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;
Date:   Tue, 22 Aug 2023 19:06:51 +0800
From:   Tio Zhang <tiozhang@didiglobal.com>
To:     <tj@kernel.org>
CC:     <linux-kernel@vger.kernel.org>, <jiangshanlai@gmail.com>,
        <zyhtheonly@gmail.com>, <zyhtheonly@yeah.net>,
        <tiozhang@didiglobal.com>
Subject: [PATCH v2] workqueue: let WORKER_CPU_INTENSIVE be included in
 watchdog
Message-ID: <20230822110609.GA3702@didi-ThinkCentre-M930t-N000>
Mail-Followup-To: tj@kernel.org, linux-kernel@vger.kernel.org,
        jiangshanlai@gmail.com, zyhtheonly@gmail.com, zyhtheonly@yeah.net
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Disposition: inline
In-Reply-To: <202308142140.cf9be57a-oliver.sang@intel.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

When a pool has a worker with WORKER_CPU_INTENSIVE set but other workers
are not that busy, the pool->worklist will mostly be empty, which leads
the intensive work always having a chance of escaping from the watchdog's
check. This may cause watchdog miss finding out a forever running work
in WQ_CPU_INTENSIVE.

Also, after commit '616db8779b1e3f93075df691432cccc5ef3c3ba0',
workers with potentially intensive works will automatically be converted
into WORKER_CPU_INTENSIVE. This might let watchdog to miss all work
potentially running forever.

Signed-off-by: Tio Zhang <tiozhang@didiglobal.com>
---
 kernel/workqueue.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 02a8f402eeb5..564d96c38d4d 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -6277,13 +6277,29 @@ static void wq_watchdog_timer_fn(struct timer_list *unused)
 	if (!thresh)
 		return;
 
-	rcu_read_lock();
+	mutex_lock(&wq_pool_mutex);
 
 	for_each_pool(pool, pi) {
+		struct worker *worker;
 		unsigned long pool_ts, touched, ts;
+		bool check_intensive = false;
 
 		pool->cpu_stall = false;
-		if (list_empty(&pool->worklist))
+
+		/* Not sure if we should let WORKER_UNBOUND to
+		 * be included? Since let a unbound work to last
+		 * more than e,g, 30 seconds seem also unacceptable.
+		 */
+		mutex_lock(&wq_pool_attach_mutex);
+		for_each_pool_worker(worker, pool) {
+			if (worker->flags & WORKER_CPU_INTENSIVE) {
+				check_intensive = true;
+				break;
+			}
+		}
+		mutex_unlock(&wq_pool_attach_mutex);
+
+		if (list_empty(&pool->worklist) && !check_intensive)
 			continue;
 
 		/*
@@ -6320,7 +6336,7 @@ static void wq_watchdog_timer_fn(struct timer_list *unused)
 
 	}
 
-	rcu_read_unlock();
+	mutex_unlock(&wq_pool_mutex);
 
 	if (lockup_detected)
 		show_all_workqueues();
-- 
2.17.1