This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / kernel / wait.c
diff --git a/kernel/wait.c b/kernel/wait.c
new file mode 100644 (file)
index 0000000..791681c
--- /dev/null
@@ -0,0 +1,246 @@
+/*
+ * Generic waiting primitives.
+ *
+ * (C) 2004 William Irwin, Oracle
+ */
+#include <linux/config.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/wait.h>
+#include <linux/hash.h>
+
+void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
+{
+       unsigned long flags;
+
+       wait->flags &= ~WQ_FLAG_EXCLUSIVE;
+       spin_lock_irqsave(&q->lock, flags);
+       __add_wait_queue(q, wait);
+       spin_unlock_irqrestore(&q->lock, flags);
+}
+EXPORT_SYMBOL(add_wait_queue);
+
+void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
+{
+       unsigned long flags;
+
+       wait->flags |= WQ_FLAG_EXCLUSIVE;
+       spin_lock_irqsave(&q->lock, flags);
+       __add_wait_queue_tail(q, wait);
+       spin_unlock_irqrestore(&q->lock, flags);
+}
+EXPORT_SYMBOL(add_wait_queue_exclusive);
+
+void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&q->lock, flags);
+       __remove_wait_queue(q, wait);
+       spin_unlock_irqrestore(&q->lock, flags);
+}
+EXPORT_SYMBOL(remove_wait_queue);
+
+
+/*
+ * Note: we use "set_current_state()" _after_ the wait-queue add,
+ * because we need a memory barrier there on SMP, so that any
+ * wake-function that tests for the wait-queue being active
+ * will be guaranteed to see waitqueue addition _or_ subsequent
+ * tests in this thread will see the wakeup having taken place.
+ *
+ * The spin_unlock() itself is semi-permeable and only protects
+ * one way (it only protects stuff inside the critical region and
+ * stops them from bleeding out - it would still allow subsequent
+ * loads to move into the the critical region).
+ */
+void fastcall
+prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
+{
+       unsigned long flags;
+
+       wait->flags &= ~WQ_FLAG_EXCLUSIVE;
+       spin_lock_irqsave(&q->lock, flags);
+       if (list_empty(&wait->task_list))
+               __add_wait_queue(q, wait);
+       /*
+        * don't alter the task state if this is just going to
+        * queue an async wait queue callback
+        */
+       if (is_sync_wait(wait))
+               set_current_state(state);
+       spin_unlock_irqrestore(&q->lock, flags);
+}
+EXPORT_SYMBOL(prepare_to_wait);
+
+void fastcall
+prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
+{
+       unsigned long flags;
+
+       wait->flags |= WQ_FLAG_EXCLUSIVE;
+       spin_lock_irqsave(&q->lock, flags);
+       if (list_empty(&wait->task_list))
+               __add_wait_queue_tail(q, wait);
+       /*
+        * don't alter the task state if this is just going to
+        * queue an async wait queue callback
+        */
+       if (is_sync_wait(wait))
+               set_current_state(state);
+       spin_unlock_irqrestore(&q->lock, flags);
+}
+EXPORT_SYMBOL(prepare_to_wait_exclusive);
+
+void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
+{
+       unsigned long flags;
+
+       __set_current_state(TASK_RUNNING);
+       /*
+        * We can check for list emptiness outside the lock
+        * IFF:
+        *  - we use the "careful" check that verifies both
+        *    the next and prev pointers, so that there cannot
+        *    be any half-pending updates in progress on other
+        *    CPU's that we haven't seen yet (and that might
+        *    still change the stack area.
+        * and
+        *  - all other users take the lock (ie we can only
+        *    have _one_ other CPU that looks at or modifies
+        *    the list).
+        */
+       if (!list_empty_careful(&wait->task_list)) {
+               spin_lock_irqsave(&q->lock, flags);
+               list_del_init(&wait->task_list);
+               spin_unlock_irqrestore(&q->lock, flags);
+       }
+}
+EXPORT_SYMBOL(finish_wait);
+
+int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
+{
+       int ret = default_wake_function(wait, mode, sync, key);
+
+       if (ret)
+               list_del_init(&wait->task_list);
+       return ret;
+}
+EXPORT_SYMBOL(autoremove_wake_function);
+
+int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
+{
+       struct wait_bit_key *key = arg;
+       struct wait_bit_queue *wait_bit
+               = container_of(wait, struct wait_bit_queue, wait);
+
+       if (wait_bit->key.flags != key->flags ||
+                       wait_bit->key.bit_nr != key->bit_nr ||
+                       test_bit(key->bit_nr, key->flags))
+               return 0;
+       else
+               return autoremove_wake_function(wait, mode, sync, key);
+}
+EXPORT_SYMBOL(wake_bit_function);
+
+/*
+ * To allow interruptible waiting and asynchronous (i.e. nonblocking)
+ * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
+ * permitted return codes. Nonzero return codes halt waiting and return.
+ */
+int __sched fastcall
+__wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
+                       int (*action)(void *), unsigned mode)
+{
+       int ret = 0;
+
+       do {
+               prepare_to_wait(wq, &q->wait, mode);
+               if (test_bit(q->key.bit_nr, q->key.flags))
+                       ret = (*action)(q->key.flags);
+       } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
+       finish_wait(wq, &q->wait);
+       return ret;
+}
+EXPORT_SYMBOL(__wait_on_bit);
+
+int __sched fastcall out_of_line_wait_on_bit(void *word, int bit,
+                                       int (*action)(void *), unsigned mode)
+{
+       wait_queue_head_t *wq = bit_waitqueue(word, bit);
+       DEFINE_WAIT_BIT(wait, word, bit);
+
+       return __wait_on_bit(wq, &wait, action, mode);
+}
+EXPORT_SYMBOL(out_of_line_wait_on_bit);
+
+int __sched fastcall
+__wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
+                       int (*action)(void *), unsigned mode)
+{
+       int ret = 0;
+
+       do {
+               prepare_to_wait_exclusive(wq, &q->wait, mode);
+               if (test_bit(q->key.bit_nr, q->key.flags)) {
+                       if ((ret = (*action)(q->key.flags)))
+                               break;
+               }
+       } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
+       finish_wait(wq, &q->wait);
+       return ret;
+}
+EXPORT_SYMBOL(__wait_on_bit_lock);
+
+int __sched fastcall out_of_line_wait_on_bit_lock(void *word, int bit,
+                                       int (*action)(void *), unsigned mode)
+{
+       wait_queue_head_t *wq = bit_waitqueue(word, bit);
+       DEFINE_WAIT_BIT(wait, word, bit);
+
+       return __wait_on_bit_lock(wq, &wait, action, mode);
+}
+EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
+
+void fastcall __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
+{
+       struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
+       if (waitqueue_active(wq))
+               __wake_up(wq, TASK_INTERRUPTIBLE|TASK_UNINTERRUPTIBLE, 1, &key);
+}
+EXPORT_SYMBOL(__wake_up_bit);
+
+/**
+ * wake_up_bit - wake up a waiter on a bit
+ * @word: the word being waited on, a kernel virtual address
+ * @bit: the bit of the word being waited on
+ *
+ * There is a standard hashed waitqueue table for generic use. This
+ * is the part of the hashtable's accessor API that wakes up waiters
+ * on a bit. For instance, if one were to have waiters on a bitflag,
+ * one would call wake_up_bit() after clearing the bit.
+ *
+ * In order for this to function properly, as it uses waitqueue_active()
+ * internally, some kind of memory barrier must be done prior to calling
+ * this. Typically, this will be smp_mb__after_clear_bit(), but in some
+ * cases where bitflags are manipulated non-atomically under a lock, one
+ * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
+ * because spin_unlock() does not guarantee a memory barrier.
+ */
+void fastcall wake_up_bit(void *word, int bit)
+{
+       __wake_up_bit(bit_waitqueue(word, bit), word, bit);
+}
+EXPORT_SYMBOL(wake_up_bit);
+
+fastcall wait_queue_head_t *bit_waitqueue(void *word, int bit)
+{
+       const int shift = BITS_PER_LONG == 32 ? 5 : 6;
+       const struct zone *zone = page_zone(virt_to_page(word));
+       unsigned long val = (unsigned long)word << shift | bit;
+
+       return &zone->wait_table[hash_long(val, zone->wait_table_bits)];
+}
+EXPORT_SYMBOL(bit_waitqueue);