Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* rwsem.c: R/W semaphores: contention handling functions |
2 | * | |
3 | * Written by David Howells (dhowells@redhat.com). | |
4 | * Derived from arch/i386/kernel/semaphore.c | |
ce6711f3 AS |
5 | * |
6 | * Writer lock-stealing by Alex Shi <alex.shi@intel.com> | |
fe6e674c | 7 | * and Michel Lespinasse <walken@google.com> |
4fc828e2 DB |
8 | * |
9 | * Optimistic spinning by Tim Chen <tim.c.chen@intel.com> | |
10 | * and Davidlohr Bueso <davidlohr@hp.com>. Based on mutexes. | |
1da177e4 LT |
11 | */ |
12 | #include <linux/rwsem.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/init.h> | |
8bc3bcc9 | 15 | #include <linux/export.h> |
4fc828e2 | 16 | #include <linux/sched/rt.h> |
7a215f89 | 17 | #include <linux/osq_lock.h> |
4fc828e2 | 18 | |
7a215f89 | 19 | #include "rwsem.h" |
1da177e4 | 20 | |
3cf2f34e TC |
21 | /* |
22 | * Guide to the rw_semaphore's count field for common values. | |
23 | * (32-bit case illustrated, similar for 64-bit) | |
24 | * | |
25 | * 0x0000000X (1) X readers active or attempting lock, no writer waiting | |
26 | * X = #active_readers + #readers attempting to lock | |
27 | * (X*ACTIVE_BIAS) | |
28 | * | |
29 | * 0x00000000 rwsem is unlocked, and no one is waiting for the lock or | |
30 | * attempting to read lock or write lock. | |
31 | * | |
32 | * 0xffff000X (1) X readers active or attempting lock, with waiters for lock | |
33 | * X = #active readers + # readers attempting lock | |
34 | * (X*ACTIVE_BIAS + WAITING_BIAS) | |
35 | * (2) 1 writer attempting lock, no waiters for lock | |
36 | * X-1 = #active readers + #readers attempting lock | |
37 | * ((X-1)*ACTIVE_BIAS + ACTIVE_WRITE_BIAS) | |
38 | * (3) 1 writer active, no waiters for lock | |
39 | * X-1 = #active readers + #readers attempting lock | |
40 | * ((X-1)*ACTIVE_BIAS + ACTIVE_WRITE_BIAS) | |
41 | * | |
42 | * 0xffff0001 (1) 1 reader active or attempting lock, waiters for lock | |
43 | * (WAITING_BIAS + ACTIVE_BIAS) | |
44 | * (2) 1 writer active or attempting lock, no waiters for lock | |
45 | * (ACTIVE_WRITE_BIAS) | |
46 | * | |
47 | * 0xffff0000 (1) There are writers or readers queued but none active | |
48 | * or in the process of attempting lock. | |
49 | * (WAITING_BIAS) | |
50 | * Note: writer can attempt to steal lock for this count by adding | |
51 | * ACTIVE_WRITE_BIAS in cmpxchg and checking the old count | |
52 | * | |
53 | * 0xfffe0001 (1) 1 writer active, or attempting lock. Waiters on queue. | |
54 | * (ACTIVE_WRITE_BIAS + WAITING_BIAS) | |
55 | * | |
56 | * Note: Readers attempt to lock by adding ACTIVE_BIAS in down_read and checking | |
57 | * the count becomes more than 0 for successful lock acquisition, | |
58 | * i.e. the case where there are only readers or nobody has lock. | |
59 | * (1st and 2nd case above). | |
60 | * | |
61 | * Writers attempt to lock by adding ACTIVE_WRITE_BIAS in down_write and | |
62 | * checking the count becomes ACTIVE_WRITE_BIAS for successful lock | |
63 | * acquisition (i.e. nobody else has lock or attempts lock). If | |
64 | * unsuccessful, in rwsem_down_write_failed, we'll check to see if there | |
65 | * are only waiters but none active (5th case above), and attempt to | |
66 | * steal the lock. | |
67 | * | |
68 | */ | |
69 | ||
4ea2176d IM |
70 | /* |
71 | * Initialize an rwsem: | |
72 | */ | |
73 | void __init_rwsem(struct rw_semaphore *sem, const char *name, | |
74 | struct lock_class_key *key) | |
75 | { | |
76 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
77 | /* | |
78 | * Make sure we are not reinitializing a held semaphore: | |
79 | */ | |
80 | debug_check_no_locks_freed((void *)sem, sizeof(*sem)); | |
4dfbb9d8 | 81 | lockdep_init_map(&sem->dep_map, name, key, 0); |
4ea2176d IM |
82 | #endif |
83 | sem->count = RWSEM_UNLOCKED_VALUE; | |
ddb6c9b5 | 84 | raw_spin_lock_init(&sem->wait_lock); |
4ea2176d | 85 | INIT_LIST_HEAD(&sem->wait_list); |
5db6c6fe | 86 | #ifdef CONFIG_RWSEM_SPIN_ON_OWNER |
4fc828e2 | 87 | sem->owner = NULL; |
4d9d951e | 88 | osq_lock_init(&sem->osq); |
4fc828e2 | 89 | #endif |
4ea2176d IM |
90 | } |
91 | ||
92 | EXPORT_SYMBOL(__init_rwsem); | |
93 | ||
e2d57f78 ML |
94 | enum rwsem_waiter_type { |
95 | RWSEM_WAITING_FOR_WRITE, | |
96 | RWSEM_WAITING_FOR_READ | |
97 | }; | |
98 | ||
1da177e4 LT |
99 | struct rwsem_waiter { |
100 | struct list_head list; | |
101 | struct task_struct *task; | |
e2d57f78 | 102 | enum rwsem_waiter_type type; |
1da177e4 LT |
103 | }; |
104 | ||
fe6e674c ML |
105 | enum rwsem_wake_type { |
106 | RWSEM_WAKE_ANY, /* Wake whatever's at head of wait list */ | |
107 | RWSEM_WAKE_READERS, /* Wake readers only */ | |
108 | RWSEM_WAKE_READ_OWNED /* Waker thread holds the read lock */ | |
109 | }; | |
70bdc6e0 | 110 | |
1da177e4 LT |
111 | /* |
112 | * handle the lock release when processes blocked on it that can now run | |
113 | * - if we come here from up_xxxx(), then: | |
114 | * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed) | |
115 | * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so) | |
345af7bf | 116 | * - there must be someone on the queue |
1da177e4 LT |
117 | * - the spinlock must be held by the caller |
118 | * - woken process blocks are discarded from the list after having task zeroed | |
119 | * - writers are only woken if downgrading is false | |
120 | */ | |
70bdc6e0 | 121 | static struct rw_semaphore * |
fe6e674c | 122 | __rwsem_do_wake(struct rw_semaphore *sem, enum rwsem_wake_type wake_type) |
1da177e4 LT |
123 | { |
124 | struct rwsem_waiter *waiter; | |
125 | struct task_struct *tsk; | |
126 | struct list_head *next; | |
b5f54181 | 127 | long oldcount, woken, loop, adjustment; |
1da177e4 | 128 | |
345af7bf | 129 | waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list); |
8cf5322c | 130 | if (waiter->type == RWSEM_WAITING_FOR_WRITE) { |
fe6e674c | 131 | if (wake_type == RWSEM_WAKE_ANY) |
8cf5322c ML |
132 | /* Wake writer at the front of the queue, but do not |
133 | * grant it the lock yet as we want other writers | |
134 | * to be able to steal it. Readers, on the other hand, | |
135 | * will block as they will notice the queued writer. | |
136 | */ | |
137 | wake_up_process(waiter->task); | |
345af7bf | 138 | goto out; |
8cf5322c | 139 | } |
1da177e4 | 140 | |
fe6e674c ML |
141 | /* Writers might steal the lock before we grant it to the next reader. |
142 | * We prefer to do the first reader grant before counting readers | |
143 | * so we can bail out early if a writer stole the lock. | |
70bdc6e0 | 144 | */ |
fe6e674c ML |
145 | adjustment = 0; |
146 | if (wake_type != RWSEM_WAKE_READ_OWNED) { | |
147 | adjustment = RWSEM_ACTIVE_READ_BIAS; | |
148 | try_reader_grant: | |
149 | oldcount = rwsem_atomic_update(adjustment, sem) - adjustment; | |
150 | if (unlikely(oldcount < RWSEM_WAITING_BIAS)) { | |
151 | /* A writer stole the lock. Undo our reader grant. */ | |
152 | if (rwsem_atomic_update(-adjustment, sem) & | |
153 | RWSEM_ACTIVE_MASK) | |
154 | goto out; | |
155 | /* Last active locker left. Retry waking readers. */ | |
156 | goto try_reader_grant; | |
157 | } | |
158 | } | |
1da177e4 | 159 | |
345af7bf ML |
160 | /* Grant an infinite number of read locks to the readers at the front |
161 | * of the queue. Note we increment the 'active part' of the count by | |
162 | * the number of readers before waking any processes up. | |
1da177e4 | 163 | */ |
1da177e4 LT |
164 | woken = 0; |
165 | do { | |
166 | woken++; | |
167 | ||
168 | if (waiter->list.next == &sem->wait_list) | |
169 | break; | |
170 | ||
171 | waiter = list_entry(waiter->list.next, | |
172 | struct rwsem_waiter, list); | |
173 | ||
e2d57f78 | 174 | } while (waiter->type != RWSEM_WAITING_FOR_WRITE); |
1da177e4 | 175 | |
fe6e674c | 176 | adjustment = woken * RWSEM_ACTIVE_READ_BIAS - adjustment; |
e2d57f78 | 177 | if (waiter->type != RWSEM_WAITING_FOR_WRITE) |
fd41b334 ML |
178 | /* hit end of list above */ |
179 | adjustment -= RWSEM_WAITING_BIAS; | |
1da177e4 | 180 | |
fe6e674c ML |
181 | if (adjustment) |
182 | rwsem_atomic_add(adjustment, sem); | |
1da177e4 LT |
183 | |
184 | next = sem->wait_list.next; | |
8cf5322c ML |
185 | loop = woken; |
186 | do { | |
1da177e4 LT |
187 | waiter = list_entry(next, struct rwsem_waiter, list); |
188 | next = waiter->list.next; | |
189 | tsk = waiter->task; | |
49e4b2bc DB |
190 | /* |
191 | * Make sure we do not wakeup the next reader before | |
192 | * setting the nil condition to grant the next reader; | |
193 | * otherwise we could miss the wakeup on the other | |
194 | * side and end up sleeping again. See the pairing | |
195 | * in rwsem_down_read_failed(). | |
196 | */ | |
d59dd462 | 197 | smp_mb(); |
1da177e4 LT |
198 | waiter->task = NULL; |
199 | wake_up_process(tsk); | |
200 | put_task_struct(tsk); | |
8cf5322c | 201 | } while (--loop); |
1da177e4 LT |
202 | |
203 | sem->wait_list.next = next; | |
204 | next->prev = &sem->wait_list; | |
205 | ||
206 | out: | |
1da177e4 | 207 | return sem; |
ce6711f3 AS |
208 | } |
209 | ||
1da177e4 | 210 | /* |
4fc828e2 | 211 | * Wait for the read lock to be granted |
1da177e4 | 212 | */ |
3ebae4f3 | 213 | __visible |
1e78277c | 214 | struct rw_semaphore __sched *rwsem_down_read_failed(struct rw_semaphore *sem) |
1da177e4 | 215 | { |
b5f54181 | 216 | long count, adjustment = -RWSEM_ACTIVE_READ_BIAS; |
a8618a0e | 217 | struct rwsem_waiter waiter; |
1da177e4 | 218 | struct task_struct *tsk = current; |
1da177e4 | 219 | |
1da177e4 | 220 | /* set up my own style of waitqueue */ |
a8618a0e | 221 | waiter.task = tsk; |
da16922c | 222 | waiter.type = RWSEM_WAITING_FOR_READ; |
1da177e4 LT |
223 | get_task_struct(tsk); |
224 | ||
f7dd1cee | 225 | raw_spin_lock_irq(&sem->wait_lock); |
fd41b334 ML |
226 | if (list_empty(&sem->wait_list)) |
227 | adjustment += RWSEM_WAITING_BIAS; | |
a8618a0e | 228 | list_add_tail(&waiter.list, &sem->wait_list); |
1da177e4 | 229 | |
70bdc6e0 | 230 | /* we're now waiting on the lock, but no longer actively locking */ |
1da177e4 LT |
231 | count = rwsem_atomic_update(adjustment, sem); |
232 | ||
25c39325 ML |
233 | /* If there are no active locks, wake the front queued process(es). |
234 | * | |
235 | * If there are no writers and we are first in the queue, | |
236 | * wake our own waiter to join the existing active readers ! | |
237 | */ | |
238 | if (count == RWSEM_WAITING_BIAS || | |
239 | (count > RWSEM_WAITING_BIAS && | |
240 | adjustment != -RWSEM_ACTIVE_READ_BIAS)) | |
fe6e674c | 241 | sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY); |
1da177e4 | 242 | |
ddb6c9b5 | 243 | raw_spin_unlock_irq(&sem->wait_lock); |
1da177e4 LT |
244 | |
245 | /* wait to be given the lock */ | |
f7dd1cee ML |
246 | while (true) { |
247 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
a8618a0e | 248 | if (!waiter.task) |
1da177e4 LT |
249 | break; |
250 | schedule(); | |
1da177e4 LT |
251 | } |
252 | ||
73105994 | 253 | __set_task_state(tsk, TASK_RUNNING); |
1da177e4 LT |
254 | return sem; |
255 | } | |
db0e716a | 256 | EXPORT_SYMBOL(rwsem_down_read_failed); |
1da177e4 | 257 | |
4fc828e2 DB |
258 | static inline bool rwsem_try_write_lock(long count, struct rw_semaphore *sem) |
259 | { | |
debfab74 JL |
260 | /* |
261 | * Try acquiring the write lock. Check count first in order | |
262 | * to reduce unnecessary expensive cmpxchg() operations. | |
263 | */ | |
264 | if (count == RWSEM_WAITING_BIAS && | |
265 | cmpxchg(&sem->count, RWSEM_WAITING_BIAS, | |
266 | RWSEM_ACTIVE_WRITE_BIAS) == RWSEM_WAITING_BIAS) { | |
267 | if (!list_is_singular(&sem->wait_list)) | |
268 | rwsem_atomic_update(RWSEM_WAITING_BIAS, sem); | |
7a215f89 | 269 | rwsem_set_owner(sem); |
debfab74 | 270 | return true; |
4fc828e2 | 271 | } |
debfab74 | 272 | |
4fc828e2 DB |
273 | return false; |
274 | } | |
275 | ||
5db6c6fe | 276 | #ifdef CONFIG_RWSEM_SPIN_ON_OWNER |
1da177e4 | 277 | /* |
4fc828e2 DB |
278 | * Try to acquire write lock before the writer has been put on wait queue. |
279 | */ | |
280 | static inline bool rwsem_try_write_lock_unqueued(struct rw_semaphore *sem) | |
281 | { | |
4d3199e4 | 282 | long old, count = READ_ONCE(sem->count); |
4fc828e2 DB |
283 | |
284 | while (true) { | |
285 | if (!(count == 0 || count == RWSEM_WAITING_BIAS)) | |
286 | return false; | |
287 | ||
288 | old = cmpxchg(&sem->count, count, count + RWSEM_ACTIVE_WRITE_BIAS); | |
7a215f89 DB |
289 | if (old == count) { |
290 | rwsem_set_owner(sem); | |
4fc828e2 | 291 | return true; |
7a215f89 | 292 | } |
4fc828e2 DB |
293 | |
294 | count = old; | |
295 | } | |
296 | } | |
297 | ||
298 | static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem) | |
299 | { | |
300 | struct task_struct *owner; | |
1a993670 | 301 | bool ret = true; |
4fc828e2 DB |
302 | |
303 | if (need_resched()) | |
37e95624 | 304 | return false; |
4fc828e2 DB |
305 | |
306 | rcu_read_lock(); | |
4d3199e4 | 307 | owner = READ_ONCE(sem->owner); |
1a993670 | 308 | if (!owner) { |
4d3199e4 | 309 | long count = READ_ONCE(sem->count); |
1a993670 DB |
310 | /* |
311 | * If sem->owner is not set, yet we have just recently entered the | |
312 | * slowpath with the lock being active, then there is a possibility | |
313 | * reader(s) may have the lock. To be safe, bail spinning in these | |
314 | * situations. | |
315 | */ | |
316 | if (count & RWSEM_ACTIVE_MASK) | |
317 | ret = false; | |
318 | goto done; | |
319 | } | |
4fc828e2 | 320 | |
1a993670 DB |
321 | ret = owner->on_cpu; |
322 | done: | |
323 | rcu_read_unlock(); | |
324 | return ret; | |
4fc828e2 DB |
325 | } |
326 | ||
4fc828e2 DB |
327 | static noinline |
328 | bool rwsem_spin_on_owner(struct rw_semaphore *sem, struct task_struct *owner) | |
329 | { | |
b3fd4f03 DB |
330 | long count; |
331 | ||
4fc828e2 | 332 | rcu_read_lock(); |
9198f6ed JL |
333 | while (sem->owner == owner) { |
334 | /* | |
335 | * Ensure we emit the owner->on_cpu, dereference _after_ | |
336 | * checking sem->owner still matches owner, if that fails, | |
337 | * owner might point to free()d memory, if it still matches, | |
338 | * the rcu_read_lock() ensures the memory stays valid. | |
339 | */ | |
340 | barrier(); | |
341 | ||
342 | /* abort spinning when need_resched or owner is not running */ | |
343 | if (!owner->on_cpu || need_resched()) { | |
b3fd4f03 DB |
344 | rcu_read_unlock(); |
345 | return false; | |
346 | } | |
4fc828e2 | 347 | |
3a6bfbc9 | 348 | cpu_relax_lowlatency(); |
4fc828e2 DB |
349 | } |
350 | rcu_read_unlock(); | |
351 | ||
b3fd4f03 DB |
352 | if (READ_ONCE(sem->owner)) |
353 | return true; /* new owner, continue spinning */ | |
354 | ||
4fc828e2 | 355 | /* |
b3fd4f03 DB |
356 | * When the owner is not set, the lock could be free or |
357 | * held by readers. Check the counter to verify the | |
358 | * state. | |
4fc828e2 | 359 | */ |
b3fd4f03 DB |
360 | count = READ_ONCE(sem->count); |
361 | return (count == 0 || count == RWSEM_WAITING_BIAS); | |
4fc828e2 DB |
362 | } |
363 | ||
364 | static bool rwsem_optimistic_spin(struct rw_semaphore *sem) | |
365 | { | |
366 | struct task_struct *owner; | |
367 | bool taken = false; | |
368 | ||
369 | preempt_disable(); | |
370 | ||
371 | /* sem->wait_lock should not be held when doing optimistic spinning */ | |
372 | if (!rwsem_can_spin_on_owner(sem)) | |
373 | goto done; | |
374 | ||
375 | if (!osq_lock(&sem->osq)) | |
376 | goto done; | |
377 | ||
378 | while (true) { | |
4d3199e4 | 379 | owner = READ_ONCE(sem->owner); |
4fc828e2 DB |
380 | if (owner && !rwsem_spin_on_owner(sem, owner)) |
381 | break; | |
382 | ||
383 | /* wait_lock will be acquired if write_lock is obtained */ | |
384 | if (rwsem_try_write_lock_unqueued(sem)) { | |
385 | taken = true; | |
386 | break; | |
387 | } | |
388 | ||
389 | /* | |
390 | * When there's no owner, we might have preempted between the | |
391 | * owner acquiring the lock and setting the owner field. If | |
392 | * we're an RT task that will live-lock because we won't let | |
393 | * the owner complete. | |
394 | */ | |
395 | if (!owner && (need_resched() || rt_task(current))) | |
396 | break; | |
397 | ||
398 | /* | |
399 | * The cpu_relax() call is a compiler barrier which forces | |
400 | * everything in this loop to be re-loaded. We don't need | |
401 | * memory barriers as we'll eventually observe the right | |
402 | * values at the cost of a few extra spins. | |
403 | */ | |
3a6bfbc9 | 404 | cpu_relax_lowlatency(); |
4fc828e2 DB |
405 | } |
406 | osq_unlock(&sem->osq); | |
407 | done: | |
408 | preempt_enable(); | |
409 | return taken; | |
410 | } | |
411 | ||
59aabfc7 WL |
412 | /* |
413 | * Return true if the rwsem has active spinner | |
414 | */ | |
415 | static inline bool rwsem_has_spinner(struct rw_semaphore *sem) | |
416 | { | |
417 | return osq_is_locked(&sem->osq); | |
418 | } | |
419 | ||
4fc828e2 DB |
420 | #else |
421 | static bool rwsem_optimistic_spin(struct rw_semaphore *sem) | |
422 | { | |
423 | return false; | |
424 | } | |
59aabfc7 WL |
425 | |
426 | static inline bool rwsem_has_spinner(struct rw_semaphore *sem) | |
427 | { | |
428 | return false; | |
429 | } | |
4fc828e2 DB |
430 | #endif |
431 | ||
432 | /* | |
433 | * Wait until we successfully acquire the write lock | |
1da177e4 | 434 | */ |
3ebae4f3 | 435 | __visible |
d1233754 | 436 | struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem) |
1da177e4 | 437 | { |
4fc828e2 DB |
438 | long count; |
439 | bool waiting = true; /* any queued threads before us */ | |
1e78277c | 440 | struct rwsem_waiter waiter; |
1e78277c | 441 | |
4fc828e2 DB |
442 | /* undo write bias from down_write operation, stop active locking */ |
443 | count = rwsem_atomic_update(-RWSEM_ACTIVE_WRITE_BIAS, sem); | |
444 | ||
445 | /* do optimistic spinning and steal lock if possible */ | |
446 | if (rwsem_optimistic_spin(sem)) | |
447 | return sem; | |
448 | ||
449 | /* | |
450 | * Optimistic spinning failed, proceed to the slowpath | |
451 | * and block until we can acquire the sem. | |
452 | */ | |
453 | waiter.task = current; | |
023fe4f7 | 454 | waiter.type = RWSEM_WAITING_FOR_WRITE; |
1e78277c ML |
455 | |
456 | raw_spin_lock_irq(&sem->wait_lock); | |
4fc828e2 DB |
457 | |
458 | /* account for this before adding a new element to the list */ | |
1e78277c | 459 | if (list_empty(&sem->wait_list)) |
4fc828e2 DB |
460 | waiting = false; |
461 | ||
1e78277c ML |
462 | list_add_tail(&waiter.list, &sem->wait_list); |
463 | ||
464 | /* we're now waiting on the lock, but no longer actively locking */ | |
4fc828e2 | 465 | if (waiting) { |
4d3199e4 | 466 | count = READ_ONCE(sem->count); |
1e78277c | 467 | |
4fc828e2 | 468 | /* |
0cc3d011 AM |
469 | * If there were already threads queued before us and there are |
470 | * no active writers, the lock must be read owned; so we try to | |
471 | * wake any read locks that were queued ahead of us. | |
4fc828e2 DB |
472 | */ |
473 | if (count > RWSEM_WAITING_BIAS) | |
474 | sem = __rwsem_do_wake(sem, RWSEM_WAKE_READERS); | |
475 | ||
476 | } else | |
477 | count = rwsem_atomic_update(RWSEM_WAITING_BIAS, sem); | |
1e78277c | 478 | |
023fe4f7 | 479 | /* wait until we successfully acquire the lock */ |
4fc828e2 | 480 | set_current_state(TASK_UNINTERRUPTIBLE); |
1e78277c | 481 | while (true) { |
4fc828e2 DB |
482 | if (rwsem_try_write_lock(count, sem)) |
483 | break; | |
1e78277c | 484 | raw_spin_unlock_irq(&sem->wait_lock); |
a7d2c573 ML |
485 | |
486 | /* Block until there are no active lockers. */ | |
487 | do { | |
488 | schedule(); | |
4fc828e2 | 489 | set_current_state(TASK_UNINTERRUPTIBLE); |
9b0fc9c0 | 490 | } while ((count = sem->count) & RWSEM_ACTIVE_MASK); |
a7d2c573 | 491 | |
023fe4f7 | 492 | raw_spin_lock_irq(&sem->wait_lock); |
1e78277c | 493 | } |
4fc828e2 | 494 | __set_current_state(TASK_RUNNING); |
1e78277c | 495 | |
023fe4f7 ML |
496 | list_del(&waiter.list); |
497 | raw_spin_unlock_irq(&sem->wait_lock); | |
1e78277c ML |
498 | |
499 | return sem; | |
1da177e4 | 500 | } |
db0e716a | 501 | EXPORT_SYMBOL(rwsem_down_write_failed); |
1da177e4 LT |
502 | |
503 | /* | |
504 | * handle waking up a waiter on the semaphore | |
505 | * - up_read/up_write has decremented the active part of count if we come here | |
506 | */ | |
3ebae4f3 | 507 | __visible |
d1233754 | 508 | struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem) |
1da177e4 LT |
509 | { |
510 | unsigned long flags; | |
511 | ||
59aabfc7 WL |
512 | /* |
513 | * If a spinner is present, it is not necessary to do the wakeup. | |
514 | * Try to do wakeup only if the trylock succeeds to minimize | |
515 | * spinlock contention which may introduce too much delay in the | |
516 | * unlock operation. | |
517 | * | |
518 | * spinning writer up_write/up_read caller | |
519 | * --------------- ----------------------- | |
520 | * [S] osq_unlock() [L] osq | |
521 | * MB RMB | |
522 | * [RmW] rwsem_try_write_lock() [RmW] spin_trylock(wait_lock) | |
523 | * | |
524 | * Here, it is important to make sure that there won't be a missed | |
525 | * wakeup while the rwsem is free and the only spinning writer goes | |
526 | * to sleep without taking the rwsem. Even when the spinning writer | |
527 | * is just going to break out of the waiting loop, it will still do | |
528 | * a trylock in rwsem_down_write_failed() before sleeping. IOW, if | |
529 | * rwsem_has_spinner() is true, it will guarantee at least one | |
530 | * trylock attempt on the rwsem later on. | |
531 | */ | |
532 | if (rwsem_has_spinner(sem)) { | |
533 | /* | |
534 | * The smp_rmb() here is to make sure that the spinner | |
535 | * state is consulted before reading the wait_lock. | |
536 | */ | |
537 | smp_rmb(); | |
538 | if (!raw_spin_trylock_irqsave(&sem->wait_lock, flags)) | |
539 | return sem; | |
540 | goto locked; | |
541 | } | |
ddb6c9b5 | 542 | raw_spin_lock_irqsave(&sem->wait_lock, flags); |
59aabfc7 | 543 | locked: |
1da177e4 LT |
544 | |
545 | /* do nothing if list empty */ | |
546 | if (!list_empty(&sem->wait_list)) | |
70bdc6e0 | 547 | sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY); |
1da177e4 | 548 | |
ddb6c9b5 | 549 | raw_spin_unlock_irqrestore(&sem->wait_lock, flags); |
1da177e4 | 550 | |
1da177e4 LT |
551 | return sem; |
552 | } | |
db0e716a | 553 | EXPORT_SYMBOL(rwsem_wake); |
1da177e4 LT |
554 | |
555 | /* | |
556 | * downgrade a write lock into a read lock | |
557 | * - caller incremented waiting part of count and discovered it still negative | |
558 | * - just wake up any readers at the front of the queue | |
559 | */ | |
3ebae4f3 | 560 | __visible |
d1233754 | 561 | struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem) |
1da177e4 LT |
562 | { |
563 | unsigned long flags; | |
564 | ||
ddb6c9b5 | 565 | raw_spin_lock_irqsave(&sem->wait_lock, flags); |
1da177e4 LT |
566 | |
567 | /* do nothing if list empty */ | |
568 | if (!list_empty(&sem->wait_list)) | |
70bdc6e0 | 569 | sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED); |
1da177e4 | 570 | |
ddb6c9b5 | 571 | raw_spin_unlock_irqrestore(&sem->wait_lock, flags); |
1da177e4 | 572 | |
1da177e4 LT |
573 | return sem; |
574 | } | |
1da177e4 | 575 | EXPORT_SYMBOL(rwsem_downgrade_wake); |