[PATCH v2] workqueue: make sysfs of unbound kworker cpumask more clever

[PATCH v2] workqueue: make sysfs of unbound kworker cpumask more clever

[menglong8.dong]

From: Menglong Dong <imagedong@tencent.com>

Some unfriendly component, such as dpdk, write the same mask to
unbound kworker cpumask again and again. Every time it write to
this interface some work is queue to cpu, even though the mask
is same with the original mask.

So, fix it by return success and do nothing if the cpumask is
equal with the old one.

Signed-off-by: Mengen Sun <mengensun@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
---
v2:
- reorganize the code to avoid unnecessary saved_cpumask alloc
---
 kernel/workqueue.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 1b3eb1e9531f..76988f39ed5a 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -5384,9 +5384,6 @@ int workqueue_set_unbound_cpumask(cpumask_var_t cpumask)
 	int ret = -EINVAL;
 	cpumask_var_t saved_cpumask;
 
-	if (!zalloc_cpumask_var(&saved_cpumask, GFP_KERNEL))
-		return -ENOMEM;
-
 	/*
 	 * Not excluding isolated cpus on purpose.
 	 * If the user wishes to include them, we allow that.
@@ -5394,6 +5391,15 @@ int workqueue_set_unbound_cpumask(cpumask_var_t cpumask)
 	cpumask_and(cpumask, cpumask, cpu_possible_mask);
 	if (!cpumask_empty(cpumask)) {
 		apply_wqattrs_lock();
+		if (cpumask_equal(cpumask, wq_unbound_cpumask)) {
+			ret = 0;
+			goto out_unlock;
+		}
+
+		if (!zalloc_cpumask_var(&saved_cpumask, GFP_KERNEL)) {
+			ret = -ENOMEM;
+			goto out_unlock;
+		}
 
 		/* save the old wq_unbound_cpumask. */
 		cpumask_copy(saved_cpumask, wq_unbound_cpumask);
@@ -5406,10 +5412,11 @@ int workqueue_set_unbound_cpumask(cpumask_var_t cpumask)
 		if (ret < 0)
 			cpumask_copy(wq_unbound_cpumask, saved_cpumask);
 
+		free_cpumask_var(saved_cpumask);
+out_unlock:
 		apply_wqattrs_unlock();
 	}
 
-	free_cpumask_var(saved_cpumask);
 	return ret;
 }
 
-- 
2.27.0

Re: [PATCH v2] workqueue: make sysfs of unbound kworker cpumask more clever

[Tejun Heo]

On Sun, Oct 17, 2021 at 08:04:02PM +0800, menglong8.dong@gmail.com wrote:
> From: Menglong Dong <imagedong@tencent.com>
> 
> Some unfriendly component, such as dpdk, write the same mask to
> unbound kworker cpumask again and again. Every time it write to
> this interface some work is queue to cpu, even though the mask
> is same with the original mask.
> 
> So, fix it by return success and do nothing if the cpumask is
> equal with the old one.
> 
> Signed-off-by: Mengen Sun <mengensun@tencent.com>
> Signed-off-by: Menglong Dong <imagedong@tencent.com>

Applied to wq/for-5.16.

Thanks.

-- 
tejun