Re: [PATCH 15/18] serial: max310x: replace ENOTSUPP with preferred EOPNOTSUPP (checkpatch)

[Kent Gibson <warthog618@gmail.com>]

On Thu, Jan 18, 2024 at 10:59:34AM +0200, Andy Shevchenko wrote:
> On Thu, Jan 18, 2024 at 1:59 AM Hugo Villeneuve <hugo@hugovil.com> wrote:
> > On Thu, 18 Jan 2024 01:24:11 +0200
> > Andy Shevchenko <andy.shevchenko@gmail.com> wrote:
> > > On Thu, Jan 18, 2024 at 12:39 AM Hugo Villeneuve <hugo@hugovil.com> wrote:
>
> ...
>
> > > > Fixes the following checkpatch warning:
> > > >
> > > >     WARNING: ENOTSUPP is not a SUSV4 error code, prefer EOPNOTSUPP
> > >
> > > NAK.
> > > It's a false positive.
> > >
> > > > According to include/linux/errno.h, ENOTSUPP is
> > > > "Defined for the NFSv3 protocol", so replace it with preferred EOPNOTSUPP.
> > >
> > > The GPIO subsystem uses this internal error code internally. User
> > > space won't get it, so users may not see this one.
> >
> > Hi Andy,
> > I will drop the patch then.
> >
> > What about adding a comment to prevent future fixes?
> >
> > -               return -ENOTSUPP;
> > +               return -ENOTSUPP; /*
> > +                                  * ENOTSUPP is used for backward compatibility
> > +                                  * with GPIO subsystem.
> > +                                  */
>
> It's kinda useless to add it to a single (GPIO) driver.
> Rather it needs to be mentioned somewhere between
> https://www.kernel.org/doc/html/latest/driver-api/gpio/index.html.
>
> +Cc: Kent, Bart. It seems we have a handful of drivers violating this
> (basically following what checkpatch says) and GPIO not documenting
> this specific error code and its scope. Did I miss anything?
>

You are correct - the GPIO subsystem is expecting ENOTSUPP if the config
is not supported.  In some cases it absorbs the failure or emulates the
feature instead (open drain/source, debounce). Returning EOPNOTSUPP
would be unfortunate, so checkpatch is not being helpful here.

And don't get me started on the gpio_chip interface contract being too
vague.

There are a handful of ways this could be addressed (documentation,
checkpatch, handle either, switch to EOPNOTSUPP, ... or some combination),
but making that call is definitely in Bart's court.

Cheers,
Kent.