Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 24 Jul 2020 10:46:50 +0100
From:   Qais Yousef <qais.yousef@arm.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     Ingo Molnar <mingo@redhat.com>,
        Doug Anderson <dianders@chromium.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>,
        Patrick Bellasi <patrick.bellasi@matbug.net>,
        Chris Redpath <chris.redpath@arm.com>,
        Lukasz Luba <lukasz.luba@arm.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v7 3/3] sched/uclamp: Fix a deadlock when enabling uclamp
 static key
Message-ID: <20200724094650.hgya5j7i7lbhrocy@e107158-lin.cambridge.arm.com>
References: <20200716110347.19553-1-qais.yousef@arm.com>
 <20200716110347.19553-4-qais.yousef@arm.com>
 <20200724091244.GX10769@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20200724091244.GX10769@hirez.programming.kicks-ass.net>
User-Agent: NeoMutt/20171215
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 07/24/20 11:12, Peter Zijlstra wrote:
> On Thu, Jul 16, 2020 at 12:03:47PM +0100, Qais Yousef wrote:
> 
> I've trimmed the Changelog to read like:

+1

Should we mention the ordering issue too? Or maybe I misinterpreted the
'Possible unsafe locking scenario' part?

> 
> ---
> Subject: sched/uclamp: Fix a deadlock when enabling uclamp static key
> From: Qais Yousef <qais.yousef@arm.com>
> Date: Thu, 16 Jul 2020 12:03:47 +0100
> 
> From: Qais Yousef <qais.yousef@arm.com>
> 
> The following splat was caught when setting uclamp value of a task:
> 
>   BUG: sleeping function called from invalid context at ./include/linux/percpu-rwsem.h:49
> 
>    cpus_read_lock+0x68/0x130
>    static_key_enable+0x1c/0x38
>    __sched_setscheduler+0x900/0xad8
> 
> Fix by ensuring we enable the key outside of the critical section in
> __sched_setscheduler()
> 
> Fixes: 46609ce22703 ("sched/uclamp: Protect uclamp fast path code with static key")
> Signed-off-by: Qais Yousef <qais.yousef@arm.com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> Link: https://lkml.kernel.org/r/20200716110347.19553-4-qais.yousef@arm.com
> ---
> 
> And placed this patch first in the series
> 
> That said; don't we have a slight problem with enabling the key late
> like this? It means the uclamp will not actually take effect immediately
> and we'll have to wait for the next context switch ... whenever that
> might be.

The enabling is racy inherently. Your suggestion below is better though.
I should have scrolled more screens up!

> 
> Should we not have enabled the key early, instead of late?
> 
> something like so perhaps?

This should work, but you'll need to sprinkle ifdef around the key. Or move it
to uclamp_validate()

---
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e1578c3ad40c..cd3b7a25ac59 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -1345,6 +1345,21 @@ static int uclamp_validate(struct task_struct *p,
 	if (upper_bound > SCHED_CAPACITY_SCALE)
 		return -EINVAL;
 
+	/*
+	 * Enable uclamp static key outside the critical section of
+	 * __sched_setschuler().
+	 *
+	 * static_branch_enable() will hold cpu_hotplug_lock; if done from
+	 * critical section, __setscheduler_uclamp(), which holds other locks
+	 * (rq->lock namely), it could lead to deadlock scenarios as both are
+	 * popular locks and could be acquired from different paths in
+	 * different orders.
+	 *
+	 * Besides cpu_hotplug_lock could sleep, which is not allowed inside
+	 * the critical section of __sched_setscheduler().
+	 */
+	static_branch_enable(&sched_uclamp_used);
+
 	return 0;
 }
 
@@ -1378,8 +1393,6 @@ static void __setscheduler_uclamp(struct task_struct *p,
 	if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)))
 		return;
 
-	static_branch_enable(&sched_uclamp_used);
-
 	if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) {
 		uclamp_se_set(&p->uclamp_req[UCLAMP_MIN],
 			      attr->sched_util_min, true);