workqueue: don't use wq_select_unbound_cpu() for bound works
authorHillf Danton <hdanton@sina.com>
Sat, 25 Jan 2020 01:14:45 +0000 (20:14 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 18 Mar 2020 06:14:20 +0000 (07:14 +0100)
commit aa202f1f56960c60e7befaa0f49c72b8fa11b0a8 upstream.

wq_select_unbound_cpu() is designed for unbound workqueues only, but
it's wrongly called when using a bound workqueue too.

Fixing this ensures work queued to a bound workqueue with
cpu=WORK_CPU_UNBOUND always runs on the local CPU.

Before, that would happen only if wq_unbound_cpumask happened to include
it (likely almost always the case), or was empty, or we got lucky with
forced round-robin placement.  So restricting
/sys/devices/virtual/workqueue/cpumask to a small subset of a machine's
CPUs would cause some bound work items to run unexpectedly there.

Fixes: ef557180447f ("workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs")
Cc: stable@vger.kernel.org # v4.5+
Signed-off-by: Hillf Danton <hdanton@sina.com>
[dj: massage changelog]
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/workqueue.c

index 4939084..eef77c8 100644 (file)
@@ -1384,14 +1384,16 @@ static void __queue_work(int cpu, struct workqueue_struct *wq,
            WARN_ON_ONCE(!is_chained_work(wq)))
                return;
 retry:
-       if (req_cpu == WORK_CPU_UNBOUND)
-               cpu = wq_select_unbound_cpu(raw_smp_processor_id());
-
        /* pwq which will be used unless @work is executing elsewhere */
-       if (!(wq->flags & WQ_UNBOUND))
-               pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
-       else
+       if (wq->flags & WQ_UNBOUND) {
+               if (req_cpu == WORK_CPU_UNBOUND)
+                       cpu = wq_select_unbound_cpu(raw_smp_processor_id());
                pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
+       } else {
+               if (req_cpu == WORK_CPU_UNBOUND)
+                       cpu = raw_smp_processor_id();
+               pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
+       }
 
        /*
         * If @work was previously on a different pool, it might still be