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;
From:   Waiman Long <longman@redhat.com>
To:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
        Boqun Feng <boqun.feng@gmail.com>
Cc:     linux-kernel@vger.kernel.org, Waiman Long <longman@redhat.com>
Subject: [PATCH v2] locking/qspinlock: Do spin-wait in slowpath if preemptible
Date:   Tue, 20 Sep 2022 15:55:42 -0400
Message-Id: <20220920195542.1548164-1-longman@redhat.com>
MIME-Version: 1.0
Content-type: text/plain
Content-Transfer-Encoding: 8bit
Precedence: bulk

There are some code paths in the kernel where arch_spin_lock() will be
called directly when the lock isn't expected to be contended and critical
section is short. For example, tracing_saved_cmdlines_size_read()
in kernel/trace/trace.c does that.

In most cases, preemption is also not disabled. This creates a problem
for the qspinlock slowpath which expects preemption to be disabled
to guarantee the safe use of per cpu qnodes structure. To work around
these special use cases, add a preemption count check in the slowpath
and do a simple spin-wait when preemption isn't disabled.

Fixes: a33fda35e3a7 ("Introduce a simple generic 4-byte queued spinlock")
Signed-off-by: Waiman Long <longman@redhat.com>
---
 [v2] Move down spin-wait to after the pending bit wait.

 kernel/locking/qspinlock.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index 65a9a10caa6f..d0159038084d 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -398,6 +398,23 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 queue:
 	lockevent_inc(lock_slowpath);
 pv_queue:
+#ifdef CONFIG_PREEMPT_COUNT
+	/*
+	 * As arch_spin_lock() can be called directly in some use cases
+	 * where the lock isn't expected to be contended, critical section
+	 * is short and preemption isn't disabled, we can't use qnodes in
+	 * this case as state may be screwed up in case preemption happens
+	 * or preemption warning may be printed (CONFIG_DEBUG_PREEMPT).
+	 * Just do a simple spin-wait in this case as the lock shouldn't be
+	 * contended for long.
+	 */
+	if (unlikely(!preempt_count())) {
+		while (!queued_spin_trylock(lock))
+			cpu_relax();
+		return;
+	}
+#endif
+
 	node = this_cpu_ptr(&qnodes[0].mcs);
 	idx = node->count++;
 	tail = encode_tail(smp_processor_id(), idx);
-- 
2.31.1