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-bp-static.h"
  39 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
  40 #include "urcu-bp.h"
  41
  42 /* Sleep delay in us */
  43 #define RCU_SLEEP_DELAY         1000
  44 #define ARENA_INIT_ALLOC        16
  45
  46 void __attribute__((destructor)) rcu_bp_exit(void);
  47
  48 static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
  49
  50 #ifdef DEBUG_YIELD
  51 unsigned int yield_active;
  52 unsigned int __thread rand_yield;
  53 #endif
  54
  55 /*
  56  * Global grace period counter.
  57  * Contains the current RCU_GP_CTR_PHASE.
  58  * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
  59  * Written to only by writer with mutex taken. Read by both writer and readers.
  60  */
  61 long rcu_gp_ctr = RCU_GP_COUNT;
  62
  63 /*
  64  * Pointer to registry elements. Written to only by each individual reader. Read
  65  * by both the reader and the writers.
  66  */
  67 struct rcu_reader __thread *rcu_reader;
  68
  69 static CDS_LIST_HEAD(registry);
  70
  71 struct registry_arena {
  72         void *p;
  73         size_t len;
  74         size_t used;
  75 };
  76
  77 static struct registry_arena registry_arena;
  78
  79 static void rcu_gc_registry(void);
  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                 perror("Error in pthread mutex lock");
  89                 exit(-1);
  90         }
  91 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
  92         while ((ret = pthread_mutex_trylock(mutex)) != 0) {
  93                 if (ret != EBUSY && ret != EINTR) {
  94                         printf("ret = %d, errno = %d\n", ret, errno);
  95                         perror("Error in pthread mutex lock");
  96                         exit(-1);
  97                 }
  98                 poll(NULL,0,10);
  99         }
 100 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
 101 }
 102
 103 static void mutex_unlock(pthread_mutex_t *mutex)
 104 {
 105         int ret;
 106
 107         ret = pthread_mutex_unlock(mutex);
 108         if (ret) {
 109                 perror("Error in pthread mutex unlock");
 110                 exit(-1);
 111         }
 112 }
 113
 114 void update_counter_and_wait(void)
 115 {
 116         CDS_LIST_HEAD(qsreaders);
 117         int wait_loops = 0;
 118         struct rcu_reader *index, *tmp;
 119
 120         /* Switch parity: 0 -> 1, 1 -> 0 */
 121         CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE);
 122
 123         /*
 124          * Must commit qparity update to memory before waiting for other parity
 125          * quiescent state. Failure to do so could result in the writer waiting
 126          * forever while new readers are always accessing data (no progress).
 127          * Ensured by CMM_STORE_SHARED and CMM_LOAD_SHARED.
 128          */
 129
 130         /*
 131          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 132          * model easier to understand. It does not have a big performance impact
 133          * anyway, given this is the write-side.
 134          */
 135         cmm_smp_mb();
 136
 137         /*
 138          * Wait for each thread rcu_reader.ctr count to become 0.
 139          */
 140         for (;;) {
 141                 wait_loops++;
 142                 cds_list_for_each_entry_safe(index, tmp, &registry, node) {
 143                         if (!rcu_old_gp_ongoing(&index->ctr))
 144                                 cds_list_move(&index->node, &qsreaders);
 145                 }
 146
 147                 if (cds_list_empty(&registry)) {
 148                         break;
 149                 } else {
 150                         if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS)
 151                                 usleep(RCU_SLEEP_DELAY);
 152                         else
 153                                 caa_cpu_relax();
 154                 }
 155         }
 156         /* put back the reader list in the registry */
 157         cds_list_splice(&qsreaders, &registry);
 158 }
 159
 160 void synchronize_rcu(void)
 161 {
 162         sigset_t newmask, oldmask;
 163         int ret;
 164
 165         ret = sigemptyset(&newmask);
 166         assert(!ret);
 167         ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask);
 168         assert(!ret);
 169
 170         mutex_lock(&rcu_gp_lock);
 171
 172         if (cds_list_empty(&registry))
 173                 goto out;
 174
 175         /* All threads should read qparity before accessing data structure
 176          * where new ptr points to. */
 177         /* Write new ptr before changing the qparity */
 178         cmm_smp_mb();
 179
 180         /* Remove old registry elements */
 181         rcu_gc_registry();
 182
 183         /*
 184          * Wait for previous parity to be empty of readers.
 185          */
 186         update_counter_and_wait();      /* 0 -> 1, wait readers in parity 0 */
 187
 188         /*
 189          * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
 190          * model easier to understand. It does not have a big performance impact
 191          * anyway, given this is the write-side.
 192          */
 193         cmm_smp_mb();
 194
 195         /*
 196          * Wait for previous parity to be empty of readers.
 197          */
 198         update_counter_and_wait();      /* 1 -> 0, wait readers in parity 1 */
 199
 200         /*
 201          * Finish waiting for reader threads before letting the old ptr being
 202          * freed.
 203          */
 204         cmm_smp_mb();
 205 out:
 206         mutex_unlock(&rcu_gp_lock);
 207         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 208         assert(!ret);
 209 }
 210
 211 /*
 212  * library wrappers to be used by non-LGPL compatible source code.
 213  */
 214
 215 void rcu_read_lock(void)
 216 {
 217         _rcu_read_lock();
 218 }
 219
 220 void rcu_read_unlock(void)
 221 {
 222         _rcu_read_unlock();
 223 }
 224
 225 /*
 226  * only grow for now.
 227  */
 228 static void resize_arena(struct registry_arena *arena, size_t len)
 229 {
 230         void *new_arena;
 231
 232         if (!arena->p)
 233                 new_arena = mmap(arena->p, len,
 234                                  PROT_READ | PROT_WRITE,
 235                                  MAP_ANONYMOUS | MAP_PRIVATE,
 236                                  -1, 0);
 237         else
 238                 new_arena = mremap(arena->p, arena->len,
 239                                    len, MREMAP_MAYMOVE);
 240         assert(new_arena != MAP_FAILED);
 241
 242         /*
 243          * re-used the same region ?
 244          */
 245         if (new_arena == arena->p)
 246                 return;
 247
 248         memcpy(new_arena, arena->p, arena->len);
 249         bzero(new_arena + arena->len, len - arena->len);
 250         arena->p = new_arena;
 251 }
 252
 253 /* Called with signals off and mutex locked */
 254 static void add_thread(void)
 255 {
 256         struct rcu_reader *rcu_reader_reg;
 257
 258         if (registry_arena.len
 259             < registry_arena.used + sizeof(struct rcu_reader))
 260                 resize_arena(&registry_arena,
 261                 max(registry_arena.len << 1, ARENA_INIT_ALLOC));
 262         /*
 263          * Find a free spot.
 264          */
 265         for (rcu_reader_reg = registry_arena.p;
 266              (void *)rcu_reader_reg < registry_arena.p + registry_arena.len;
 267              rcu_reader_reg++) {
 268                 if (!rcu_reader_reg->alloc)
 269                         break;
 270         }
 271         rcu_reader_reg->alloc = 1;
 272         registry_arena.used += sizeof(struct rcu_reader);
 273
 274         /* Add to registry */
 275         rcu_reader_reg->tid = pthread_self();
 276         assert(rcu_reader_reg->ctr == 0);
 277         cds_list_add(&rcu_reader_reg->node, &registry);
 278         rcu_reader = rcu_reader_reg;
 279 }
 280
 281 /* Called with signals off and mutex locked */
 282 static void rcu_gc_registry(void)
 283 {
 284         struct rcu_reader *rcu_reader_reg;
 285         pthread_t tid;
 286         int ret;
 287
 288         for (rcu_reader_reg = registry_arena.p;
 289              (void *)rcu_reader_reg < registry_arena.p + registry_arena.len;
 290              rcu_reader_reg++) {
 291                 if (!rcu_reader_reg->alloc)
 292                         continue;
 293                 tid = rcu_reader_reg->tid;
 294                 ret = pthread_kill(tid, 0);
 295                 assert(ret != EINVAL);
 296                 if (ret == ESRCH) {
 297                         cds_list_del(&rcu_reader_reg->node);
 298                         rcu_reader_reg->ctr = 0;
 299                         rcu_reader_reg->alloc = 0;
 300                         registry_arena.used -= sizeof(struct rcu_reader);
 301                 }
 302         }
 303 }
 304
 305 /* Disable signals, take mutex, add to registry */
 306 void rcu_bp_register(void)
 307 {
 308         sigset_t newmask, oldmask;
 309         int ret;
 310
 311         ret = sigemptyset(&newmask);
 312         assert(!ret);
 313         ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask);
 314         assert(!ret);
 315
 316         /*
 317          * Check if a signal concurrently registered our thread since
 318          * the check in rcu_read_lock(). */
 319         if (rcu_reader)
 320                 goto end;
 321
 322         mutex_lock(&rcu_gp_lock);
 323         add_thread();
 324         mutex_unlock(&rcu_gp_lock);
 325 end:
 326         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 327         assert(!ret);
 328 }
 329
 330 void rcu_bp_exit()
 331 {
 332         munmap(registry_arena.p, registry_arena.len);
 333 }