Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Chuyi Zhou <zhouchuyi@bytedance.com>
To:     peterz@infradead.org, juri.lelli@redhat.com, mingo@redhat.com,
        vincent.guittot@linaro.org
Cc:     linux-kernel@vger.kernel.org, joshdon@google.com,
        Chuyi Zhou <zhouchuyi@bytedance.com>
Subject: [PATCH] sched/fair: favor non-idle group in tick preemption
Date:   Thu, 27 Oct 2022 16:16:30 +0800
Message-Id: <20221027081630.34081-1-zhouchuyi@bytedance.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

The non-idle se dominates competition vs the idle se when they
are belong to the same group. We ensure that idle groups would not
preempt non-idle group in wakeup preemption(see check_preempt_wakeup()).
However, this can happen in tick preemption, since check_preempt_tick()
dose not check current/se is idle or not. This patch adds this check.

Signed-off-by: Chuyi Zhou <zhouchuyi@bytedance.com>
---
 kernel/sched/fair.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e4a0b8bd941c..f3324b8753b3 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4750,6 +4750,7 @@ static void
 check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
 {
 	unsigned long ideal_runtime, delta_exec;
+	int cse_is_idle, pse_is_idle;
 	struct sched_entity *se;
 	s64 delta;
 
@@ -4779,8 +4780,17 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
 	if (delta < 0)
 		return;
 
-	if (delta > ideal_runtime)
+	if (delta > ideal_runtime) {
+		/*
+		 * Favor non-idle group even in tick preemption
+		 */
+		cse_is_idle = se_is_idle(curr);
+		pse_is_idle = se_is_idle(se);
+		if (unlikely(!cse_is_idle && pse_is_idle))
+			return;
+
 		resched_curr(rq_of(cfs_rq));
+	}
 }
 
 static void
-- 
2.20.1