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 #include <stdio.h>
  28 #include <pthread.h>
  29 #include <signal.h>
  30 #include <assert.h>
  31 #include <stdlib.h>
  32 #include <stdint.h>
  33 #include <string.h>
  34 #include <errno.h>
  35 #include <poll.h>
  36
  37 #include "urcu/map/urcu-qsbr.h"
  38
  39 #define BUILD_QSBR_LIB
  40 #include "urcu/static/urcu-qsbr.h"
  41 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  42 #include "urcu-qsbr.h"
  43
  44 void __attribute__((destructor)) rcu_exit(void);
  45
  46 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  47
  48 int32_t gp_futex;
  49
  50 /*
  51  * Global grace period counter.
  52  */
  53 unsigned long rcu_gp_ctr = RCU_GP_ONLINE;
  54
  55 /*
  56  * Active attempts to check for reader Q.S. before calling futex().
  57  */
  58 #define RCU_QS_ACTIVE_ATTEMPTS 100
  59
  60 /*
  61  * Written to only by each individual reader. Read by both the reader and the
  62  * writers.
  63  */
  64 struct rcu_reader __thread rcu_reader;
  65
  66 #ifdef DEBUG_YIELD
  67 unsigned int yield_active;
  68 unsigned int __thread rand_yield;
  69 #endif
  70
  71 static CDS_LIST_HEAD(registry);
  72
  73 static void mutex_lock(pthread_mutex_t *mutex)
  74 {
  75         int ret;
  76
  77 #ifndef DISTRUST_SIGNALS_EXTREME
  78         ret = pthread_mutex_lock(mutex);
  79         if (ret) {
  80                 perror("Error in pthread mutex lock");
  81                 exit(-1);
  82         }
  83 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
  84         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
  85                 if (ret != EBUSY && ret != EINTR) {
  86                         printf("ret = %d, errno = %d\n", ret, errno);
  87                         perror("Error in pthread mutex lock");
  88                         exit(-1);
  89                 }
  90                 poll(NULL,0,10);
  91         }
  92 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
  93 }
  94
  95 static void mutex_unlock(pthread_mutex_t *mutex)
  96 {
  97         int ret;
  98
  99         ret = pthread_mutex_unlock(mutex);
 100         if (ret) {
 101                 perror("Error in pthread mutex unlock");
 102                 exit(-1);
 103         }
 104 }
 105
 106 /*
 107  * synchronize_rcu() waiting. Single thread.
 108  */
 109 static void wait_gp(void)
 110 {
 111         /* Read reader_gp before read futex */
 112         cmm_smp_rmb();
 113         if (uatomic_read(&gp_futex) == -1)
 114                 futex_noasync(&gp_futex, FUTEX_WAIT, -1,
 115                       NULL, NULL, 0);
 116 }
 117
 118 static void update_counter_and_wait(void)
 119 {
 120         CDS_LIST_HEAD(qsreaders);
 121         int wait_loops = 0;
 122         struct rcu_reader *index, *tmp;
 123
 124 #if (CAA_BITS_PER_LONG < 64)
 125         /* Switch parity: 0 -> 1, 1 -> 0 */
 126         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 127 #else   /* !(CAA_BITS_PER_LONG < 64) */
 128         /* Increment current G.P. */
 129         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
 130 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 131
 132         /*
 133          * Must commit rcu_gp_ctr update to memory before waiting for
 134          * quiescent state. Failure to do so could result in the writer
 135          * waiting forever while new readers are always accessing data
 136          * (no progress). Enforce compiler-order of store to rcu_gp_ctr
 137          * before load rcu_reader ctr.
 138          */
 139         cmm_barrier();
 140
 141         /*
 142          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 143          * model easier to understand. It does not have a big performance impact
 144          * anyway, given this is the write-side.
 145          */
 146         cmm_smp_mb();
 147
 148         /*
 149          * Wait for each thread rcu_reader_qs_gp count to become 0.
 150          */
 151         for (;;) {
 152                 wait_loops++;
 153                 if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 154                         uatomic_set(&gp_futex, -1);
 155                         /*
 156                          * Write futex before write waiting (the other side
 157                          * reads them in the opposite order).
 158                          */
 159                         cmm_smp_wmb();
 160                         cds_list_for_each_entry(index, &registry, node) {
 161                                 _CMM_STORE_SHARED(index->waiting, 1);
 162                         }
 163                         /* Write futex before read reader_gp */
 164                         cmm_smp_mb();
 165                 }
 166                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 167                         if (!rcu_gp_ongoing(&index->ctr))
 168                                 cds_list_move(&index->node, &qsreaders);
 169                 }
 170
 171                 if (cds_list_empty(&registry)) {
 172                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 173                                 /* Read reader_gp before write futex */
 174                                 cmm_smp_mb();
 175                                 uatomic_set(&gp_futex, 0);
 176                         }
 177                         break;
 178                 } else {
 179                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 180                                 wait_gp();
 181                         } else {
 182 #ifndef HAS_INCOHERENT_CACHES
 183                                 caa_cpu_relax();
 184 #else /* #ifndef HAS_INCOHERENT_CACHES */
 185                                 cmm_smp_mb();
 186 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
 187                         }
 188                 }
 189         }
 190         /* put back the reader list in the registry */
 191         cds_list_splice(&qsreaders, &registry);
 192 }
 193
 194 /*
 195  * Using a two-subphases algorithm for architectures with smaller than 64-bit
 196  * long-size to ensure we do not encounter an overflow bug.
 197  */
 198
 199 #if (CAA_BITS_PER_LONG < 64)
 200 void synchronize_rcu(void)
 201 {
 202         unsigned long was_online;
 203
 204         was_online = rcu_reader.ctr;
 205
 206         /* All threads should read qparity before accessing data structure
 207          * where new ptr points to.
 208          */
 209         /* Write new ptr before changing the qparity */
 210         cmm_smp_mb();
 211
 212         /*
 213          * Mark the writer thread offline to make sure we don't wait for
 214          * our own quiescent state. This allows using synchronize_rcu()
 215          * in threads registered as readers.
 216          */
 217         if (was_online)
 218                 CMM_STORE_SHARED(rcu_reader.ctr, 0);
 219
 220         mutex_lock(&rcu_gp_lock);
 221
 222         if (cds_list_empty(&registry))
 223                 goto out;
 224
 225         /*
 226          * Wait for previous parity to be empty of readers.
 227          */
 228         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 229
 230         /*
 231          * Must finish waiting for quiescent state for parity 0 before
 232          * committing next rcu_gp_ctr update to memory. Failure to
 233          * do so could result in the writer waiting forever while new
 234          * readers are always accessing data (no progress).  Enforce
 235          * compiler-order of load rcu_reader ctr before store to
 236          * rcu_gp_ctr.
 237          */
 238         cmm_barrier();
 239
 240         /*
 241          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 242          * model easier to understand. It does not have a big performance impact
 243          * anyway, given this is the write-side.
 244          */
 245         cmm_smp_mb();
 246
 247         /*
 248          * Wait for previous parity to be empty of readers.
 249          */
 250         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 251 out:
 252         mutex_unlock(&rcu_gp_lock);
 253
 254         /*
 255          * Finish waiting for reader threads before letting the old ptr being
 256          * freed.
 257          */
 258         if (was_online)
 259                 _CMM_STORE_SHARED(rcu_reader.ctr,
 260                                   CMM_LOAD_SHARED(rcu_gp_ctr));
 261         cmm_smp_mb();
 262 }
 263 #else /* !(CAA_BITS_PER_LONG < 64) */
 264 void synchronize_rcu(void)
 265 {
 266         unsigned long was_online;
 267
 268         was_online = rcu_reader.ctr;
 269
 270         /*
 271          * Mark the writer thread offline to make sure we don't wait for
 272          * our own quiescent state. This allows using synchronize_rcu()
 273          * in threads registered as readers.
 274          */
 275         cmm_smp_mb();
 276         if (was_online)
 277                 CMM_STORE_SHARED(rcu_reader.ctr, 0);
 278
 279         mutex_lock(&rcu_gp_lock);
 280         if (cds_list_empty(&registry))
 281                 goto out;
 282         update_counter_and_wait();
 283 out:
 284         mutex_unlock(&rcu_gp_lock);
 285
 286         if (was_online)
 287                 _CMM_STORE_SHARED(rcu_reader.ctr,
 288                                   CMM_LOAD_SHARED(rcu_gp_ctr));
 289         cmm_smp_mb();
 290 }
 291 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 292
 293 /*
 294  * library wrappers to be used by non-LGPL compatible source code.
 295  */
 296
 297 void rcu_read_lock(void)
 298 {
 299         _rcu_read_lock();
 300 }
 301
 302 void rcu_read_unlock(void)
 303 {
 304         _rcu_read_unlock();
 305 }
 306
 307 void rcu_quiescent_state(void)
 308 {
 309         _rcu_quiescent_state();
 310 }
 311
 312 void rcu_thread_offline(void)
 313 {
 314         _rcu_thread_offline();
 315 }
 316
 317 void rcu_thread_online(void)
 318 {
 319         _rcu_thread_online();
 320 }
 321
 322 void rcu_register_thread(void)
 323 {
 324         rcu_reader.tid = pthread_self();
 325         assert(rcu_reader.ctr == 0);
 326
 327         mutex_lock(&rcu_gp_lock);
 328         cds_list_add(&rcu_reader.node, &registry);
 329         mutex_unlock(&rcu_gp_lock);
 330         _rcu_thread_online();
 331 }
 332
 333 void rcu_unregister_thread(void)
 334 {
 335         /*
 336          * We have to make the thread offline otherwise we end up dealocking
 337          * with a waiting writer.
 338          */
 339         _rcu_thread_offline();
 340         mutex_lock(&rcu_gp_lock);
 341         cds_list_del(&rcu_reader.node);
 342         mutex_unlock(&rcu_gp_lock);
 343 }
 344
 345 void rcu_exit(void)
 346 {
 347         assert(cds_list_empty(&registry));
 348 }
 349
 350 #include "urcu-call-rcu-impl.h"
 351 #include "urcu-defer-impl.h"