sched/topology: avoid calling synchronize_rcu()

[PATCH] sched/topology: avoid calling synchronize_rcu()

Posted by cgel.zte@gmail.com 4 years, 3 months ago

From: Lv Ruyi (CGEL ZTE) <lv.ruyi@zte.com.cn>

Kfree_rcu() usually results in even simpler code than does
synchronize_rcu() without synchronize_rcu()'s multi-millisecond
latency, so replace synchronize_rcu() with kfree_rcu().

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: Lv Ruyi (CGEL ZTE) <lv.ruyi@zte.com.cn>
---
 kernel/sched/topology.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 974212620fa1..0ce302b4d732 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1970,8 +1970,7 @@ static void sched_reset_numa(void)
 	if (distances || masks) {
 		int i, j;
 
-		synchronize_rcu();
-		kfree(distances);
+		kfree_rcu(distances);
 		for (i = 0; i < nr_levels && masks; i++) {
 			if (!masks[i])
 				continue;
-- 
2.25.1

Re: [PATCH] sched/topology: avoid calling synchronize_rcu()

Posted by Valentin Schneider 4 years, 3 months ago

On 02/03/22 01:14, cgel.zte@gmail.com wrote:
> From: Lv Ruyi (CGEL ZTE) <lv.ruyi@zte.com.cn>
>
> Kfree_rcu() usually results in even simpler code than does
> synchronize_rcu() without synchronize_rcu()'s multi-millisecond
> latency, so replace synchronize_rcu() with kfree_rcu().
>
> Reported-by: Zeal Robot <zealci@zte.com.cn>
> Signed-off-by: Lv Ruyi (CGEL ZTE) <lv.ruyi@zte.com.cn>
> ---
>  kernel/sched/topology.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index 974212620fa1..0ce302b4d732 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -1970,8 +1970,7 @@ static void sched_reset_numa(void)
>       if (distances || masks) {
>               int i, j;
>
> -		synchronize_rcu();
> -		kfree(distances);
> +		kfree_rcu(distances);

What about the freeing of the cpumasks below? You don't want to free those
before the end of the grace period either. With that in mind, I think the
current synchronize_rcu() + batch of kfree()'s is fine, it's a slow path
anyway (hotplug).


>               for (i = 0; i < nr_levels && masks; i++) {
>                       if (!masks[i])
>                               continue;
> --
> 2.25.1