diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 481ede3eac13..3f7ee102868a 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -5867,8 +5867,15 @@ ftrace_graph_release(struct inode *inode, struct file *file)
 
 		mutex_unlock(&graph_lock);
 
-		/* Wait till all users are no longer using the old hash */
-		synchronize_rcu();
+		/*
+		 * We need to do a hard force of sched synchronization.
+		 * This is because we use preempt_disable() to do RCU, but
+		 * the function tracers can be called where RCU is not watching
+		 * (like before user_exit()). We can not rely on the RCU
+		 * infrastructure to do the synchronization, thus we must do it
+		 * ourselves.
+		 */
+		schedule_on_each_cpu(ftrace_sync);
 
 		free_ftrace_hash(old_hash);
 	}
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 8c52f5de9384..3c75d29bd861 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -979,6 +979,7 @@ static inline int ftrace_graph_addr(struct ftrace_graph_ent *trace)
 	 * Have to open code "rcu_dereference_sched()" because the
 	 * function graph tracer can be called when RCU is not
 	 * "watching".
+	 * Protected with schedule_on_each_cpu(ftrace_sync)
 	 */
 	hash = rcu_dereference_protected(ftrace_graph_hash, !preemptible());
 
@@ -1031,6 +1032,7 @@ static inline int ftrace_graph_notrace_addr(unsigned long addr)
 	 * Have to open code "rcu_dereference_sched()" because the
 	 * function graph tracer can be called when RCU is not
 	 * "watching".
+	 * Protected with schedule_on_each_cpu(ftrace_sync)
 	 */
 	notrace_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
 						 !preemptible());