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 #include "urcu-pointer.h"
  43 #include "urcu/tls-compat.h"
  44
  45 #include "urcu-die.h"
  46
  47 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  48 #undef _LGPL_SOURCE
  49 #include "urcu-qsbr.h"
  50 #define _LGPL_SOURCE
  51
  52 void __attribute__((destructor)) rcu_exit(void);
  53
  54 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  55
  56 int32_t gp_futex;
  57
  58 /*
  59  * Global grace period counter.
  60  */
  61 unsigned long rcu_gp_ctr = RCU_GP_ONLINE;
  62
  63 /*
  64  * Active attempts to check for reader Q.S. before calling futex().
  65  */
  66 #define RCU_QS_ACTIVE_ATTEMPTS 100
  67
  68 /*
  69  * Written to only by each individual reader. Read by both the reader and the
  70  * writers.
  71  */
  72 __DEFINE_URCU_TLS_GLOBAL(struct rcu_reader, rcu_reader);
  73
  74 #ifdef DEBUG_YIELD
  75 unsigned int yield_active;
  76 __DEFINE_URCU_TLS_GLOBAL(unsigned int, rand_yield);
  77 #endif
  78
  79 static CDS_LIST_HEAD(registry);
  80
  81 static void mutex_lock(pthread_mutex_t *mutex)
  82 {
  83         int ret;
  84
  85 #ifndef DISTRUST_SIGNALS_EXTREME
  86         ret = pthread_mutex_lock(mutex);
  87         if (ret)
  88                 urcu_die(ret);
  89 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
  90         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
  91                 if (ret != EBUSY && ret != EINTR)
  92                         urcu_die(ret);
  93                 poll(NULL,0,10);
  94         }
  95 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
  96 }
  97
  98 static void mutex_unlock(pthread_mutex_t *mutex)
  99 {
 100         int ret;
 101
 102         ret = pthread_mutex_unlock(mutex);
 103         if (ret)
 104                 urcu_die(ret);
 105 }
 106
 107 /*
 108  * synchronize_rcu() waiting. Single thread.
 109  */
 110 static void wait_gp(void)
 111 {
 112         /* Read reader_gp before read futex */
 113         cmm_smp_rmb();
 114         if (uatomic_read(&gp_futex) == -1)
 115                 futex_noasync(&gp_futex, FUTEX_WAIT, -1,
 116                       NULL, NULL, 0);
 117 }
 118
 119 static void update_counter_and_wait(void)
 120 {
 121         CDS_LIST_HEAD(qsreaders);
 122         unsigned int wait_loops = 0;
 123         struct rcu_reader *index, *tmp;
 124
 125 #if (CAA_BITS_PER_LONG < 64)
 126         /* Switch parity: 0 -> 1, 1 -> 0 */
 127         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 128 #else   /* !(CAA_BITS_PER_LONG < 64) */
 129         /* Increment current G.P. */
 130         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
 131 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 132
 133         /*
 134          * Must commit rcu_gp_ctr update to memory before waiting for
 135          * quiescent state. Failure to do so could result in the writer
 136          * waiting forever while new readers are always accessing data
 137          * (no progress). Enforce compiler-order of store to rcu_gp_ctr
 138          * before load URCU_TLS(rcu_reader).ctr.
 139          */
 140         cmm_barrier();
 141
 142         /*
 143          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 144          * model easier to understand. It does not have a big performance impact
 145          * anyway, given this is the write-side.
 146          */
 147         cmm_smp_mb();
 148
 149         /*
 150          * Wait for each thread rcu_reader_qs_gp count to become 0.
 151          */
 152         for (;;) {
 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                 } else {
 166                         wait_loops++;
 167                 }
 168                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 169                         if (!rcu_gp_ongoing(&index->ctr))
 170                                 cds_list_move(&index->node, &qsreaders);
 171                 }
 172
 173                 if (cds_list_empty(&registry)) {
 174                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 175                                 /* Read reader_gp before write futex */
 176                                 cmm_smp_mb();
 177                                 uatomic_set(&gp_futex, 0);
 178                         }
 179                         break;
 180                 } else {
 181                         if (wait_loops >= RCU_QS_ACTIVE_ATTEMPTS) {
 182                                 wait_gp();
 183                         } else {
 184 #ifndef HAS_INCOHERENT_CACHES
 185                                 caa_cpu_relax();
 186 #else /* #ifndef HAS_INCOHERENT_CACHES */
 187                                 cmm_smp_mb();
 188 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
 189                         }
 190                 }
 191         }
 192         /* put back the reader list in the registry */
 193         cds_list_splice(&qsreaders, &registry);
 194 }
 195
 196 /*
 197  * Using a two-subphases algorithm for architectures with smaller than 64-bit
 198  * long-size to ensure we do not encounter an overflow bug.
 199  */
 200
 201 #if (CAA_BITS_PER_LONG < 64)
 202 void synchronize_rcu(void)
 203 {
 204         unsigned long was_online;
 205
 206         was_online = URCU_TLS(rcu_reader).ctr;
 207
 208         /* All threads should read qparity before accessing data structure
 209          * where new ptr points to.  In the "then" case, rcu_thread_offline
 210          * includes a memory barrier.
 211          *
 212          * Mark the writer thread offline to make sure we don't wait for
 213          * our own quiescent state. This allows using synchronize_rcu()
 214          * in threads registered as readers.
 215          */
 216         if (was_online)
 217                 rcu_thread_offline();
 218         else
 219                 cmm_smp_mb();
 220
 221         mutex_lock(&rcu_gp_lock);
 222
 223         if (cds_list_empty(&registry))
 224                 goto out;
 225
 226         /*
 227          * Wait for previous parity to be empty of readers.
 228          */
 229         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 230
 231         /*
 232          * Must finish waiting for quiescent state for parity 0 before
 233          * committing next rcu_gp_ctr update to memory. Failure to
 234          * do so could result in the writer waiting forever while new
 235          * readers are always accessing data (no progress).  Enforce
 236          * compiler-order of load URCU_TLS(rcu_reader).ctr before store to
 237          * rcu_gp_ctr.
 238          */
 239         cmm_barrier();
 240
 241         /*
 242          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 243          * model easier to understand. It does not have a big performance impact
 244          * anyway, given this is the write-side.
 245          */
 246         cmm_smp_mb();
 247
 248         /*
 249          * Wait for previous parity to be empty of readers.
 250          */
 251         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 252 out:
 253         mutex_unlock(&rcu_gp_lock);
 254
 255         /*
 256          * Finish waiting for reader threads before letting the old ptr being
 257          * freed.
 258          */
 259         if (was_online)
 260                 rcu_thread_online();
 261         else
 262                 cmm_smp_mb();
 263 }
 264 #else /* !(CAA_BITS_PER_LONG < 64) */
 265 void synchronize_rcu(void)
 266 {
 267         unsigned long was_online;
 268
 269         was_online = URCU_TLS(rcu_reader).ctr;
 270
 271         /*
 272          * Mark the writer thread offline to make sure we don't wait for
 273          * our own quiescent state. This allows using synchronize_rcu()
 274          * in threads registered as readers.
 275          */
 276         if (was_online)
 277                 rcu_thread_offline();
 278         else
 279                 cmm_smp_mb();
 280
 281         mutex_lock(&rcu_gp_lock);
 282         if (cds_list_empty(&registry))
 283                 goto out;
 284         update_counter_and_wait();
 285 out:
 286         mutex_unlock(&rcu_gp_lock);
 287
 288         if (was_online)
 289                 rcu_thread_online();
 290         else
 291                 cmm_smp_mb();
 292 }
 293 #endif  /* !(CAA_BITS_PER_LONG < 64) */
 294
 295 /*
 296  * library wrappers to be used by non-LGPL compatible source code.
 297  */
 298
 299 void rcu_read_lock(void)
 300 {
 301         _rcu_read_lock();
 302 }
 303
 304 void rcu_read_unlock(void)
 305 {
 306         _rcu_read_unlock();
 307 }
 308
 309 void rcu_quiescent_state(void)
 310 {
 311         _rcu_quiescent_state();
 312 }
 313
 314 void rcu_thread_offline(void)
 315 {
 316         _rcu_thread_offline();
 317 }
 318
 319 void rcu_thread_online(void)
 320 {
 321         _rcu_thread_online();
 322 }
 323
 324 void rcu_register_thread(void)
 325 {
 326         URCU_TLS(rcu_reader).tid = pthread_self();
 327         assert(URCU_TLS(rcu_reader).ctr == 0);
 328
 329         mutex_lock(&rcu_gp_lock);
 330         cds_list_add(&URCU_TLS(rcu_reader).node, &registry);
 331         mutex_unlock(&rcu_gp_lock);
 332         _rcu_thread_online();
 333 }
 334
 335 void rcu_unregister_thread(void)
 336 {
 337         /*
 338          * We have to make the thread offline otherwise we end up dealocking
 339          * with a waiting writer.
 340          */
 341         _rcu_thread_offline();
 342         mutex_lock(&rcu_gp_lock);
 343         cds_list_del(&URCU_TLS(rcu_reader).node);
 344         mutex_unlock(&rcu_gp_lock);
 345 }
 346
 347 void rcu_exit(void)
 348 {
 349         /*
 350          * Assertion disabled because call_rcu threads are now rcu
 351          * readers, and left running at exit.
 352          * assert(cds_list_empty(&registry));
 353          */
 354 }
 355
 356 DEFINE_RCU_FLAVOR(rcu_flavor);
 357
 358 #include "urcu-call-rcu-impl.h"
 359 #include "urcu-defer-impl.h"