Received-SPF: pass (google.com: domain of linux-kernel+bounces-195356-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
From: Ze Gao <zegao2021@gmail.com>
To: Frederic Weisbecker <frederic@kernel.org>,
	Joel Fernandes <joel@joelfernandes.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@elte.hu>
Cc: Cruz Zhao <CruzZhao@linux.alibaba.com>,
	linux-kernel@vger.kernel.org,
	Ze Gao <zegao@tencent.com>
Subject: [RFC PATCH 1/2] timer: Use is_idle_task() check instead of idle_cpu() on irq exit
Date: Thu, 30 May 2024 08:24:00 -0400
Message-ID: <20240530122401.3826022-2-zegao@tencent.com>
In-Reply-To: <20240530122401.3826022-1-zegao@tencent.com>
References: <20240530122401.3826022-1-zegao@tencent.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

idle_cpu() was initially introduced in irq_enter()/exit() to check
whether an irq interrupts an idle cpu or not since commit
79bf2bb335b8 ("[PATCH] tick-management: dyntick / highres functionality")
and at that time, it's implemented via a simple check if the curr
of task of that rq is idle or not. And then commit 6378ddb59215 ("time:
track accurate idle time with tick_sched.idle_sleeptime") uses the same
check to do accurate idle time accounting.

But since commit 908a3283728d ("sched: Fix idle_cpu()"), idle_cpu()
takes scheduler stats into consideration and becomes more constrained,
and therefore it tells more than if we have interrupted an idle
process but also whether a cpu is going to be idle or not since it
takes queued tasks and queued to be woken tasks into account.

However for tick user, it is too much as now we only rely on this check
to do nohz idle time accounting in tick_nohz_start_idle() just in case
that tick_nohz_stop_idle() is called upon irq_enter() if we actually
rupture an idle cpu(process). The use of idle_cpu() simply complicates
things here, and the introduction of sched_core_idle_cpu() in
commit 548796e2e70b ("sched/core: introduce sched_core_idle_cpu()")
proves this.

The use of is_idle_task() just like in commit 0a8a2e78b7ee ("timer: Fix
bad idle check on irq entry") helps to save one unnecessary fix for idle
time accounting for the newly force idle state. Note this also preps for
the remove of sched_core_idle_cpu() in the following patch.

Signed-off-by: Ze Gao <zegao@tencent.com>
---
 kernel/softirq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/softirq.c b/kernel/softirq.c
index 02582017759a..24c7bf3c3f6c 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -617,7 +617,7 @@ static inline void tick_irq_exit(void)
 	int cpu = smp_processor_id();
 
 	/* Make sure that timer wheel updates are propagated */
-	if ((sched_core_idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
+	if ((is_idle_task(current) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
 		if (!in_hardirq())
 			tick_nohz_irq_exit();
 	}
-- 
2.41.0