workqueue: Use rcuwait for wq_manager_wait

The workqueue code has it's internal spinlock (pool::lock) and also
implicit spinlock usage in the wq_manager waitqueue. These spinlocks
are converted to 'sleeping' spinlocks on a RT-kernel.

Workqueue functions can be invoked from contexts which are truly atomic
even on a PREEMPT_RT enabled kernel. Taking sleeping locks from such
contexts is forbidden.

pool::lock can be converted to a raw spinlock as the lock held times
are short. But the workqueue manager waitqueue is handled inside of
pool::lock held regions which again violates the lock nesting rules
of raw and regular spinlocks.

The manager waitqueue has no special requirements like custom wakeup
callbacks or mass wakeups. While it does not use exclusive wait mode
explicitly there is no strict requirement to queue the waiters in a
particular order as there is only one waiter at a time.

This allows to replace the waitqueue with rcuwait which solves the
locking problem because rcuwait relies on existing locking.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Tejun Heo <tj@kernel.org>

1 files changed, 19 insertions, 5 deletions

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 7a1fc9fe6314..7c3566f8e4ca 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -301,7 +301,8 @@ static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf;
 static DEFINE_MUTEX(wq_pool_mutex);	/* protects pools and workqueues list */
 static DEFINE_MUTEX(wq_pool_attach_mutex); /* protects worker attach/detach */
 static DEFINE_SPINLOCK(wq_mayday_lock);	/* protects wq->maydays list */
-static DECLARE_WAIT_QUEUE_HEAD(wq_manager_wait); /* wait for manager to go away */
+/* wait for manager to go away */
+static struct rcuwait manager_wait = __RCUWAIT_INITIALIZER(manager_wait);
 
 static LIST_HEAD(workqueues);		/* PR: list of all workqueues */
 static bool workqueue_freezing;		/* PL: have wqs started freezing? */
@@ -2140,7 +2141,7 @@ static bool manage_workers(struct worker *worker)
 
 	pool->manager = NULL;
 	pool->flags &= ~POOL_MANAGER_ACTIVE;
-	wake_up(&wq_manager_wait);
+	rcuwait_wake_up(&manager_wait);
 	return true;
 }
 
@@ -3503,6 +3504,18 @@ static void rcu_free_pool(struct rcu_head *rcu)
 	kfree(pool);
 }
 
+/* This returns with the lock held on success (pool manager is inactive). */
+static bool wq_manager_inactive(struct worker_pool *pool)
+{
+	spin_lock_irq(&pool->lock);
+
+	if (pool->flags & POOL_MANAGER_ACTIVE) {
+		spin_unlock_irq(&pool->lock);
+		return false;
+	}
+	return true;
+}
+
 /**
  * put_unbound_pool - put a worker_pool
  * @pool: worker_pool to put
@@ -3538,10 +3551,11 @@ static void put_unbound_pool(struct worker_pool *pool)
 	 * Become the manager and destroy all workers.  This prevents
 	 * @pool's workers from blocking on attach_mutex.  We're the last
 	 * manager and @pool gets freed with the flag set.
+	 * Because of how wq_manager_inactive() works, we will hold the
+	 * spinlock after a successful wait.
 	 */
-	spin_lock_irq(&pool->lock);
-	wait_event_lock_irq(wq_manager_wait,
-			    !(pool->flags & POOL_MANAGER_ACTIVE), pool->lock);
+	rcuwait_wait_event(&manager_wait, wq_manager_inactive(pool),
+			   TASK_UNINTERRUPTIBLE);
 	pool->flags |= POOL_MANAGER_ACTIVE;
 
 	while ((worker = first_idle_worker(pool)))