[PATCH] cpuidle: fix fallback mechanism for suspend to idle in absence of enter_freeze

[PATCH] cpuidle: fix fallback mechanism for suspend to idle in absence of enter_freeze

[Sudeep Holla]

Commit 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback
from call_cpuidle()") made find_deepest_state() return non-negative
value and check all the states with index > 0. Also a result,
find_deepest_state() returns 0 even when enter_freeze callbacks are not
implemented and enter_freeze_proper is called which ends up crashing
the kernel.

This patch updates the check for index > 0 in cpuidle_enter_freeze and
cpuidle_idle_call(when idle_should_freeze is true) to restore the
suspend-to-idle functionality in absence of enter_freeze callback.

Fixes: 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback from call_cpuidle()")
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
---
 drivers/cpuidle/cpuidle.c | 2 +-
 kernel/sched/idle.c       | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

Hi Rafael,

I assume you prefer to retain find_deepest_state return non-negative
values, so I took this approach for fixing the bug. Do you think we
need to support enter_freeze_proper for index 0 ?

Regards,
Sudeep

diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
index 046423b0c5ca..f996efc56605 100644
--- a/drivers/cpuidle/cpuidle.c
+++ b/drivers/cpuidle/cpuidle.c
@@ -153,7 +153,7 @@ int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
 	 * be frozen safely.
 	 */
 	index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
-	if (index >= 0)
+	if (index > 0)
 		enter_freeze_proper(drv, dev, index);

 	return index;
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index de0e786b2667..544a7133cbd1 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -162,7 +162,7 @@ static void cpuidle_idle_call(void)
 	 */
 	if (idle_should_freeze()) {
 		entered_state = cpuidle_enter_freeze(drv, dev);
-		if (entered_state >= 0) {
+		if (entered_state > 0) {
 			local_irq_enable();
 			goto exit_idle;
 		}
--
1.9.1

Re: [PATCH] cpuidle: fix fallback mechanism for suspend to idle in absence of enter_freeze

[Rafael J. Wysocki]

On Thursday, January 21, 2016 11:19:29 AM Sudeep Holla wrote:
> Commit 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback
> from call_cpuidle()") made find_deepest_state() return non-negative
> value and check all the states with index > 0. Also a result,
> find_deepest_state() returns 0 even when enter_freeze callbacks are not
> implemented and enter_freeze_proper is called which ends up crashing
> the kernel.
> 
> This patch updates the check for index > 0 in cpuidle_enter_freeze and
> cpuidle_idle_call(when idle_should_freeze is true) to restore the
> suspend-to-idle functionality in absence of enter_freeze callback.
> 
> Fixes: 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback from call_cpuidle()")
> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> ---
>  drivers/cpuidle/cpuidle.c | 2 +-
>  kernel/sched/idle.c       | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> Hi Rafael,

Hi,

Sorry for the breakage.

> I assume you prefer to retain find_deepest_state return non-negative
> values, so I took this approach for fixing the bug. Do you think we
> need to support enter_freeze_proper for index 0 ?

Zero is a special case on x86, so supporting enter_freeze_proper() for it
is not necessary.

If you think we can also make 0 a special case on ARM, the others should
not object to that either.

> diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
> index 046423b0c5ca..f996efc56605 100644
> --- a/drivers/cpuidle/cpuidle.c
> +++ b/drivers/cpuidle/cpuidle.c
> @@ -153,7 +153,7 @@ int cpuidle_enter_freeze(struct cpuidle_driver *drv, struct cpuidle_device *dev)
>  	 * be frozen safely.
>  	 */
>  	index = find_deepest_state(drv, dev, UINT_MAX, 0, true);
> -	if (index >= 0)
> +	if (index > 0)
>  		enter_freeze_proper(drv, dev, index);
> 
>  	return index;
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index de0e786b2667..544a7133cbd1 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -162,7 +162,7 @@ static void cpuidle_idle_call(void)
>  	 */
>  	if (idle_should_freeze()) {
>  		entered_state = cpuidle_enter_freeze(drv, dev);
> -		if (entered_state >= 0) {
> +		if (entered_state > 0) {
>  			local_irq_enable();
>  			goto exit_idle;
>  		}
> --
> 1.9.1
> 

Thanks,
Rafael

Re: [PATCH] cpuidle: fix fallback mechanism for suspend to idle in absence of enter_freeze

[Sudeep Holla]

On 22/01/16 01:15, Rafael J. Wysocki wrote:
> On Thursday, January 21, 2016 11:19:29 AM Sudeep Holla wrote:
>> Commit 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback
>> from call_cpuidle()") made find_deepest_state() return non-negative
>> value and check all the states with index > 0. Also a result,
>> find_deepest_state() returns 0 even when enter_freeze callbacks are not
>> implemented and enter_freeze_proper is called which ends up crashing
>> the kernel.
>>
>> This patch updates the check for index > 0 in cpuidle_enter_freeze and
>> cpuidle_idle_call(when idle_should_freeze is true) to restore the
>> suspend-to-idle functionality in absence of enter_freeze callback.
>>
>> Fixes: 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback from call_cpuidle()")
>> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
>> Cc: Ingo Molnar <mingo@redhat.com>
>> Cc: Peter Zijlstra <peterz@infradead.org>
>> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
>> ---
>>   drivers/cpuidle/cpuidle.c | 2 +-
>>   kernel/sched/idle.c       | 2 +-
>>   2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> Hi Rafael,
>
> Hi,
>
> Sorry for the breakage.
>

Sorry that I missed to test suspend-to-idle before it got merged.

>> I assume you prefer to retain find_deepest_state return non-negative
>> values, so I took this approach for fixing the bug. Do you think we
>> need to support enter_freeze_proper for index 0 ?
>
> Zero is a special case on x86, so supporting enter_freeze_proper() for it
> is not necessary.
>

Even on ARM, 0 is used for WFI only, which will not be used for freeze.

> If you think we can also make 0 a special case on ARM, the others should
> not object to that either.
>

Makes sense and since it's already reserved for WFI on ARM, it should be
fine. If there are no objections, can you pick up this fix ?

-- 
Regards,
Sudeep

Re: [PATCH] cpuidle: fix fallback mechanism for suspend to idle in absence of enter_freeze

[Rafael J. Wysocki]

On Friday, January 22, 2016 09:20:15 AM Sudeep Holla wrote:
> 
> On 22/01/16 01:15, Rafael J. Wysocki wrote:
> > On Thursday, January 21, 2016 11:19:29 AM Sudeep Holla wrote:
> >> Commit 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback
> >> from call_cpuidle()") made find_deepest_state() return non-negative
> >> value and check all the states with index > 0. Also a result,
> >> find_deepest_state() returns 0 even when enter_freeze callbacks are not
> >> implemented and enter_freeze_proper is called which ends up crashing
> >> the kernel.
> >>
> >> This patch updates the check for index > 0 in cpuidle_enter_freeze and
> >> cpuidle_idle_call(when idle_should_freeze is true) to restore the
> >> suspend-to-idle functionality in absence of enter_freeze callback.
> >>
> >> Fixes: 51164251f5c3 ("sched / idle: Drop default_idle_call() fallback from call_cpuidle()")
> >> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
> >> Cc: Ingo Molnar <mingo@redhat.com>
> >> Cc: Peter Zijlstra <peterz@infradead.org>
> >> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> >> ---
> >>   drivers/cpuidle/cpuidle.c | 2 +-
> >>   kernel/sched/idle.c       | 2 +-
> >>   2 files changed, 2 insertions(+), 2 deletions(-)
> >>
> >> Hi Rafael,
> >
> > Hi,
> >
> > Sorry for the breakage.
> >
> 
> Sorry that I missed to test suspend-to-idle before it got merged.
> 
> >> I assume you prefer to retain find_deepest_state return non-negative
> >> values, so I took this approach for fixing the bug. Do you think we
> >> need to support enter_freeze_proper for index 0 ?
> >
> > Zero is a special case on x86, so supporting enter_freeze_proper() for it
> > is not necessary.
> >
> 
> Even on ARM, 0 is used for WFI only, which will not be used for freeze.
> 
> > If you think we can also make 0 a special case on ARM, the others should
> > not object to that either.
> >
> 
> Makes sense and since it's already reserved for WFI on ARM, it should be
> fine. If there are no objections, can you pick up this fix ?

Yes, I've already done that.

Thanks,
Rafael