Received-SPF: pass (google.com: domain of linux-kernel+bounces-122469-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: mingyang.cui <mingyang.cui@horizon.ai>
To: <mingo@redhat.com>, <peterz@infradead.org>, <juri.lelli@redhat.com>,
	<vincent.guittot@linaro.org>, <dietmar.eggemann@arm.com>,
	<rostedt@goodmis.org>, <bsegall@google.com>, <mgorman@suse.de>,
	<bristot@redhat.com>
CC: <linux-kernel@vger.kernel.org>, <kernel-team@android.com>,
	<stable@vger.kernel.org>, <tkjos@google.com>, <pjt@google.com>,
	<quentin.perret@arm.com>, <Patrick.Bellasi@arm.com>, <Chris.Redpath@arm.com>,
	<Morten.Rasmussen@arm.com>, <joaodias@google.com>, mingyang.cui
	<mingyang.cui@horizon.ai>
Subject: [PATCH] sched/fair: Fix forked task check in vruntime_normalized
Date: Thu, 28 Mar 2024 14:27:57 +0800
Message-ID: <20240328062757.29803-1-mingyang.cui@horizon.ai>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain

When rt_mutex_setprio changes a task's scheduling class to RT,
sometimes the task's vruntime is not updated correctly upon
return to the fair class.
Specifically, the following is being observed:
- task has just been created and running for a short time
- task sleep while still in the fair class
- task is boosted to RT via rt_mutex_setprio, which changes
  the task to RT and calls check_class_changed.
- check_class_changed leads to detach_task_cfs_rq, at which point
  the vruntime_normalized check sees that the task's sum_exec_runtime
  is zero, which results in skipping the subtraction of the
  rq's min_vruntime from the task's vruntime
- later, when the prio is deboosted and the task is moved back
  to the fair class, the fair rq's min_vruntime is added to
  the task's vruntime, even though it wasn't subtracted earlier.

Since the task's vruntime is about double that of other tasks in cfs_rq,
the task to be unable to run for a long time when there are continuous
runnable tasks in cfs_rq.

The immediate result is inflation of the task's vruntime, giving
it lower priority (starving it if there's enough available work).
The longer-term effect is inflation of all vruntimes because the
task's vruntime becomes the rq's min_vruntime when the higher
priority tasks go idle. That leads to a vicious cycle, where
the vruntime inflation repeatedly doubled.

The root cause of the problem is that the vruntime_normalized made a
misjudgment. Since the sum_exec_runtime of some tasks that were just
created and run for a short time is zero, the vruntime_normalized
mistakenly thinks that they are tasks that have just been forked.
Therefore, sum_exec_runtime is not subtracted from the vruntime of the
task.

So, we fix this bug by adding a check condition for newly forked task.

Signed-off-by: mingyang.cui <mingyang.cui@horizon.ai>
---
 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 73a89fbd81be..3d0c14f3731f 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -11112,7 +11112,7 @@ static inline bool vruntime_normalized(struct task_struct *p)
 	 * - A task which has been woken up by try_to_wake_up() and
 	 *   waiting for actually being woken up by sched_ttwu_pending().
 	 */
-	if (!se->sum_exec_runtime ||
+	if (!se->sum_exec_runtime && p->state == TASK_NEW ||
 	    (p->state == TASK_WAKING && p->sched_remote_wakeup))
 		return true;
 
-- 
2.34.1