urcu-qsbr.c

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