kernel/wait.c

   1 /*
   2  * Generic waiting primitives.
   3  *
   4  * (C) 2004 Nadia Yvette Chambers, Oracle
   5  */
   6 #include <linux/init.h>
   7 #include <linux/export.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, const char *name, struct lock_class_key *key)
  14 {
  15         spin_lock_init(&q->lock);
  16         lockdep_set_class_and_name(&q->lock, key, name);
  17         INIT_LIST_HEAD(&q->task_list);
  18 }
  19
  20 EXPORT_SYMBOL(__init_waitqueue_head);
  21
  22 void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  23 {
  24         unsigned long flags;
  25
  26         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  27         spin_lock_irqsave(&q->lock, flags);
  28         __add_wait_queue(q, wait);
  29         spin_unlock_irqrestore(&q->lock, flags);
  30 }
  31 EXPORT_SYMBOL(add_wait_queue);
  32
  33 void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
  34 {
  35         unsigned long flags;
  36
  37         wait->flags |= WQ_FLAG_EXCLUSIVE;
  38         spin_lock_irqsave(&q->lock, flags);
  39         __add_wait_queue_tail(q, wait);
  40         spin_unlock_irqrestore(&q->lock, flags);
  41 }
  42 EXPORT_SYMBOL(add_wait_queue_exclusive);
  43
  44 void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
  45 {
  46         unsigned long flags;
  47
  48         spin_lock_irqsave(&q->lock, flags);
  49         __remove_wait_queue(q, wait);
  50         spin_unlock_irqrestore(&q->lock, flags);
  51 }
  52 EXPORT_SYMBOL(remove_wait_queue);
  53
  54
  55 /*
  56  * Note: we use "set_current_state()" _after_ the wait-queue add,
  57  * because we need a memory barrier there on SMP, so that any
  58  * wake-function that tests for the wait-queue being active
  59  * will be guaranteed to see waitqueue addition _or_ subsequent
  60  * tests in this thread will see the wakeup having taken place.
  61  *
  62  * The spin_unlock() itself is semi-permeable and only protects
  63  * one way (it only protects stuff inside the critical region and
  64  * stops them from bleeding out - it would still allow subsequent
  65  * loads to move into the critical region).
  66  */
  67 void
  68 prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
  69 {
  70         unsigned long flags;
  71
  72         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
  73         spin_lock_irqsave(&q->lock, flags);
  74         if (list_empty(&wait->task_list))
  75                 __add_wait_queue(q, wait);
  76         set_current_state(state);
  77         spin_unlock_irqrestore(&q->lock, flags);
  78 }
  79 EXPORT_SYMBOL(prepare_to_wait);
  80
  81 void
  82 prepare_to_wait_exclusive(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_tail(q, wait);
  90         set_current_state(state);
  91         spin_unlock_irqrestore(&q->lock, flags);
  92 }
  93 EXPORT_SYMBOL(prepare_to_wait_exclusive);
  94
  95 /**
  96  * finish_wait - clean up after waiting in a queue
  97  * @q: waitqueue waited on
  98  * @wait: wait descriptor
  99  *
 100  * Sets current thread back to running state and removes
 101  * the wait descriptor from the given waitqueue if still
 102  * queued.
 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 /**
 131  * abort_exclusive_wait - abort exclusive waiting in a queue
 132  * @q: waitqueue waited on
 133  * @wait: wait descriptor
 134  * @mode: runstate of the waiter to be woken
 135  * @key: key to identify a wait bit queue or %NULL
 136  *
 137  * Sets current thread back to running state and removes
 138  * the wait descriptor from the given waitqueue if still
 139  * queued.
 140  *
 141  * Wakes up the next waiter if the caller is concurrently
 142  * woken up through the queue.
 143  *
 144  * This prevents waiter starvation where an exclusive waiter
 145  * aborts and is woken up concurrently and no one wakes up
 146  * the next waiter.
 147  */
 148 void abort_exclusive_wait(wait_queue_head_t *q, wait_queue_t *wait,
 149                         unsigned int mode, void *key)
 150 {
 151         unsigned long flags;
 152
 153         __set_current_state(TASK_RUNNING);
 154         spin_lock_irqsave(&q->lock, flags);
 155         if (!list_empty(&wait->task_list))
 156                 list_del_init(&wait->task_list);
 157         else if (waitqueue_active(q))
 158                 __wake_up_locked_key(q, mode, key);
 159         spin_unlock_irqrestore(&q->lock, flags);
 160 }
 161 EXPORT_SYMBOL(abort_exclusive_wait);
 162
 163 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 164 {
 165         int ret = default_wake_function(wait, mode, sync, key);
 166
 167         if (ret)
 168                 list_del_init(&wait->task_list);
 169         return ret;
 170 }
 171 EXPORT_SYMBOL(autoremove_wake_function);
 172
 173 int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
 174 {
 175         struct wait_bit_key *key = arg;
 176         struct wait_bit_queue *wait_bit
 177                 = container_of(wait, struct wait_bit_queue, wait);
 178
 179         if (wait_bit->key.flags != key->flags ||
 180                         wait_bit->key.bit_nr != key->bit_nr ||
 181                         test_bit(key->bit_nr, key->flags))
 182                 return 0;
 183         else
 184                 return autoremove_wake_function(wait, mode, sync, key);
 185 }
 186 EXPORT_SYMBOL(wake_bit_function);
 187
 188 /*
 189  * To allow interruptible waiting and asynchronous (i.e. nonblocking)
 190  * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
 191  * permitted return codes. Nonzero return codes halt waiting and return.
 192  */
 193 int __sched
 194 __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
 195                         int (*action)(void *), unsigned mode)
 196 {
 197         int ret = 0;
 198
 199         do {
 200                 prepare_to_wait(wq, &q->wait, mode);
 201                 if (test_bit(q->key.bit_nr, q->key.flags))
 202                         ret = (*action)(q->key.flags);
 203         } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
 204         finish_wait(wq, &q->wait);
 205         return ret;
 206 }
 207 EXPORT_SYMBOL(__wait_on_bit);
 208
 209 int __sched out_of_line_wait_on_bit(void *word, int bit,
 210                                         int (*action)(void *), unsigned mode)
 211 {
 212         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 213         DEFINE_WAIT_BIT(wait, word, bit);
 214
 215         return __wait_on_bit(wq, &wait, action, mode);
 216 }
 217 EXPORT_SYMBOL(out_of_line_wait_on_bit);
 218
 219 int __sched
 220 __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
 221                         int (*action)(void *), unsigned mode)
 222 {
 223         do {
 224                 int ret;
 225
 226                 prepare_to_wait_exclusive(wq, &q->wait, mode);
 227                 if (!test_bit(q->key.bit_nr, q->key.flags))
 228                         continue;
 229                 ret = action(q->key.flags);
 230                 if (!ret)
 231                         continue;
 232                 abort_exclusive_wait(wq, &q->wait, mode, &q->key);
 233                 return ret;
 234         } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
 235         finish_wait(wq, &q->wait);
 236         return 0;
 237 }
 238 EXPORT_SYMBOL(__wait_on_bit_lock);
 239
 240 int __sched out_of_line_wait_on_bit_lock(void *word, int bit,
 241                                         int (*action)(void *), unsigned mode)
 242 {
 243         wait_queue_head_t *wq = bit_waitqueue(word, bit);
 244         DEFINE_WAIT_BIT(wait, word, bit);
 245
 246         return __wait_on_bit_lock(wq, &wait, action, mode);
 247 }
 248 EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
 249
 250 void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
 251 {
 252         struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
 253         if (waitqueue_active(wq))
 254                 __wake_up(wq, TASK_NORMAL, 1, &key);
 255 }
 256 EXPORT_SYMBOL(__wake_up_bit);
 257
 258 /**
 259  * wake_up_bit - wake up a waiter on a bit
 260  * @word: the word being waited on, a kernel virtual address
 261  * @bit: the bit of the word being waited on
 262  *
 263  * There is a standard hashed waitqueue table for generic use. This
 264  * is the part of the hashtable's accessor API that wakes up waiters
 265  * on a bit. For instance, if one were to have waiters on a bitflag,
 266  * one would call wake_up_bit() after clearing the bit.
 267  *
 268  * In order for this to function properly, as it uses waitqueue_active()
 269  * internally, some kind of memory barrier must be done prior to calling
 270  * this. Typically, this will be smp_mb__after_clear_bit(), but in some
 271  * cases where bitflags are manipulated non-atomically under a lock, one
 272  * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
 273  * because spin_unlock() does not guarantee a memory barrier.
 274  */
 275 void wake_up_bit(void *word, int bit)
 276 {
 277         __wake_up_bit(bit_waitqueue(word, bit), word, bit);
 278 }
 279 EXPORT_SYMBOL(wake_up_bit);
 280
 281 wait_queue_head_t *bit_waitqueue(void *word, int bit)
 282 {
 283         const int shift = BITS_PER_LONG == 32 ? 5 : 6;
 284         const struct zone *zone = page_zone(virt_to_page(word));
 285         unsigned long val = (unsigned long)word << shift | bit;
 286
 287         return &zone->wait_table[hash_long(val, zone->wait_table_bits)];
 288 }
 289 EXPORT_SYMBOL(bit_waitqueue);
 290
 291 /*
 292  * Manipulate the atomic_t address to produce a better bit waitqueue table hash
 293  * index (we're keying off bit -1, but that would produce a horrible hash
 294  * value).
 295  */
 296 static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p)
 297 {
 298         if (BITS_PER_LONG == 64) {
 299                 unsigned long q = (unsigned long)p;
 300                 return bit_waitqueue((void *)(q & ~1), q & 1);
 301         }
 302         return bit_waitqueue(p, 0);
 303 }
 304
 305 static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync,
 306                                   void *arg)
 307 {
 308         struct wait_bit_key *key = arg;
 309         struct wait_bit_queue *wait_bit
 310                 = container_of(wait, struct wait_bit_queue, wait);
 311         atomic_t *val = key->flags;
 312
 313         if (wait_bit->key.flags != key->flags ||
 314             wait_bit->key.bit_nr != key->bit_nr ||
 315             atomic_read(val) != 0)
 316                 return 0;
 317         return autoremove_wake_function(wait, mode, sync, key);
 318 }
 319
 320 /*
 321  * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
 322  * the actions of __wait_on_atomic_t() are permitted return codes.  Nonzero
 323  * return codes halt waiting and return.
 324  */
 325 static __sched
 326 int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q,
 327                        int (*action)(atomic_t *), unsigned mode)
 328 {
 329         atomic_t *val;
 330         int ret = 0;
 331
 332         do {
 333                 prepare_to_wait(wq, &q->wait, mode);
 334                 val = q->key.flags;
 335                 if (atomic_read(val) == 0)
 336                         ret = (*action)(val);
 337         } while (!ret && atomic_read(val) != 0);
 338         finish_wait(wq, &q->wait);
 339         return ret;
 340 }
 341
 342 #define DEFINE_WAIT_ATOMIC_T(name, p)                                   \
 343         struct wait_bit_queue name = {                                  \
 344                 .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p),              \
 345                 .wait   = {                                             \
 346                         .private        = current,                      \
 347                         .func           = wake_atomic_t_function,       \
 348                         .task_list      =                               \
 349                                 LIST_HEAD_INIT((name).wait.task_list),  \
 350                 },                                                      \
 351         }
 352
 353 __sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *),
 354                                          unsigned mode)
 355 {
 356         wait_queue_head_t *wq = atomic_t_waitqueue(p);
 357         DEFINE_WAIT_ATOMIC_T(wait, p);
 358
 359         return __wait_on_atomic_t(wq, &wait, action, mode);
 360 }
 361 EXPORT_SYMBOL(out_of_line_wait_on_atomic_t);
 362
 363 /**
 364  * wake_up_atomic_t - Wake up a waiter on a atomic_t
 365  * @word: The word being waited on, a kernel virtual address
 366  * @bit: The bit of the word being waited on
 367  *
 368  * Wake up anyone waiting for the atomic_t to go to zero.
 369  *
 370  * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
 371  * check is done by the waiter's wake function, not the by the waker itself).
 372  */
 373 void wake_up_atomic_t(atomic_t *p)
 374 {
 375         __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR);
 376 }
 377 EXPORT_SYMBOL(wake_up_atomic_t);