Re: [PATCH v2 10/10] pinctrl: tegra: Switch to use DEFINE_NOIRQ_DEV_PM_OPS() helper

[Paul Cercueil]

Hi Thierry,

Le mardi 18 juillet 2023 à 15:20 +0200, Thierry Reding a écrit :
> On Tue, Jul 18, 2023 at 01:55:05PM +0200, Paul Cercueil wrote:
> > Le mardi 18 juillet 2023 à 13:41 +0200, Thierry Reding a écrit :
> > > On Tue, Jul 18, 2023 at 10:42:47AM +0200, Paul Cercueil wrote:
> > > > Hi Thierry,
> > > > 
> > > > Le mardi 18 juillet 2023 à 09:45 +0200, Thierry Reding a
> > > > écrit :
> > > > > On Mon, Jul 17, 2023 at 09:14:12PM +0200, Paul Cercueil
> > > > > wrote:
> > > > > > Hi Andy,
> > > > > > 
> > > > > > Le lundi 17 juillet 2023 à 20:28 +0300, Andy Shevchenko a
> > > > > > écrit :
> > > > > > > Since pm.h provides a helper for system no-IRQ PM
> > > > > > > callbacks,
> > > > > > > switch the driver to use it instead of open coded
> > > > > > > variant.
> > > > > > > 
> > > > > > > Signed-off-by: Andy Shevchenko
> > > > > > > <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > > > > > > ---
> > > > > > >  drivers/pinctrl/tegra/pinctrl-tegra.c | 5 +----
> > > > > > >  1 file changed, 1 insertion(+), 4 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/drivers/pinctrl/tegra/pinctrl-tegra.c
> > > > > > > b/drivers/pinctrl/tegra/pinctrl-tegra.c
> > > > > > > index 4547cf66d03b..734c71ef005b 100644
> > > > > > > --- a/drivers/pinctrl/tegra/pinctrl-tegra.c
> > > > > > > +++ b/drivers/pinctrl/tegra/pinctrl-tegra.c
> > > > > > > @@ -747,10 +747,7 @@ static int
> > > > > > > tegra_pinctrl_resume(struct
> > > > > > > device
> > > > > > > *dev)
> > > > > > >         return 0;
> > > > > > >  }
> > > > > > >  
> > > > > > > -const struct dev_pm_ops tegra_pinctrl_pm = {
> > > > > > > -       .suspend_noirq = &tegra_pinctrl_suspend,
> > > > > > > -       .resume_noirq = &tegra_pinctrl_resume
> > > > > > > -};
> > > > > > > +DEFINE_NOIRQ_DEV_PM_OPS(tegra_pinctrl_pm,
> > > > > > > tegra_pinctrl_suspend,
> > > > > > > tegra_pinctrl_resume);
> > > > > > >  
> > > > > > >  static bool tegra_pinctrl_gpio_node_has_range(struct
> > > > > > > tegra_pmx
> > > > > > > *pmx)
> > > > > > >  {
> > > > > > 
> > > > > > Another driver where using EXPORT_GPL_DEV_PM_OPS() would
> > > > > > make
> > > > > > more
> > > > > > sense.
> > > > > 
> > > > > We don't currently export these PM ops because none of the
> > > > > Tegra
> > > > > pinctrl
> > > > > drivers can be built as a module.
> > > > 
> > > > This doesn't change anything. You'd want to use
> > > > EXPORT_GPL_DEV_PM_OPS
> > > > (or better, the namespaced version) so that the PM ops can be
> > > > defined
> > > > in one file and referenced in another, while still having them
> > > > garbage-
> > > > collected when CONFIG_PM is disabled.
> > > 
> > > Looking at the definition of EXPORT_GPL_DEV_PM_OPS(), it will
> > > cause
> > > an
> > > EXPORT_SYMBOL_GPL() to be added. So there very well is a change.
> > > And
> > > it's a completely bogus change because no module is ever going to
> > > use
> > > that symbol. If we were to ever support building the pinctrl
> > > driver
> > > as
> > > a module, then this would perhaps make sense, but we don't.
> > 
> > In this particular case the EXPORT_SYMBOL_GPL() isn't really
> > important,
> > the rest of EXPORT_GPL_DEV_PM_OPS() is.
> > 
> > I don't think having a symbol exported it is a big deal, TBH, if
> > you
> > use the namespaced version. If you really don't want that, we need
> > a
> > version of EXPORT_GPL_DEV_PM_OPS() that doesn't export the symbol.
> 
> I do think it's a big deal to export a symbol if there's no reason to
> do
> so.
> 
> And please, can we stop adding these macros for every possible
> scenario?

Yes, as you can read from my other responses, I am not really keen on
having a multiplication of these macros.

> Maybe I'm just getting old, but I find it increasingly difficult to
> understand what all of these are supposed to be. I get that people
> want
> to get rid of boilerplate, but I think we need to more carefully
> balance
> boilerplate vs. simplicity.

The EXPORT_GPL_DEV_PM_OPS() macro does more than get rid of
boilerplate, it gets rid of dead code.

If we take this driver as an example, before the patch the
"tegra_pinctrl_pm" struct, as well as the "tegra_pinctrl_suspend" and
"tegra_pinctrl_resume" functions were always compiled in, even if
CONFIG_PM_SLEEP is disabled in the config.

The status-quo before the introduction of the new PM macros was to just
wrap the dev_pm_ops struct + callbacks with a #ifdef CONFIG_PM_SLEEP.
This was pretty bad as the code was then conditionally compiled. With
the new PM macros this code is always compiled, independently of any
Kconfig option; and thanks to that, bugs and regressions are
subsequently easier to catch.

Cheers,
-Paul

> I'm seeing the same thing with stuff like those mass conversions to
> atrocities like devm_platform_ioremap_resource() and
> devm_platform_get_and_ioremap_resource(). There's so much churn
> involved
> in getting those merged for usually saving a single line of code. And
> it's not even mass conversions, but people tend to send these as one
> patch per driver, which doesn't exactly help (except perhaps for
> patch
> statistics).
> 
> Thierry