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