diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index c6bd67b3fcf6e89ca8c04cb6c4939ef196252d64..548d8d4e86d03503ea335d627db55ea3277a347f 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -4595,10 +4595,19 @@ void cgroup_post_fork(struct task_struct *child)
 {
 	if (use_task_css_set_links) {
 		write_lock(&css_set_lock);
-		task_lock(child);
-		if (list_empty(&child->cg_list))
+		if (list_empty(&child->cg_list)) {
+			/*
+			 * It's safe to use child->cgroups without task_lock()
+			 * here because we are protected through
+			 * threadgroup_change_begin() against concurrent
+			 * css_set change in cgroup_task_migrate(). Also
+			 * the task can't exit at that point until
+			 * wake_up_new_task() is called, so we are protected
+			 * against cgroup_exit() setting child->cgroup to
+			 * init_css_set.
+			 */
 			list_add(&child->cg_list, &child->cgroups->tasks);
-		task_unlock(child);
+		}
 		write_unlock(&css_set_lock);
 	}
 }