Re: [PATCH 1/3] gpio: tqmx86: really make IRQ optional

[Andy Shevchenko]

On Thu, Jun 24, 2021 at 4:44 PM Matthias Schiffer
<matthias.schiffer@xxxxxxxxxxxxxxx> wrote:
> On Wed, 2021-03-31 at 17:03 +0300, Andy Shevchenko wrote:
> > On Wed, Mar 31, 2021 at 4:36 PM Matthias Schiffer
> > <matthias.schiffer@xxxxxxxxxxxxxxx> wrote:
> > > On Wed, 2021-03-31 at 15:39 +0300, Andy Shevchenko wrote:
> > > > On Wed, Mar 31, 2021 at 3:37 PM Matthias Schiffer
> > > > <matthias.schiffer@xxxxxxxxxxxxxxx> wrote:
> > > > > On Wed, 2021-03-31 at 15:29 +0300, Andy Shevchenko wrote:

...

> > > > > I don't understand which part of the code is dead now. I assume the
> > > > > `return irq` case is still useful for unexpected errors, or things like
> > > > > EPROBE_DEFER? I'm not sure if EPROBE_DEFER is relevant for this driver,
> > > > > but just ignoring the error code completely doesn't seem right to me.
> > > >
> > > > platform_get_irq() AFAIK won't ever return such a code.
> > > > So, basically your conditional is always false.
> > > >
> > > > I would like to see the code path which makes my comment wrong.
> > >
> > > EPROBE_DEFER appears a few times in platform_get_irq_optional()
> > > (drivers/base/platform.c), but it's possible that this is only relevant
> > > for OF-based platforms and not x86.
> >
> > Ah, okay, that's something I haven't paid attention to.
> >
> > So the root cause of the your case is platform_get_irq_optional|()
> > return code. I'm wondering why it can't return 0 instead of absent
> > IRQ? Perhaps you need to fix it instead of lurking into each caller.
>
> what's the plan here? "driver core: platform: Make
> platform_get_irq_optional() optional" had to be reverted because it
> broke existing users of platform_get_irq_optional(). I'm not convinced
> that a slightly more convenient API is worth going through the trouble
> of fixing them all - I know we don't care much about out-of-tree
> modules, but subtly changing the behaviour of such a function doesn't
> seem like a good idea to me even if we review all in-tree users.

Why? The problem with this function is either naming or semantics.
It should be fixed and that will require revisiting all current users anyway.

> Should I just rebase my patches with the existing ENXIO handing (and
> fix up the other issues that were noted), or do you intend to give the
> platform_get_irq_optional() revamp another try?

I do intend to give another try, but if you want to be independent of
that, just make sure that in any new / revisited user of
platform_get_irq_optional() the 0 is taken into consideration as an
optional case.


-- 
With Best Regards,
Andy Shevchenko