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/wfqueue.h"
  39 #include "urcu/map/urcu-qsbr.h"
  40 #define BUILD_QSBR_LIB
  41 #include "urcu/static/urcu-qsbr.h"
  42
  43 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  44 #undef _LGPL_SOURCE
  45 #include "urcu-qsbr.h"
  46 #define _LGPL_SOURCE
  47
  48 void __attribute__((destructor)) rcu_exit(void);
  49
  50 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  51
  52 int32_t gp_futex;
  53
  54 /*
  55  * Global grace period counter.
  56  */
  57 unsigned long rcu_gp_ctr = RCU_GP_ONLINE;
  58
  59 /*
  60  * Active attempts to check for reader Q.S. before calling futex().
  61  */
  62 #define RCU_QS_ACTIVE_ATTEMPTS 100
  63
  64 /*
  65  * Written to only by each individual reader. Read by both the reader and the
  66  * writers.
  67  */
  68 struct rcu_reader __thread rcu_reader;
  69
  70 #ifdef DEBUG_YIELD
  71 unsigned int yield_active;
  72 unsigned int __thread rand_yield;
  73 #endif
  74
  75 static CDS_LIST_HEAD(registry);
  76
  77 static void mutex_lock(pthread_mutex_t *mutex)
  78 {
  79         int ret;
  80
  81 #ifndef DISTRUST_SIGNALS_EXTREME
  82         ret = pthread_mutex_lock(mutex);
  83         if (ret) {
  84                 perror("Error in pthread mutex lock");
  85                 exit(-1);
  86         }
  87 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
  88         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
  89                 if (ret != EBUSY && ret != EINTR) {
  90                         printf("ret = %d, errno = %d\n", ret, errno);
  91                         perror("Error in pthread mutex lock");
  92                         exit(-1);
  93                 }
  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                 perror("Error in pthread mutex unlock");
 106                 exit(-1);
 107         }
 108 }
 109
 110 /*
 111  * synchronize_rcu() waiting. Single thread.
 112  */
 113 static void wait_gp(void)
 114 {
 115         /* Read reader_gp before read futex */
 116         cmm_smp_rmb();
 117         if (uatomic_read(&gp_futex) == -1)
 118                 futex_noasync(&gp_futex, FUTEX_WAIT, -1,
 119                       NULL, NULL, 0);
 120 }
 121
 122 static void update_counter_and_wait(void)
 123 {
 124         CDS_LIST_HEAD(qsreaders);
 125         int wait_loops = 0;
 126         struct rcu_reader *index, *tmp;
 127
 128 #if (CAA_BITS_PER_LONG < 64)
 129         /* Switch parity: 0 -> 1, 1 -> 0 */
 130         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 131 #else   /* !(CAA_BITS_PER_LONG < 64) */
 132         /* Increment current G.P. */
 133         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
 134 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 135
 136         /*
 137          * Must commit rcu_gp_ctr update to memory before waiting for
 138          * quiescent state. Failure to do so could result in the writer
 139          * waiting forever while new readers are always accessing data
 140          * (no progress). Enforce compiler-order of store to rcu_gp_ctr
 141          * before load rcu_reader ctr.
 142          */
 143         cmm_barrier();
 144
 145         /*
 146          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 147          * model easier to understand. It does not have a big performance impact
 148          * anyway, given this is the write-side.
 149          */
 150         cmm_smp_mb();
 151
 152         /*
 153          * Wait for each thread rcu_reader_qs_gp count to become 0.
 154          */
 155         for (;;) {
 156                 wait_loops++;
 157                 if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 158                         uatomic_set(&gp_futex, -1);
 159                         /*
 160                          * Write futex before write waiting (the other side
 161                          * reads them in the opposite order).
 162                          */
 163                         cmm_smp_wmb();
 164                         cds_list_for_each_entry(index, &registry, node) {
 165                                 _CMM_STORE_SHARED(index->waiting, 1);
 166                         }
 167                         /* Write futex before read reader_gp */
 168                         cmm_smp_mb();
 169                 }
 170                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 171                         if (!rcu_gp_ongoing(&index->ctr))
 172                                 cds_list_move(&index->node, &qsreaders);
 173                 }
 174
 175                 if (cds_list_empty(&registry)) {
 176                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 177                                 /* Read reader_gp before write futex */
 178                                 cmm_smp_mb();
 179                                 uatomic_set(&gp_futex, 0);
 180                         }
 181                         break;
 182                 } else {
 183                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 184                                 wait_gp();
 185                         } else {
 186 #ifndef HAS_INCOHERENT_CACHES
 187                                 caa_cpu_relax();
 188 #else /* #ifndef HAS_INCOHERENT_CACHES */
 189                                 cmm_smp_mb();
 190 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
 191                         }
 192                 }
 193         }
 194         /* put back the reader list in the registry */
 195         cds_list_splice(&qsreaders, &registry);
 196 }
 197
 198 /*
 199  * Using a two-subphases algorithm for architectures with smaller than 64-bit
 200  * long-size to ensure we do not encounter an overflow bug.
 201  */
 202
 203 #if (CAA_BITS_PER_LONG < 64)
 204 void synchronize_rcu(void)
 205 {
 206         unsigned long was_online;
 207
 208         was_online = rcu_reader.ctr;
 209
 210         /* All threads should read qparity before accessing data structure
 211          * where new ptr points to.  In the "then" case, rcu_thread_offline
 212          * includes a memory barrier.
 213          *
 214          * Mark the writer thread offline to make sure we don't wait for
 215          * our own quiescent state. This allows using synchronize_rcu()
 216          * in threads registered as readers.
 217          */
 218         if (was_online)
 219                 rcu_thread_offline();
 220         else
 221                 cmm_smp_mb();
 222
 223         mutex_lock(&rcu_gp_lock);
 224
 225         if (cds_list_empty(&registry))
 226                 goto out;
 227
 228         /*
 229          * Wait for previous parity to be empty of readers.
 230          */
 231         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 232
 233         /*
 234          * Must finish waiting for quiescent state for parity 0 before
 235          * committing next rcu_gp_ctr update to memory. Failure to
 236          * do so could result in the writer waiting forever while new
 237          * readers are always accessing data (no progress).  Enforce
 238          * compiler-order of load rcu_reader ctr before store to
 239          * rcu_gp_ctr.
 240          */
 241         cmm_barrier();
 242
 243         /*
 244          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 245          * model easier to understand. It does not have a big performance impact
 246          * anyway, given this is the write-side.
 247          */
 248         cmm_smp_mb();
 249
 250         /*
 251          * Wait for previous parity to be empty of readers.
 252          */
 253         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 254 out:
 255         mutex_unlock(&rcu_gp_lock);
 256
 257         /*
 258          * Finish waiting for reader threads before letting the old ptr being
 259          * freed.
 260          */
 261         if (was_online)
 262                 rcu_thread_online();
 263         else
 264                 cmm_smp_mb();
 265 }
 266 #else /* !(CAA_BITS_PER_LONG < 64) */
 267 void synchronize_rcu(void)
 268 {
 269         unsigned long was_online;
 270
 271         was_online = rcu_reader.ctr;
 272
 273         /*
 274          * Mark the writer thread offline to make sure we don't wait for
 275          * our own quiescent state. This allows using synchronize_rcu()
 276          * in threads registered as readers.
 277          */
 278         if (was_online)
 279                 rcu_thread_offline();
 280         else
 281                 cmm_smp_mb();
 282
 283         mutex_lock(&rcu_gp_lock);
 284         if (cds_list_empty(&registry))
 285                 goto out;
 286         update_counter_and_wait();
 287 out:
 288         mutex_unlock(&rcu_gp_lock);
 289
 290         if (was_online)
 291                 rcu_thread_online();
 292         else
 293                 cmm_smp_mb();
 294 }
 295 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 296
 297 /*
 298  * library wrappers to be used by non-LGPL compatible source code.
 299  */
 300
 301 void rcu_read_lock(void)
 302 {
 303         _rcu_read_lock();
 304 }
 305
 306 void rcu_read_unlock(void)
 307 {
 308         _rcu_read_unlock();
 309 }
 310
 311 void rcu_quiescent_state(void)
 312 {
 313         _rcu_quiescent_state();
 314 }
 315
 316 void rcu_thread_offline(void)
 317 {
 318         _rcu_thread_offline();
 319 }
 320
 321 void rcu_thread_online(void)
 322 {
 323         _rcu_thread_online();
 324 }
 325
 326 void rcu_register_thread(void)
 327 {
 328         rcu_reader.tid = pthread_self();
 329         assert(rcu_reader.ctr == 0);
 330
 331         mutex_lock(&rcu_gp_lock);
 332         cds_list_add(&rcu_reader.node, &registry);
 333         mutex_unlock(&rcu_gp_lock);
 334         _rcu_thread_online();
 335 }
 336
 337 void rcu_unregister_thread(void)
 338 {
 339         /*
 340          * We have to make the thread offline otherwise we end up dealocking
 341          * with a waiting writer.
 342          */
 343         _rcu_thread_offline();
 344         mutex_lock(&rcu_gp_lock);
 345         cds_list_del(&rcu_reader.node);
 346         mutex_unlock(&rcu_gp_lock);
 347 }
 348
 349 void rcu_exit(void)
 350 {
 351         assert(cds_list_empty(&registry));
 352 }
 353
 354 #include "urcu-call-rcu-impl.h"
 355 #include "urcu-defer-impl.h"