Re: [PATCH v5 02/14] gpio: brcmstb: Use modern PM macros

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

On Mon, Nov 24, 2025 at 04:05:29PM +0100, Jonas Gorski wrote:
> On Mon, Nov 24, 2025 at 3:49 PM Andy Shevchenko
> <andriy.shevchenko@intel.com> wrote:
> > On Mon, Nov 24, 2025 at 03:20:00PM +0100, Jonas Gorski wrote:
> > > On Mon, Nov 24, 2025 at 2:52 PM Andy Shevchenko
> > > <andy.shevchenko@gmail.com> wrote:
> > > > On Mon, Nov 24, 2025 at 2:40 PM Jonas Gorski <jonas.gorski@gmail.com> wrote:
> > > > > On Mon, Nov 24, 2025 at 1:39 AM Jisheng Zhang <jszhang@kernel.org> wrote:

...

> > > > > >  static const struct dev_pm_ops brcmstb_gpio_pm_ops = {
> > > > > > -       .suspend_noirq  = brcmstb_gpio_suspend,
> > > > > > -       .resume_noirq = brcmstb_gpio_resume,
> > > > > > +       .suspend_noirq = pm_sleep_ptr(brcmstb_gpio_suspend),
> > > > > > +       .resume_noirq = pm_sleep_ptr(brcmstb_gpio_resume),
> > > > > >  };

...

> > > > > > -               .pm = &brcmstb_gpio_pm_ops,
> > > > > > +               .pm = pm_sleep_ptr(&brcmstb_gpio_pm_ops),
> > > > >
> > > > > won't this cause a "brcmstb_gpio_pm_ops is unused" compile warning for
> > > > > !CONFIG_PM_SLEEP?
> > > > >
> > > > > You probably need to add a __maybe_unused to brcmstb_gpio_pm_ops
> > > > > (which incidentally DEFINE_NOIRQ_DEV_PM_OPS() also doesn't set, but
> > > > > all other *_DEV_PM_OPS() macros do).
> >
> > Do they? I mean the modern ones and not that are deprecated.
> >
> > > > Shouldn't it be covered by the same trick as pm_sleep_ptr() does for functions?
> > >
> > > pm_sleep_ptr() becomes NULL for !CONFIG_PM_SLEEP, so there is no
> > > reference then anymore to brcmstb_gpio_pm_ops. You would need a
> > > wrapper for brcmstb_gpio_pm_ops itself to conditionally define it to
> > > avoid the warning, or add __maybe_unused to it to silence it.
> >
> > PTR_IF() magic is exactly to make sure compiler will have a visibility while
> > dropping a dead code. Did I miss anything?
> 
> No, I just was working with old assumptions, so my bad. I faintly
> remember that they used to work that way, but maybe I also
> misremember. TIL. So disregard my comment.

NP. I'm glad everything is clear now about them.

> > > Note how SIMPLE_DEV_PM_OPS() and UNIVERSAL_DEV_PM_OPS() tag the struct
> > > with it (for that reason I assume).
> >
> > Both are deprecated. Not a good orienteer.
> > None of the new approach uses __maybe_unused. (See DEFINE_*() macros in pm.h.)
> 
> Maybe that some were using it was confusing me into thinking it is required.

-- 
With Best Regards,
Andy Shevchenko