[PATCH cgroup/for-next 3/4] cgroup/cpuset: Optimize cpuset_attach_task()

[Waiman Long <longman@redhat.com>]

Within cpuset_attach(), cpuset_attach_task() is called only if either the
CPU and/or the memory setting are updated. If only one of the settings
is updated, cpuset_attach_task() still updates both CPU and memory node
setting of each task. Further optimize it by checking attach_cpus_updated
and attach_mems_updated for v2 to skip the unnecessary update.

While at it, also move the mpol_rebind_mm() call for mm group leader
to cpuset_attach_task(). This change shouldn't affect the cpuset_fork()
caller as the newly cloned task isn't the group leader. For that caller,
it is assumed that both CPU and memory nodes are updated to keep the
existing behavior.

Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 36 ++++++++++++++++++++++++++----------
 1 file changed, 26 insertions(+), 10 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 68392cf6429b..8ced1fa0900f 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -3132,8 +3132,13 @@ static nodemask_t cpuset_attach_nodemask_to;
 
 static void cpuset_attach_task(struct cpuset *cs, struct task_struct *task)
 {
+	struct mm_struct *mm;
+
 	lockdep_assert_cpuset_lock_held();
 
+	if (cpuset_v2() && !attach_cpus_updated)
+		goto update_mem;
+
 	if (cs != &top_cpuset)
 		guarantee_active_cpus(task, cpus_attach);
 	else
@@ -3145,8 +3150,21 @@ static void cpuset_attach_task(struct cpuset *cs, struct task_struct *task)
 	 */
 	WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach));
 
+update_mem:
+	if (cpuset_v2() && !attach_mems_updated)
+		return;
+
 	cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to);
 	cpuset1_update_task_spread_flags(cs, task);
+
+	if (task != task->group_leader)
+		return;
+
+	mm = get_task_mm(task);
+	if (mm) {
+		mpol_rebind_mm(mm, &cs->effective_mems);
+		mmput(mm);
+	}
 }
 
 static void cpuset_attach(struct cgroup_taskset *tset)
@@ -3187,15 +3205,13 @@ static void cpuset_attach(struct cgroup_taskset *tset)
 	 * not set.
 	 */
 	cpuset_attach_nodemask_to = cs->effective_mems;
-	if (!is_memory_migrate(cs) && !attach_mems_updated)
+	if (!is_memory_migrate(cs))
 		goto out;
 
 	cgroup_taskset_for_each_leader(leader, css, tset) {
 		struct mm_struct *mm = get_task_mm(leader);
 
 		if (mm) {
-			mpol_rebind_mm(mm, &cpuset_attach_nodemask_to);
-
 			/*
 			 * old_mems_allowed is the same with mems_allowed
 			 * here, except if this task is being moved
@@ -3204,18 +3220,15 @@ static void cpuset_attach(struct cgroup_taskset *tset)
 			 * @old_mems_allowed is the right nodesets that we
 			 * migrate mm from.
 			 */
-			if (is_memory_migrate(cs)) {
-				cpuset_migrate_mm(mm, &oldcs->old_mems_allowed,
-						  &cpuset_attach_nodemask_to);
-				queue_task_work = true;
-			} else
-				mmput(mm);
+			cpuset_migrate_mm(mm, &oldcs->old_mems_allowed,
+					  &cpuset_attach_nodemask_to);
+			queue_task_work = true;
 		}
 	}
 
-out:
 	if (queue_task_work)
 		schedule_flush_migrate_mm();
+out:
 	cs->old_mems_allowed = cpuset_attach_nodemask_to;
 
 	if (cs->nr_migrate_dl_tasks) {
@@ -3666,7 +3679,10 @@ static void cpuset_fork(struct task_struct *task)
 	/* CLONE_INTO_CGROUP */
 	mutex_lock(&cpuset_mutex);
 	guarantee_online_mems(cs, &cpuset_attach_nodemask_to);
+	/* Assume CPUs and memory nodes are updated */
+	attach_cpus_updated = attach_mems_updated = true;
 	cpuset_attach_task(cs, task);
+	attach_cpus_updated = attach_mems_updated = false;
 
 	dec_attach_in_progress_locked(cs);
 	mutex_unlock(&cpuset_mutex);
-- 
2.54.0