workqueue: Update the rescuer's affinity only when it is detached

When a rescuer is attached to a pool, its affinity should be only
managed by the pool.

But updating the detached rescuer's affinity is still meaningful so
that it will not disrupt isolated CPUs when it is to be waken up.

But the commit d64f2fa ("kernel/workqueue: Let rescuers follow
unbound wq cpumask changes") updates the affinity unconditionally, and
causes some issues

1) it also changes the affinity when the rescuer is already attached to
   a pool, which violates the affinity management.

2) the said commit tries to update the affinity of the rescuers, but it
   misses the rescuers of the PERCPU workqueues, and isolated CPUs can
   be possibly disrupted by these rescuers when they are summoned.

3) The affinity to set to the rescuers should be consistent in all paths
   when a rescuer is in detached state. The affinity could be either
   wq_unbound_cpumask or unbound_effective_cpumask(wq). Related paths:
       rescuer's worker_detach_from_pool()
       update wq_unbound_cpumask
       update wq's cpumask
       init_rescuer()
   Both affinities are Ok as long as they are consistent in all paths.
   But using unbound_effective_cpumask(wq) requres much more code to
   maintain the consistency, and it doesn't make much sense since the
   affinity is only effective when the rescuer is not processing works.
   wq_unbound_cpumask is more favorable.

Fix the 1) issue by testing rescuer->pool before updating with
wq_pool_attach_mutex held.

Fix the 2) issue by moving the rescuer's affinity updating code to
the place updating wq_unbound_cpumask and make it also update for
PERCPU workqueues.

Partially cleanup the 3) consistency issue by using wq_unbound_cpumask.
So that the path of "updating wq's cpumask" doesn't need to maintain it.
and both the paths of "updating wq_unbound_cpumask" and "rescuer's
worker_detach_from_pool()" use wq_unbound_cpumask.

Cleanup for init_rescuer()'s consistency for affinity can be done in
future.

Cc: Juri Lelli <[email protected]>
Cc: Waiman Long <[email protected]>
Signed-off-by: Lai Jiangshan <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>

Author: Lai Jiangshan

kernel/workqueue.c

@@ -5370,11 +5370,6 @@ static void apply_wqattrs_commit(struct apply_wqattrs_ctx *ctx)
 	/* update node_nr_active->max */
 	wq_update_node_max_active(ctx->wq, -1);
 
-	/* rescuer needs to respect wq cpumask changes */
-	if (ctx->wq->rescuer)
-		set_cpus_allowed_ptr(ctx->wq->rescuer->task,
-				     unbound_effective_cpumask(ctx->wq));
-
 	mutex_unlock(&ctx->wq->mutex);
 }
 
@@ -6933,6 +6928,11 @@ static int workqueue_apply_unbound_cpumask(const cpumask_var_t unbound_cpumask)
 	if (!ret) {
 		mutex_lock(&wq_pool_attach_mutex);
 		cpumask_copy(wq_unbound_cpumask, unbound_cpumask);
+		/* rescuer needs to respect cpumask changes when it is not attached */
+		list_for_each_entry(wq, &workqueues, list) {
+			if (wq->rescuer && !wq->rescuer->pool)
+				unbind_worker(wq->rescuer);
+		}
 		mutex_unlock(&wq_pool_attach_mutex);
 	}
 	return ret;