Commit | Line | Data |
---|---|---|
9f1621ca | 1 | /* |
7ac06cef | 2 | * urcu-qsbr.c |
9f1621ca | 3 | * |
7ac06cef | 4 | * Userspace RCU QSBR library |
9f1621ca | 5 | * |
6982d6d7 | 6 | * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
9f1621ca MD |
7 | * Copyright (c) 2009 Paul E. McKenney, IBM Corporation. |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | * | |
23 | * IBM's contributions to this file may be relicensed under LGPLv2 or later. | |
24 | */ | |
25 | ||
c1d2c60b | 26 | #define _GNU_SOURCE |
71c811bf | 27 | #define _LGPL_SOURCE |
9f1621ca MD |
28 | #include <stdio.h> |
29 | #include <pthread.h> | |
30 | #include <signal.h> | |
31 | #include <assert.h> | |
32 | #include <stdlib.h> | |
6d841bc2 | 33 | #include <stdint.h> |
9f1621ca MD |
34 | #include <string.h> |
35 | #include <errno.h> | |
36 | #include <poll.h> | |
37 | ||
d73fb81f | 38 | #include "urcu/wfcqueue.h" |
57760d44 | 39 | #include "urcu/map/urcu-qsbr.h" |
727f819d | 40 | #define BUILD_QSBR_LIB |
af7c2dbe | 41 | #include "urcu/static/urcu-qsbr.h" |
618b2595 | 42 | #include "urcu-pointer.h" |
bd252a04 | 43 | #include "urcu/tls-compat.h" |
71c811bf | 44 | |
4a6d7378 | 45 | #include "urcu-die.h" |
cba82d7b | 46 | #include "urcu-wait.h" |
4a6d7378 | 47 | |
9f1621ca | 48 | /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */ |
71c811bf | 49 | #undef _LGPL_SOURCE |
7ac06cef | 50 | #include "urcu-qsbr.h" |
71c811bf | 51 | #define _LGPL_SOURCE |
9f1621ca | 52 | |
f6d18c64 MD |
53 | void __attribute__((destructor)) rcu_exit(void); |
54 | ||
6abb4bd5 | 55 | static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER; |
9f1621ca | 56 | |
1de4df4b | 57 | int32_t rcu_gp_futex; |
bc6c15bb | 58 | |
9f1621ca MD |
59 | /* |
60 | * Global grace period counter. | |
61 | */ | |
02be5561 | 62 | unsigned long rcu_gp_ctr = RCU_GP_ONLINE; |
9f1621ca | 63 | |
408f6d92 PB |
64 | /* |
65 | * Active attempts to check for reader Q.S. before calling futex(). | |
66 | */ | |
67 | #define RCU_QS_ACTIVE_ATTEMPTS 100 | |
68 | ||
9f1621ca MD |
69 | /* |
70 | * Written to only by each individual reader. Read by both the reader and the | |
71 | * writers. | |
72 | */ | |
bd252a04 | 73 | DEFINE_URCU_TLS(struct rcu_reader, rcu_reader); |
9f1621ca MD |
74 | |
75 | #ifdef DEBUG_YIELD | |
1de4df4b MD |
76 | unsigned int rcu_yield_active; |
77 | DEFINE_URCU_TLS(unsigned int, rcu_rand_yield); | |
9f1621ca MD |
78 | #endif |
79 | ||
16aa9ee8 | 80 | static CDS_LIST_HEAD(registry); |
9f1621ca | 81 | |
6362f68f | 82 | /* |
bf6822a6 | 83 | * Queue keeping threads awaiting to wait for a grace period. Contains |
6362f68f MD |
84 | * struct gp_waiters_thread objects. |
85 | */ | |
bf6822a6 | 86 | static DEFINE_URCU_WAIT_QUEUE(gp_waiters); |
6362f68f | 87 | |
6abb4bd5 | 88 | static void mutex_lock(pthread_mutex_t *mutex) |
9f1621ca MD |
89 | { |
90 | int ret; | |
91 | ||
92 | #ifndef DISTRUST_SIGNALS_EXTREME | |
6abb4bd5 | 93 | ret = pthread_mutex_lock(mutex); |
4a6d7378 MD |
94 | if (ret) |
95 | urcu_die(ret); | |
9f1621ca | 96 | #else /* #ifndef DISTRUST_SIGNALS_EXTREME */ |
6abb4bd5 | 97 | while ((ret = pthread_mutex_trylock(mutex)) != 0) { |
4a6d7378 MD |
98 | if (ret != EBUSY && ret != EINTR) |
99 | urcu_die(ret); | |
9f1621ca MD |
100 | poll(NULL,0,10); |
101 | } | |
102 | #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */ | |
103 | } | |
104 | ||
6abb4bd5 | 105 | static void mutex_unlock(pthread_mutex_t *mutex) |
9f1621ca MD |
106 | { |
107 | int ret; | |
108 | ||
6abb4bd5 | 109 | ret = pthread_mutex_unlock(mutex); |
4a6d7378 MD |
110 | if (ret) |
111 | urcu_die(ret); | |
9f1621ca MD |
112 | } |
113 | ||
bc6c15bb MD |
114 | /* |
115 | * synchronize_rcu() waiting. Single thread. | |
116 | */ | |
4d703340 | 117 | static void wait_gp(void) |
bc6c15bb | 118 | { |
4d703340 | 119 | /* Read reader_gp before read futex */ |
5481ddb3 | 120 | cmm_smp_rmb(); |
1de4df4b MD |
121 | if (uatomic_read(&rcu_gp_futex) == -1) |
122 | futex_noasync(&rcu_gp_futex, FUTEX_WAIT, -1, | |
4d703340 | 123 | NULL, NULL, 0); |
bc6c15bb MD |
124 | } |
125 | ||
708d89f0 MD |
126 | static void wait_for_readers(struct cds_list_head *input_readers, |
127 | struct cds_list_head *cur_snap_readers, | |
128 | struct cds_list_head *qsreaders) | |
9f1621ca | 129 | { |
4d703340 | 130 | int wait_loops = 0; |
02be5561 | 131 | struct rcu_reader *index, *tmp; |
9f1621ca | 132 | |
9f1621ca | 133 | /* |
f6b42f9c MD |
134 | * Wait for each thread URCU_TLS(rcu_reader).ctr to either |
135 | * indicate quiescence (offline), or for them to observe the | |
136 | * current rcu_gp_ctr value. | |
9f1621ca | 137 | */ |
4d703340 MD |
138 | for (;;) { |
139 | wait_loops++; | |
83a2c421 | 140 | if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) { |
1de4df4b | 141 | uatomic_set(&rcu_gp_futex, -1); |
83a2c421 PB |
142 | /* |
143 | * Write futex before write waiting (the other side | |
144 | * reads them in the opposite order). | |
145 | */ | |
146 | cmm_smp_wmb(); | |
708d89f0 | 147 | cds_list_for_each_entry(index, input_readers, node) { |
83a2c421 PB |
148 | _CMM_STORE_SHARED(index->waiting, 1); |
149 | } | |
4d703340 | 150 | /* Write futex before read reader_gp */ |
5481ddb3 | 151 | cmm_smp_mb(); |
4d703340 | 152 | } |
708d89f0 MD |
153 | cds_list_for_each_entry_safe(index, tmp, input_readers, node) { |
154 | switch (rcu_reader_state(&index->ctr)) { | |
155 | case RCU_READER_ACTIVE_CURRENT: | |
156 | if (cur_snap_readers) { | |
157 | cds_list_move(&index->node, | |
158 | cur_snap_readers); | |
159 | break; | |
160 | } | |
161 | /* Fall-through */ | |
162 | case RCU_READER_INACTIVE: | |
163 | cds_list_move(&index->node, qsreaders); | |
164 | break; | |
165 | case RCU_READER_ACTIVE_OLD: | |
166 | /* | |
167 | * Old snapshot. Leaving node in | |
168 | * input_readers will make us busy-loop | |
169 | * until the snapshot becomes current or | |
170 | * the reader becomes inactive. | |
171 | */ | |
172 | break; | |
173 | } | |
4d703340 | 174 | } |
bc6c15bb | 175 | |
708d89f0 | 176 | if (cds_list_empty(input_readers)) { |
83a2c421 | 177 | if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) { |
4d703340 | 178 | /* Read reader_gp before write futex */ |
5481ddb3 | 179 | cmm_smp_mb(); |
1de4df4b | 180 | uatomic_set(&rcu_gp_futex, 0); |
4d703340 MD |
181 | } |
182 | break; | |
183 | } else { | |
83a2c421 | 184 | if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) { |
4d703340 | 185 | wait_gp(); |
bc6c15bb | 186 | } else { |
9f1621ca | 187 | #ifndef HAS_INCOHERENT_CACHES |
06f22bdb | 188 | caa_cpu_relax(); |
9f1621ca | 189 | #else /* #ifndef HAS_INCOHERENT_CACHES */ |
5481ddb3 | 190 | cmm_smp_mb(); |
9f1621ca | 191 | #endif /* #else #ifndef HAS_INCOHERENT_CACHES */ |
bc6c15bb MD |
192 | } |
193 | } | |
9f1621ca MD |
194 | } |
195 | } | |
196 | ||
47d2f29e MD |
197 | /* |
198 | * Using a two-subphases algorithm for architectures with smaller than 64-bit | |
199 | * long-size to ensure we do not encounter an overflow bug. | |
200 | */ | |
201 | ||
b39e1761 | 202 | #if (CAA_BITS_PER_LONG < 64) |
47d2f29e MD |
203 | void synchronize_rcu(void) |
204 | { | |
708d89f0 MD |
205 | CDS_LIST_HEAD(cur_snap_readers); |
206 | CDS_LIST_HEAD(qsreaders); | |
bc49c323 | 207 | unsigned long was_online; |
bf6822a6 MD |
208 | DEFINE_URCU_WAIT_NODE(wait, URCU_WAIT_WAITING); |
209 | struct urcu_waiters waiters; | |
bc49c323 | 210 | |
bd252a04 | 211 | was_online = URCU_TLS(rcu_reader).ctr; |
bc49c323 | 212 | |
47d2f29e | 213 | /* All threads should read qparity before accessing data structure |
27b940e7 PB |
214 | * where new ptr points to. In the "then" case, rcu_thread_offline |
215 | * includes a memory barrier. | |
216 | * | |
bc49c323 | 217 | * Mark the writer thread offline to make sure we don't wait for |
5e77fc1f PM |
218 | * our own quiescent state. This allows using synchronize_rcu() |
219 | * in threads registered as readers. | |
bc49c323 | 220 | */ |
27b940e7 PB |
221 | if (was_online) |
222 | rcu_thread_offline(); | |
223 | else | |
224 | cmm_smp_mb(); | |
bc49c323 | 225 | |
6362f68f | 226 | /* |
bf6822a6 | 227 | * Add ourself to gp_waiters queue of threads awaiting to wait |
6362f68f | 228 | * for a grace period. Proceed to perform the grace period only |
bf6822a6 | 229 | * if we are the first thread added into the queue. |
6362f68f | 230 | */ |
bf6822a6 MD |
231 | if (urcu_wait_add(&gp_waiters, &wait) != 0) { |
232 | /* Not first in queue: will be awakened by another thread. */ | |
233 | urcu_adaptative_busy_wait(&wait); | |
6362f68f MD |
234 | goto gp_end; |
235 | } | |
bf6822a6 MD |
236 | /* We won't need to wake ourself up */ |
237 | urcu_wait_set_state(&wait, URCU_WAIT_RUNNING); | |
6362f68f | 238 | |
6abb4bd5 | 239 | mutex_lock(&rcu_gp_lock); |
47d2f29e | 240 | |
6362f68f | 241 | /* |
bf6822a6 | 242 | * Move all waiters into our local queue. |
6362f68f | 243 | */ |
bf6822a6 | 244 | urcu_move_waiters(&waiters, &gp_waiters); |
6362f68f | 245 | |
16aa9ee8 | 246 | if (cds_list_empty(®istry)) |
2dfb8b5e | 247 | goto out; |
47d2f29e MD |
248 | |
249 | /* | |
f6b42f9c | 250 | * Wait for readers to observe original parity or be quiescent. |
47d2f29e | 251 | */ |
708d89f0 | 252 | wait_for_readers(®istry, &cur_snap_readers, &qsreaders); |
47d2f29e MD |
253 | |
254 | /* | |
f6b42f9c MD |
255 | * Must finish waiting for quiescent state for original parity |
256 | * before committing next rcu_gp_ctr update to memory. Failure | |
257 | * to do so could result in the writer waiting forever while new | |
5e77fc1f | 258 | * readers are always accessing data (no progress). Enforce |
f6b42f9c MD |
259 | * compiler-order of load URCU_TLS(rcu_reader).ctr before store |
260 | * to rcu_gp_ctr. | |
47d2f29e | 261 | */ |
5481ddb3 | 262 | cmm_barrier(); |
47d2f29e | 263 | |
47d2f29e | 264 | /* |
5481ddb3 | 265 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the |
2dfb8b5e MD |
266 | * model easier to understand. It does not have a big performance impact |
267 | * anyway, given this is the write-side. | |
47d2f29e | 268 | */ |
5481ddb3 | 269 | cmm_smp_mb(); |
47d2f29e | 270 | |
f6b42f9c MD |
271 | /* Switch parity: 0 -> 1, 1 -> 0 */ |
272 | CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR); | |
273 | ||
47d2f29e | 274 | /* |
f6b42f9c MD |
275 | * Must commit rcu_gp_ctr update to memory before waiting for |
276 | * quiescent state. Failure to do so could result in the writer | |
277 | * waiting forever while new readers are always accessing data | |
278 | * (no progress). Enforce compiler-order of store to rcu_gp_ctr | |
279 | * before load URCU_TLS(rcu_reader).ctr. | |
47d2f29e | 280 | */ |
f6b42f9c MD |
281 | cmm_barrier(); |
282 | ||
283 | /* | |
284 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the | |
285 | * model easier to understand. It does not have a big performance impact | |
286 | * anyway, given this is the write-side. | |
287 | */ | |
288 | cmm_smp_mb(); | |
289 | ||
290 | /* | |
291 | * Wait for readers to observe new parity or be quiescent. | |
292 | */ | |
708d89f0 MD |
293 | wait_for_readers(&cur_snap_readers, NULL, &qsreaders); |
294 | ||
295 | /* | |
296 | * Put quiescent reader list back into registry. | |
297 | */ | |
298 | cds_list_splice(&qsreaders, ®istry); | |
2dfb8b5e | 299 | out: |
6abb4bd5 | 300 | mutex_unlock(&rcu_gp_lock); |
bf6822a6 | 301 | urcu_wake_all_waiters(&waiters); |
6362f68f | 302 | gp_end: |
bc49c323 MD |
303 | /* |
304 | * Finish waiting for reader threads before letting the old ptr being | |
47d2f29e MD |
305 | * freed. |
306 | */ | |
bc49c323 | 307 | if (was_online) |
27b940e7 PB |
308 | rcu_thread_online(); |
309 | else | |
310 | cmm_smp_mb(); | |
47d2f29e | 311 | } |
b39e1761 | 312 | #else /* !(CAA_BITS_PER_LONG < 64) */ |
9f1621ca MD |
313 | void synchronize_rcu(void) |
314 | { | |
708d89f0 | 315 | CDS_LIST_HEAD(qsreaders); |
f0f7dbdd | 316 | unsigned long was_online; |
bf6822a6 MD |
317 | DEFINE_URCU_WAIT_NODE(wait, URCU_WAIT_WAITING); |
318 | struct urcu_waiters waiters; | |
ff2f67a0 | 319 | |
bd252a04 | 320 | was_online = URCU_TLS(rcu_reader).ctr; |
ff2f67a0 MD |
321 | |
322 | /* | |
323 | * Mark the writer thread offline to make sure we don't wait for | |
5e77fc1f PM |
324 | * our own quiescent state. This allows using synchronize_rcu() |
325 | * in threads registered as readers. | |
ff2f67a0 | 326 | */ |
27b940e7 PB |
327 | if (was_online) |
328 | rcu_thread_offline(); | |
329 | else | |
330 | cmm_smp_mb(); | |
ff2f67a0 | 331 | |
6362f68f | 332 | /* |
bf6822a6 | 333 | * Add ourself to gp_waiters queue of threads awaiting to wait |
6362f68f | 334 | * for a grace period. Proceed to perform the grace period only |
bf6822a6 | 335 | * if we are the first thread added into the queue. |
6362f68f | 336 | */ |
bf6822a6 MD |
337 | if (urcu_wait_add(&gp_waiters, &wait) != 0) { |
338 | /* Not first in queue: will be awakened by another thread. */ | |
339 | urcu_adaptative_busy_wait(&wait); | |
6362f68f MD |
340 | goto gp_end; |
341 | } | |
bf6822a6 MD |
342 | /* We won't need to wake ourself up */ |
343 | urcu_wait_set_state(&wait, URCU_WAIT_RUNNING); | |
6362f68f | 344 | |
6abb4bd5 | 345 | mutex_lock(&rcu_gp_lock); |
6362f68f MD |
346 | |
347 | /* | |
bf6822a6 | 348 | * Move all waiters into our local queue. |
6362f68f | 349 | */ |
bf6822a6 | 350 | urcu_move_waiters(&waiters, &gp_waiters); |
6362f68f | 351 | |
16aa9ee8 | 352 | if (cds_list_empty(®istry)) |
2dfb8b5e | 353 | goto out; |
f6b42f9c MD |
354 | |
355 | /* Increment current G.P. */ | |
356 | CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR); | |
357 | ||
358 | /* | |
359 | * Must commit rcu_gp_ctr update to memory before waiting for | |
360 | * quiescent state. Failure to do so could result in the writer | |
361 | * waiting forever while new readers are always accessing data | |
362 | * (no progress). Enforce compiler-order of store to rcu_gp_ctr | |
363 | * before load URCU_TLS(rcu_reader).ctr. | |
364 | */ | |
365 | cmm_barrier(); | |
366 | ||
367 | /* | |
368 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the | |
369 | * model easier to understand. It does not have a big performance impact | |
370 | * anyway, given this is the write-side. | |
371 | */ | |
372 | cmm_smp_mb(); | |
373 | ||
374 | /* | |
375 | * Wait for readers to observe new count of be quiescent. | |
376 | */ | |
708d89f0 MD |
377 | wait_for_readers(®istry, NULL, &qsreaders); |
378 | ||
379 | /* | |
380 | * Put quiescent reader list back into registry. | |
381 | */ | |
382 | cds_list_splice(&qsreaders, ®istry); | |
2dfb8b5e | 383 | out: |
6abb4bd5 | 384 | mutex_unlock(&rcu_gp_lock); |
bf6822a6 | 385 | urcu_wake_all_waiters(&waiters); |
6362f68f | 386 | gp_end: |
ff2f67a0 | 387 | if (was_online) |
27b940e7 PB |
388 | rcu_thread_online(); |
389 | else | |
390 | cmm_smp_mb(); | |
9f1621ca | 391 | } |
b39e1761 | 392 | #endif /* !(CAA_BITS_PER_LONG < 64) */ |
9f1621ca MD |
393 | |
394 | /* | |
395 | * library wrappers to be used by non-LGPL compatible source code. | |
396 | */ | |
397 | ||
398 | void rcu_read_lock(void) | |
399 | { | |
400 | _rcu_read_lock(); | |
401 | } | |
402 | ||
403 | void rcu_read_unlock(void) | |
404 | { | |
405 | _rcu_read_unlock(); | |
406 | } | |
407 | ||
7ac06cef MD |
408 | void rcu_quiescent_state(void) |
409 | { | |
410 | _rcu_quiescent_state(); | |
411 | } | |
412 | ||
413 | void rcu_thread_offline(void) | |
414 | { | |
415 | _rcu_thread_offline(); | |
416 | } | |
417 | ||
418 | void rcu_thread_online(void) | |
419 | { | |
420 | _rcu_thread_online(); | |
421 | } | |
422 | ||
9f1621ca MD |
423 | void rcu_register_thread(void) |
424 | { | |
bd252a04 MD |
425 | URCU_TLS(rcu_reader).tid = pthread_self(); |
426 | assert(URCU_TLS(rcu_reader).ctr == 0); | |
4f8e3380 | 427 | |
6abb4bd5 | 428 | mutex_lock(&rcu_gp_lock); |
bd252a04 | 429 | cds_list_add(&URCU_TLS(rcu_reader).node, ®istry); |
6abb4bd5 | 430 | mutex_unlock(&rcu_gp_lock); |
5f373c84 | 431 | _rcu_thread_online(); |
9f1621ca MD |
432 | } |
433 | ||
434 | void rcu_unregister_thread(void) | |
435 | { | |
76f3022f MD |
436 | /* |
437 | * We have to make the thread offline otherwise we end up dealocking | |
438 | * with a waiting writer. | |
439 | */ | |
440 | _rcu_thread_offline(); | |
6abb4bd5 | 441 | mutex_lock(&rcu_gp_lock); |
bd252a04 | 442 | cds_list_del(&URCU_TLS(rcu_reader).node); |
6abb4bd5 | 443 | mutex_unlock(&rcu_gp_lock); |
9f1621ca | 444 | } |
f6d18c64 MD |
445 | |
446 | void rcu_exit(void) | |
447 | { | |
01cadde4 MD |
448 | /* |
449 | * Assertion disabled because call_rcu threads are now rcu | |
450 | * readers, and left running at exit. | |
451 | * assert(cds_list_empty(®istry)); | |
452 | */ | |
f6d18c64 | 453 | } |
5e77fc1f | 454 | |
5e6b23a6 | 455 | DEFINE_RCU_FLAVOR(rcu_flavor); |
541d828d | 456 | |
5e77fc1f | 457 | #include "urcu-call-rcu-impl.h" |
0376e7b2 | 458 | #include "urcu-defer-impl.h" |