[PATCH v2 0/2] cpufreq: Fix global boost state updates

[PATCH v2 0/2] cpufreq: Fix global boost state updates

[Christian Loehle]

Preserve the global boost state on failure and serialize concurrent
global boost writes.

Changes in v2:
- Split rollback and serialization into separate patches, with the
  corresponding Fixes tags (Zhongqiu Han)
- Carry Pengjie Zhang's Reviewed-by onto both patches since code
  is unchanged.

v1: https://lore.kernel.org/r/20261104161657.546370-1-christian.loehle@arm.com

Christian Loehle (2):
  cpufreq: Preserve global boost state on failed updates
  cpufreq: Serialize global boost updates

[PATCH v2 1/2] cpufreq: Preserve global boost state on failed updates

[Christian Loehle]

cpufreq_boost_trigger_state() restores the opposite of the requested
state when updating policy boost state fails. For example, a failed
write of 0 when global boost is already disabled enables the global
flag without enabling any policies.

Save the original global boost state and restore it on failure.

Fixes: 121baab7b88e ("cpufreq: Force sync policy boost with global boost on sysfs update")
Signed-off-by: Christian Loehle <christian.loehle@arm.com>
Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
---
 drivers/cpufreq/cpufreq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 54dde8419bdc..23a77f5524df 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -2814,6 +2814,7 @@ static int cpufreq_boost_trigger_state(int state)
 {
 	struct cpufreq_policy *policy;
 	unsigned long flags;
+	bool boost_enabled;
 	int ret = -EOPNOTSUPP;
 
 	/*
@@ -2822,6 +2823,7 @@ static int cpufreq_boost_trigger_state(int state)
 	 */
 
 	write_lock_irqsave(&cpufreq_driver_lock, flags);
+	boost_enabled = cpufreq_driver->boost_enabled;
 	cpufreq_driver->boost_enabled = state;
 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
 
@@ -2841,7 +2843,7 @@ static int cpufreq_boost_trigger_state(int state)
 		return 0;
 
 	write_lock_irqsave(&cpufreq_driver_lock, flags);
-	cpufreq_driver->boost_enabled = !state;
+	cpufreq_driver->boost_enabled = boost_enabled;
 	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
 
 	pr_err("%s: Cannot %s BOOST\n",
-- 
2.34.1

[PATCH v2 2/2] cpufreq: Serialize global boost updates

[Christian Loehle]

Serialize concurrent global boost sysfs writes with a mutex.

Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
Signed-off-by: Christian Loehle <christian.loehle@arm.com>
Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
---
 drivers/cpufreq/cpufreq.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 23a77f5524df..d95c849f79c5 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
 }
 EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
 
+static DEFINE_MUTEX(cpufreq_boost_mutex);
+
 static int cpufreq_boost_trigger_state(int state)
 {
 	struct cpufreq_policy *policy;
@@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
 	bool boost_enabled;
 	int ret = -EOPNOTSUPP;
 
+	guard(mutex)(&cpufreq_boost_mutex);
+
 	/*
 	 * Don't compare 'cpufreq_driver->boost_enabled' with 'state' here to
 	 * make sure all policies are in sync with global boost flag.
-- 
2.34.1

Re: [PATCH v2 1/2] cpufreq: Preserve global boost state on failed updates

[Zhongqiu Han]

On 11/5/2026 5:03 PM, Christian Loehle wrote:
> cpufreq_boost_trigger_state() restores the opposite of the requested
> state when updating policy boost state fails. For example, a failed
> write of 0 when global boost is already disabled enables the global
> flag without enabling any policies.
> 
> Save the original global boost state and restore it on failure.
> 
> Fixes: 121baab7b88e ("cpufreq: Force sync policy boost with global boost on sysfs update")
> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>

Reviewed-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>

> ---
>   drivers/cpufreq/cpufreq.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 54dde8419bdc..23a77f5524df 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -2814,6 +2814,7 @@ static int cpufreq_boost_trigger_state(int state)
>   {
>   	struct cpufreq_policy *policy;
>   	unsigned long flags;
> +	bool boost_enabled;
>   	int ret = -EOPNOTSUPP;
>   
>   	/*
> @@ -2822,6 +2823,7 @@ static int cpufreq_boost_trigger_state(int state)
>   	 */
>   
>   	write_lock_irqsave(&cpufreq_driver_lock, flags);
> +	boost_enabled = cpufreq_driver->boost_enabled;
>   	cpufreq_driver->boost_enabled = state;
>   	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
>   
> @@ -2841,7 +2843,7 @@ static int cpufreq_boost_trigger_state(int state)
>   		return 0;
>   
>   	write_lock_irqsave(&cpufreq_driver_lock, flags);
> -	cpufreq_driver->boost_enabled = !state;
> +	cpufreq_driver->boost_enabled = boost_enabled;
>   	write_unlock_irqrestore(&cpufreq_driver_lock, flags);
>   
>   	pr_err("%s: Cannot %s BOOST\n",


-- 
Thx and BRs,
Zhongqiu Han

Re: [PATCH v2 1/2] cpufreq: Preserve global boost state on failed updates

[Sumit Gupta]

On 05/11/26 14:33, Christian Loehle wrote:
> External email: Use caution opening links or attachments
> 
> 
> cpufreq_boost_trigger_state() restores the opposite of the requested
> state when updating policy boost state fails. For example, a failed
> write of 0 when global boost is already disabled enables the global
> flag without enabling any policies.
> 
> Save the original global boost state and restore it on failure.
> 
> Fixes: 121baab7b88e ("cpufreq: Force sync policy boost with global boost on sysfs update")
> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
> ---

Reviewed-by: Sumit Gupta <sumitg@nvidia.com>


>   drivers/cpufreq/cpufreq.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 54dde8419bdc..23a77f5524df 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -2814,6 +2814,7 @@ static int cpufreq_boost_trigger_state(int state)
>   {
>          struct cpufreq_policy *policy;
>          unsigned long flags;
> +       bool boost_enabled;
>          int ret = -EOPNOTSUPP;
> 
>          /*
> @@ -2822,6 +2823,7 @@ static int cpufreq_boost_trigger_state(int state)
>           */
> 
>          write_lock_irqsave(&cpufreq_driver_lock, flags);
> +       boost_enabled = cpufreq_driver->boost_enabled;
>          cpufreq_driver->boost_enabled = state;
>          write_unlock_irqrestore(&cpufreq_driver_lock, flags);
> 
> @@ -2841,7 +2843,7 @@ static int cpufreq_boost_trigger_state(int state)
>                  return 0;
> 
>          write_lock_irqsave(&cpufreq_driver_lock, flags);
> -       cpufreq_driver->boost_enabled = !state;
> +       cpufreq_driver->boost_enabled = boost_enabled;
>          write_unlock_irqrestore(&cpufreq_driver_lock, flags);
> 
>          pr_err("%s: Cannot %s BOOST\n",
> --
> 2.34.1
> 

Re: [PATCH v2 2/2] cpufreq: Serialize global boost updates

[Sumit Gupta]

On 05/11/26 14:33, Christian Loehle wrote:
> External email: Use caution opening links or attachments
> 
> 
> Serialize concurrent global boost sysfs writes with a mutex.
> 
> Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
> ---
>   drivers/cpufreq/cpufreq.c | 4 ++++
>   1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 23a77f5524df..d95c849f79c5 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
>   }
>   EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
> 
> +static DEFINE_MUTEX(cpufreq_boost_mutex);
> +
>   static int cpufreq_boost_trigger_state(int state)
>   {
>          struct cpufreq_policy *policy;
> @@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
>          bool boost_enabled;
>          int ret = -EOPNOTSUPP;
> 
> +       guard(mutex)(&cpufreq_boost_mutex);
> +


The policy->rwsem case from [1] seems to be still open. The per-policy
boost attribute calls policy_set_boost() with policy->rwsem held, but
cpufreq_boost_trigger_state() and cpufreq_online() don't, so they can
race with per-policy writers like ospm_nominal_freq in [1].

Could cpufreq_boost_trigger_state() take the policy lock, as
cpufreq_suspend() does?

   scoped_guard(cpufreq_policy_write, policy)
       ret = policy_set_boost(policy, state);

[1] https://lore.kernel.org/lkml/20261030172745.2045743-4-sumitg@nvidia.com/

Thanks,
Sumit

Re: [PATCH v2 2/2] cpufreq: Serialize global boost updates

[Zhongqiu Han]

On 11/5/2026 10:41 PM, Sumit Gupta wrote:
> 
> 
> On 05/11/26 14:33, Christian Loehle wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> Serialize concurrent global boost sysfs writes with a mutex.
>>
>> Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
>> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
>> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
>> ---
>>   drivers/cpufreq/cpufreq.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>> index 23a77f5524df..d95c849f79c5 100644
>> --- a/drivers/cpufreq/cpufreq.c
>> +++ b/drivers/cpufreq/cpufreq.c
>> @@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy 
>> *policy, int state)
>>   }
>>   EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
>>
>> +static DEFINE_MUTEX(cpufreq_boost_mutex);
>> +
>>   static int cpufreq_boost_trigger_state(int state)
>>   {
>>          struct cpufreq_policy *policy;
>> @@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
>>          bool boost_enabled;
>>          int ret = -EOPNOTSUPP;
>>
>> +       guard(mutex)(&cpufreq_boost_mutex);
>> +
> 
> 
> The policy->rwsem case from [1] seems to be still open. The per-policy
> boost attribute calls policy_set_boost() with policy->rwsem held, but
> cpufreq_boost_trigger_state() and cpufreq_online() don't, so they can
> race with per-policy writers like ospm_nominal_freq in [1].
> 
> Could cpufreq_boost_trigger_state() take the policy lock, as
> cpufreq_suspend() does?
> 
>    scoped_guard(cpufreq_policy_write, policy)
>        ret = policy_set_boost(policy, state);
> 
> [1] https://lore.kernel.org/lkml/20261030172745.2045743-4- 
> sumitg@nvidia.com/
> 

Hi Sumit,

Could you please check whether the following patch addresses the issue
you mentioned?

https://lore.kernel.org/all/be3739f8-e668-4711-912c-91459d77ed35@arm.com/



> Thanks,
> Sumit
> 
> 


-- 
Thx and BRs,
Zhongqiu Han

Re: [PATCH v2 2/2] cpufreq: Serialize global boost updates

[Zhongqiu Han]

On 11/5/2026 5:03 PM, Christian Loehle wrote:
> Serialize concurrent global boost sysfs writes with a mutex.
> 
> Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>

The change looks good to me. It might be worth expanding the commit
message a little to describe the race between concurrent global boost
sysfs writes and why serialization is needed.

With that minor comment:

Reviewed-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>

> ---
>   drivers/cpufreq/cpufreq.c | 4 ++++
>   1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 23a77f5524df..d95c849f79c5 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
>   }
>   EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
>   
> +static DEFINE_MUTEX(cpufreq_boost_mutex);
> +
>   static int cpufreq_boost_trigger_state(int state)
>   {
>   	struct cpufreq_policy *policy;
> @@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
>   	bool boost_enabled;
>   	int ret = -EOPNOTSUPP;
>   
> +	guard(mutex)(&cpufreq_boost_mutex);
> +
>   	/*
>   	 * Don't compare 'cpufreq_driver->boost_enabled' with 'state' here to
>   	 * make sure all policies are in sync with global boost flag.


-- 
Thx and BRs,
Zhongqiu Han

Re: [PATCH v2 2/2] cpufreq: Serialize global boost updates

[Sumit Gupta]

On 05/11/26 20:31, Zhongqiu Han wrote:
> External email: Use caution opening links or attachments
> 
> 
> On 11/5/2026 10:41 PM, Sumit Gupta wrote:
>>
>>
>> On 05/11/26 14:33, Christian Loehle wrote:
>>> External email: Use caution opening links or attachments
>>>
>>>
>>> Serialize concurrent global boost sysfs writes with a mutex.
>>>
>>> Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
>>> Signed-off-by: Christian Loehle <christian.loehle@arm.com>
>>> Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
>>> ---
>>>   drivers/cpufreq/cpufreq.c | 4 ++++
>>>   1 file changed, 4 insertions(+)
>>>
>>> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
>>> index 23a77f5524df..d95c849f79c5 100644
>>> --- a/drivers/cpufreq/cpufreq.c
>>> +++ b/drivers/cpufreq/cpufreq.c
>>> @@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy
>>> *policy, int state)
>>>   }
>>>   EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
>>>
>>> +static DEFINE_MUTEX(cpufreq_boost_mutex);
>>> +
>>>   static int cpufreq_boost_trigger_state(int state)
>>>   {
>>>          struct cpufreq_policy *policy;
>>> @@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
>>>          bool boost_enabled;
>>>          int ret = -EOPNOTSUPP;
>>>
>>> +       guard(mutex)(&cpufreq_boost_mutex);
>>> +
>>
>>
>> The policy->rwsem case from [1] seems to be still open. The per-policy
>> boost attribute calls policy_set_boost() with policy->rwsem held, but
>> cpufreq_boost_trigger_state() and cpufreq_online() don't, so they can
>> race with per-policy writers like ospm_nominal_freq in [1].
>>
>> Could cpufreq_boost_trigger_state() take the policy lock, as
>> cpufreq_suspend() does?
>>
>>    scoped_guard(cpufreq_policy_write, policy)
>>        ret = policy_set_boost(policy, state);
>>
>> [1] https://lore.kernel.org/lkml/20261030172745.2045743-4-
>> sumitg@nvidia.com/
>>
> 
> Hi Sumit,
> 
> Could you please check whether the following patch addresses the issue
> you mentioned?
> 
> https://lore.kernel.org/all/be3739f8-e668-4711-912c-91459d77ed35@arm.com/
> 
> 
Yes, that covers it. I missed that. Please ignore my comment.

Reviewed-by: Sumit Gupta <sumitg@nvidia.com>

Thanks,
Sumit

Re: [PATCH v2 2/2] cpufreq: Serialize global boost updates

[Rafael J. Wysocki (Intel)]

On Thu, Nov 5, 2026 at 4:01 PM Zhongqiu Han
<zhongqiu.han@oss.qualcomm.com> wrote:
>
> On 11/5/2026 5:03 PM, Christian Loehle wrote:
> > Serialize concurrent global boost sysfs writes with a mutex.
> >
> > Fixes: 6f19efc0a1ca ("cpufreq: Add boost frequency support in core")
> > Signed-off-by: Christian Loehle <christian.loehle@arm.com>
> > Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>
>
> The change looks good to me. It might be worth expanding the commit
> message a little to describe the race between concurrent global boost
> sysfs writes and why serialization is needed.

Yeah, it would be good to say why this change is necessary.

> With that minor comment:
>
> Reviewed-by: Zhongqiu Han <zhongqiu.han@oss.qualcomm.com>
>
> > ---
> >   drivers/cpufreq/cpufreq.c | 4 ++++
> >   1 file changed, 4 insertions(+)
> >
> > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > index 23a77f5524df..d95c849f79c5 100644
> > --- a/drivers/cpufreq/cpufreq.c
> > +++ b/drivers/cpufreq/cpufreq.c
> > @@ -2810,6 +2810,8 @@ int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
> >   }
> >   EXPORT_SYMBOL_GPL(cpufreq_boost_set_sw);
> >
> > +static DEFINE_MUTEX(cpufreq_boost_mutex);
> > +
> >   static int cpufreq_boost_trigger_state(int state)
> >   {
> >       struct cpufreq_policy *policy;
> > @@ -2817,6 +2819,8 @@ static int cpufreq_boost_trigger_state(int state)
> >       bool boost_enabled;
> >       int ret = -EOPNOTSUPP;
> >
> > +     guard(mutex)(&cpufreq_boost_mutex);
> > +
> >       /*
> >        * Don't compare 'cpufreq_driver->boost_enabled' with 'state' here to
> >        * make sure all policies are in sync with global boost flag.
>
>
> --
> Thx and BRs,
> Zhongqiu Han