Re: [PATCH] pinctrl: pinconf-generic: check error value EOPNOTSUPP

[Andy Shevchenko]

On Fri, Apr 05, 2024 at 05:07:00PM +0100, Sudeep Holla wrote:
> On Fri, Apr 05, 2024 at 06:55:34PM +0300, Andy Shevchenko wrote:
> > On Fri, Apr 05, 2024 at 04:47:19PM +0100, Sudeep Holla wrote:

..

> > > Well, I don't agree with that 100% now since this is GPIO/pinmux sub-system
> > > practice only.
> >
> > 	git grep -lw ENOTSUPP
> >
> > utterly disagrees with you.
> >
> 
> /me more confused. Though I haven't dig deeper to chech how many of these
> EOPNOTSUPP uses are intended for userspace.
> 
> $git grep -lw ENOTSUPP | wc -l
> 713
> git grep -lw EOPNOTSUPP | wc -l
> 2946

I (mis?) interpret your words that only GPIO/pin control uses ENOTSUPP internally.

> > > What if we change the source/root error cause(SCMI) in this
> > > case and keep GPIO/pinmux happy today but tomorrow when this needs to be
> > > used in some other subsystem which uses EOPNOTSUPP by default/consistently.
> >
> > This is different case. For that we may shadow error codes with explicit
> > comments.
> 
> Sure as along as that is acceptable.
> 
> > > Now how do we address that then, hence I mentioned I am not 100% in agreement
> > > now while I was before knowing that this is GPIO/pinmux strategy.
> > >
> > > I don't know how to proceed now 🙁.
> >
> > KISS principle? There are only 10+ drivers to fix (I showed a rough list)
> > to use ENOTSUPP instead of 100s+ otherwise.
> 
> Again I assume you are referring to just GPIO/pinmux subsystem right.

Yes, I am.

> As the number of occurrence of EOPNOTSUPP in the kernel overall is quite
> large.

Of course that is out of scope of GPIO/pin control design.

> I was thinking of changing the SCMI error map from EOPNOTSUPP to ENOTSUPP,
> but for now I think it is better to just handle the mapping in the pinmux
> part of SCMI that pinmux subsystem interacts with.

Sure. Wherever you prefer, the only expectation that GPIO / pin control callbacks
will return into GPIO / pin control _core_ ENOTSUPP.

> In future if more
> subsystem expect ENOTSUPP, then we can change it. I hope this aligns with
> KISS principle as we are just fixing for the case that is know to cause
> issue rather than changing all probably regressing and then having to
> fix them all.

Exactly!

> Thanks for the time and explanation.

Thanks for discussion!

-- 
With Best Regards,
Andy Shevchenko