The commit 6d25be5782e4 ("sched/core, workqueues: Distangle worker accounting from rq lock") changed the schedule callbacks for workqueue and removed the local-wake-up functionality. Now the wakingup of workers is done by normal fashion and workers not yet migrated to the specific CPU in concurrency managed pool can also be woken up by workers that already bound to the specific cpu now. So this advanced kicking of the idle workers to migrate them to the associated CPU is unneeded now.
Showing the changes to kernel/workqueue.c (1 of 1 file). Show all
Unified · Split
| @@ -5057,17 +5057,6 @@ static void rebind_workers(struct worker_pool *pool) · 5056 unmodified lines | ||
| 5057 | 5057 | for_each_pool_worker(worker, pool) { |
| 5058 | 5058 | unsigned int worker_flags = worker->flags; |
| 5059 | 5059 | |
| 5060 | − /* | |
| 5061 | − * A bound idle worker should actually be on the runqueue | |
| 5062 | − * of the associated CPU for local wake-ups targeting it to | |
| 5063 | − * work. Kick all idle workers so that they migrate to the | |
| 5064 | − * associated CPU. Doing this in the same loop as | |
| 5065 | − * replacing UNBOUND with REBOUND is safe as no worker will | |
| 5066 | − * be bound before @pool->lock is released. | |
| 5067 | − */ | |
| 5068 | − if (worker_flags & WORKER_IDLE) | |
| 5069 | − wake_up_process(worker->task); | |
| 5070 | − | |
| 5071 | 5060 | /* |
| 5072 | 5061 | * We want to clear UNBOUND but can't directly call |
| 5073 | 5062 | * worker_clr_flags() or adjust nr_running. Atomically |