Re: [PATCH RESEND] sched: Move some scheduler fields to new static branch API

[Hongyan Xia <hongyan.xia@transsion.com>]

On 8/19/2026 8:19 PM, Shrikanth Hegde wrote:
> 
> Hi Hongyan.
> 
> On 8/19/26 1:39 PM, Hongyan Xia wrote:
>> From: Hongyan Xia <hongyan.xia@transsion.com>
>>
>> __cfs_bandwidth_used uses struct static_key directly which is
>> deprecated. Fix.
> 
> A bit of context on why it is deprecated would help.

Sure. Will add some context in the next rev.

>>
>> sk_dynamic_* uses static_key_{enable/disable}(), which aren't really
>> deprecated, but take the opportunity to move to the new static_branch_*
>> APIs to be consistent.
>>
> 
> Mark had a series to remove a few of them completely. IIRC only lazy
> check will remain.
> 
> https://apc01.safelinks.protection.outlook.com/? 
> url=https%3A%2F%2Flore.kernel.org%2Fall%2F20260803191731.3244294-1- 
> mark.rutland%40arm.com%2F&data=05%7C02%7Chongyan.xia%40transsion.com%7C65194a86312643aed75b08defdec3f5a%7C2e8503a62d0143338e366ab7c8cd7ae2%7C0%7C0%7C639227388298975336%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=KbAcWOiQg%2Bir4%2FzKW2KG66IkxSnbq79m2I9IUlLe4os%3D&reserved=0
> 

I see. These two do conflict. Would be nice if Mark's patch gets merged 
first, then it makes this diff smaller.

>> No functional change.
>>
>> Signed-off-by: Hongyan Xia <hongyan.xia@transsion.com>
>> ---
>> Changed in RESEND:
>> - Separate the original series into individual patches. They aren't easy
>>    to review as a series.
>> - Move sk_dynamic_* to the new API as well.
>>
>>   kernel/sched/core.c | 18 +++++++++---------
>>   kernel/sched/fair.c |  8 ++++----
>>   2 files changed, 13 insertions(+), 13 deletions(-)
>>
>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>> index 96226707c2f6..5c07d53e43b5 100644
>> --- a/kernel/sched/core.c
>> +++ b/kernel/sched/core.c
>> @@ -7933,15 +7933,15 @@ int sched_dynamic_mode(const char *str)
>>       return -EINVAL;
>>   }
>>
>> -# define preempt_dynamic_key_enable(f)       
>> static_key_enable(&sk_dynamic_##f.key)
>> -# define preempt_dynamic_key_disable(f)      
>> static_key_disable(&sk_dynamic_##f.key)
>> +# define preempt_dynamic_branch_enable(f)    
>> static_branch_enable(&sk_dynamic_##f)
>> +# define preempt_dynamic_branch_disable(f)   
>> static_branch_disable(&sk_dynamic_##f)
> 
> Also, doc doesn't about mention about static_branch_enable vs 
> static_key_enable.
> is static_key_enable also deprecated.

Right, like mentioned in the commit message, this change here is a 
nice-to-have, not really deprecation. The trend is to move from untyped 
static_key_* variants to typed static_branch_* ones, so I took the 
opportunity to fix it.

>>
>>   # if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
>>   #  define preempt_dynamic_enable(f) static_call_update(f, 
>> f##_dynamic_enabled)
>>   #  define preempt_dynamic_disable(f)        static_call_update(f, 
>> f##_dynamic_disabled)
>>   # elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
>> -#  define preempt_dynamic_enable(f)  preempt_dynamic_key_enable(f)
>> -#  define preempt_dynamic_disable(f) preempt_dynamic_key_disable(f)
>> +#  define preempt_dynamic_enable(f)  preempt_dynamic_branch_enable(f)
>> +#  define preempt_dynamic_disable(f) preempt_dynamic_branch_disable(f)
>>   # else
>>   #  error "Unsupported PREEMPT_DYNAMIC mechanism"
>>   # endif
>> @@ -7959,7 +7959,7 @@ static void __sched_dynamic_update(int mode)
>>       preempt_dynamic_enable(preempt_schedule);
>>       preempt_dynamic_enable(preempt_schedule_notrace);
>>       preempt_dynamic_enable(irqentry_exit_cond_resched);
>> -     preempt_dynamic_key_disable(preempt_lazy);
>> +     preempt_dynamic_branch_disable(preempt_lazy);
>>
>>       switch (mode) {
>>       case preempt_dynamic_none:
>> @@ -7968,7 +7968,7 @@ static void __sched_dynamic_update(int mode)
>>               preempt_dynamic_disable(preempt_schedule);
>>               preempt_dynamic_disable(preempt_schedule_notrace);
>>               preempt_dynamic_disable(irqentry_exit_cond_resched);
>> -             preempt_dynamic_key_disable(preempt_lazy);
>> +             preempt_dynamic_branch_disable(preempt_lazy);
>>               if (mode != preempt_dynamic_mode)
>>                       pr_info("Dynamic Preempt: none\n");
>>               break;
>> @@ -7979,7 +7979,7 @@ static void __sched_dynamic_update(int mode)
>>               preempt_dynamic_disable(preempt_schedule);
>>               preempt_dynamic_disable(preempt_schedule_notrace);
>>               preempt_dynamic_disable(irqentry_exit_cond_resched);
>> -             preempt_dynamic_key_disable(preempt_lazy);
>> +             preempt_dynamic_branch_disable(preempt_lazy);
>>               if (mode != preempt_dynamic_mode)
>>                       pr_info("Dynamic Preempt: voluntary\n");
>>               break;
>> @@ -7990,7 +7990,7 @@ static void __sched_dynamic_update(int mode)
>>               preempt_dynamic_enable(preempt_schedule);
>>               preempt_dynamic_enable(preempt_schedule_notrace);
>>               preempt_dynamic_enable(irqentry_exit_cond_resched);
>> -             preempt_dynamic_key_disable(preempt_lazy);
>> +             preempt_dynamic_branch_disable(preempt_lazy);
>>               if (mode != preempt_dynamic_mode)
>>                       pr_info("Dynamic Preempt: full\n");
>>               break;
>> @@ -8001,7 +8001,7 @@ static void __sched_dynamic_update(int mode)
>>               preempt_dynamic_enable(preempt_schedule);
>>               preempt_dynamic_enable(preempt_schedule_notrace);
>>               preempt_dynamic_enable(irqentry_exit_cond_resched);
>> -             preempt_dynamic_key_enable(preempt_lazy);
>> +             preempt_dynamic_branch_enable(preempt_lazy);
>>               if (mode != preempt_dynamic_mode)
>>                       pr_info("Dynamic Preempt: lazy\n");
>>               break;
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 762dd8a4647c..4b65be5ec471 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -6442,21 +6442,21 @@ entity_tick(struct cfs_rq *cfs_rq, struct 
>> sched_entity *curr, int queued)
>>   #ifdef CONFIG_CFS_BANDWIDTH
>>
>>   #ifdef CONFIG_JUMP_LABEL
>> -static struct static_key __cfs_bandwidth_used;
>> +static DEFINE_STATIC_KEY_FALSE(__cfs_bandwidth_used);
>>
>>   static inline bool cfs_bandwidth_used(void)
>>   {
>> -     return static_key_false(&__cfs_bandwidth_used);
>> +     return static_branch_unlikely(&__cfs_bandwidth_used);
>>   }
>>
> 
> 
> We will still have two left.
> 
> core.c: if (static_key_false((&paravirt_steal_rq_enabled))) {
> cputime.c:      if (static_key_false(&paravirt_steal_enabled)) {
> 

That has been sent, at

https://lore.kernel.org/all/20260819081207.12150-1-hongyan.xia@transsion.com/

That touches all archs and may need review from Xen folks, so I split it 
into a separate patch.

>>   void cfs_bandwidth_usage_inc(void)
>>   {
>> -     static_key_slow_inc_cpuslocked(&__cfs_bandwidth_used);
>> +     static_branch_inc_cpuslocked(&__cfs_bandwidth_used);
>>   }
>>
>>   void cfs_bandwidth_usage_dec(void)
>>   {
>> -     static_key_slow_dec_cpuslocked(&__cfs_bandwidth_used);
>> +     static_branch_dec_cpuslocked(&__cfs_bandwidth_used);
>>   }
>>   #else /* !CONFIG_JUMP_LABEL: */
>>   static bool cfs_bandwidth_used(void)
>