sche/fair: static cpumasks for load balance

[PATCH] sche/fair: static cpumasks for load balance

Posted by Bing Huang 1 year, 11 months ago

The both cpu mask load_balance_mask and select_idle_mask just only used
in fair.c, but allocation in core.c in CONFIG_CPUMASK_OFFSTACK=y case,
and global via declare per cpu variations. More or less, it looks wired.

Signed-off-by: Bing Huang <huangbing@kylinos.cn>
---
 kernel/sched/core.c  | 13 +++----------
 kernel/sched/fair.c  | 18 ++++++++++++++++--
 kernel/sched/sched.h |  4 ++++
 3 files changed, 23 insertions(+), 12 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index d58c0389eb23..64934c5e236d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -9448,9 +9448,6 @@ LIST_HEAD(task_groups);
 static struct kmem_cache *task_group_cache __read_mostly;
 #endif
 
-DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
-DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
-
 void __init sched_init(void)
 {
 	unsigned long ptr = 0;
@@ -9494,14 +9491,10 @@ void __init sched_init(void)
 
 #endif /* CONFIG_RT_GROUP_SCHED */
 	}
+
 #ifdef CONFIG_CPUMASK_OFFSTACK
-	for_each_possible_cpu(i) {
-		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
-			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
-		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
-			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
-	}
-#endif /* CONFIG_CPUMASK_OFFSTACK */
+	init_working_cpumask();
+#endif
 
 	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
 
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index a68482d66535..3a4b0dd1beb5 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5831,8 +5831,8 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 #ifdef CONFIG_SMP
 
 /* Working cpumask for: load_balance, load_balance_newidle. */
-DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
-DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
+static DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
+static DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
 
 #ifdef CONFIG_NO_HZ_COMMON
 
@@ -7052,6 +7052,20 @@ balance_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 }
 #endif /* CONFIG_SMP */
 
+#ifdef CONFIG_CPUMASK_OFFSTACK
+void __init init_working_cpumask(void)
+{
+	int i;
+
+	for_each_possible_cpu(i) {
+		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
+				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
+		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
+				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
+	}
+}
+#endif
+
 static unsigned long wakeup_gran(struct sched_entity *se)
 {
 	unsigned long gran = sysctl_sched_wakeup_granularity;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 8dccb34eb190..9065be3def51 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -3087,4 +3087,8 @@ extern int sched_dynamic_mode(const char *str);
 extern void sched_dynamic_update(int mode);
 #endif
 
+#ifdef CONFIG_CPUMASK_OFFSTACK
+extern void __init init_working_cpumask(void);
+#endif /* CONFIG_CPUMASK_OFFSTACK */
+
 #endif /* _KERNEL_SCHED_SCHED_H */
-- 
2.25.1


No virus found
		Checked by Hillstone Network AntiVirus

Re: [PATCH] sche/fair: static cpumasks for load balance

Posted by Peter Zijlstra 1 year, 10 months ago

On Mon, May 23, 2022 at 05:42:16PM +0800, Bing Huang wrote:
> The both cpu mask load_balance_mask and select_idle_mask just only used
> in fair.c, but allocation in core.c in CONFIG_CPUMASK_OFFSTACK=y case,
> and global via declare per cpu variations. More or less, it looks wired.
> 
> Signed-off-by: Bing Huang <huangbing@kylinos.cn>
> ---
>  kernel/sched/core.c  | 13 +++----------
>  kernel/sched/fair.c  | 18 ++++++++++++++++--
>  kernel/sched/sched.h |  4 ++++
>  3 files changed, 23 insertions(+), 12 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index d58c0389eb23..64934c5e236d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -9448,9 +9448,6 @@ LIST_HEAD(task_groups);
>  static struct kmem_cache *task_group_cache __read_mostly;
>  #endif
>  
> -DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
> -DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
> -
>  void __init sched_init(void)
>  {
>  	unsigned long ptr = 0;
> @@ -9494,14 +9491,10 @@ void __init sched_init(void)
>  
>  #endif /* CONFIG_RT_GROUP_SCHED */
>  	}
> +
>  #ifdef CONFIG_CPUMASK_OFFSTACK
> -	for_each_possible_cpu(i) {
> -		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
> -			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> -		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
> -			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> -	}
> -#endif /* CONFIG_CPUMASK_OFFSTACK */
> +	init_working_cpumask();
> +#endif

We already have init_sched_fair_class(), can't you use that?

>  
>  	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
>  
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index a68482d66535..3a4b0dd1beb5 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -5831,8 +5831,8 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
>  #ifdef CONFIG_SMP
>  
>  /* Working cpumask for: load_balance, load_balance_newidle. */
> -DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
> -DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
> +static DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
> +static DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
>  
>  #ifdef CONFIG_NO_HZ_COMMON
>  
> @@ -7052,6 +7052,20 @@ balance_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
>  }
>  #endif /* CONFIG_SMP */
>  
> +#ifdef CONFIG_CPUMASK_OFFSTACK
> +void __init init_working_cpumask(void)
> +{
> +	int i;
> +
> +	for_each_possible_cpu(i) {
> +		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
> +				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> +		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
> +				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> +	}
> +}
> +#endif

If you stick that #ifdef inside the function, you can remove it
everwhere else. Less #ifdef is more better.

Re: [PATCH] sche/fair: static cpumasks for load balance

Posted by huangbing 1 year, 10 months ago


On 5/26/22 01:04, Peter Zijlstra wrote:
> On Mon, May 23, 2022 at 05:42:16PM +0800, Bing Huang wrote:
>> The both cpu mask load_balance_mask and select_idle_mask just only used
>> in fair.c, but allocation in core.c in CONFIG_CPUMASK_OFFSTACK=y case,
>> and global via declare per cpu variations. More or less, it looks wired.
>>
>> Signed-off-by: Bing Huang <huangbing@kylinos.cn>
>> ---
>>   kernel/sched/core.c  | 13 +++----------
>>   kernel/sched/fair.c  | 18 ++++++++++++++++--
>>   kernel/sched/sched.h |  4 ++++
>>   3 files changed, 23 insertions(+), 12 deletions(-)
>>
>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>> index d58c0389eb23..64934c5e236d 100644
>> --- a/kernel/sched/core.c
>> +++ b/kernel/sched/core.c
>> @@ -9448,9 +9448,6 @@ LIST_HEAD(task_groups);
>>   static struct kmem_cache *task_group_cache __read_mostly;
>>   #endif
>>   
>> -DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
>> -DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
>> -
>>   void __init sched_init(void)
>>   {
>>   	unsigned long ptr = 0;
>> @@ -9494,14 +9491,10 @@ void __init sched_init(void)
>>   
>>   #endif /* CONFIG_RT_GROUP_SCHED */
>>   	}
>> +
>>   #ifdef CONFIG_CPUMASK_OFFSTACK
>> -	for_each_possible_cpu(i) {
>> -		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
>> -			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
>> -		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
>> -			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
>> -	}
>> -#endif /* CONFIG_CPUMASK_OFFSTACK */
>> +	init_working_cpumask();
>> +#endif
> We already have init_sched_fair_class(), can't you use that?

OK, thanks for your review. I got it.

In addition, I notice that load_balance_mask and select_idle_mask are 
defined
in CONFIG_SMP=y case in fair.c, so could be they moved inside 
CONFIG_SMP, or just outside ?

>
>>   
>>   	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
>>   
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index a68482d66535..3a4b0dd1beb5 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -5831,8 +5831,8 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
>>   #ifdef CONFIG_SMP
>>   
>>   /* Working cpumask for: load_balance, load_balance_newidle. */
>> -DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
>> -DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
>> +static DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
>> +static DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);
>>   
>>   #ifdef CONFIG_NO_HZ_COMMON
>>   
>> @@ -7052,6 +7052,20 @@ balance_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
>>   }
>>   #endif /* CONFIG_SMP */
>>   
>> +#ifdef CONFIG_CPUMASK_OFFSTACK
>> +void __init init_working_cpumask(void)
>> +{
>> +	int i;
>> +
>> +	for_each_possible_cpu(i) {
>> +		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
>> +				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
>> +		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
>> +				cpumask_size(), GFP_KERNEL, cpu_to_node(i));
>> +	}
>> +}
>> +#endif
> If you stick that #ifdef inside the function, you can remove it
> everwhere else. Less #ifdef is more better.
I will remove it, move them inside init_sched_fair_class().