Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
Date:   Wed, 11 Oct 2023 15:47:23 +0200
From:   Frederic Weisbecker <frederic@kernel.org>
To:     "Paul E. McKenney" <paulmck@kernel.org>
Cc:     Joel Fernandes <joel@joelfernandes.org>,
        "Liam R. Howlett" <Liam.Howlett@oracle.com>,
        Naresh Kamboju <naresh.kamboju@linaro.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        stable@vger.kernel.org, patches@lists.linux.dev,
        linux-kernel@vger.kernel.org, torvalds@linux-foundation.org,
        akpm@linux-foundation.org, linux@roeck-us.net, shuah@kernel.org,
        patches@kernelci.org, lkft-triage@lists.linaro.org, pavel@denx.de,
        jonathanh@nvidia.com, f.fainelli@gmail.com,
        sudipm.mukherjee@gmail.com, srw@sladewatkins.net, rwarsow@gmx.de,
        conor@kernel.org, Chengming Zhou <zhouchengming@bytedance.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Ovidiu Panait <ovidiu.panait@windriver.com>,
        Ingo Molnar <mingo@kernel.org>, rcu <rcu@vger.kernel.org>
Subject: Re: [PATCH 5.15 000/183] 5.15.134-rc1 review
Message-ID: <ZSana69n6RWgCnqi@localhost.localdomain>
References: <20231004175203.943277832@linuxfoundation.org>
 <CA+G9fYunnEUT2evdabX1KOTiryP1heNHWDH4LWZCt2SVRmnKOA@mail.gmail.com>
 <20231006162038.d3q7sl34b4ouvjxf@revolver>
 <57c1ff4d-f138-4f89-8add-c96fb3ba6701@paulmck-laptop>
 <20231006175714.begtgj6wrs46ukmo@revolver>
 <7652477c-a37c-4509-9dc9-7f9d1dc08291@paulmck-laptop>
 <CAEXW_YS16NxPxg52T=3FcyZ2qocj36zKyhPnEQL3nBTbD-qJ-A@mail.gmail.com>
 <9470dab6-dee5-4505-95a2-f6782b648726@paulmck-laptop>
 <433f5823-059c-4b51-8d18-8b356a5a507f@paulmck-laptop>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <433f5823-059c-4b51-8d18-8b356a5a507f@paulmck-laptop>
Precedence: bulk

Le Tue, Oct 10, 2023 at 06:34:35PM -0700, Paul E. McKenney a ?crit :
> If this problem is real, fixes include:
> 
> o	Revert Liam's patch and make Tiny RCU's call_rcu() deal with
> 	the problem.  This is overhead and non-tinyness, but to Joel's
> 	point, it might be best.

But what is calling call_rcu() or start_poll_synchronize_rcu() so
early that the CPU is not even online? (that's before boot_cpu_init() !)

Deferring PF_IDLE setting might pave the way for more issues like this one,
present or future. Though is_idle_task() returning true when the task is not
in the idle loop but is playing the init/0 role is debatable.

An alternative for tiny RCU is to force waking up ksoftirqd when call_rcu()
is in the idle task. Since rcu_qs() during the context switch raises a softirq
anyway. It's more overhead for start_poll_synchronize_rcu() though but do we
expect much RCU polling in idle?

diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
index a92bce40b04b..6ab15233e2be 100644
--- a/include/linux/interrupt.h
+++ b/include/linux/interrupt.h
@@ -604,6 +604,7 @@ extern void __raise_softirq_irqoff(unsigned int nr);
 
 extern void raise_softirq_irqoff(unsigned int nr);
 extern void raise_softirq(unsigned int nr);
+extern void raise_ksoftirqd_irqsoff(unsigned int nr);
 
 DECLARE_PER_CPU(struct task_struct *, ksoftirqd);
 
diff --git a/kernel/rcu/tiny.c b/kernel/rcu/tiny.c
index 42f7589e51e0..872dab8b8b53 100644
--- a/kernel/rcu/tiny.c
+++ b/kernel/rcu/tiny.c
@@ -189,12 +189,12 @@ void call_rcu(struct rcu_head *head, rcu_callback_t func)
 	local_irq_save(flags);
 	*rcu_ctrlblk.curtail = head;
 	rcu_ctrlblk.curtail = &head->next;
-	local_irq_restore(flags);
 
 	if (unlikely(is_idle_task(current))) {
 		/* force scheduling for rcu_qs() */
-		resched_cpu(0);
+		raise_ksoftirqd_irqsoff(RCU_SOFTIRQ);
 	}
+	local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(call_rcu);
 
@@ -225,10 +225,13 @@ EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
 unsigned long start_poll_synchronize_rcu(void)
 {
 	unsigned long gp_seq = get_state_synchronize_rcu();
+	unsigned long flags;
 
 	if (unlikely(is_idle_task(current))) {
+		local_irq_save(flags);
 		/* force scheduling for rcu_qs() */
-		resched_cpu(0);
+		raise_ksoftirqd_irqsoff(RCU_SOFTIRQ);
+		local_irq_restore(flags);
 	}
 	return gp_seq;
 }
diff --git a/kernel/softirq.c b/kernel/softirq.c
index 210cf5f8d92c..ef105cbdc705 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -695,6 +695,14 @@ void __raise_softirq_irqoff(unsigned int nr)
 	or_softirq_pending(1UL << nr);
 }
 
+#ifdef CONFIG_RCU_TINY
+void raise_ksoftirqd(unsigned int nr)
+{
+	__raise_softirq_irqoff(nr);
+	wakeup_softirqd();
+}
+#endif
+
 void open_softirq(int nr, void (*action)(struct softirq_action *))
 {
 	softirq_vec[nr].action = action;