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 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <signal.h>
  29 #include <assert.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <errno.h>
  33 #include <poll.h>
  34
  35 #include "urcu-qsbr-map.h"
  36
  37 #define BUILD_QSBR_LIB
  38 #include "urcu-qsbr-static.h"
  39 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  40 #include "urcu-qsbr.h"
  41
  42 void __attribute__((destructor)) rcu_exit(void);
  43
  44 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  45
  46 int gp_futex;
  47
  48 /*
  49  * Global grace period counter.
  50  */
  51 unsigned long rcu_gp_ctr = RCU_GP_ONLINE;
  52
  53 /*
  54  * Written to only by each individual reader. Read by both the reader and the
  55  * writers.
  56  */
  57 struct rcu_reader __thread rcu_reader;
  58
  59 #ifdef DEBUG_YIELD
  60 unsigned int yield_active;
  61 unsigned int __thread rand_yield;
  62 #endif
  63
  64 static CDS_LIST_HEAD(registry);
  65
  66 static void mutex_lock(pthread_mutex_t *mutex)
  67 {
  68         int ret;
  69
  70 #ifndef DISTRUST_SIGNALS_EXTREME
  71         ret = pthread_mutex_lock(mutex);
  72         if (ret) {
  73                 perror("Error in pthread mutex lock");
  74                 exit(-1);
  75         }
  76 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
  77         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
  78                 if (ret != EBUSY && ret != EINTR) {
  79                         printf("ret = %d, errno = %d\n", ret, errno);
  80                         perror("Error in pthread mutex lock");
  81                         exit(-1);
  82                 }
  83                 poll(NULL,0,10);
  84         }
  85 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
  86 }
  87
  88 static void mutex_unlock(pthread_mutex_t *mutex)
  89 {
  90         int ret;
  91
  92         ret = pthread_mutex_unlock(mutex);
  93         if (ret) {
  94                 perror("Error in pthread mutex unlock");
  95                 exit(-1);
  96         }
  97 }
  98
  99 /*
 100  * synchronize_rcu() waiting. Single thread.
 101  */
 102 static void wait_gp(void)
 103 {
 104         /* Read reader_gp before read futex */
 105         cmm_smp_rmb();
 106         if (uatomic_read(&gp_futex) == -1)
 107                 futex_noasync(&gp_futex, FUTEX_WAIT, -1,
 108                       NULL, NULL, 0);
 109 }
 110
 111 static void update_counter_and_wait(void)
 112 {
 113         CDS_LIST_HEAD(qsreaders);
 114         int wait_loops = 0;
 115         struct rcu_reader *index, *tmp;
 116
 117 #if (CAA_BITS_PER_LONG < 64)
 118         /* Switch parity: 0 -> 1, 1 -> 0 */
 119         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 120 #else   /* !(CAA_BITS_PER_LONG < 64) */
 121         /* Increment current G.P. */
 122         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
 123 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 124
 125         /*
 126          * Must commit rcu_gp_ctr update to memory before waiting for
 127          * quiescent state. Failure to do so could result in the writer
 128          * waiting forever while new readers are always accessing data
 129          * (no progress). Enforce compiler-order of store to rcu_gp_ctr
 130          * before load rcu_reader ctr.
 131          */
 132         cmm_barrier();
 133
 134         /*
 135          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 136          * model easier to understand. It does not have a big performance impact
 137          * anyway, given this is the write-side.
 138          */
 139         cmm_smp_mb();
 140
 141         /*
 142          * Wait for each thread rcu_reader_qs_gp count to become 0.
 143          */
 144         for (;;) {
 145                 wait_loops++;
 146                 if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 147                         uatomic_dec(&gp_futex);
 148                         /* Write futex before read reader_gp */
 149                         cmm_smp_mb();
 150                 }
 151
 152                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 153                         if (!rcu_gp_ongoing(&index->ctr))
 154                                 cds_list_move(&index->node, &qsreaders);
 155                 }
 156
 157                 if (cds_list_empty(&registry)) {
 158                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 159                                 /* Read reader_gp before write futex */
 160                                 cmm_smp_mb();
 161                                 uatomic_set(&gp_futex, 0);
 162                         }
 163                         break;
 164                 } else {
 165                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 166                                 wait_gp();
 167                         } else {
 168 #ifndef HAS_INCOHERENT_CACHES
 169                                 caa_cpu_relax();
 170 #else /* #ifndef HAS_INCOHERENT_CACHES */
 171                                 cmm_smp_mb();
 172 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
 173                         }
 174                 }
 175         }
 176         /* put back the reader list in the registry */
 177         cds_list_splice(&qsreaders, &registry);
 178 }
 179
 180 /*
 181  * Using a two-subphases algorithm for architectures with smaller than 64-bit
 182  * long-size to ensure we do not encounter an overflow bug.
 183  */
 184
 185 #if (CAA_BITS_PER_LONG < 64)
 186 void synchronize_rcu(void)
 187 {
 188         unsigned long was_online;
 189
 190         was_online = rcu_reader.ctr;
 191
 192         /* All threads should read qparity before accessing data structure
 193          * where new ptr points to.
 194          */
 195         /* Write new ptr before changing the qparity */
 196         cmm_smp_mb();
 197
 198         /*
 199          * Mark the writer thread offline to make sure we don't wait for
 200          * our own quiescent state. This allows using synchronize_rcu()
 201          * in threads registered as readers.
 202          */
 203         if (was_online)
 204                 CMM_STORE_SHARED(rcu_reader.ctr, 0);
 205
 206         mutex_lock(&rcu_gp_lock);
 207
 208         if (cds_list_empty(&registry))
 209                 goto out;
 210
 211         /*
 212          * Wait for previous parity to be empty of readers.
 213          */
 214         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 215
 216         /*
 217          * Must finish waiting for quiescent state for parity 0 before
 218          * committing next rcu_gp_ctr update to memory. Failure to
 219          * do so could result in the writer waiting forever while new
 220          * readers are always accessing data (no progress).  Enforce
 221          * compiler-order of load rcu_reader ctr before store to
 222          * rcu_gp_ctr.
 223          */
 224         cmm_barrier();
 225
 226         /*
 227          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 228          * model easier to understand. It does not have a big performance impact
 229          * anyway, given this is the write-side.
 230          */
 231         cmm_smp_mb();
 232
 233         /*
 234          * Wait for previous parity to be empty of readers.
 235          */
 236         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 237 out:
 238         mutex_unlock(&rcu_gp_lock);
 239
 240         /*
 241          * Finish waiting for reader threads before letting the old ptr being
 242          * freed.
 243          */
 244         if (was_online)
 245                 _CMM_STORE_SHARED(rcu_reader.ctr,
 246                                   CMM_LOAD_SHARED(rcu_gp_ctr));
 247         cmm_smp_mb();
 248 }
 249 #else /* !(CAA_BITS_PER_LONG < 64) */
 250 void synchronize_rcu(void)
 251 {
 252         unsigned long was_online;
 253
 254         was_online = rcu_reader.ctr;
 255
 256         /*
 257          * Mark the writer thread offline to make sure we don't wait for
 258          * our own quiescent state. This allows using synchronize_rcu()
 259          * in threads registered as readers.
 260          */
 261         cmm_smp_mb();
 262         if (was_online)
 263                 CMM_STORE_SHARED(rcu_reader.ctr, 0);
 264
 265         mutex_lock(&rcu_gp_lock);
 266         if (cds_list_empty(&registry))
 267                 goto out;
 268         update_counter_and_wait();
 269 out:
 270         mutex_unlock(&rcu_gp_lock);
 271
 272         if (was_online)
 273                 _CMM_STORE_SHARED(rcu_reader.ctr,
 274                                   CMM_LOAD_SHARED(rcu_gp_ctr));
 275         cmm_smp_mb();
 276 }
 277 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 278
 279 /*
 280  * library wrappers to be used by non-LGPL compatible source code.
 281  */
 282
 283 void rcu_read_lock(void)
 284 {
 285         _rcu_read_lock();
 286 }
 287
 288 void rcu_read_unlock(void)
 289 {
 290         _rcu_read_unlock();
 291 }
 292
 293 void rcu_quiescent_state(void)
 294 {
 295         _rcu_quiescent_state();
 296 }
 297
 298 void rcu_thread_offline(void)
 299 {
 300         _rcu_thread_offline();
 301 }
 302
 303 void rcu_thread_online(void)
 304 {
 305         _rcu_thread_online();
 306 }
 307
 308 void rcu_register_thread(void)
 309 {
 310         rcu_reader.tid = pthread_self();
 311         assert(rcu_reader.ctr == 0);
 312
 313         mutex_lock(&rcu_gp_lock);
 314         cds_list_add(&rcu_reader.node, &registry);
 315         mutex_unlock(&rcu_gp_lock);
 316         _rcu_thread_online();
 317 }
 318
 319 void rcu_unregister_thread(void)
 320 {
 321         /*
 322          * We have to make the thread offline otherwise we end up dealocking
 323          * with a waiting writer.
 324          */
 325         _rcu_thread_offline();
 326         mutex_lock(&rcu_gp_lock);
 327         cds_list_del(&rcu_reader.node);
 328         mutex_unlock(&rcu_gp_lock);
 329 }
 330
 331 void rcu_exit(void)
 332 {
 333         assert(cds_list_empty(&registry));
 334 }
 335
 336 #include "urcu-call-rcu-impl.h"