workqueue: Disable per-cpu CPU hog detection when wq_cpu_intensive_thresh_us is 0
authorZqiang <qiang.zhang1211@gmail.com>
Thu, 25 May 2023 04:00:38 +0000 (12:00 +0800)
committerTejun Heo <tj@kernel.org>
Thu, 25 May 2023 20:46:53 +0000 (10:46 -1000)
If workqueue.cpu_intensive_thresh_us is set to 0, the detection mechanism
for CPU-hogging per-cpu work item will keep triggering spuriously:

  workqueue: process_srcu hogged CPU for >0us 4 times, consider switching to WQ_UNBOUND
  workqueue: gc_worker hogged CPU for >0us 4 times, consider switching to WQ_UNBOUND
  workqueue: gc_worker hogged CPU for >0us 8 times, consider switching to WQ_UNBOUND
  workqueue: wait_rcu_exp_gp hogged CPU for >0us 4 times, consider switching to WQ_UNBOUND
  workqueue: kfree_rcu_monitor hogged CPU for >0us 4 times, consider switching to WQ_UNBOUND
  workqueue: kfree_rcu_monitor hogged CPU for >0us 8 times, consider switching to WQ_UNBOUND
  workqueue: reg_todo hogged CPU for >0us 4 times, consider switching to WQ_UNBOUND

This commit therefore disables the CPU-hog detection mechanism when
workqueue.cpu_intensive_thresh_us is set to 0.

tj: Patch description updated and the condition check on
    cpu_intensive_thresh_us separated into a separate if statement for
    readability.

Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/workqueue.c

index 3ad6806..41b8388 100644 (file)
@@ -1139,6 +1139,9 @@ void wq_worker_tick(struct task_struct *task)
 
        pwq->stats[PWQ_STAT_CPU_TIME] += TICK_USEC;
 
+       if (!wq_cpu_intensive_thresh_us)
+               return;
+
        /*
         * If the current worker is concurrency managed and hogged the CPU for
         * longer than wq_cpu_intensive_thresh_us, it's automatically marked