(offset, fwd->valid_mask))
> > >>>> +               return GPIO_LINE_DIRECTION_IN;
> > >>>
> > >>> Can you remind me why we choose a valid return for invalid line? From
> > >>> a pure code perspective this should return an error.
> > >>
> > >> I reproduced gpiolib behavior. During gpiochip registration, we get the
> > >> direction of all lines. In the case the line is not valid, it is marked
> > >> as input if direction_input operation exists, otherwise it is marked as
> > >> output. [1]
> > >>
> > >> But in fact we could return an error and the core will mark the line as
> > >> input. Maybe ENODEV ?
> > >
> > > I am fine with this error code, but do we have similar cases already
> > > in the kernel? Do they use the same or different error code(s)?
> >
> > I dumped all get_direction() operations in drivers/gpio and
> > drivers/pinctrl and returned values are:
> > - GPIO_LINE_DIRECTION_OUT and GPIO_LINE_DIRECTION_IN (make sense).
> > - -EINVAL (for example [1]).
> > - -EBADE in gpiochip_get_direction() [2].
> > - regmap_read() return code.
> >
> > But from my point of view -EINVAL and -EBADE do not match our case.
>
> Hmm... I believe we need a GPIO maintainer to have a look at this.
>

I went with -EBADE in GPIO core to indicate that the underlying driver
borked and returned an invalid value. I'm not sure if this is the
right one here. I'm not against using -ENODEV.

Bart