Re: [PATCH v3] PM: Fix symbol export for _SIMPLE_ variants of _PM_OPS()

[Raag Jadav]

On Wed, Sep 20, 2023 at 10:53:23AM +0200, Paul Cercueil wrote:
> Le mercredi 20 septembre 2023 à 14:11 +0530, Raag Jadav a écrit :
> > Currently EXPORT_*_SIMPLE_DEV_PM_OPS() use EXPORT_*_DEV_PM_OPS() set
> > of macros to export dev_pm_ops symbol, which export the symbol in
> > case
> > CONFIG_PM=y but don't take CONFIG_PM_SLEEP into consideration.
> > 
> > Since _SIMPLE_ variants of _PM_OPS() do not include runtime PM
> > handles
> > and are only used in case CONFIG_PM_SLEEP=y, we should not be
> > exporting
> > their dev_pm_ops symbol in case CONFIG_PM_SLEEP=n.
> > 
> > This can be fixed by having two distinct set of export macros for
> > both
> > _RUNTIME_ and _SIMPLE_ variants of _PM_OPS(), such that the export of
> > dev_pm_ops symbol used in each variant depends on CONFIG_PM and
> > CONFIG_PM_SLEEP respectively.
> > 
> > Introduce _DEV_SLEEP_PM_OPS() set of export macros for _SIMPLE_
> > variants
> > of _PM_OPS(), which export dev_pm_ops symbol only in case
> > CONFIG_PM_SLEEP=y
> > and discard it otherwise.
> > 
> > Fixes: 34e1ed189fab ("PM: Improve EXPORT_*_DEV_PM_OPS macros")
> > Signed-off-by: Raag Jadav <raag.jadav@xxxxxxxxx>
> > ---
> > PS: This is a standalone fix and works without updating any drivers.
> 
> I had to double-check that, to make sure that none of the drivers using
> these macros also use pm_ptr() instead of pm_sleep_ptr() to access the
> exported dev_pm_ops.
> 
> I did not check extensively but everything seems to use pm_sleep_ptr(),
> so it looks safe enough.

I have tested it against -rc2 without any problems.

> > Changes since v2:
> > - Drop redundant patches
> > 
> > Changes since v1:
> > - Update drivers to new set of macros
> > 
> >  include/linux/pm.h | 43 +++++++++++++++++++++++++++++--------------
> >  1 file changed, 29 insertions(+), 14 deletions(-)
> > 
> > diff --git a/include/linux/pm.h b/include/linux/pm.h
> > index 1400c37b29c7..99a8146fa479 100644
> > --- a/include/linux/pm.h
> > +++ b/include/linux/pm.h
> > @@ -374,24 +374,39 @@ const struct dev_pm_ops name = { \
> >         RUNTIME_PM_OPS(runtime_suspend_fn, runtime_resume_fn,
> > idle_fn) \
> >  }
> >  
> > -#ifdef CONFIG_PM
> > -#define _EXPORT_DEV_PM_OPS(name, license,
> > ns)                          \
> > +#define _EXPORT_PM_OPS(name, license,
> > ns)                              \
> >         const struct dev_pm_ops
> > name;                                   \
> >         __EXPORT_SYMBOL(name, license,
> > ns);                             \
> >         const struct dev_pm_ops name
> > -#define EXPORT_PM_FN_GPL(name)         EXPORT_SYMBOL_GPL(name)
> > -#define EXPORT_PM_FN_NS_GPL(name, ns)  EXPORT_SYMBOL_NS_GPL(name,
> > ns)
> > -#else
> > -#define _EXPORT_DEV_PM_OPS(name, license,
> > ns)                          \
> > +
> > +#define _PM_OPS(name, license,
> 
> This macro creates a dev_pm_ops that's meant to be garbage-collected by
> the compiler; so maybe name it _USELESS_PM_OPS() or something like
> that.

_USELESS_PM_OPS() sounds a bit heavy handed ;)
Gives the impression that the macro itelf is not used anywhere in code.

Something like _DISCARD_PM_OPS() makes more sense.

Raag