[PATCH v2 0/2] cpuidle: Do not return from cpuidle_play_dead() on callback failures

[PATCH v2 0/2] cpuidle: Do not return from cpuidle_play_dead() on callback failures

[Rafael J. Wysocki]

Hi Everyone,

This series of 2 patches supersedes

https://lore.kernel.org/linux-pm/4992010.31r3eYUQgx@rjwysocki.net/

in order to address review feedback.

Patch [1/2] modifies cpuidle_play_dead() as per the subject and patch [2/2]
changes the :enter_dead() cpuidle state callback return type to void.

Thanks!

[PATCH v2 1/2] cpuidle: Do not return from cpuidle_play_dead() on callback failures

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

If the :enter_dead() idle state callback fails for a certain state,
there may be still a shallower state for which it will work.

Because the only caller of cpuidle_play_dead(), native_play_dead(),
falls back to hlt_play_dead() if it returns an error, it should
better try all of the idle states for which :enter_dead() is present
before failing, so change it accordingly.

Also notice that the :enter_dead() state callback is not expected
to return on success (the CPU should be "dead" then), so make
cpuidle_play_dead() ignore its return value.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Tested-by: Mario Limonciello <mario.limonciello@amd.com> # 6.12-rc7
---

v1 -> v2:
   * Make cpuidle_play_dead() never return 0.
   * Add tags from Mario (I have added them because the change of the patch
     should not make a practical difference, but if you want them to be
     dropped, please let me know).

---
 drivers/cpuidle/cpuidle.c |   10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

Index: linux-pm/drivers/cpuidle/cpuidle.c
===================================================================
--- linux-pm.orig/drivers/cpuidle/cpuidle.c
+++ linux-pm/drivers/cpuidle/cpuidle.c
@@ -69,11 +69,15 @@ int cpuidle_play_dead(void)
 	if (!drv)
 		return -ENODEV;
 
-	/* Find lowest-power state that supports long-term idle */
-	for (i = drv->state_count - 1; i >= 0; i--)
+	for (i = drv->state_count - 1; i >= 0; i--) {
 		if (drv->states[i].enter_dead)
-			return drv->states[i].enter_dead(dev, i);
+			drv->states[i].enter_dead(dev, i);
+	}
 
+	/*
+	 * If :enter_dead() is successful, it will never return, so reaching
+	 * here means that all of them failed above or were not present.
+	 */
 	return -ENODEV;
 }
 

Re: [PATCH v2 1/2] cpuidle: Do not return from cpuidle_play_dead() on callback failures

[Gautham R. Shenoy]

Hello Rafael,

On Fri, Nov 15, 2024 at 09:58:31PM +0100, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> If the :enter_dead() idle state callback fails for a certain state,
> there may be still a shallower state for which it will work.
> 
> Because the only caller of cpuidle_play_dead(), native_play_dead(),
> falls back to hlt_play_dead() if it returns an error, it should
> better try all of the idle states for which :enter_dead() is present
> before failing, so change it accordingly.
> 
> Also notice that the :enter_dead() state callback is not expected
> to return on success (the CPU should be "dead" then), so make
> cpuidle_play_dead() ignore its return value.
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
> Tested-by: Mario Limonciello <mario.limonciello@amd.com> # 6.12-rc7

Thanks for fixing this.

Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>

--
Thanks and Regards
gautham.
> ---
> 
> v1 -> v2:
>    * Make cpuidle_play_dead() never return 0.
>    * Add tags from Mario (I have added them because the change of the patch
>      should not make a practical difference, but if you want them to be
>      dropped, please let me know).
> 
> ---
>  drivers/cpuidle/cpuidle.c |   10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> Index: linux-pm/drivers/cpuidle/cpuidle.c
> ===================================================================
> --- linux-pm.orig/drivers/cpuidle/cpuidle.c
> +++ linux-pm/drivers/cpuidle/cpuidle.c
> @@ -69,11 +69,15 @@ int cpuidle_play_dead(void)
>  	if (!drv)
>  		return -ENODEV;
>  
> -	/* Find lowest-power state that supports long-term idle */
> -	for (i = drv->state_count - 1; i >= 0; i--)
> +	for (i = drv->state_count - 1; i >= 0; i--) {
>  		if (drv->states[i].enter_dead)
> -			return drv->states[i].enter_dead(dev, i);
> +			drv->states[i].enter_dead(dev, i);
> +	}
>  
> +	/*
> +	 * If :enter_dead() is successful, it will never return, so reaching
> +	 * here means that all of them failed above or were not present.
> +	 */
>  	return -ENODEV;
>  }
>  
> 
> 
> 

[PATCH v2 2/2] cpuidle: Change :enter_dead() driver callback return type to void

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

After a previous change, cpuidle_play_dead(), which is the only caller
of idle state :enter_dead() callbacks, ignores their return values, so
they may as well be void.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
---

v1 -> v2: New patch

Interestingly enough, the only user of :enter_dead() idle state callbacks
in the current mainline turns out to be ACPI idle.

---
 drivers/acpi/processor_idle.c |    7 ++-----
 include/linux/cpuidle.h       |    2 +-
 2 files changed, 3 insertions(+), 6 deletions(-)

Index: linux-pm/include/linux/cpuidle.h
===================================================================
--- linux-pm.orig/include/linux/cpuidle.h
+++ linux-pm/include/linux/cpuidle.h
@@ -61,7 +61,7 @@ struct cpuidle_state {
 			struct cpuidle_driver *drv,
 			int index);
 
-	int (*enter_dead) (struct cpuidle_device *dev, int index);
+	void (*enter_dead) (struct cpuidle_device *dev, int index);
 
 	/*
 	 * CPUs execute ->enter_s2idle with the local tick or entire timekeeping
Index: linux-pm/drivers/acpi/processor_idle.c
===================================================================
--- linux-pm.orig/drivers/acpi/processor_idle.c
+++ linux-pm/drivers/acpi/processor_idle.c
@@ -578,7 +578,7 @@ static void __cpuidle acpi_idle_do_entry
  * @dev: the target CPU
  * @index: the index of suggested state
  */
-static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
+static void acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 {
 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 
@@ -591,11 +591,8 @@ static int acpi_idle_play_dead(struct cp
 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 			io_idle(cx->address);
 		} else
-			return -ENODEV;
+			return;
 	}
-
-	/* Never reached */
-	return 0;
 }
 
 static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)

Re: [PATCH v2 2/2] cpuidle: Change :enter_dead() driver callback return type to void

[Gautham R. Shenoy]

Hello Rafael,

On Fri, Nov 15, 2024 at 10:00:25PM +0100, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> After a previous change, cpuidle_play_dead(), which is the only caller
> of idle state :enter_dead() callbacks, ignores their return values, so
> they may as well be void.
> 
> Suggested-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

> ---
> 
> v1 -> v2: New patch
> 
> Interestingly enough, the only user of :enter_dead() idle state callbacks
> in the current mainline turns out to be ACPI idle.

For that matter, the only user of cpuidle_play_dead() is the
native_play_dead(). Was that always the case? 

Some of the other architectures select the deepest available idle
state at boot time, and enter that state when a CPU is offlined.

In any case I am ok with this.

Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
--
Thanks and Regards
gautham.
> 
> ---
>  drivers/acpi/processor_idle.c |    7 ++-----
>  include/linux/cpuidle.h       |    2 +-
>  2 files changed, 3 insertions(+), 6 deletions(-)
> 
> Index: linux-pm/include/linux/cpuidle.h
> ===================================================================
> --- linux-pm.orig/include/linux/cpuidle.h
> +++ linux-pm/include/linux/cpuidle.h
> @@ -61,7 +61,7 @@ struct cpuidle_state {
>  			struct cpuidle_driver *drv,
>  			int index);
>  
> -	int (*enter_dead) (struct cpuidle_device *dev, int index);
> +	void (*enter_dead) (struct cpuidle_device *dev, int index);
>  
>  	/*
>  	 * CPUs execute ->enter_s2idle with the local tick or entire timekeeping
> Index: linux-pm/drivers/acpi/processor_idle.c
> ===================================================================
> --- linux-pm.orig/drivers/acpi/processor_idle.c
> +++ linux-pm/drivers/acpi/processor_idle.c
> @@ -578,7 +578,7 @@ static void __cpuidle acpi_idle_do_entry
>   * @dev: the target CPU
>   * @index: the index of suggested state
>   */
> -static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
> +static void acpi_idle_play_dead(struct cpuidle_device *dev, int index)
>  {
>  	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
>  
> @@ -591,11 +591,8 @@ static int acpi_idle_play_dead(struct cp
>  		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
>  			io_idle(cx->address);
>  		} else
> -			return -ENODEV;
> +			return;
>  	}
> -
> -	/* Never reached */
> -	return 0;
>  }
>  
>  static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
> 
> 
> 

Re: [PATCH v2 2/2] cpuidle: Change :enter_dead() driver callback return type to void

[Mario Limonciello]

On 11/19/2024 09:09, Gautham R. Shenoy wrote:
> Hello Rafael,
> 
> On Fri, Nov 15, 2024 at 10:00:25PM +0100, Rafael J. Wysocki wrote:
>> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>>
>> After a previous change, cpuidle_play_dead(), which is the only caller
>> of idle state :enter_dead() callbacks, ignores their return values, so
>> they may as well be void.
>>
>> Suggested-by: Peter Zijlstra <peterz@infradead.org>
>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
>> ---
>>
>> v1 -> v2: New patch
>>
>> Interestingly enough, the only user of :enter_dead() idle state callbacks
>> in the current mainline turns out to be ACPI idle.
> 
> For that matter, the only user of cpuidle_play_dead() is the
> native_play_dead(). Was that always the case?
> 
> Some of the other architectures select the deepest available idle
> state at boot time, and enter that state when a CPU is offlined.
> 
> In any case I am ok with this.
> 
> Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>

> --
> Thanks and Regards
> gautham.
>>
>> ---
>>   drivers/acpi/processor_idle.c |    7 ++-----
>>   include/linux/cpuidle.h       |    2 +-
>>   2 files changed, 3 insertions(+), 6 deletions(-)
>>
>> Index: linux-pm/include/linux/cpuidle.h
>> ===================================================================
>> --- linux-pm.orig/include/linux/cpuidle.h
>> +++ linux-pm/include/linux/cpuidle.h
>> @@ -61,7 +61,7 @@ struct cpuidle_state {
>>   			struct cpuidle_driver *drv,
>>   			int index);
>>   
>> -	int (*enter_dead) (struct cpuidle_device *dev, int index);
>> +	void (*enter_dead) (struct cpuidle_device *dev, int index);
>>   
>>   	/*
>>   	 * CPUs execute ->enter_s2idle with the local tick or entire timekeeping
>> Index: linux-pm/drivers/acpi/processor_idle.c
>> ===================================================================
>> --- linux-pm.orig/drivers/acpi/processor_idle.c
>> +++ linux-pm/drivers/acpi/processor_idle.c
>> @@ -578,7 +578,7 @@ static void __cpuidle acpi_idle_do_entry
>>    * @dev: the target CPU
>>    * @index: the index of suggested state
>>    */
>> -static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
>> +static void acpi_idle_play_dead(struct cpuidle_device *dev, int index)
>>   {
>>   	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
>>   
>> @@ -591,11 +591,8 @@ static int acpi_idle_play_dead(struct cp
>>   		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
>>   			io_idle(cx->address);
>>   		} else
>> -			return -ENODEV;
>> +			return;
>>   	}
>> -
>> -	/* Never reached */
>> -	return 0;
>>   }
>>   
>>   static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
>>
>>
>>

Re: [PATCH v2 2/2] cpuidle: Change :enter_dead() driver callback return type to void

[Rafael J. Wysocki]

Hi Gautham,

On Tue, Nov 19, 2024 at 4:09 PM Gautham R. Shenoy
<gautham.shenoy@amd.com> wrote:
>
> Hello Rafael,
>
> On Fri, Nov 15, 2024 at 10:00:25PM +0100, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> >
> > After a previous change, cpuidle_play_dead(), which is the only caller
> > of idle state :enter_dead() callbacks, ignores their return values, so
> > they may as well be void.
> >
> > Suggested-by: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>
> > ---
> >
> > v1 -> v2: New patch
> >
> > Interestingly enough, the only user of :enter_dead() idle state callbacks
> > in the current mainline turns out to be ACPI idle.
>
> For that matter, the only user of cpuidle_play_dead() is the
> native_play_dead(). Was that always the case?

At least as long ago as of 4.20.

> Some of the other architectures select the deepest available idle
> state at boot time, and enter that state when a CPU is offlined.

That's fine as long as the state to use is already known at that time.

> In any case I am ok with this.
>
> Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
> --
> Thanks and Regards
> gautham.
> >
> > ---
> >  drivers/acpi/processor_idle.c |    7 ++-----
> >  include/linux/cpuidle.h       |    2 +-
> >  2 files changed, 3 insertions(+), 6 deletions(-)
> >
> > Index: linux-pm/include/linux/cpuidle.h
> > ===================================================================
> > --- linux-pm.orig/include/linux/cpuidle.h
> > +++ linux-pm/include/linux/cpuidle.h
> > @@ -61,7 +61,7 @@ struct cpuidle_state {
> >                       struct cpuidle_driver *drv,
> >                       int index);
> >
> > -     int (*enter_dead) (struct cpuidle_device *dev, int index);
> > +     void (*enter_dead) (struct cpuidle_device *dev, int index);
> >
> >       /*
> >        * CPUs execute ->enter_s2idle with the local tick or entire timekeeping
> > Index: linux-pm/drivers/acpi/processor_idle.c
> > ===================================================================
> > --- linux-pm.orig/drivers/acpi/processor_idle.c
> > +++ linux-pm/drivers/acpi/processor_idle.c
> > @@ -578,7 +578,7 @@ static void __cpuidle acpi_idle_do_entry
> >   * @dev: the target CPU
> >   * @index: the index of suggested state
> >   */
> > -static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
> > +static void acpi_idle_play_dead(struct cpuidle_device *dev, int index)
> >  {
> >       struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
> >
> > @@ -591,11 +591,8 @@ static int acpi_idle_play_dead(struct cp
> >               else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
> >                       io_idle(cx->address);
> >               } else
> > -                     return -ENODEV;
> > +                     return;
> >       }
> > -
> > -     /* Never reached */
> > -     return 0;
> >  }
> >
> >  static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
> >
> >
> >
>