kernel/wait.c

   1 /*
   2  * Generic waiting primitives.
   3  *
   4  * (C) 2004 William Irwin, Oracle
   5  */
   6 #include <linux/config.h>
   7 #include <linux/init.h>
   8 #include <linux/module.h>
   9 #include <linux/sched.h>
  10 #include <linux/mm.h>
  11 #include <linux/wait.h>
  12 #include <linux/hash.h>
  13
  14 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  15 {
  16         unsigned long flags;
  17
  18         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  19         spin_lock_irqsave(&q->lock, flags);
  20         __add_wait_queue(q, wait);
  21         spin_unlock_irqrestore(&q->lock, flags);
  22 }
  23 EXPORT_SYMBOL(add_wait_queue);
  24
  25 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
  26 {
  27         unsigned long flags;
  28
  29         wait->flags |= WQ_FLAG_EXCLUSIVE;
  30         spin_lock_irqsave(&q->lock, flags);
  31         __add_wait_queue_tail(q, wait);
  32         spin_unlock_irqrestore(&q->lock, flags);
  33 }
  34 EXPORT_SYMBOL(add_wait_queue_exclusive);
  35
  36 int fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  37 {
  38         unsigned long flags;
  39         struct list_head *list;
  40         int seen, retval;
  41
  42         spin_lock_irqsave(&q->lock, flags);
  43         list = &q->task_list;
  44         seen = 0;
  45         retval = -1;
  46
  47         do {
  48                 struct list_head *next;
  49                 if (list == &wait->task_list)
  50                         seen++;
  51                 next = list->next;
  52                 if (next->prev != list) {
  53                         seen += 2;
  54                         break;
  55                 }
  56                 list = next;
  57         } while (list != &q->task_list);
  58
  59         if (seen == 1) {
  60                 __remove_wait_queue(q, wait);
  61                 retval = 0;
  62         }
  63         spin_unlock_irqrestore(&q->lock, flags);
  64         return retval;
  65 }
  66 EXPORT_SYMBOL(remove_wait_queue);
  67
  68
  69 /*
  70  * Note: we use "set_current_state()" _after_ the wait-queue add,
  71  * because we need a memory barrier there on SMP, so that any
  72  * wake-function that tests for the wait-queue being active
  73  * will be guaranteed to see waitqueue addition _or_ subsequent
  74  * tests in this thread will see the wakeup having taken place.
  75  *
  76  * The spin_unlock() itself is semi-permeable and only protects
  77  * one way (it only protects stuff inside the critical region and
  78  * stops them from bleeding out - it would still allow subsequent
  79  * loads to move into the the critical region).
  80  */
  81 void fastcall
  82 prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
  83 {
  84         unsigned long flags;
  85
  86         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  87         spin_lock_irqsave(&q->lock, flags);
  88         if (list_empty(&wait->task_list))
  89                 __add_wait_queue(q, wait);
  90         /*
  91          * don't alter the task state if this is just going to
  92          * queue an async wait queue callback
  93          */
  94         if (is_sync_wait(wait))
  95                 set_current_state(state);
  96         spin_unlock_irqrestore(&q->lock, flags);
  97 }
  98 EXPORT_SYMBOL(prepare_to_wait);
  99
 100 void fastcall
 101 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 102 {
 103         unsigned long flags;
 104
 105         wait->flags |= WQ_FLAG_EXCLUSIVE;
 106         spin_lock_irqsave(&q->lock, flags);
 107         if (list_empty(&wait->task_list))
 108                 __add_wait_queue_tail(q, wait);
 109         /*
 110          * don't alter the task state if this is just going to
 111          * queue an async wait queue callback
 112          */
 113         if (is_sync_wait(wait))
 114                 set_current_state(state);
 115         spin_unlock_irqrestore(&q->lock, flags);
 116 }
 117 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 118
 119 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 120 {
 121         unsigned long flags;
 122
 123         __set_current_state(TASK_RUNNING);
 124         /*
 125          * We can check for list emptiness outside the lock
 126          * IFF:
 127          *  - we use the "careful" check that verifies both
 128          *    the next and prev pointers, so that there cannot
 129          *    be any half-pending updates in progress on other
 130          *    CPU's that we haven't seen yet (and that might
 131          *    still change the stack area.
 132          * and
 133          *  - all other users take the lock (ie we can only
 134          *    have _one_ other CPU that looks at or modifies
 135          *    the list).
 136          */
 137         if (!list_empty_careful(&wait->task_list)) {
 138                 spin_lock_irqsave(&q->lock, flags);
 139                 list_del_init(&wait->task_list);
 140                 spin_unlock_irqrestore(&q->lock, flags);
 141         }
 142 }
 143 EXPORT_SYMBOL(finish_wait);
 144
 145 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 146 {
 147         int ret = default_wake_function(wait, mode, sync, key);
 148
 149         if (ret)
 150                 list_del_init(&wait->task_list);
 151         return ret;
 152 }
 153 EXPORT_SYMBOL(autoremove_wake_function);
 154
 155 int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
 156 {
 157         struct wait_bit_key *key = arg;
 158         struct wait_bit_queue *wait_bit
 159                 = container_of(wait, struct wait_bit_queue, wait);
 160
 161         if (wait_bit->key.flags != key->flags ||
 162                         wait_bit->key.bit_nr != key->bit_nr ||
 163                         test_bit(key->bit_nr, key->flags))
 164                 return 0;
 165         else
 166                 return autoremove_wake_function(wait, mode, sync, key);
 167 }
 168 EXPORT_SYMBOL(wake_bit_function);
 169
 170 /*
 171  * To allow interruptible waiting and asynchronous (i.e. nonblocking)
 172  * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
 173  * permitted return codes. Nonzero return codes halt waiting and return.
 174  */
 175 int __sched fastcall
 176 __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
 177                         int (*action)(void *), unsigned mode)
 178 {
 179         int ret = 0;
 180
 181         do {
 182                 prepare_to_wait(wq, &q->wait, mode);
 183                 if (test_bit(q->key.bit_nr, q->key.flags))
 184                         ret = (*action)(q->key.flags);
 185         } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
 186         finish_wait(wq, &q->wait);
 187         return ret;
 188 }
 189 EXPORT_SYMBOL(__wait_on_bit);
 190
 191 int __sched fastcall out_of_line_wait_on_bit(void *word, int bit,
 192                                         int (*action)(void *), unsigned mode)
 193 {
 194         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 195         DEFINE_WAIT_BIT(wait, word, bit);
 196
 197         return __wait_on_bit(wq, &wait, action, mode);
 198 }
 199 EXPORT_SYMBOL(out_of_line_wait_on_bit);
 200
 201 int __sched fastcall
 202 __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
 203                         int (*action)(void *), unsigned mode)
 204 {
 205         int ret = 0;
 206
 207         do {
 208                 prepare_to_wait_exclusive(wq, &q->wait, mode);
 209                 if (test_bit(q->key.bit_nr, q->key.flags)) {
 210                         if ((ret = (*action)(q->key.flags)))
 211                                 break;
 212                 }
 213         } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
 214         finish_wait(wq, &q->wait);
 215         return ret;
 216 }
 217 EXPORT_SYMBOL(__wait_on_bit_lock);
 218
 219 int __sched fastcall out_of_line_wait_on_bit_lock(void *word, int bit,
 220                                         int (*action)(void *), unsigned mode)
 221 {
 222         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 223         DEFINE_WAIT_BIT(wait, word, bit);
 224
 225         return __wait_on_bit_lock(wq, &wait, action, mode);
 226 }
 227 EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
 228
 229 void fastcall __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
 230 {
 231         struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
 232         if (waitqueue_active(wq))
 233                 __wake_up(wq, TASK_INTERRUPTIBLE|TASK_UNINTERRUPTIBLE, 1, &key);
 234 }
 235 EXPORT_SYMBOL(__wake_up_bit);
 236
 237 /**
 238  * wake_up_bit - wake up a waiter on a bit
 239  * @word: the word being waited on, a kernel virtual address
 240  * @bit: the bit of the word being waited on
 241  *
 242  * There is a standard hashed waitqueue table for generic use. This
 243  * is the part of the hashtable's accessor API that wakes up waiters
 244  * on a bit. For instance, if one were to have waiters on a bitflag,
 245  * one would call wake_up_bit() after clearing the bit.
 246  *
 247  * In order for this to function properly, as it uses waitqueue_active()
 248  * internally, some kind of memory barrier must be done prior to calling
 249  * this. Typically, this will be smp_mb__after_clear_bit(), but in some
 250  * cases where bitflags are manipulated non-atomically under a lock, one
 251  * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
 252  * because spin_unlock() does not guarantee a memory barrier.
 253  */
 254 void fastcall wake_up_bit(void *word, int bit)
 255 {
 256         __wake_up_bit(bit_waitqueue(word, bit), word, bit);
 257 }
 258 EXPORT_SYMBOL(wake_up_bit);
 259
 260 fastcall wait_queue_head_t *bit_waitqueue(void *word, int bit)
 261 {
 262         const int shift = BITS_PER_LONG == 32 ? 5 : 6;
 263         const struct zone *zone = page_zone(virt_to_page(word));
 264         unsigned long val = (unsigned long)word << shift | bit;
 265
 266         return &zone->wait_table[hash_long(val, zone->wait_table_bits)];
 267 }
 268 EXPORT_SYMBOL(bit_waitqueue);