Re: [PATCH v1 2/3] PCI/sysfs: Use PM runtime class macro for auto cleanup in reset_method_store()

[Rafael J. Wysocki]

On Fri, Sep 26, 2025 at 8:13 PM <dan.j.williams@xxxxxxxxx> wrote:
>
> Rafael J. Wysocki wrote:
> > On Fri, Sep 26, 2025 at 3:49 PM Jonathan Cameron
> > <jonathan.cameron@xxxxxxxxxx> wrote:
> > >
> > > On Fri, 19 Sep 2025 18:38:42 +0200
> > > "Rafael J. Wysocki" <rafael@xxxxxxxxxx> wrote:
> > >
> > > > From: Takashi Iwai <tiwai@xxxxxxx>
> > > >
> > > > The newly introduced class macro can simplify the code.
> > > >
> > > > Also, add the proper error handling for the PM runtime get.
> > > >
> > > > Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>
> > > > [ rjw: Adjust subject and error handling ]
> > > > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > > > ---
> > > >  drivers/pci/pci-sysfs.c |    5 +++--
> > > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > >
> > > > --- a/drivers/pci/pci-sysfs.c
> > > > +++ b/drivers/pci/pci-sysfs.c
> > > > @@ -1475,8 +1475,9 @@ static ssize_t reset_method_store(struct
> > > >               return count;
> > > >       }
> > > >
> > > > -     pm_runtime_get_sync(dev);
> > > > -     struct device *pmdev __free(pm_runtime_put) = dev;
> > > > +     CLASS(pm_runtime_resume_and_get, pmdev)(dev);
> > > > +     if (IS_ERR(pmdev))
> > > > +             return -ENXIO;
> > > Hi Rafael,
> > >
> > > Why this approach rather than treating runtime pm state like a conditional
> > > lock (we use it much like one) and using ACQUIRE() / ACQUIRE_ERR()?
> >
> > Mostly because devices are not locks.
> >
> > > Ultimately that's a wrapper around the same infrastructure but
> > > perhaps neater as it removes need to have that explicit magic pmdev.
> >
> > You'll need to have a magic pmdev or similar regardless IIUC.
> >
> > Say there is
> >
> > DEFINE_GUARD(pm_runtime_active, struct device *,
> > pm_runtime_get_sync(_T), pm_runtime_put(_T))
> > DEFINE_GUARD_COND(pm_runtime_active, _try, pm_runtime_resume_and_get(_T))
> >
> > so the user of this will do
> >
> > ACQUIRE(pm_runtime_active_try, pm)(dev);
> > if (ACQUIRE_ERR(pm_runtime_active_try, &pm))
> >         return -ENXIO;
>
> FWIW this looks better to me than the open-coded CLASS(). The pattern,
> admittedly coding-style bending, we are using in drivers/cxl/ for
> compactness and error code fidelity is:
>
>    ACQUIRE(pm_runtime_active_try, pm)(dev);
>    if ((ret = ACQUIRE_ERR(pm_runtime_active_try, &pm)))
>            return ret;

I prefer somewhat more traditional

ret = ACQUIRE_ERR(pm_runtime_active_try, &pm);
if (ret < 0)
        return ret;

It would be nice to be able to hide the pm variable somehow, but this
is not too bad the way it looks now.

> > and there's a "magic" pm though pm is not a struct device pointer.
> >
> > Maybe it's nicer.  I guess people may be more used to dealing with int
> > error variables.
> >
> > Let me try this and see how far I can get with this.
> >
> > > +CC Dan as he can probably remember the discussions around ACQUIRE()
> > > vs the way you have here better than I can.
>
> Yes, effectively a new open-coded CLASS() prompted the ACQUIRE()
> proposal [1]. This pm-active-state reference management indeed looks
> more like a guard() of the active state than an object constructor
> auto-unwind-on-error case.
>
> [1]: http://lore.kernel.org/20250507072145.3614298-1-dan.j.williams@xxxxxxxxx

OK, so please see
https://lore.kernel.org/linux-pm/6196611.lOV4Wx5bFT@rafael.j.wysocki/