urcu.c

   1 /*
   2  * urcu.c
   3  *
   4  * Userspace RCU 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 _BSD_SOURCE
  27 #define _GNU_SOURCE
  28 #define _LGPL_SOURCE
  29 #include <stdio.h>
  30 #include <pthread.h>
  31 #include <signal.h>
  32 #include <assert.h>
  33 #include <stdlib.h>
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <errno.h>
  37 #include <poll.h>
  38
  39 #include "urcu/wfqueue.h"
  40 #include "urcu/map/urcu.h"
  41 #include "urcu/static/urcu.h"
  42 #include "urcu-pointer.h"
  43 #include "urcu/tls-compat.h"
  44
  45 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  46 #undef _LGPL_SOURCE
  47 #include "urcu.h"
  48 #define _LGPL_SOURCE
  49
  50 /*
  51  * If a reader is really non-cooperative and refuses to commit its
  52  * rcu_active_readers count to memory (there is no barrier in the reader
  53  * per-se), kick it after a few loops waiting for it.
  54  */
  55 #define KICK_READER_LOOPS 10000
  56
  57 /*
  58  * Active attempts to check for reader Q.S. before calling futex().
  59  */
  60 #define RCU_QS_ACTIVE_ATTEMPTS 100
  61
  62 #ifdef RCU_MEMBARRIER
  63 static int init_done;
  64 int has_sys_membarrier;
  65
  66 void __attribute__((constructor)) rcu_init(void);
  67 #endif
  68
  69 #ifdef RCU_MB
  70 void rcu_init(void)
  71 {
  72 }
  73 #endif
  74
  75 #ifdef RCU_SIGNAL
  76 static int init_done;
  77
  78 void __attribute__((constructor)) rcu_init(void);
  79 void __attribute__((destructor)) rcu_exit(void);
  80 #endif
  81
  82 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  83
  84 int32_t gp_futex;
  85
  86 /*
  87  * Global grace period counter.
  88  * Contains the current RCU_GP_CTR_PHASE.
  89  * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
  90  * Written to only by writer with mutex taken. Read by both writer and readers.
  91  */
  92 unsigned long rcu_gp_ctr = RCU_GP_COUNT;
  93
  94 /*
  95  * Written to only by each individual reader. Read by both the reader and the
  96  * writers.
  97  */
  98 DEFINE_URCU_TLS(struct rcu_reader, rcu_reader);
  99
 100 #ifdef DEBUG_YIELD
 101 unsigned int yield_active;
 102 DEFINE_URCU_TLS(unsigned int, rand_yield);
 103 #endif
 104
 105 static CDS_LIST_HEAD(registry);
 106
 107 static void mutex_lock(pthread_mutex_t *mutex)
 108 {
 109         int ret;
 110
 111 #ifndef DISTRUST_SIGNALS_EXTREME
 112         ret = pthread_mutex_lock(mutex);
 113         if (ret) {
 114                 perror("Error in pthread mutex lock");
 115                 exit(-1);
 116         }
 117 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
 118         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
 119                 if (ret != EBUSY && ret != EINTR) {
 120                         printf("ret = %d, errno = %d\n", ret, errno);
 121                         perror("Error in pthread mutex lock");
 122                         exit(-1);
 123                 }
 124                 if (CMM_LOAD_SHARED(URCU_TLS(rcu_reader).need_mb)) {
 125                         cmm_smp_mb();
 126                         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).need_mb, 0);
 127                         cmm_smp_mb();
 128                 }
 129                 poll(NULL,0,10);
 130         }
 131 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
 132 }
 133
 134 static void mutex_unlock(pthread_mutex_t *mutex)
 135 {
 136         int ret;
 137
 138         ret = pthread_mutex_unlock(mutex);
 139         if (ret) {
 140                 perror("Error in pthread mutex unlock");
 141                 exit(-1);
 142         }
 143 }
 144
 145 #ifdef RCU_MEMBARRIER
 146 static void smp_mb_master(int group)
 147 {
 148         if (caa_likely(has_sys_membarrier))
 149                 membarrier(MEMBARRIER_EXPEDITED);
 150         else
 151                 cmm_smp_mb();
 152 }
 153 #endif
 154
 155 #ifdef RCU_MB
 156 static void smp_mb_master(int group)
 157 {
 158         cmm_smp_mb();
 159 }
 160 #endif
 161
 162 #ifdef RCU_SIGNAL
 163 static void force_mb_all_readers(void)
 164 {
 165         struct rcu_reader *index;
 166
 167         /*
 168          * Ask for each threads to execute a cmm_smp_mb() so we can consider the
 169          * compiler barriers around rcu read lock as real memory barriers.
 170          */
 171         if (cds_list_empty(&registry))
 172                 return;
 173         /*
 174          * pthread_kill has a cmm_smp_mb(). But beware, we assume it performs
 175          * a cache flush on architectures with non-coherent cache. Let's play
 176          * safe and don't assume anything : we use cmm_smp_mc() to make sure the
 177          * cache flush is enforced.
 178          */
 179         cds_list_for_each_entry(index, &registry, node) {
 180                 CMM_STORE_SHARED(index->need_mb, 1);
 181                 pthread_kill(index->tid, SIGRCU);
 182         }
 183         /*
 184          * Wait for sighandler (and thus mb()) to execute on every thread.
 185          *
 186          * Note that the pthread_kill() will never be executed on systems
 187          * that correctly deliver signals in a timely manner.  However, it
 188          * is not uncommon for kernels to have bugs that can result in
 189          * lost or unduly delayed signals.
 190          *
 191          * If you are seeing the below pthread_kill() executing much at
 192          * all, we suggest testing the underlying kernel and filing the
 193          * relevant bug report.  For Linux kernels, we recommend getting
 194          * the Linux Test Project (LTP).
 195          */
 196         cds_list_for_each_entry(index, &registry, node) {
 197                 while (CMM_LOAD_SHARED(index->need_mb)) {
 198                         pthread_kill(index->tid, SIGRCU);
 199                         poll(NULL, 0, 1);
 200                 }
 201         }
 202         cmm_smp_mb();   /* read ->need_mb before ending the barrier */
 203 }
 204
 205 static void smp_mb_master(int group)
 206 {
 207         force_mb_all_readers();
 208 }
 209 #endif /* #ifdef RCU_SIGNAL */
 210
 211 /*
 212  * synchronize_rcu() waiting. Single thread.
 213  */
 214 static void wait_gp(void)
 215 {
 216         /* Read reader_gp before read futex */
 217         smp_mb_master(RCU_MB_GROUP);
 218         if (uatomic_read(&gp_futex) == -1)
 219                 futex_async(&gp_futex, FUTEX_WAIT, -1,
 220                       NULL, NULL, 0);
 221 }
 222
 223 void update_counter_and_wait(void)
 224 {
 225         CDS_LIST_HEAD(qsreaders);
 226         int wait_loops = 0;
 227         struct rcu_reader *index, *tmp;
 228
 229         /* Switch parity: 0 -> 1, 1 -> 0 */
 230         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE);
 231
 232         /*
 233          * Must commit rcu_gp_ctr update to memory before waiting for quiescent
 234          * state. Failure to do so could result in the writer waiting forever
 235          * while new readers are always accessing data (no progress). Enforce
 236          * compiler-order of store to rcu_gp_ctr before load rcu_reader ctr.
 237          */
 238         cmm_barrier();
 239
 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 each thread URCU_TLS(rcu_reader).ctr count to become 0.
 250          */
 251         for (;;) {
 252                 wait_loops++;
 253                 if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 254                         uatomic_dec(&gp_futex);
 255                         /* Write futex before read reader_gp */
 256                         smp_mb_master(RCU_MB_GROUP);
 257                 }
 258
 259                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 260                         if (!rcu_gp_ongoing(&index->ctr))
 261                                 cds_list_move(&index->node, &qsreaders);
 262                 }
 263
 264 #ifndef HAS_INCOHERENT_CACHES
 265                 if (cds_list_empty(&registry)) {
 266                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 267                                 /* Read reader_gp before write futex */
 268                                 smp_mb_master(RCU_MB_GROUP);
 269                                 uatomic_set(&gp_futex, 0);
 270                         }
 271                         break;
 272                 } else {
 273                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS)
 274                                 wait_gp();
 275                         else
 276                                 caa_cpu_relax();
 277                 }
 278 #else /* #ifndef HAS_INCOHERENT_CACHES */
 279                 /*
 280                  * BUSY-LOOP. Force the reader thread to commit its
 281                  * URCU_TLS(rcu_reader).ctr update to memory if we wait
 282                  * for too long.
 283                  */
 284                 if (cds_list_empty(&registry)) {
 285                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
 286                                 /* Read reader_gp before write futex */
 287                                 smp_mb_master(RCU_MB_GROUP);
 288                                 uatomic_set(&gp_futex, 0);
 289                         }
 290                         break;
 291                 } else {
 292                         switch (wait_loops) {
 293                         case RCU_QS_ACTIVE_ATTEMPTS:
 294                                 wait_gp();
 295                                 break; /* only escape switch */
 296                         case KICK_READER_LOOPS:
 297                                 smp_mb_master(RCU_MB_GROUP);
 298                                 wait_loops = 0;
 299                                 break; /* only escape switch */
 300                         default:
 301                                 caa_cpu_relax();
 302                         }
 303                 }
 304 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
 305         }
 306         /* put back the reader list in the registry */
 307         cds_list_splice(&qsreaders, &registry);
 308 }
 309
 310 void synchronize_rcu(void)
 311 {
 312         mutex_lock(&rcu_gp_lock);
 313
 314         if (cds_list_empty(&registry))
 315                 goto out;
 316
 317         /* All threads should read qparity before accessing data structure
 318          * where new ptr points to. Must be done within rcu_gp_lock because it
 319          * iterates on reader threads.*/
 320         /* Write new ptr before changing the qparity */
 321         smp_mb_master(RCU_MB_GROUP);
 322
 323         /*
 324          * Wait for previous parity to be empty of readers.
 325          */
 326         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 327
 328         /*
 329          * Must finish waiting for quiescent state for parity 0 before
 330          * committing next rcu_gp_ctr update to memory. Failure to do so could
 331          * result in the writer waiting forever while new readers are always
 332          * accessing data (no progress).  Enforce compiler-order of load
 333          * URCU_TLS(rcu_reader).ctr before store to rcu_gp_ctr.
 334          */
 335         cmm_barrier();
 336
 337         /*
 338          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 339          * model easier to understand. It does not have a big performance impact
 340          * anyway, given this is the write-side.
 341          */
 342         cmm_smp_mb();
 343
 344         /*
 345          * Wait for previous parity to be empty of readers.
 346          */
 347         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 348
 349         /* Finish waiting for reader threads before letting the old ptr being
 350          * freed. Must be done within rcu_gp_lock because it iterates on reader
 351          * threads. */
 352         smp_mb_master(RCU_MB_GROUP);
 353 out:
 354         mutex_unlock(&rcu_gp_lock);
 355 }
 356
 357 /*
 358  * library wrappers to be used by non-LGPL compatible source code.
 359  */
 360
 361 void rcu_read_lock(void)
 362 {
 363         _rcu_read_lock();
 364 }
 365
 366 void rcu_read_unlock(void)
 367 {
 368         _rcu_read_unlock();
 369 }
 370
 371 void rcu_register_thread(void)
 372 {
 373         URCU_TLS(rcu_reader).tid = pthread_self();
 374         assert(URCU_TLS(rcu_reader).need_mb == 0);
 375         assert(!(URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK));
 376
 377         mutex_lock(&rcu_gp_lock);
 378         rcu_init();     /* In case gcc does not support constructor attribute */
 379         cds_list_add(&URCU_TLS(rcu_reader).node, &registry);
 380         mutex_unlock(&rcu_gp_lock);
 381 }
 382
 383 void rcu_unregister_thread(void)
 384 {
 385         mutex_lock(&rcu_gp_lock);
 386         cds_list_del(&URCU_TLS(rcu_reader).node);
 387         mutex_unlock(&rcu_gp_lock);
 388 }
 389
 390 #ifdef RCU_MEMBARRIER
 391 void rcu_init(void)
 392 {
 393         if (init_done)
 394                 return;
 395         init_done = 1;
 396         if (!membarrier(MEMBARRIER_EXPEDITED | MEMBARRIER_QUERY))
 397                 has_sys_membarrier = 1;
 398 }
 399 #endif
 400
 401 #ifdef RCU_SIGNAL
 402 static void sigrcu_handler(int signo, siginfo_t *siginfo, void *context)
 403 {
 404         /*
 405          * Executing this cmm_smp_mb() is the only purpose of this signal handler.
 406          * It punctually promotes cmm_barrier() into cmm_smp_mb() on every thread it is
 407          * executed on.
 408          */
 409         cmm_smp_mb();
 410         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).need_mb, 0);
 411         cmm_smp_mb();
 412 }
 413
 414 /*
 415  * rcu_init constructor. Called when the library is linked, but also when
 416  * reader threads are calling rcu_register_thread().
 417  * Should only be called by a single thread at a given time. This is ensured by
 418  * holing the rcu_gp_lock from rcu_register_thread() or by running at library
 419  * load time, which should not be executed by multiple threads nor concurrently
 420  * with rcu_register_thread() anyway.
 421  */
 422 void rcu_init(void)
 423 {
 424         struct sigaction act;
 425         int ret;
 426
 427         if (init_done)
 428                 return;
 429         init_done = 1;
 430
 431         act.sa_sigaction = sigrcu_handler;
 432         act.sa_flags = SA_SIGINFO | SA_RESTART;
 433         sigemptyset(&act.sa_mask);
 434         ret = sigaction(SIGRCU, &act, NULL);
 435         if (ret) {
 436                 perror("Error in sigaction");
 437                 exit(-1);
 438         }
 439 }
 440
 441 void rcu_exit(void)
 442 {
 443         struct sigaction act;
 444         int ret;
 445
 446         ret = sigaction(SIGRCU, NULL, &act);
 447         if (ret) {
 448                 perror("Error in sigaction");
 449                 exit(-1);
 450         }
 451         assert(act.sa_sigaction == sigrcu_handler);
 452         assert(cds_list_empty(&registry));
 453 }
 454
 455 #endif /* #ifdef RCU_SIGNAL */
 456
 457 DEFINE_RCU_FLAVOR(rcu_flavor);
 458
 459 #include "urcu-call-rcu-impl.h"
 460 #include "urcu-defer-impl.h"