Re: [PATCH] driver core: platform: Rename platform_get_irq_optional() to platform_get_irq_silent()

[Andy Shevchenko]

On Thu, Jan 20, 2022 at 08:57:18AM +0100, Uwe Kleine-König wrote:
> On Wed, Jan 19, 2022 at 08:51:29PM +0200, Andy Shevchenko wrote:
> > On Sat, Jan 15, 2022 at 04:45:39PM +0100, Uwe Kleine-König wrote:
> > > On Fri, Jan 14, 2022 at 03:04:38PM +0200, Andy Shevchenko wrote:
> > > > On Thu, Jan 13, 2022 at 08:43:58PM +0100, Uwe Kleine-König wrote:
> > > > > > It'd certainly be good to name anything that doesn't correspond to one
> > > > > > of the existing semantics for the API (!) something different rather
> > > > > > than adding yet another potentially overloaded meaning.
> > > > > 
> > > > > It seems we're (at least) three who agree about this. Here is a patch
> > > > > fixing the name.
> > > > 
> > > > And similar number of people are on the other side.
> > > 
> > > If someone already opposed to the renaming (and not only the name) I
> > > must have missed that.
> > > 
> > > So you think it's a good idea to keep the name
> > > platform_get_irq_optional() despite the "not found" value returned by it
> > > isn't usable as if it were a normal irq number?
> > 
> > I meant that on the other side people who are in favour of Sergey's patch.
> > Since that I commented already that I opposed the renaming being a standalone
> > change.
> > 
> > Do you agree that we have several issues with platform_get_irq*() APIs?
> > 
> > 1. The unfortunate naming
> 
> unfortunate naming for the currently implemented semantic, yes.

Yes.

> > 2. The vIRQ0 handling: a) WARN() followed by b) returned value 0
> 
> I'm happy with the vIRQ0 handling. Today platform_get_irq() and it's
> silent variant returns either a valid and usuable irq number or a
> negative error value. That's totally fine.

It might return 0.
Actually it seems that the WARN() can only be issued in two cases:
- SPARC with vIRQ0 in one of the array member
- fallback to ACPI for GPIO IRQ resource with index 0

But the latter is bogus, because it would mean a bug in the ACPI code.

The bottom line here is the SPARC case. Anybody familiar with the platform
can shed a light on this. If there is no such case, we may remove warning
along with ret = 0 case from platfrom_get_irq().

> > 3. The specific cookie for "IRQ not found, while no error happened" case
> 
> Not sure what you mean here. I have no problem that a situation I can
> cope with is called an error for the query function. I just do error
> handling and continue happily. So the part "while no error happened" is
> irrelevant to me.

I meant that instead of using special error code, 0 is very much good for
the cases when IRQ is not found. It allows to distinguish -ENXIO from the
low layer from -ENXIO with this magic meaning.

> Additionally I see the problems:
> 
> 4. The semantic as implemented in Sergey's patch isn't better than the
> current one.

I disagree on this. See above on why.

> platform_get_irq*() is still considerably different from
> (clk|gpiod)_get* because the not-found value for the _optional variant
> isn't usuable for the irq case. For clk and gpio I get rid of a whole if
> branch, for irq I only change the if-condition. (And if that change is
> considered good or bad seems to be subjective.)

You are mixing up two things:
 - semantics of the pointer
 - semantics of the cookie

Like I said previously the mistake is in putting an equal sign between apples
and oranges (or in terms of Python, which is a good example here, None and
False objects, where in both case 0 is magic and Python `if X`, `while `X` will
work in the same way, the `typeof(X)` is different semantically).

> For the idea to add a warning to platform_get_irq_optional for all but
> -ENXIO (and -EPROBE_DEFER), I see the problem:
> 
> 5. platform_get_irq*() issuing an error message is only correct most of
> the time and given proper error handling in the caller (which might be
> able to handle not only -ENXIO but maybe also -EINVAL[1]) the error message
> is irritating. Today platform_get_irq() emits an error message for all
> but -EPROBE_DEFER. As soon as we find a driver that handles -EINVAL we
> need a function platform_get_irq_variant1 to be silent for -EINVAL,
> -EPROBE_DEFER and -ENXIO (or platform_get_irq_variant2 that is only
> silent for -EINVAL and -EPROBE_DEFER?)
> 
> IMHO a query function should always be silent and let the caller do the
> error handling. And if it's only because
> 
> 	mydev: IRQ index 0 not found
> 
> is worse than
> 
> 	mydev: neither TX irq not a muxed RX/TX irq found
> 
> . Also "index 0" is irritating for devices that are expected to have
> only a single irq (i.e. the majority of all devices).

Yeah, ack the #5.

> Yes, I admit, we can safe some code by pushing the error message in a
> query function. But that doesn't only have advantages.

> [1] Looking through the source I wonder: What are the errors that can happen
>     in platform_get_irq*()? (calling everything but a valid irq number
>     an error) Looking at many callers, they only seem to expect "not
>     found" and some "probe defer" (even platform_get_irq() interprets
>     everything but -EPROBE_DEFER as "IRQ index %u not found\n".)
>     IMHO before we should consider to introduce a platform_get_irq*()
>     variant with improved semantics, some cleanup in the internals of
>     the irq lookup are necessary.

-- 
With Best Regards,
Andy Shevchenko