]> git.karo-electronics.de Git - linux-beck.git/commitdiff
workqueue: rename manager_mutex to assoc_mutex
authorLai Jiangshan <laijs@cn.fujitsu.com>
Tue, 18 Sep 2012 16:59:23 +0000 (09:59 -0700)
committerTejun Heo <tj@kernel.org>
Tue, 18 Sep 2012 16:59:23 +0000 (09:59 -0700)
Now that manager_mutex's role has changed from synchronizing manager
role to excluding hotplug against manager, the name is misleading.

As it is protecting the CPU-association of the gcwq now, rename it to
assoc_mutex.

This patch is pure rename and doesn't introduce any functional change.

tj: Updated comments and description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/workqueue.c

index cdc6bfc84b784c8c90eb8e5441a4f46f3e088982..e651239f1ece1f18df234e5a23f09881f4657f7e 100644 (file)
@@ -58,7 +58,7 @@ enum {
         * be executing on any CPU.  The gcwq behaves as an unbound one.
         *
         * Note that DISASSOCIATED can be flipped only while holding
-        * managership of all pools on the gcwq to avoid changing binding
+        * assoc_mutex of all pools on the gcwq to avoid changing binding
         * state while create_worker() is in progress.
         */
        GCWQ_DISASSOCIATED      = 1 << 0,       /* cpu can't serve workers */
@@ -165,7 +165,7 @@ 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            assoc_mutex;    /* protect GCWQ_DISASSOCIATED */
        struct ida              worker_ida;     /* L: for worker IDs */
 };
 
@@ -1681,7 +1681,7 @@ static void rebind_workers(struct global_cwq *gcwq)
        lockdep_assert_held(&gcwq->lock);
 
        for_each_worker_pool(pool, gcwq)
-               lockdep_assert_held(&pool->manager_mutex);
+               lockdep_assert_held(&pool->assoc_mutex);
 
        /* dequeue and kick idle ones */
        for_each_worker_pool(pool, gcwq) {
@@ -2081,22 +2081,22 @@ static bool manage_workers(struct worker *worker)
         * 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.  Use @pool->manager_mutex to synchronize
+        * extreme circumstances.  Use @pool->assoc_mutex to synchronize
         * manager against CPU hotplug.
         *
-        * manager_mutex would always be free unless CPU hotplug is in
+        * assoc_mutex would always be free unless CPU hotplug is in
         * progress.  trylock first without dropping @gcwq->lock.
         */
-       if (unlikely(!mutex_trylock(&pool->manager_mutex))) {
+       if (unlikely(!mutex_trylock(&pool->assoc_mutex))) {
                spin_unlock_irq(&pool->gcwq->lock);
-               mutex_lock(&pool->manager_mutex);
+               mutex_lock(&pool->assoc_mutex);
                /*
                 * CPU hotplug could have happened while we were waiting
-                * for manager_mutex.  Hotplug itself can't handle us
+                * for assoc_mutex.  Hotplug itself can't handle us
                 * because manager isn't either on idle or busy list, and
                 * @gcwq's state and ours could have deviated.
                 *
-                * As hotplug is now excluded via manager_mutex, we can
+                * As hotplug is now excluded via assoc_mutex, we can
                 * simply try to bind.  It will succeed or fail depending
                 * on @gcwq's current state.  Try it and adjust
                 * %WORKER_UNBOUND accordingly.
@@ -2119,7 +2119,7 @@ static bool manage_workers(struct worker *worker)
        ret |= maybe_create_worker(pool);
 
        pool->flags &= ~POOL_MANAGING_WORKERS;
-       mutex_unlock(&pool->manager_mutex);
+       mutex_unlock(&pool->assoc_mutex);
        return ret;
 }
 
@@ -3474,23 +3474,23 @@ EXPORT_SYMBOL_GPL(work_busy);
  */
 
 /* claim manager positions of all pools */
-static void gcwq_claim_management_and_lock(struct global_cwq *gcwq)
+static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq)
 {
        struct worker_pool *pool;
 
        for_each_worker_pool(pool, gcwq)
-               mutex_lock_nested(&pool->manager_mutex, pool - gcwq->pools);
+               mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools);
        spin_lock_irq(&gcwq->lock);
 }
 
 /* release manager positions */
-static void gcwq_release_management_and_unlock(struct global_cwq *gcwq)
+static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq)
 {
        struct worker_pool *pool;
 
        spin_unlock_irq(&gcwq->lock);
        for_each_worker_pool(pool, gcwq)
-               mutex_unlock(&pool->manager_mutex);
+               mutex_unlock(&pool->assoc_mutex);
 }
 
 static void gcwq_unbind_fn(struct work_struct *work)
@@ -3503,7 +3503,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
 
        BUG_ON(gcwq->cpu != smp_processor_id());
 
-       gcwq_claim_management_and_lock(gcwq);
+       gcwq_claim_assoc_and_lock(gcwq);
 
        /*
         * We've claimed all manager positions.  Make all workers unbound
@@ -3520,7 +3520,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
 
        gcwq->flags |= GCWQ_DISASSOCIATED;
 
-       gcwq_release_management_and_unlock(gcwq);
+       gcwq_release_assoc_and_unlock(gcwq);
 
        /*
         * Call schedule() so that we cross rq->lock and thus can guarantee
@@ -3576,10 +3576,10 @@ static int __devinit workqueue_cpu_up_callback(struct notifier_block *nfb,
 
        case CPU_DOWN_FAILED:
        case CPU_ONLINE:
-               gcwq_claim_management_and_lock(gcwq);
+               gcwq_claim_assoc_and_lock(gcwq);
                gcwq->flags &= ~GCWQ_DISASSOCIATED;
                rebind_workers(gcwq);
-               gcwq_release_management_and_unlock(gcwq);
+               gcwq_release_assoc_and_unlock(gcwq);
                break;
        }
        return NOTIFY_OK;
@@ -3833,7 +3833,7 @@ static int __init init_workqueues(void)
                        setup_timer(&pool->mayday_timer, gcwq_mayday_timeout,
                                    (unsigned long)pool);
 
-                       mutex_init(&pool->manager_mutex);
+                       mutex_init(&pool->assoc_mutex);
                        ida_init(&pool->worker_ida);
                }
        }