Received-SPF: pass (google.com: domain of linux-kernel+bounces-101056-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: Kassey Li <quic_yingangl@quicinc.com>
To: <mingo@redhat.com>, <peterz@infradead.org>, <linux-kernel@vger.kernel.org>
CC: <quic_yingangl@quicinc.com>
Subject: [PATCH] sched/headers: do not set last_queued to 0 in arrive
Date: Wed, 13 Mar 2024 10:39:20 +0800
Message-ID: <20240313023920.3338035-1-quic_yingangl@quicinc.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain

The pcount accounting for Task A missed in the step 4:

1. Task A enqueued
2. Task A arrive and hit CPU
3. Task B arrive and hit CPU, preempted Task A, Task A is still in rq
   as TASK_RUNNING
4. Task A arrive and hit CPU again.

This change leaves enqueue/dequeue on last_queued only, and
correct the pcount accounting.

Signed-off-by: Kassey Li <quic_yingangl@quicinc.com>
---
 kernel/sched/stats.h | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h
index 38f3698f5e5b..3decd2261875 100644
--- a/kernel/sched/stats.h
+++ b/kernel/sched/stats.h
@@ -229,8 +229,17 @@ static void sched_info_arrive(struct rq *rq, struct task_struct *t)
 		return;
 
 	now = rq_clock(rq);
-	delta = now - t->sched_info.last_queued;
-	t->sched_info.last_queued = 0;
+
+	/*
+	 * last_arrival > last_queued means a task in the rq, it is not the
+	 * first time hits the CPU.
+	 */
+
+	if(unlikely(t->sched_info.last_arrival > t->sched_info.last_queued))
+		delta = now - t->sched_info.last_arrival;
+	else
+		delta = now - t->sched_info.last_queued;
+
 	t->sched_info.run_delay += delta;
 	t->sched_info.last_arrival = now;
 	t->sched_info.pcount++;
-- 
2.25.1