Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
From:   Abel Wu <wuyun.abel@bytedance.com>
To:     Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Juri Lelli <juri.lelli@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>
Cc:     linux-kernel@vger.kernel.org
Subject: [RFC PATCH 5/5] sched/fair: favor cpu capacity for idle tasks
Date:   Thu, 17 Feb 2022 23:44:01 +0800
Message-Id: <20220217154403.6497-6-wuyun.abel@bytedance.com>
In-Reply-To: <20220217154403.6497-1-wuyun.abel@bytedance.com>
References: <20220217154403.6497-1-wuyun.abel@bytedance.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Unlike select_idle_sibling() in which we need to find a
not-so-bad candidate ASAP, the slowpath gives us more
tolerance: ignore sched-idle cpus for idle tasks since
they prefer cpu capacity rather than latency, and besides
spreading out idle tasks also good for latency of normal
tasks.

Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
---
 kernel/sched/fair.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 1d8f396e6f41..57f1d8c43228 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6007,6 +6007,7 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu);
 static int
 find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
 {
+	bool ignore_si = task_h_idle(p);
 	unsigned long load, min_load = ULONG_MAX;
 	unsigned int min_exit_latency = UINT_MAX;
 	u64 latest_idle_timestamp = 0;
@@ -6025,7 +6026,13 @@ find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this
 		if (!sched_core_cookie_match(rq, p))
 			continue;
 
-		if (sched_idle_cpu(i))
+		/*
+		 * The idle tasks prefer cpu capacity rather than
+		 * latency. Spreading out idle tasks also good for
+		 * latency of normal tasks since they won't suffer
+		 * high cpu wakeup delay.
+		 */
+		if (!ignore_si && sched_idle_cpu(i))
 			return i;
 
 		if (available_idle_cpu(i)) {
-- 
2.11.0