Merge "Merge remote-tracking branch 'origin/dev/dev-kernel-3.18-lz4-zram' into dev...
[linux-3.10.git] / lib / rwsem.c
index 9a675fa..19c5fa9 100644 (file)
@@ -4,6 +4,7 @@
  * Derived from arch/i386/kernel/semaphore.c
  *
  * Writer lock-stealing by Alex Shi <alex.shi@intel.com>
+ * and Michel Lespinasse <walken@google.com>
  */
 #include <linux/rwsem.h>
 #include <linux/sched.h>
@@ -41,13 +42,11 @@ struct rwsem_waiter {
        enum rwsem_waiter_type type;
 };
 
-/* Wake types for __rwsem_do_wake().  Note that RWSEM_WAKE_NO_ACTIVE and
- * RWSEM_WAKE_READ_OWNED imply that the spinlock must have been kept held
- * since the rwsem value was observed.
- */
-#define RWSEM_WAKE_ANY        0 /* Wake whatever's at head of wait list */
-#define RWSEM_WAKE_NO_ACTIVE  1 /* rwsem was observed with no active thread */
-#define RWSEM_WAKE_READ_OWNED 2 /* rwsem was observed to be read owned */
+enum rwsem_wake_type {
+       RWSEM_WAKE_ANY,         /* Wake whatever's at head of wait list */
+       RWSEM_WAKE_READERS,     /* Wake readers only */
+       RWSEM_WAKE_READ_OWNED   /* Waker thread holds the read lock */
+};
 
 /*
  * handle the lock release when processes blocked on it that can now run
@@ -60,16 +59,16 @@ struct rwsem_waiter {
  * - writers are only woken if downgrading is false
  */
 static struct rw_semaphore *
-__rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
+__rwsem_do_wake(struct rw_semaphore *sem, enum rwsem_wake_type wake_type)
 {
        struct rwsem_waiter *waiter;
        struct task_struct *tsk;
        struct list_head *next;
-       signed long woken, loop, adjustment;
+       long oldcount, woken, loop, adjustment;
 
        waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
        if (waiter->type == RWSEM_WAITING_FOR_WRITE) {
-               if (wake_type != RWSEM_WAKE_READ_OWNED)
+               if (wake_type == RWSEM_WAKE_ANY)
                        /* Wake writer at the front of the queue, but do not
                         * grant it the lock yet as we want other writers
                         * to be able to steal it.  Readers, on the other hand,
@@ -79,24 +78,24 @@ __rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
                goto out;
        }
 
-       /* If we come here from up_xxxx(), another thread might have reached
-        * rwsem_down_failed_common() before we acquired the spinlock and
-        * woken up a waiter, making it now active.  We prefer to check for
-        * this first in order to not spend too much time with the spinlock
-        * held if we're not going to be able to wake up readers in the end.
-        *
-        * Note that we do not need to update the rwsem count: any writer
-        * trying to acquire rwsem will run rwsem_down_write_failed() due
-        * to the waiting threads and block trying to acquire the spinlock.
-        *
-        * We use a dummy atomic update in order to acquire the cache line
-        * exclusively since we expect to succeed and run the final rwsem
-        * count adjustment pretty soon.
+       /* Writers might steal the lock before we grant it to the next reader.
+        * We prefer to do the first reader grant before counting readers
+        * so we can bail out early if a writer stole the lock.
         */
-       if (wake_type == RWSEM_WAKE_ANY &&
-           rwsem_atomic_update(0, sem) < RWSEM_WAITING_BIAS)
-               /* Someone grabbed the sem for write already */
-               goto out;
+       adjustment = 0;
+       if (wake_type != RWSEM_WAKE_READ_OWNED) {
+               adjustment = RWSEM_ACTIVE_READ_BIAS;
+ try_reader_grant:
+               oldcount = rwsem_atomic_update(adjustment, sem) - adjustment;
+               if (unlikely(oldcount < RWSEM_WAITING_BIAS)) {
+                       /* A writer stole the lock. Undo our reader grant. */
+                       if (rwsem_atomic_update(-adjustment, sem) &
+                                               RWSEM_ACTIVE_MASK)
+                               goto out;
+                       /* Last active locker left. Retry waking readers. */
+                       goto try_reader_grant;
+               }
+       }
 
        /* Grant an infinite number of read locks to the readers at the front
         * of the queue.  Note we increment the 'active part' of the count by
@@ -114,12 +113,13 @@ __rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
 
        } while (waiter->type != RWSEM_WAITING_FOR_WRITE);
 
-       adjustment = woken * RWSEM_ACTIVE_READ_BIAS;
+       adjustment = woken * RWSEM_ACTIVE_READ_BIAS - adjustment;
        if (waiter->type != RWSEM_WAITING_FOR_WRITE)
                /* hit end of list above */
                adjustment -= RWSEM_WAITING_BIAS;
 
-       rwsem_atomic_add(adjustment, sem);
+       if (adjustment)
+               rwsem_atomic_add(adjustment, sem);
 
        next = sem->wait_list.next;
        loop = woken;
@@ -145,10 +145,9 @@ __rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
  */
 struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
 {
-       signed long adjustment = -RWSEM_ACTIVE_READ_BIAS;
+       long count, adjustment = -RWSEM_ACTIVE_READ_BIAS;
        struct rwsem_waiter waiter;
        struct task_struct *tsk = current;
-       signed long count;
 
        /* set up my own style of waitqueue */
        waiter.task = tsk;
@@ -163,9 +162,15 @@ struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
        /* 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). */
-       if (count == RWSEM_WAITING_BIAS)
-               sem = __rwsem_do_wake(sem, RWSEM_WAKE_NO_ACTIVE);
+       /* If there are no active locks, wake the front queued process(es).
+        *
+        * If there are no writers and we are first in the queue,
+        * wake our own waiter to join the existing active readers !
+        */
+       if (count == RWSEM_WAITING_BIAS ||
+           (count > RWSEM_WAITING_BIAS &&
+            adjustment != -RWSEM_ACTIVE_READ_BIAS))
+               sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY);
 
        raw_spin_unlock_irq(&sem->wait_lock);
 
@@ -187,10 +192,9 @@ struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
  */
 struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
 {
-       signed long adjustment = -RWSEM_ACTIVE_WRITE_BIAS;
+       long count, 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;
@@ -209,7 +213,7 @@ struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
         * any read locks that were queued ahead of us. */
        if (count > RWSEM_WAITING_BIAS &&
            adjustment == -RWSEM_ACTIVE_WRITE_BIAS)
-               sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
+               sem = __rwsem_do_wake(sem, RWSEM_WAKE_READERS);
 
        /* wait until we successfully acquire the lock */
        set_task_state(tsk, TASK_UNINTERRUPTIBLE);
@@ -219,7 +223,9 @@ struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
                        count = RWSEM_ACTIVE_WRITE_BIAS;
                        if (!list_is_singular(&sem->wait_list))
                                count += RWSEM_WAITING_BIAS;
-                       if (cmpxchg(&sem->count, RWSEM_WAITING_BIAS, count) ==
+
+                       if (sem->count == RWSEM_WAITING_BIAS &&
+                           cmpxchg(&sem->count, RWSEM_WAITING_BIAS, count) ==
                                                        RWSEM_WAITING_BIAS)
                                break;
                }