workqueue: use manager lock only to protect worker_idr
authorLai Jiangshan <laijs@cn.fujitsu.com>
Tue, 20 May 2014 09:46:27 +0000 (17:46 +0800)
committerTejun Heo <tj@kernel.org>
Tue, 20 May 2014 14:59:30 +0000 (10:59 -0400)
worker_idr is highly bound to managers and is always/only accessed in manager
lock context. So we don't need pool->lock for it.

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

index c841108..910d963 100644 (file)
@@ -124,8 +124,7 @@ enum {
  *    cpu or grabbing pool->lock is enough for read access.  If
  *    POOL_DISASSOCIATED is set, it's identical to L.
  *
- * MG: pool->manager_mutex and pool->lock protected.  Writes require both
- *     locks.  Reads can happen under either lock.
+ * M: pool->manager_mutex protected.
  *
  * PL: wq_pool_mutex protected.
  *
@@ -164,7 +163,7 @@ struct worker_pool {
        /* see manage_workers() for details on the two manager mutexes */
        struct mutex            manager_arb;    /* manager arbitration */
        struct mutex            manager_mutex;  /* manager exclusion */
-       struct idr              worker_idr;     /* MG: worker IDs and iteration */
+       struct idr              worker_idr;     /* M: worker IDs and iteration */
 
        struct workqueue_attrs  *attrs;         /* I: worker attributes */
        struct hlist_node       hash_node;      /* PL: unbound_pool_hash node */
@@ -340,16 +339,6 @@ static void copy_workqueue_attrs(struct workqueue_attrs *to,
                           lockdep_is_held(&wq->mutex),                 \
                           "sched RCU or wq->mutex should be held")
 
-#ifdef CONFIG_LOCKDEP
-#define assert_manager_or_pool_lock(pool)                              \
-       WARN_ONCE(debug_locks &&                                        \
-                 !lockdep_is_held(&(pool)->manager_mutex) &&           \
-                 !lockdep_is_held(&(pool)->lock),                      \
-                 "pool->manager_mutex or ->lock should be held")
-#else
-#define assert_manager_or_pool_lock(pool)      do { } while (0)
-#endif
-
 #define for_each_cpu_worker_pool(pool, cpu)                            \
        for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0];               \
             (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \
@@ -378,14 +367,14 @@ static void copy_workqueue_attrs(struct workqueue_attrs *to,
  * @wi: integer used for iteration
  * @pool: worker_pool to iterate workers of
  *
- * This must be called with either @pool->manager_mutex or ->lock held.
+ * This must be called with @pool->manager_mutex.
  *
  * The if/else clause exists only for the lockdep assertion and can be
  * ignored.
  */
 #define for_each_pool_worker(worker, wi, pool)                         \
        idr_for_each_entry(&(pool)->worker_idr, (worker), (wi))         \
-               if (({ assert_manager_or_pool_lock((pool)); false; })) { } \
+               if (({ lockdep_assert_held(&pool->manager_mutex); false; })) { } \
                else
 
 /**
@@ -1725,13 +1714,7 @@ static struct worker *create_worker(struct worker_pool *pool)
         * ID is needed to determine kthread name.  Allocate ID first
         * without installing the pointer.
         */
-       idr_preload(GFP_KERNEL);
-       spin_lock_irq(&pool->lock);
-
-       id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_NOWAIT);
-
-       spin_unlock_irq(&pool->lock);
-       idr_preload_end();
+       id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_KERNEL);
        if (id < 0)
                goto fail;
 
@@ -1773,18 +1756,13 @@ static struct worker *create_worker(struct worker_pool *pool)
                worker->flags |= WORKER_UNBOUND;
 
        /* successful, commit the pointer to idr */
-       spin_lock_irq(&pool->lock);
        idr_replace(&pool->worker_idr, worker, worker->id);
-       spin_unlock_irq(&pool->lock);
 
        return worker;
 
 fail:
-       if (id >= 0) {
-               spin_lock_irq(&pool->lock);
+       if (id >= 0)
                idr_remove(&pool->worker_idr, id);
-               spin_unlock_irq(&pool->lock);
-       }
        kfree(worker);
        return NULL;
 }