urcu-bp.c

   1 /*
   2  * urcu-bp.c
   3  *
   4  * Userspace RCU library, "bulletproof" version.
   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 #include <unistd.h>
  36 #include <sys/mman.h>
  37
  38 #include "urcu/map/urcu-bp.h"
  39
  40 #include "urcu/static/urcu-bp.h"
  41 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  42 #include "urcu-bp.h"
  43
  44 #ifndef MAP_ANONYMOUS
  45 #define MAP_ANONYMOUS MAP_ANON
  46 #endif
  47
  48 #ifndef __linux__
  49
  50 #define MREMAP_MAYMOVE  1
  51 #define MREMAP_FIXED    2
  52
  53 /*
  54  * mremap wrapper for non-Linux systems. Maps a RW, anonymous private mapping.
  55  * This is not generic.
  56 */
  57 void *mremap(void *old_address, size_t old_size, size_t new_size, int flags)
  58 {
  59         void *new_address;
  60
  61         assert(flags & MREMAP_MAYMOVE);
  62         assert(!(flags & MREMAP_FIXED));
  63         new_address = mmap(old_address, new_size,
  64                            PROT_READ | PROT_WRITE,
  65                            MAP_ANONYMOUS | MAP_PRIVATE,
  66                            -1, 0);
  67         if (new_address == MAP_FAILED)
  68                 return MAP_FAILED;
  69         if (old_address) {
  70                 memcpy(new_address, old_address, old_size);
  71                 munmap(old_address, old_size);
  72         }
  73         return new_address;
  74 }
  75 #endif
  76
  77 /* Sleep delay in us */
  78 #define RCU_SLEEP_DELAY         1000
  79 #define ARENA_INIT_ALLOC        16
  80
  81 void __attribute__((destructor)) rcu_bp_exit(void);
  82
  83 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  84
  85 #ifdef DEBUG_YIELD
  86 unsigned int yield_active;
  87 unsigned int __thread rand_yield;
  88 #endif
  89
  90 /*
  91  * Global grace period counter.
  92  * Contains the current RCU_GP_CTR_PHASE.
  93  * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
  94  * Written to only by writer with mutex taken. Read by both writer and readers.
  95  */
  96 long rcu_gp_ctr = RCU_GP_COUNT;
  97
  98 /*
  99  * Pointer to registry elements. Written to only by each individual reader. Read
 100  * by both the reader and the writers.
 101  */
 102 struct rcu_reader __thread *rcu_reader;
 103
 104 static CDS_LIST_HEAD(registry);
 105
 106 struct registry_arena {
 107         void *p;
 108         size_t len;
 109         size_t used;
 110 };
 111
 112 static struct registry_arena registry_arena;
 113
 114 /* Saved fork signal mask, protected by rcu_gp_lock */
 115 static sigset_t saved_fork_signal_mask;
 116
 117 static void rcu_gc_registry(void);
 118
 119 static void mutex_lock(pthread_mutex_t *mutex)
 120 {
 121         int ret;
 122
 123 #ifndef DISTRUST_SIGNALS_EXTREME
 124         ret = pthread_mutex_lock(mutex);
 125         if (ret) {
 126                 perror("Error in pthread mutex lock");
 127                 exit(-1);
 128         }
 129 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
 130         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
 131                 if (ret != EBUSY && ret != EINTR) {
 132                         printf("ret = %d, errno = %d\n", ret, errno);
 133                         perror("Error in pthread mutex lock");
 134                         exit(-1);
 135                 }
 136                 poll(NULL,0,10);
 137         }
 138 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
 139 }
 140
 141 static void mutex_unlock(pthread_mutex_t *mutex)
 142 {
 143         int ret;
 144
 145         ret = pthread_mutex_unlock(mutex);
 146         if (ret) {
 147                 perror("Error in pthread mutex unlock");
 148                 exit(-1);
 149         }
 150 }
 151
 152 void update_counter_and_wait(void)
 153 {
 154         CDS_LIST_HEAD(qsreaders);
 155         int wait_loops = 0;
 156         struct rcu_reader *index, *tmp;
 157
 158         /* Switch parity: 0 -> 1, 1 -> 0 */
 159         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE);
 160
 161         /*
 162          * Must commit qparity update to memory before waiting for other parity
 163          * quiescent state. Failure to do so could result in the writer waiting
 164          * forever while new readers are always accessing data (no progress).
 165          * Ensured by CMM_STORE_SHARED and CMM_LOAD_SHARED.
 166          */
 167
 168         /*
 169          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 170          * model easier to understand. It does not have a big performance impact
 171          * anyway, given this is the write-side.
 172          */
 173         cmm_smp_mb();
 174
 175         /*
 176          * Wait for each thread rcu_reader.ctr count to become 0.
 177          */
 178         for (;;) {
 179                 wait_loops++;
 180                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 181                         if (!rcu_old_gp_ongoing(&index->ctr))
 182                                 cds_list_move(&index->node, &qsreaders);
 183                 }
 184
 185                 if (cds_list_empty(&registry)) {
 186                         break;
 187                 } else {
 188                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS)
 189                                 usleep(RCU_SLEEP_DELAY);
 190                         else
 191                                 caa_cpu_relax();
 192                 }
 193         }
 194         /* put back the reader list in the registry */
 195         cds_list_splice(&qsreaders, &registry);
 196 }
 197
 198 void synchronize_rcu(void)
 199 {
 200         sigset_t newmask, oldmask;
 201         int ret;
 202
 203         ret = sigemptyset(&newmask);
 204         assert(!ret);
 205         ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask);
 206         assert(!ret);
 207
 208         mutex_lock(&rcu_gp_lock);
 209
 210         if (cds_list_empty(&registry))
 211                 goto out;
 212
 213         /* All threads should read qparity before accessing data structure
 214          * where new ptr points to. */
 215         /* Write new ptr before changing the qparity */
 216         cmm_smp_mb();
 217
 218         /* Remove old registry elements */
 219         rcu_gc_registry();
 220
 221         /*
 222          * Wait for previous parity to be empty of readers.
 223          */
 224         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 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
 238         /*
 239          * Finish waiting for reader threads before letting the old ptr being
 240          * freed.
 241          */
 242         cmm_smp_mb();
 243 out:
 244         mutex_unlock(&rcu_gp_lock);
 245         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 246         assert(!ret);
 247 }
 248
 249 /*
 250  * library wrappers to be used by non-LGPL compatible source code.
 251  */
 252
 253 void rcu_read_lock(void)
 254 {
 255         _rcu_read_lock();
 256 }
 257
 258 void rcu_read_unlock(void)
 259 {
 260         _rcu_read_unlock();
 261 }
 262
 263 /*
 264  * only grow for now.
 265  */
 266 static void resize_arena(struct registry_arena *arena, size_t len)
 267 {
 268         void *new_arena;
 269
 270         if (!arena->p)
 271                 new_arena = mmap(arena->p, len,
 272                                  PROT_READ | PROT_WRITE,
 273                                  MAP_ANONYMOUS | MAP_PRIVATE,
 274                                  -1, 0);
 275         else
 276                 new_arena = mremap(arena->p, arena->len,
 277                                    len, MREMAP_MAYMOVE);
 278         assert(new_arena != MAP_FAILED);
 279
 280         /*
 281          * re-used the same region ?
 282          */
 283         if (new_arena == arena->p)
 284                 return;
 285
 286         bzero(new_arena + arena->len, len - arena->len);
 287         arena->p = new_arena;
 288 }
 289
 290 /* Called with signals off and mutex locked */
 291 static void add_thread(void)
 292 {
 293         struct rcu_reader *rcu_reader_reg;
 294
 295         if (registry_arena.len
 296             < registry_arena.used + sizeof(struct rcu_reader))
 297                 resize_arena(&registry_arena,
 298                 max(registry_arena.len << 1, ARENA_INIT_ALLOC));
 299         /*
 300          * Find a free spot.
 301          */
 302         for (rcu_reader_reg = registry_arena.p;
 303              (void *)rcu_reader_reg < registry_arena.p + registry_arena.len;
 304              rcu_reader_reg++) {
 305                 if (!rcu_reader_reg->alloc)
 306                         break;
 307         }
 308         rcu_reader_reg->alloc = 1;
 309         registry_arena.used += sizeof(struct rcu_reader);
 310
 311         /* Add to registry */
 312         rcu_reader_reg->tid = pthread_self();
 313         assert(rcu_reader_reg->ctr == 0);
 314         cds_list_add(&rcu_reader_reg->node, &registry);
 315         rcu_reader = rcu_reader_reg;
 316 }
 317
 318 /* Called with signals off and mutex locked */
 319 static void rcu_gc_registry(void)
 320 {
 321         struct rcu_reader *rcu_reader_reg;
 322         pthread_t tid;
 323         int ret;
 324
 325         for (rcu_reader_reg = registry_arena.p;
 326              (void *)rcu_reader_reg < registry_arena.p + registry_arena.len;
 327              rcu_reader_reg++) {
 328                 if (!rcu_reader_reg->alloc)
 329                         continue;
 330                 tid = rcu_reader_reg->tid;
 331                 ret = pthread_kill(tid, 0);
 332                 assert(ret != EINVAL);
 333                 if (ret == ESRCH) {
 334                         cds_list_del(&rcu_reader_reg->node);
 335                         rcu_reader_reg->ctr = 0;
 336                         rcu_reader_reg->alloc = 0;
 337                         registry_arena.used -= sizeof(struct rcu_reader);
 338                 }
 339         }
 340 }
 341
 342 /* Disable signals, take mutex, add to registry */
 343 void rcu_bp_register(void)
 344 {
 345         sigset_t newmask, oldmask;
 346         int ret;
 347
 348         ret = sigemptyset(&newmask);
 349         assert(!ret);
 350         ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask);
 351         assert(!ret);
 352
 353         /*
 354          * Check if a signal concurrently registered our thread since
 355          * the check in rcu_read_lock(). */
 356         if (rcu_reader)
 357                 goto end;
 358
 359         mutex_lock(&rcu_gp_lock);
 360         add_thread();
 361         mutex_unlock(&rcu_gp_lock);
 362 end:
 363         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 364         assert(!ret);
 365 }
 366
 367 void rcu_bp_exit()
 368 {
 369         munmap(registry_arena.p, registry_arena.len);
 370 }
 371
 372 /*
 373  * Holding the rcu_gp_lock across fork will make sure we fork() don't race with
 374  * a concurrent thread executing with this same lock held. This ensures that the
 375  * registry is in a coherent state in the child.
 376  */
 377 void rcu_bp_before_fork(void)
 378 {
 379         sigset_t newmask, oldmask;
 380         int ret;
 381
 382         ret = sigemptyset(&newmask);
 383         assert(!ret);
 384         ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask);
 385         assert(!ret);
 386         mutex_lock(&rcu_gp_lock);
 387         saved_fork_signal_mask = oldmask;
 388 }
 389
 390 void rcu_bp_after_fork_parent(void)
 391 {
 392         sigset_t oldmask;
 393         int ret;
 394
 395         oldmask = saved_fork_signal_mask;
 396         mutex_unlock(&rcu_gp_lock);
 397         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 398         assert(!ret);
 399 }
 400
 401 void rcu_bp_after_fork_child(void)
 402 {
 403         sigset_t oldmask;
 404         int ret;
 405
 406         rcu_gc_registry();
 407         oldmask = saved_fork_signal_mask;
 408         mutex_unlock(&rcu_gp_lock);
 409         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 410         assert(!ret);
 411 }
 412
 413 #include "urcu-call-rcu-impl.h"
 414 #include "urcu-defer-impl.h"