rwsem: move rwsem_down_failed_common code into rwsem_down_{read,write}_failed
Michel Lespinasse [Tue, 7 May 2013 13:45:51 +0000 (06:45 -0700)]
Remove the rwsem_down_failed_common function and replace it with two
identical copies of its code in rwsem_down_{read,write}_failed.

This is because we want to make different optimizations in
rwsem_down_{read,write}_failed; we are adding this pure-duplication
step as a separate commit in order to make it easier to check the
following steps.

Signed-off-by: Michel Lespinasse <walken@google.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/rwsem.c

index 4063645..fb658af 100644 (file)
@@ -178,12 +178,12 @@ try_again_write:
 }
 
 /*
- * wait for a lock to be granted
+ * wait for the read lock to be granted
  */
-static struct rw_semaphore __sched *
-rwsem_down_failed_common(struct rw_semaphore *sem,
-                        enum rwsem_waiter_type type, signed long adjustment)
+struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
 {
+       enum rwsem_waiter_type type = RWSEM_WAITING_FOR_READ;
+       signed long adjustment = -RWSEM_ACTIVE_READ_BIAS;
        struct rwsem_waiter waiter;
        struct task_struct *tsk = current;
        signed long count;
@@ -238,21 +238,63 @@ rwsem_down_failed_common(struct rw_semaphore *sem,
 }
 
 /*
- * wait for the read lock to be granted
- */
-struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
-{
-       return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_READ,
-                                       -RWSEM_ACTIVE_READ_BIAS);
-}
-
-/*
  * wait for the write lock to be granted
  */
 struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
 {
-       return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_WRITE,
-                                       -RWSEM_ACTIVE_WRITE_BIAS);
+       enum rwsem_waiter_type type = RWSEM_WAITING_FOR_WRITE;
+       signed long adjustment = -RWSEM_ACTIVE_WRITE_BIAS;
+       struct rwsem_waiter waiter;
+       struct task_struct *tsk = current;
+       signed long count;
+
+       /* set up my own style of waitqueue */
+       waiter.task = tsk;
+       waiter.type = type;
+       get_task_struct(tsk);
+
+       raw_spin_lock_irq(&sem->wait_lock);
+       if (list_empty(&sem->wait_list))
+               adjustment += RWSEM_WAITING_BIAS;
+       list_add_tail(&waiter.list, &sem->wait_list);
+
+       /* we're now waiting on the lock, but no longer actively locking */
+       count = rwsem_atomic_update(adjustment, sem);
+
+       /* If there are no active locks, wake the front queued process(es) up.
+        *
+        * Alternatively, if we're called from a failed down_write(), there
+        * were already threads queued before us and there are no active
+        * writers, the lock must be read owned; so we try to wake any read
+        * locks that were queued ahead of us. */
+       if (count == RWSEM_WAITING_BIAS)
+               sem = __rwsem_do_wake(sem, RWSEM_WAKE_NO_ACTIVE);
+       else if (count > RWSEM_WAITING_BIAS &&
+                adjustment == -RWSEM_ACTIVE_WRITE_BIAS)
+               sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
+
+       raw_spin_unlock_irq(&sem->wait_lock);
+
+       /* wait to be given the lock */
+       while (true) {
+               set_task_state(tsk, TASK_UNINTERRUPTIBLE);
+               if (!waiter.task)
+                       break;
+
+               raw_spin_lock_irq(&sem->wait_lock);
+               /* Try to get the writer sem, may steal from the head writer: */
+               if (type == RWSEM_WAITING_FOR_WRITE)
+                       if (try_get_writer_sem(sem, &waiter)) {
+                               raw_spin_unlock_irq(&sem->wait_lock);
+                               return sem;
+                       }
+               raw_spin_unlock_irq(&sem->wait_lock);
+               schedule();
+       }
+
+       tsk->state = TASK_RUNNING;
+
+       return sem;
 }
 
 /*