Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 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 | void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) |
| 37 | { |
| 38 | unsigned long flags; |
| 39 | |
| 40 | spin_lock_irqsave(&q->lock, flags); |
| 41 | __remove_wait_queue(q, wait); |
| 42 | spin_unlock_irqrestore(&q->lock, flags); |
| 43 | } |
| 44 | EXPORT_SYMBOL(remove_wait_queue); |
| 45 | |
| 46 | |
| 47 | /* |
| 48 | * Note: we use "set_current_state()" _after_ the wait-queue add, |
| 49 | * because we need a memory barrier there on SMP, so that any |
| 50 | * wake-function that tests for the wait-queue being active |
| 51 | * will be guaranteed to see waitqueue addition _or_ subsequent |
| 52 | * tests in this thread will see the wakeup having taken place. |
| 53 | * |
| 54 | * The spin_unlock() itself is semi-permeable and only protects |
| 55 | * one way (it only protects stuff inside the critical region and |
| 56 | * stops them from bleeding out - it would still allow subsequent |
| 57 | * loads to move into the the critical region). |
| 58 | */ |
| 59 | void fastcall |
| 60 | prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state) |
| 61 | { |
| 62 | unsigned long flags; |
| 63 | |
| 64 | wait->flags &= ~WQ_FLAG_EXCLUSIVE; |
| 65 | spin_lock_irqsave(&q->lock, flags); |
| 66 | if (list_empty(&wait->task_list)) |
| 67 | __add_wait_queue(q, wait); |
| 68 | /* |
| 69 | * don't alter the task state if this is just going to |
| 70 | * queue an async wait queue callback |
| 71 | */ |
| 72 | if (is_sync_wait(wait)) |
| 73 | set_current_state(state); |
| 74 | spin_unlock_irqrestore(&q->lock, flags); |
| 75 | } |
| 76 | EXPORT_SYMBOL(prepare_to_wait); |
| 77 | |
| 78 | void fastcall |
| 79 | prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state) |
| 80 | { |
| 81 | unsigned long flags; |
| 82 | |
| 83 | wait->flags |= WQ_FLAG_EXCLUSIVE; |
| 84 | spin_lock_irqsave(&q->lock, flags); |
| 85 | if (list_empty(&wait->task_list)) |
| 86 | __add_wait_queue_tail(q, wait); |
| 87 | /* |
| 88 | * don't alter the task state if this is just going to |
| 89 | * queue an async wait queue callback |
| 90 | */ |
| 91 | if (is_sync_wait(wait)) |
| 92 | set_current_state(state); |
| 93 | spin_unlock_irqrestore(&q->lock, flags); |
| 94 | } |
| 95 | EXPORT_SYMBOL(prepare_to_wait_exclusive); |
| 96 | |
| 97 | void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait) |
| 98 | { |
| 99 | unsigned long flags; |
| 100 | |
| 101 | __set_current_state(TASK_RUNNING); |
| 102 | /* |
| 103 | * We can check for list emptiness outside the lock |
| 104 | * IFF: |
| 105 | * - we use the "careful" check that verifies both |
| 106 | * the next and prev pointers, so that there cannot |
| 107 | * be any half-pending updates in progress on other |
| 108 | * CPU's that we haven't seen yet (and that might |
| 109 | * still change the stack area. |
| 110 | * and |
| 111 | * - all other users take the lock (ie we can only |
| 112 | * have _one_ other CPU that looks at or modifies |
| 113 | * the list). |
| 114 | */ |
| 115 | if (!list_empty_careful(&wait->task_list)) { |
| 116 | spin_lock_irqsave(&q->lock, flags); |
| 117 | list_del_init(&wait->task_list); |
| 118 | spin_unlock_irqrestore(&q->lock, flags); |
| 119 | } |
| 120 | } |
| 121 | EXPORT_SYMBOL(finish_wait); |
| 122 | |
| 123 | int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key) |
| 124 | { |
| 125 | int ret = default_wake_function(wait, mode, sync, key); |
| 126 | |
| 127 | if (ret) |
| 128 | list_del_init(&wait->task_list); |
| 129 | return ret; |
| 130 | } |
| 131 | EXPORT_SYMBOL(autoremove_wake_function); |
| 132 | |
| 133 | int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg) |
| 134 | { |
| 135 | struct wait_bit_key *key = arg; |
| 136 | struct wait_bit_queue *wait_bit |
| 137 | = container_of(wait, struct wait_bit_queue, wait); |
| 138 | |
| 139 | if (wait_bit->key.flags != key->flags || |
| 140 | wait_bit->key.bit_nr != key->bit_nr || |
| 141 | test_bit(key->bit_nr, key->flags)) |
| 142 | return 0; |
| 143 | else |
| 144 | return autoremove_wake_function(wait, mode, sync, key); |
| 145 | } |
| 146 | EXPORT_SYMBOL(wake_bit_function); |
| 147 | |
| 148 | /* |
| 149 | * To allow interruptible waiting and asynchronous (i.e. nonblocking) |
| 150 | * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are |
| 151 | * permitted return codes. Nonzero return codes halt waiting and return. |
| 152 | */ |
| 153 | int __sched fastcall |
| 154 | __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q, |
| 155 | int (*action)(void *), unsigned mode) |
| 156 | { |
| 157 | int ret = 0; |
| 158 | |
| 159 | do { |
| 160 | prepare_to_wait(wq, &q->wait, mode); |
| 161 | if (test_bit(q->key.bit_nr, q->key.flags)) |
| 162 | ret = (*action)(q->key.flags); |
| 163 | } while (test_bit(q->key.bit_nr, q->key.flags) && !ret); |
| 164 | finish_wait(wq, &q->wait); |
| 165 | return ret; |
| 166 | } |
| 167 | EXPORT_SYMBOL(__wait_on_bit); |
| 168 | |
| 169 | int __sched fastcall out_of_line_wait_on_bit(void *word, int bit, |
| 170 | int (*action)(void *), unsigned mode) |
| 171 | { |
| 172 | wait_queue_head_t *wq = bit_waitqueue(word, bit); |
| 173 | DEFINE_WAIT_BIT(wait, word, bit); |
| 174 | |
| 175 | return __wait_on_bit(wq, &wait, action, mode); |
| 176 | } |
| 177 | EXPORT_SYMBOL(out_of_line_wait_on_bit); |
| 178 | |
| 179 | int __sched fastcall |
| 180 | __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q, |
| 181 | int (*action)(void *), unsigned mode) |
| 182 | { |
| 183 | int ret = 0; |
| 184 | |
| 185 | do { |
| 186 | prepare_to_wait_exclusive(wq, &q->wait, mode); |
| 187 | if (test_bit(q->key.bit_nr, q->key.flags)) { |
| 188 | if ((ret = (*action)(q->key.flags))) |
| 189 | break; |
| 190 | } |
| 191 | } while (test_and_set_bit(q->key.bit_nr, q->key.flags)); |
| 192 | finish_wait(wq, &q->wait); |
| 193 | return ret; |
| 194 | } |
| 195 | EXPORT_SYMBOL(__wait_on_bit_lock); |
| 196 | |
| 197 | int __sched fastcall out_of_line_wait_on_bit_lock(void *word, int bit, |
| 198 | int (*action)(void *), unsigned mode) |
| 199 | { |
| 200 | wait_queue_head_t *wq = bit_waitqueue(word, bit); |
| 201 | DEFINE_WAIT_BIT(wait, word, bit); |
| 202 | |
| 203 | return __wait_on_bit_lock(wq, &wait, action, mode); |
| 204 | } |
| 205 | EXPORT_SYMBOL(out_of_line_wait_on_bit_lock); |
| 206 | |
| 207 | void fastcall __wake_up_bit(wait_queue_head_t *wq, void *word, int bit) |
| 208 | { |
| 209 | struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit); |
| 210 | if (waitqueue_active(wq)) |
| 211 | __wake_up(wq, TASK_INTERRUPTIBLE|TASK_UNINTERRUPTIBLE, 1, &key); |
| 212 | } |
| 213 | EXPORT_SYMBOL(__wake_up_bit); |
| 214 | |
| 215 | /** |
| 216 | * wake_up_bit - wake up a waiter on a bit |
| 217 | * @word: the word being waited on, a kernel virtual address |
| 218 | * @bit: the bit of the word being waited on |
| 219 | * |
| 220 | * There is a standard hashed waitqueue table for generic use. This |
| 221 | * is the part of the hashtable's accessor API that wakes up waiters |
| 222 | * on a bit. For instance, if one were to have waiters on a bitflag, |
| 223 | * one would call wake_up_bit() after clearing the bit. |
| 224 | * |
| 225 | * In order for this to function properly, as it uses waitqueue_active() |
| 226 | * internally, some kind of memory barrier must be done prior to calling |
| 227 | * this. Typically, this will be smp_mb__after_clear_bit(), but in some |
| 228 | * cases where bitflags are manipulated non-atomically under a lock, one |
| 229 | * may need to use a less regular barrier, such fs/inode.c's smp_mb(), |
| 230 | * because spin_unlock() does not guarantee a memory barrier. |
| 231 | */ |
| 232 | void fastcall wake_up_bit(void *word, int bit) |
| 233 | { |
| 234 | __wake_up_bit(bit_waitqueue(word, bit), word, bit); |
| 235 | } |
| 236 | EXPORT_SYMBOL(wake_up_bit); |
| 237 | |
| 238 | fastcall wait_queue_head_t *bit_waitqueue(void *word, int bit) |
| 239 | { |
| 240 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; |
| 241 | const struct zone *zone = page_zone(virt_to_page(word)); |
| 242 | unsigned long val = (unsigned long)word << shift | bit; |
| 243 | |
| 244 | return &zone->wait_table[hash_long(val, zone->wait_table_bits)]; |
| 245 | } |
| 246 | EXPORT_SYMBOL(bit_waitqueue); |