Re: [PATCH 2/3] platform/x86: intel: punit_ipc: Propagate errors from optional IRQ lookup

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Tue, Aug 11, 2026 at 10:30:53AM +0700, Bui Duc Phuc wrote:

...

> > >         irq = platform_get_irq_optional(pdev, 0);
> > >         if (irq < 0) {
> > > +               if (irq != -ENXIO)
> > > +                       return irq;
> > >                 dev_warn(&pdev->dev, "Invalid IRQ, using polling mode\n");
> >
> > I can admit that the warning message is not comprehensive, but after
> > this patch it obviously makes a regression. A previously working
> > driver for even some Linux failures (besides deferred probe) now
> > becomes unusable. Have you studied the possible error codes returned
> > by platform_get_irq_optional()? What are they? Perhaps you need to
> > improve the documentation of that API?
> 
> So I understand you'd prefer to keep the current behavior of the function,
> to make sure existing systems keep working fine?

Keeping the current behaviour. Which means that probably only deferred probe is
the error we may return.

> However, I think if this IRQs is truly meant to be optional, then any error
> other than -ENXIO should be captured and returned
> so that developers can actually investigate and fix the underlying issue.
> Silently swallowing an error and falling back to polling doesn't
> really guarantee
> the system is working correctly or safely either, does it?
> 
> That said, if you'd rather keep the current implementation as-is,
> then I think it would make more sense to switch to platform_get_irq() instead.

But then the (additional in this case) warning will be printed each time the
device is probed without proper IRQ.

> That would more accurately reflect that this driver doesn't actually
> treat the IRQ
> as optional, rather than using an "optional" lookup function while implementing
> it as if it weren't optional.

Hmm... It says it will use polling mode, to me it means that IRQ is optional.

> https://elixir.bootlin.com/linux/v7.2-rc6/source/drivers/base/platform.c#L301

-- 
With Best Regards,
Andy Shevchenko