Re: [RFC PATCH RT] PM: runtime: avoid retry loops on RT

[John Keeping <john@metanate.com>]

On Tue, 5 Oct 2021 18:38:27 +0200
"Rafael J. Wysocki" <rafael@kernel.org> wrote:

> On Tue, Oct 5, 2021 at 6:14 PM John Keeping <john@metanate.com> wrote:
> >
> > With PREEMPT_RT spin_unlock() is identical to spin_unlock_irq() so there
> > is no reason to have a special case using the former.  Furthermore,
> > spin_unlock() enables preemption meaning that a task in RESUMING or
> > SUSPENDING state may be preempted by a higher priority task running
> > pm_runtime_get_sync() leading to a livelock.
> >
> > Use the non-irq_safe path for all waiting so that the waiting task will
> > block.
> >
> > Note that this changes only the waiting behaviour of irq_safe, other
> > uses are left unchanged so that the parent device always remains active
> > in the same way as !RT.
> >
> > Signed-off-by: John Keeping <john@metanate.com>  
> 
> So basically, the idea is that the irq_safe flag should have no effect
> when CONFIG_PREEMPT_RT is set, right?
> 
> Wouldn't it be cleaner to make it not present at all in that case?

Yes, just replacing pm_runtime_irq_safe() with an empty function would
also fix it, but I'm not sure if that will have unexpected effects from
the parent device suspending/resuming, especially in terms of latency
for handling interrupts.

> > ---
> >  drivers/base/power/runtime.c | 9 +++++----
> >  1 file changed, 5 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> > index 96972d5f6ef3..5e0d349fab4e 100644
> > --- a/drivers/base/power/runtime.c
> > +++ b/drivers/base/power/runtime.c
> > @@ -347,8 +347,9 @@ static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
> >  {
> >         int retval = 0, idx;
> >         bool use_links = dev->power.links_count > 0;
> > +       bool irq_safe = dev->power.irq_safe && !IS_ENABLED(CONFIG_PREEMPT_RT);
> >
> > -       if (dev->power.irq_safe) {
> > +       if (irq_safe) {
> >                 spin_unlock(&dev->power.lock);
> >         } else {
> >                 spin_unlock_irq(&dev->power.lock);
> > @@ -376,7 +377,7 @@ static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
> >         if (cb)
> >                 retval = cb(dev);
> >
> > -       if (dev->power.irq_safe) {
> > +       if (irq_safe) {
> >                 spin_lock(&dev->power.lock);
> >         } else {
> >                 /*
> > @@ -596,7 +597,7 @@ static int rpm_suspend(struct device *dev, int rpmflags)
> >                         goto out;
> >                 }
> >
> > -               if (dev->power.irq_safe) {
> > +               if (dev->power.irq_safe && !IS_ENABLED(CONFIG_PREEMPT_RT)) {
> >                         spin_unlock(&dev->power.lock);
> >
> >                         cpu_relax();
> > @@ -777,7 +778,7 @@ static int rpm_resume(struct device *dev, int rpmflags)
> >                         goto out;
> >                 }
> >
> > -               if (dev->power.irq_safe) {
> > +               if (dev->power.irq_safe && !IS_ENABLED(CONFIG_PREEMPT_RT)) {
> >                         spin_unlock(&dev->power.lock);
> >
> >                         cpu_relax();
> > --
> > 2.33.0
> >