kernel/wait.c

   1 /*
   2  * Generic waiting primitives.
   3  *
   4  * (C) 2004 William Irwin, Oracle
   5  */
   6 #include <linux/init.h>
   7 #include <linux/module.h>
   8 #include <linux/sched.h>
   9 #include <linux/mm.h>
  10 #include <linux/wait.h>
  11 #include <linux/hash.h>
  12
  13 void init_waitqueue_head(wait_queue_head_t *q)
  14 {
  15         spin_lock_init(&q->lock);
  16         INIT_LIST_HEAD(&q->task_list);
  17 }
  18
  19 EXPORT_SYMBOL(init_waitqueue_head);
  20
  21 void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  22 {
  23         unsigned long flags;
  24
  25         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  26         spin_lock_irqsave(&q->lock, flags);
  27         __add_wait_queue(q, wait);
  28         spin_unlock_irqrestore(&q->lock, flags);
  29 }
  30 EXPORT_SYMBOL(add_wait_queue);
  31
  32 void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
  33 {
  34         unsigned long flags;
  35
  36         wait->flags |= WQ_FLAG_EXCLUSIVE;
  37         spin_lock_irqsave(&q->lock, flags);
  38         __add_wait_queue_tail(q, wait);
  39         spin_unlock_irqrestore(&q->lock, flags);
  40 }
  41 EXPORT_SYMBOL(add_wait_queue_exclusive);
  42
  43 void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  44 {
  45         unsigned long flags;
  46
  47         spin_lock_irqsave(&q->lock, flags);
  48         __remove_wait_queue(q, wait);
  49         spin_unlock_irqrestore(&q->lock, flags);
  50 }
  51 EXPORT_SYMBOL(remove_wait_queue);
  52
  53
  54 /*
  55  * Note: we use "set_current_state()" _after_ the wait-queue add,
  56  * because we need a memory barrier there on SMP, so that any
  57  * wake-function that tests for the wait-queue being active
  58  * will be guaranteed to see waitqueue addition _or_ subsequent
  59  * tests in this thread will see the wakeup having taken place.
  60  *
  61  * The spin_unlock() itself is semi-permeable and only protects
  62  * one way (it only protects stuff inside the critical region and
  63  * stops them from bleeding out - it would still allow subsequent
  64  * loads to move into the critical region).
  65  */
  66 void
  67 prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
  68 {
  69         unsigned long flags;
  70
  71         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  72         spin_lock_irqsave(&q->lock, flags);
  73         if (list_empty(&wait->task_list))
  74                 __add_wait_queue(q, wait);
  75         /*
  76          * don't alter the task state if this is just going to
  77          * queue an async wait queue callback
  78          */
  79         if (is_sync_wait(wait))
  80                 set_current_state(state);
  81         spin_unlock_irqrestore(&q->lock, flags);
  82 }
  83 EXPORT_SYMBOL(prepare_to_wait);
  84
  85 void
  86 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
  87 {
  88         unsigned long flags;
  89
  90         wait->flags |= WQ_FLAG_EXCLUSIVE;
  91         spin_lock_irqsave(&q->lock, flags);
  92         if (list_empty(&wait->task_list))
  93                 __add_wait_queue_tail(q, wait);
  94         /*
  95          * don't alter the task state if this is just going to
  96          * queue an async wait queue callback
  97          */
  98         if (is_sync_wait(wait))
  99                 set_current_state(state);
 100         spin_unlock_irqrestore(&q->lock, flags);
 101 }
 102 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 103
 104 void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 105 {
 106         unsigned long flags;
 107
 108         __set_current_state(TASK_RUNNING);
 109         /*
 110          * We can check for list emptiness outside the lock
 111          * IFF:
 112          *  - we use the "careful" check that verifies both
 113          *    the next and prev pointers, so that there cannot
 114          *    be any half-pending updates in progress on other
 115          *    CPU's that we haven't seen yet (and that might
 116          *    still change the stack area.
 117          * and
 118          *  - all other users take the lock (ie we can only
 119          *    have _one_ other CPU that looks at or modifies
 120          *    the list).
 121          */
 122         if (!list_empty_careful(&wait->task_list)) {
 123                 spin_lock_irqsave(&q->lock, flags);
 124                 list_del_init(&wait->task_list);
 125                 spin_unlock_irqrestore(&q->lock, flags);
 126         }
 127 }
 128 EXPORT_SYMBOL(finish_wait);
 129
 130 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 131 {
 132         int ret = default_wake_function(wait, mode, sync, key);
 133
 134         if (ret)
 135                 list_del_init(&wait->task_list);
 136         return ret;
 137 }
 138 EXPORT_SYMBOL(autoremove_wake_function);
 139
 140 int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
 141 {
 142         struct wait_bit_key *key = arg;
 143         struct wait_bit_queue *wait_bit
 144                 = container_of(wait, struct wait_bit_queue, wait);
 145
 146         if (wait_bit->key.flags != key->flags ||
 147                         wait_bit->key.bit_nr != key->bit_nr ||
 148                         test_bit(key->bit_nr, key->flags))
 149                 return 0;
 150         else
 151                 return autoremove_wake_function(wait, mode, sync, key);
 152 }
 153 EXPORT_SYMBOL(wake_bit_function);
 154
 155 /*
 156  * To allow interruptible waiting and asynchronous (i.e. nonblocking)
 157  * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
 158  * permitted return codes. Nonzero return codes halt waiting and return.
 159  */
 160 int __sched
 161 __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
 162                         int (*action)(void *), unsigned mode)
 163 {
 164         int ret = 0;
 165
 166         do {
 167                 prepare_to_wait(wq, &q->wait, mode);
 168                 if (test_bit(q->key.bit_nr, q->key.flags))
 169                         ret = (*action)(q->key.flags);
 170         } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
 171         finish_wait(wq, &q->wait);
 172         return ret;
 173 }
 174 EXPORT_SYMBOL(__wait_on_bit);
 175
 176 int __sched out_of_line_wait_on_bit(void *word, int bit,
 177                                         int (*action)(void *), unsigned mode)
 178 {
 179         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 180         DEFINE_WAIT_BIT(wait, word, bit);
 181
 182         return __wait_on_bit(wq, &wait, action, mode);
 183 }
 184 EXPORT_SYMBOL(out_of_line_wait_on_bit);
 185
 186 int __sched
 187 __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
 188                         int (*action)(void *), unsigned mode)
 189 {
 190         int ret = 0;
 191
 192         do {
 193                 prepare_to_wait_exclusive(wq, &q->wait, mode);
 194                 if (test_bit(q->key.bit_nr, q->key.flags)) {
 195                         if ((ret = (*action)(q->key.flags)))
 196                                 break;
 197                 }
 198         } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
 199         finish_wait(wq, &q->wait);
 200         return ret;
 201 }
 202 EXPORT_SYMBOL(__wait_on_bit_lock);
 203
 204 int __sched out_of_line_wait_on_bit_lock(void *word, int bit,
 205                                         int (*action)(void *), unsigned mode)
 206 {
 207         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 208         DEFINE_WAIT_BIT(wait, word, bit);
 209
 210         return __wait_on_bit_lock(wq, &wait, action, mode);
 211 }
 212 EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
 213
 214 void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
 215 {
 216         struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
 217         if (waitqueue_active(wq))
 218                 __wake_up(wq, TASK_NORMAL, 1, &key);
 219 }
 220 EXPORT_SYMBOL(__wake_up_bit);
 221
 222 /**
 223  * wake_up_bit - wake up a waiter on a bit
 224  * @word: the word being waited on, a kernel virtual address
 225  * @bit: the bit of the word being waited on
 226  *
 227  * There is a standard hashed waitqueue table for generic use. This
 228  * is the part of the hashtable's accessor API that wakes up waiters
 229  * on a bit. For instance, if one were to have waiters on a bitflag,
 230  * one would call wake_up_bit() after clearing the bit.
 231  *
 232  * In order for this to function properly, as it uses waitqueue_active()
 233  * internally, some kind of memory barrier must be done prior to calling
 234  * this. Typically, this will be smp_mb__after_clear_bit(), but in some
 235  * cases where bitflags are manipulated non-atomically under a lock, one
 236  * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
 237  * because spin_unlock() does not guarantee a memory barrier.
 238  */
 239 void wake_up_bit(void *word, int bit)
 240 {
 241         __wake_up_bit(bit_waitqueue(word, bit), word, bit);
 242 }
 243 EXPORT_SYMBOL(wake_up_bit);
 244
 245 wait_queue_head_t *bit_waitqueue(void *word, int bit)
 246 {
 247         const int shift = BITS_PER_LONG == 32 ? 5 : 6;
 248         const struct zone *zone = page_zone(virt_to_page(word));
 249         unsigned long val = (unsigned long)word << shift | bit;
 250
 251         return &zone->wait_table[hash_long(val, zone->wait_table_bits)];
 252 }
 253 EXPORT_SYMBOL(bit_waitqueue);