Re: [PATCH wq/for-3.6-fixes 3/3] workqueue: fix possible idle worker depletion during CPU_ONLINE

[Tejun Heo <tj@kernel.org>]

On Fri, Sep 07, 2012 at 04:05:56PM -0700, Tejun Heo wrote:
> I got it down to the following but it creates a problem where CPU
> hotplug queues a work item on worker->scheduled before the execution
> loops starts.  :(

Oops, wrong patch.  This is the right one.

Index: work/kernel/workqueue.c
===================================================================
--- work.orig/kernel/workqueue.c
+++ work/kernel/workqueue.c
@@ -66,6 +66,7 @@ enum {
 
 	/* pool flags */
 	POOL_MANAGE_WORKERS	= 1 << 0,	/* need to manage workers */
+	POOL_MANAGING_WORKERS   = 1 << 1,       /* managing workers */
 
 	/* worker flags */
 	WORKER_STARTED		= 1 << 0,	/* started */
@@ -165,7 +166,7 @@ struct worker_pool {
 	struct timer_list	idle_timer;	/* L: worker idle timeout */
 	struct timer_list	mayday_timer;	/* L: SOS timer for workers */
 
-	struct mutex		manager_mutex;	/* mutex manager should hold */
+	struct mutex		manager_mutex;	/* manager <-> CPU hotplug */
 	struct ida		worker_ida;	/* L: for worker IDs */
 };
 
@@ -652,7 +653,7 @@ static bool need_to_manage_workers(struc
 /* Do we have too many workers and should some go away? */
 static bool too_many_workers(struct worker_pool *pool)
 {
-	bool managing = mutex_is_locked(&pool->manager_mutex);
+	bool managing = pool->flags & POOL_MANAGING_WORKERS;
 	int nr_idle = pool->nr_idle + managing; /* manager is considered idle */
 	int nr_busy = pool->nr_workers - nr_idle;
 
@@ -1820,14 +1821,35 @@ static bool maybe_destroy_workers(struct
  * some action was taken.
  */
 static bool manage_workers(struct worker *worker)
+	__releases(&gcwq->lock) __acquires(&gcwq->lock)
 {
 	struct worker_pool *pool = worker->pool;
+	struct global_cwq *gcwq = pool->gcwq;
 	bool ret = false;
 
-	if (!mutex_trylock(&pool->manager_mutex))
-		return ret;
+	if (pool->flags & POOL_MANAGING_WORKERS)
+ 		return ret;
 
 	pool->flags &= ~POOL_MANAGE_WORKERS;
+	pool->flags |= POOL_MANAGING_WORKERS;
+
+	/*
+	 * To simplify both worker management and CPU hotplug, hold off
+	 * management while hotplug is in progress.  CPU hotplug path can't
+	 * grab %POOL_MANAGING_WORKERS to achieve this because that can
+	 * lead to idle worker depletion (all become busy thinking someone
+	 * else is managing) which in turn can result in deadlock under
+	 * extreme circumstances.
+	 *
+	 * manager_mutex would always be free unless CPU hotplug is in
+	 * progress.  trylock first without dropping gcwq->lock.
+	 */
+	if (unlikely(!mutex_trylock(&pool->manager_mutex))) {
+		spin_unlock_irq(&gcwq->lock);
+		mutex_lock(&pool->manager_mutex);
+		spin_lock_irq(&gcwq->lock);
+		ret = true;
+	}
 
 	/*
 	 * Destroy and then create so that may_start_working() is true
@@ -1836,6 +1858,7 @@ static bool manage_workers(struct worker
 	ret |= maybe_destroy_workers(pool);
 	ret |= maybe_create_worker(pool);
 
+	pool->flags &= ~POOL_MANAGING_WORKERS;
 	mutex_unlock(&pool->manager_mutex);
 	return ret;
 }
@@ -3393,7 +3416,7 @@ EXPORT_SYMBOL_GPL(work_busy);
  * cpu comes back online.
  */
 
-/* claim manager positions of all pools */
+/* claim manager positions of all pools, see manage_workers() for details */
 static void gcwq_claim_management_and_lock(struct global_cwq *gcwq)
 {
 	struct worker_pool *pool;


-- 
tejun