Re: [PATCH v2 1/2] cpuidle: Extract and export no-lock variants of cpuidle_unregister_device

[lihuisong (C)]

On 4/7/2026 9:36 PM, Rafael J. Wysocki wrote:
> On Tue, Apr 7, 2026 at 10:11 AM Huisong Li <lihuisong@xxxxxxxxxx> wrote:
> > The cpuidle_unregister_device() function always acquire the internal
> > cpuidle_lock (or pause/resume idle) during their execution.
> > However, in some power notification scenarios (e.g., when old idle
> > states may become unavailable), it is necessary to efficiently disable
> > cpuidle first, then remove and re-create all cpuidle devices for all
> > CPUs. To avoid frequent lock overhead and ensure atomicity across the
> > entire batch operation, the caller needs to hold the cpuidle_lock once
> > outside the loop.
> >
> > To address this, extract the core logic into the new function
> > cpuidle_unregister_device_no_lock() and export it.
> >
> > Signed-off-by: Huisong Li <lihuisong@xxxxxxxxxx>
> > ---
> >   drivers/cpuidle/cpuidle.c | 22 +++++++++++++++-------
> >   include/linux/cpuidle.h   |  2 ++
> >   2 files changed, 17 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> > index c7876e9e024f..1a55542efead 100644
> > --- a/drivers/cpuidle/cpuidle.c
> > +++ b/drivers/cpuidle/cpuidle.c
> > @@ -714,16 +714,12 @@ int cpuidle_register_device(struct cpuidle_device *dev)
> >
> >   EXPORT_SYMBOL_GPL(cpuidle_register_device);
> >
> > -/**
> > - * cpuidle_unregister_device - unregisters a CPU's idle PM feature
> > - * @dev: the cpu
> > - */
> > -void cpuidle_unregister_device(struct cpuidle_device *dev)
> > +void cpuidle_unregister_device_no_lock(struct cpuidle_device *dev)
> >   {
> >          if (!dev || dev->registered == 0)
> >                  return;
> >
> > -       cpuidle_pause_and_lock();
> > +       lockdep_assert_held(&cpuidle_lock);
> >
> >          cpuidle_disable_device(dev);
> >
> > @@ -732,10 +728,22 @@ void cpuidle_unregister_device(struct cpuidle_device *dev)
> >          __cpuidle_unregister_device(dev);
> >
> >          cpuidle_coupled_unregister_device(dev);
> > +}
> > +EXPORT_SYMBOL_GPL(cpuidle_unregister_device_no_lock);
> > +
> > +/**
> > + * cpuidle_unregister_device - unregisters a CPU's idle PM feature
> > + * @dev: the cpu
> > + */
> > +void cpuidle_unregister_device(struct cpuidle_device *dev)
> > +{
> > +       if (!dev || dev->registered == 0)
> > +               return;
> >
> > +       cpuidle_pause_and_lock();
> > +       cpuidle_unregister_device_no_lock(dev);
> >          cpuidle_resume_and_unlock();
> >   }
> > -
> >   EXPORT_SYMBOL_GPL(cpuidle_unregister_device);
> >
> >   /**
> > diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> > index 4073690504a7..2ead2d8c57e6 100644
> > --- a/include/linux/cpuidle.h
> > +++ b/include/linux/cpuidle.h
> > @@ -188,6 +188,7 @@ extern void cpuidle_driver_state_disabled(struct cpuidle_driver *drv, int idx,
> >   extern void cpuidle_unregister_driver(struct cpuidle_driver *drv);
> >   extern int cpuidle_register_device(struct cpuidle_device *dev);
> >   extern void cpuidle_unregister_device(struct cpuidle_device *dev);
> > +extern void cpuidle_unregister_device_no_lock(struct cpuidle_device *dev);
> >   extern int cpuidle_register(struct cpuidle_driver *drv,
> >                              const struct cpumask *const coupled_cpus);
> >   extern void cpuidle_unregister(struct cpuidle_driver *drv);
> > @@ -226,6 +227,7 @@ static inline void cpuidle_unregister_driver(struct cpuidle_driver *drv) { }
> >   static inline int cpuidle_register_device(struct cpuidle_device *dev)
> >   {return -ENODEV; }
> >   static inline void cpuidle_unregister_device(struct cpuidle_device *dev) { }
> > +static void cpuidle_unregister_device_no_lock(struct cpuidle_device *dev) {}
> >   static inline int cpuidle_register(struct cpuidle_driver *drv,
> >                                     const struct cpumask *const coupled_cpus)
> >   {return -ENODEV; }
> > --
> Both patches applied as 7.1 material, though I'm somewhat nervous
> about removing sysfs entries under cpus_read_lock().
Thanks.

The sysfs entries has been removed using cpuidle_disable_device() before 
this change.
From this perspective, we don't change this behavor.🙂
> Note that I've changed the subject of the second patch to "ACPI:
> processor: idle: Reset cpuidle on C-state list changes".
>
> Thanks!