cpufreq: Move cancelling of policy update work just after removing notifiers
diff mbox series

Message ID 20191021132818.23787-1-sudeep.holla@arm.com
State Mainlined, archived
Headers show
Series
  • cpufreq: Move cancelling of policy update work just after removing notifiers
Related show

Commit Message

Sudeep Holla Oct. 21, 2019, 1:28 p.m. UTC
Commit 099967699ad9 ("cpufreq: Cancel policy update work scheduled before freeing")
added cancel_work_sync(policy->update) after the frequency QoS were
removed. We can cancel the work just after taking the last CPU in the
policy offline and unregistering the notifiers as policy->update cannot
be scheduled from anywhere at this point.

However, due to other bugs, doing so still triggered the race between
freeing of policy and scheduled policy update work. Now that all those
issues are resolved, we can move this cancelling of any scheduled policy
update work just after removing min/max notifiers.

Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
---
 drivers/cpufreq/cpufreq.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Hi Rafael,

Based on Viresh's suggestion, I am posting a patch to move this
cancel_work_sync earlier though it's not a must have change.
I will leave it up to your preference.

Regards,
Sudeep

Comments

Viresh Kumar Oct. 22, 2019, 2:25 a.m. UTC | #1
On 21-10-19, 14:28, Sudeep Holla wrote:
> Commit 099967699ad9 ("cpufreq: Cancel policy update work scheduled before freeing")
> added cancel_work_sync(policy->update) after the frequency QoS were
> removed. We can cancel the work just after taking the last CPU in the
> policy offline and unregistering the notifiers as policy->update cannot
> be scheduled from anywhere at this point.
> 
> However, due to other bugs, doing so still triggered the race between
> freeing of policy and scheduled policy update work. Now that all those
> issues are resolved, we can move this cancelling of any scheduled policy
> update work just after removing min/max notifiers.
> 
> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> ---
>  drivers/cpufreq/cpufreq.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> Hi Rafael,
> 
> Based on Viresh's suggestion, I am posting a patch to move this
> cancel_work_sync earlier though it's not a must have change.

For me it is :)

> I will leave it up to your preference.
> 
> Regards,
> Sudeep
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 829a3764df1b..48a224a6b178 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -1268,6 +1268,9 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
>  	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
>  				 &policy->nb_min);
>  
> +	/* Cancel any pending policy->update work before freeing the policy. */
> +	cancel_work_sync(&policy->update);
> +
>  	if (policy->max_freq_req) {
>  		/*
>  		 * CPUFREQ_CREATE_POLICY notification is sent only after
> @@ -1279,8 +1282,6 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
>  	}
>  
>  	freq_qos_remove_request(policy->min_freq_req);
> -	/* Cancel any pending policy->update work before freeing the policy. */
> -	cancel_work_sync(&policy->update);
>  	kfree(policy->min_freq_req);
>  
>  	cpufreq_policy_put_kobj(policy);

Thanks for doing this.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Rafael J. Wysocki Oct. 22, 2019, 9:46 a.m. UTC | #2
On Tue, Oct 22, 2019 at 4:25 AM Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> On 21-10-19, 14:28, Sudeep Holla wrote:
> > Commit 099967699ad9 ("cpufreq: Cancel policy update work scheduled before freeing")
> > added cancel_work_sync(policy->update) after the frequency QoS were
> > removed. We can cancel the work just after taking the last CPU in the
> > policy offline and unregistering the notifiers as policy->update cannot
> > be scheduled from anywhere at this point.
> >
> > However, due to other bugs, doing so still triggered the race between
> > freeing of policy and scheduled policy update work. Now that all those
> > issues are resolved, we can move this cancelling of any scheduled policy
> > update work just after removing min/max notifiers.
> >
> > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> > ---
> >  drivers/cpufreq/cpufreq.c | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > Hi Rafael,
> >
> > Based on Viresh's suggestion, I am posting a patch to move this
> > cancel_work_sync earlier though it's not a must have change.
>
> For me it is :)
>
> > I will leave it up to your preference.
> >
> > Regards,
> > Sudeep
> >
> > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > index 829a3764df1b..48a224a6b178 100644
> > --- a/drivers/cpufreq/cpufreq.c
> > +++ b/drivers/cpufreq/cpufreq.c
> > @@ -1268,6 +1268,9 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> >       freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
> >                                &policy->nb_min);
> >
> > +     /* Cancel any pending policy->update work before freeing the policy. */
> > +     cancel_work_sync(&policy->update);
> > +
> >       if (policy->max_freq_req) {
> >               /*
> >                * CPUFREQ_CREATE_POLICY notification is sent only after
> > @@ -1279,8 +1282,6 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> >       }
> >
> >       freq_qos_remove_request(policy->min_freq_req);
> > -     /* Cancel any pending policy->update work before freeing the policy. */
> > -     cancel_work_sync(&policy->update);
> >       kfree(policy->min_freq_req);
> >
> >       cpufreq_policy_put_kobj(policy);
>
> Thanks for doing this.
>
> Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

Folded into the previous patch and applied.

Please double check the result in the current linux-next branch in my tree.
Viresh Kumar Oct. 22, 2019, 10:07 a.m. UTC | #3
On 22-10-19, 11:46, Rafael J. Wysocki wrote:
> On Tue, Oct 22, 2019 at 4:25 AM Viresh Kumar <viresh.kumar@linaro.org> wrote:
> >
> > On 21-10-19, 14:28, Sudeep Holla wrote:
> > > Commit 099967699ad9 ("cpufreq: Cancel policy update work scheduled before freeing")
> > > added cancel_work_sync(policy->update) after the frequency QoS were
> > > removed. We can cancel the work just after taking the last CPU in the
> > > policy offline and unregistering the notifiers as policy->update cannot
> > > be scheduled from anywhere at this point.
> > >
> > > However, due to other bugs, doing so still triggered the race between
> > > freeing of policy and scheduled policy update work. Now that all those
> > > issues are resolved, we can move this cancelling of any scheduled policy
> > > update work just after removing min/max notifiers.
> > >
> > > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> > > ---
> > >  drivers/cpufreq/cpufreq.c | 5 +++--
> > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > >
> > > Hi Rafael,
> > >
> > > Based on Viresh's suggestion, I am posting a patch to move this
> > > cancel_work_sync earlier though it's not a must have change.
> >
> > For me it is :)
> >
> > > I will leave it up to your preference.
> > >
> > > Regards,
> > > Sudeep
> > >
> > > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > > index 829a3764df1b..48a224a6b178 100644
> > > --- a/drivers/cpufreq/cpufreq.c
> > > +++ b/drivers/cpufreq/cpufreq.c
> > > @@ -1268,6 +1268,9 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> > >       freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
> > >                                &policy->nb_min);
> > >
> > > +     /* Cancel any pending policy->update work before freeing the policy. */
> > > +     cancel_work_sync(&policy->update);
> > > +
> > >       if (policy->max_freq_req) {
> > >               /*
> > >                * CPUFREQ_CREATE_POLICY notification is sent only after
> > > @@ -1279,8 +1282,6 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> > >       }
> > >
> > >       freq_qos_remove_request(policy->min_freq_req);
> > > -     /* Cancel any pending policy->update work before freeing the policy. */
> > > -     cancel_work_sync(&policy->update);
> > >       kfree(policy->min_freq_req);
> > >
> > >       cpufreq_policy_put_kobj(policy);
> >
> > Thanks for doing this.
> >
> > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> 
> Folded into the previous patch and applied.
> 
> Please double check the result in the current linux-next branch in my tree.

I would have kept the blank line after cancel_work_sync() which isn't
there anymore.
Rafael J. Wysocki Oct. 22, 2019, 4:11 p.m. UTC | #4
On Tue, Oct 22, 2019 at 12:07 PM Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> On 22-10-19, 11:46, Rafael J. Wysocki wrote:
> > On Tue, Oct 22, 2019 at 4:25 AM Viresh Kumar <viresh.kumar@linaro.org> wrote:
> > >
> > > On 21-10-19, 14:28, Sudeep Holla wrote:
> > > > Commit 099967699ad9 ("cpufreq: Cancel policy update work scheduled before freeing")
> > > > added cancel_work_sync(policy->update) after the frequency QoS were
> > > > removed. We can cancel the work just after taking the last CPU in the
> > > > policy offline and unregistering the notifiers as policy->update cannot
> > > > be scheduled from anywhere at this point.
> > > >
> > > > However, due to other bugs, doing so still triggered the race between
> > > > freeing of policy and scheduled policy update work. Now that all those
> > > > issues are resolved, we can move this cancelling of any scheduled policy
> > > > update work just after removing min/max notifiers.
> > > >
> > > > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
> > > > ---
> > > >  drivers/cpufreq/cpufreq.c | 5 +++--
> > > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > >
> > > > Hi Rafael,
> > > >
> > > > Based on Viresh's suggestion, I am posting a patch to move this
> > > > cancel_work_sync earlier though it's not a must have change.
> > >
> > > For me it is :)
> > >
> > > > I will leave it up to your preference.
> > > >
> > > > Regards,
> > > > Sudeep
> > > >
> > > > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > > > index 829a3764df1b..48a224a6b178 100644
> > > > --- a/drivers/cpufreq/cpufreq.c
> > > > +++ b/drivers/cpufreq/cpufreq.c
> > > > @@ -1268,6 +1268,9 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> > > >       freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
> > > >                                &policy->nb_min);
> > > >
> > > > +     /* Cancel any pending policy->update work before freeing the policy. */
> > > > +     cancel_work_sync(&policy->update);
> > > > +
> > > >       if (policy->max_freq_req) {
> > > >               /*
> > > >                * CPUFREQ_CREATE_POLICY notification is sent only after
> > > > @@ -1279,8 +1282,6 @@ static void cpufreq_policy_free(struct cpufreq_policy *policy)
> > > >       }
> > > >
> > > >       freq_qos_remove_request(policy->min_freq_req);
> > > > -     /* Cancel any pending policy->update work before freeing the policy. */
> > > > -     cancel_work_sync(&policy->update);
> > > >       kfree(policy->min_freq_req);
> > > >
> > > >       cpufreq_policy_put_kobj(policy);
> > >
> > > Thanks for doing this.
> > >
> > > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> >
> > Folded into the previous patch and applied.
> >
> > Please double check the result in the current linux-next branch in my tree.
>
> I would have kept the blank line after cancel_work_sync() which isn't
> there anymore.

OK, it looks better with the extra blank line, so updated.  Thanks!

Patch
diff mbox series

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 829a3764df1b..48a224a6b178 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1268,6 +1268,9 @@  static void cpufreq_policy_free(struct cpufreq_policy *policy)
 	freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
 				 &policy->nb_min);
 
+	/* Cancel any pending policy->update work before freeing the policy. */
+	cancel_work_sync(&policy->update);
+
 	if (policy->max_freq_req) {
 		/*
 		 * CPUFREQ_CREATE_POLICY notification is sent only after
@@ -1279,8 +1282,6 @@  static void cpufreq_policy_free(struct cpufreq_policy *policy)
 	}
 
 	freq_qos_remove_request(policy->min_freq_req);
-	/* Cancel any pending policy->update work before freeing the policy. */
-	cancel_work_sync(&policy->update);
 	kfree(policy->min_freq_req);
 
 	cpufreq_policy_put_kobj(policy);