tests/benchmark/test_urcu_lfs_rcu.c

   1 /*
   2  * test_urcu_lfs_rcu.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free stack
   5  *
   6  * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program 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
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #define _GNU_SOURCE
  25 #include "config.h"
  26 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <stdlib.h>
  29 #include <stdint.h>
  30 #include <stdbool.h>
  31 #include <string.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <unistd.h>
  35 #include <stdio.h>
  36 #include <assert.h>
  37 #include <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include "cpuset.h"
  42 #include "thread-id.h"
  43
  44 /* hardcoded number of CPUs */
  45 #define NR_CPUS 16384
  46
  47 #ifndef DYNAMIC_LINK_TEST
  48 #define _LGPL_SOURCE
  49 #endif
  50 #include <urcu.h>
  51
  52 /* Remove deprecation warnings from test build. */
  53 #define CDS_LFS_RCU_DEPRECATED
  54
  55 #include <urcu/cds.h>
  56
  57 static volatile int test_go, test_stop;
  58
  59 static unsigned long rduration;
  60
  61 static unsigned long duration;
  62
  63 /* read-side C.S. duration, in loops */
  64 static unsigned long wdelay;
  65
  66 static inline void loop_sleep(unsigned long loops)
  67 {
  68         while (loops-- != 0)
  69                 caa_cpu_relax();
  70 }
  71
  72 static int verbose_mode;
  73
  74 #define printf_verbose(fmt, args...)            \
  75         do {                                    \
  76                 if (verbose_mode)               \
  77                         printf(fmt, args);      \
  78         } while (0)
  79
  80 static unsigned int cpu_affinities[NR_CPUS];
  81 static unsigned int next_aff = 0;
  82 static int use_affinity = 0;
  83
  84 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  85
  86 static void set_affinity(void)
  87 {
  88 #if HAVE_SCHED_SETAFFINITY
  89         cpu_set_t mask;
  90         int cpu, ret;
  91 #endif /* HAVE_SCHED_SETAFFINITY */
  92
  93         if (!use_affinity)
  94                 return;
  95
  96 #if HAVE_SCHED_SETAFFINITY
  97         ret = pthread_mutex_lock(&affinity_mutex);
  98         if (ret) {
  99                 perror("Error in pthread mutex lock");
 100                 exit(-1);
 101         }
 102         cpu = cpu_affinities[next_aff++];
 103         ret = pthread_mutex_unlock(&affinity_mutex);
 104         if (ret) {
 105                 perror("Error in pthread mutex unlock");
 106                 exit(-1);
 107         }
 108
 109         CPU_ZERO(&mask);
 110         CPU_SET(cpu, &mask);
 111 #if SCHED_SETAFFINITY_ARGS == 2
 112         sched_setaffinity(0, &mask);
 113 #else
 114         sched_setaffinity(0, sizeof(mask), &mask);
 115 #endif
 116 #endif /* HAVE_SCHED_SETAFFINITY */
 117 }
 118
 119 /*
 120  * returns 0 if test should end.
 121  */
 122 static int test_duration_dequeue(void)
 123 {
 124         return !test_stop;
 125 }
 126
 127 static int test_duration_enqueue(void)
 128 {
 129         return !test_stop;
 130 }
 131
 132 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 133 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 134
 135 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 136 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 137
 138 static unsigned int nr_enqueuers;
 139 static unsigned int nr_dequeuers;
 140
 141 struct test {
 142         struct cds_lfs_node_rcu list;
 143         struct rcu_head rcu;
 144 };
 145
 146 static struct cds_lfs_stack_rcu s;
 147
 148 void *thr_enqueuer(void *_count)
 149 {
 150         unsigned long long *count = _count;
 151
 152         printf_verbose("thread_begin %s, tid %lu\n",
 153                         "enqueuer", urcu_get_thread_id());
 154
 155         set_affinity();
 156
 157         rcu_register_thread();
 158
 159         while (!test_go)
 160         {
 161         }
 162         cmm_smp_mb();
 163
 164         for (;;) {
 165                 struct test *node = malloc(sizeof(*node));
 166                 if (!node)
 167                         goto fail;
 168                 cds_lfs_node_init_rcu(&node->list);
 169                 /* No rcu read-side is needed for push */
 170                 cds_lfs_push_rcu(&s, &node->list);
 171                 URCU_TLS(nr_successful_enqueues)++;
 172
 173                 if (caa_unlikely(wdelay))
 174                         loop_sleep(wdelay);
 175 fail:
 176                 URCU_TLS(nr_enqueues)++;
 177                 if (caa_unlikely(!test_duration_enqueue()))
 178                         break;
 179         }
 180
 181         rcu_unregister_thread();
 182
 183         count[0] = URCU_TLS(nr_enqueues);
 184         count[1] = URCU_TLS(nr_successful_enqueues);
 185         printf_verbose("enqueuer thread_end, tid %lu, "
 186                         "enqueues %llu successful_enqueues %llu\n",
 187                         urcu_get_thread_id(),
 188                         URCU_TLS(nr_enqueues),
 189                         URCU_TLS(nr_successful_enqueues));
 190         return ((void*)1);
 191
 192 }
 193
 194 static
 195 void free_node_cb(struct rcu_head *head)
 196 {
 197         struct test *node =
 198                 caa_container_of(head, struct test, rcu);
 199         free(node);
 200 }
 201
 202 void *thr_dequeuer(void *_count)
 203 {
 204         unsigned long long *count = _count;
 205
 206         printf_verbose("thread_begin %s, tid %lu\n",
 207                         "dequeuer", urcu_get_thread_id());
 208
 209         set_affinity();
 210
 211         rcu_register_thread();
 212
 213         while (!test_go)
 214         {
 215         }
 216         cmm_smp_mb();
 217
 218         for (;;) {
 219                 struct cds_lfs_node_rcu *snode;
 220
 221                 rcu_read_lock();
 222                 snode = cds_lfs_pop_rcu(&s);
 223                 rcu_read_unlock();
 224                 if (snode) {
 225                         struct test *node;
 226
 227                         node = caa_container_of(snode, struct test, list);
 228                         call_rcu(&node->rcu, free_node_cb);
 229                         URCU_TLS(nr_successful_dequeues)++;
 230                 }
 231                 URCU_TLS(nr_dequeues)++;
 232                 if (caa_unlikely(!test_duration_dequeue()))
 233                         break;
 234                 if (caa_unlikely(rduration))
 235                         loop_sleep(rduration);
 236         }
 237
 238         rcu_unregister_thread();
 239
 240         printf_verbose("dequeuer thread_end, tid %lu, "
 241                         "dequeues %llu, successful_dequeues %llu\n",
 242                         urcu_get_thread_id(),
 243                         URCU_TLS(nr_dequeues),
 244                         URCU_TLS(nr_successful_dequeues));
 245         count[0] = URCU_TLS(nr_dequeues);
 246         count[1] = URCU_TLS(nr_successful_dequeues);
 247         return ((void*)2);
 248 }
 249
 250 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 251 {
 252         struct cds_lfs_node_rcu *snode;
 253
 254         do {
 255                 snode = cds_lfs_pop_rcu(s);
 256                 if (snode) {
 257                         struct test *node;
 258
 259                         node = caa_container_of(snode, struct test, list);
 260                         free(node);
 261                         (*nr_dequeues)++;
 262                 }
 263         } while (snode);
 264 }
 265
 266 void show_usage(int argc, char **argv)
 267 {
 268         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 269                 argv[0]);
 270         printf("OPTIONS:\n");
 271         printf("        [-d delay] (enqueuer period (in loops))\n");
 272         printf("        [-c duration] (dequeuer period (in loops))\n");
 273         printf("        [-v] (verbose output)\n");
 274         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 275         printf("\n");
 276 }
 277
 278 int main(int argc, char **argv)
 279 {
 280         int err;
 281         pthread_t *tid_enqueuer, *tid_dequeuer;
 282         void *tret;
 283         unsigned long long *count_enqueuer, *count_dequeuer;
 284         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 285         unsigned long long tot_successful_enqueues = 0,
 286                            tot_successful_dequeues = 0;
 287         unsigned long long end_dequeues = 0;
 288         int i, a;
 289
 290         if (argc < 4) {
 291                 show_usage(argc, argv);
 292                 return -1;
 293         }
 294
 295         err = sscanf(argv[1], "%u", &nr_dequeuers);
 296         if (err != 1) {
 297                 show_usage(argc, argv);
 298                 return -1;
 299         }
 300
 301         err = sscanf(argv[2], "%u", &nr_enqueuers);
 302         if (err != 1) {
 303                 show_usage(argc, argv);
 304                 return -1;
 305         }
 306
 307         err = sscanf(argv[3], "%lu", &duration);
 308         if (err != 1) {
 309                 show_usage(argc, argv);
 310                 return -1;
 311         }
 312
 313         for (i = 4; i < argc; i++) {
 314                 if (argv[i][0] != '-')
 315                         continue;
 316                 switch (argv[i][1]) {
 317                 case 'a':
 318                         if (argc < i + 2) {
 319                                 show_usage(argc, argv);
 320                                 return -1;
 321                         }
 322                         a = atoi(argv[++i]);
 323                         cpu_affinities[next_aff++] = a;
 324                         use_affinity = 1;
 325                         printf_verbose("Adding CPU %d affinity\n", a);
 326                         break;
 327                 case 'c':
 328                         if (argc < i + 2) {
 329                                 show_usage(argc, argv);
 330                                 return -1;
 331                         }
 332                         rduration = atol(argv[++i]);
 333                         break;
 334                 case 'd':
 335                         if (argc < i + 2) {
 336                                 show_usage(argc, argv);
 337                                 return -1;
 338                         }
 339                         wdelay = atol(argv[++i]);
 340                         break;
 341                 case 'v':
 342                         verbose_mode = 1;
 343                         break;
 344                 }
 345         }
 346
 347         printf_verbose("running test for %lu seconds, %u enqueuers, "
 348                        "%u dequeuers.\n",
 349                        duration, nr_enqueuers, nr_dequeuers);
 350         printf_verbose("Writer delay : %lu loops.\n", rduration);
 351         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 352         printf_verbose("thread %-6s, tid %lu\n",
 353                         "main", urcu_get_thread_id());
 354
 355         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 356         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 357         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 358         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 359         cds_lfs_init_rcu(&s);
 360         err = create_all_cpu_call_rcu_data(0);
 361         if (err) {
 362                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 363         }
 364
 365         next_aff = 0;
 366
 367         for (i = 0; i < nr_enqueuers; i++) {
 368                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 369                                      &count_enqueuer[2 * i]);
 370                 if (err != 0)
 371                         exit(1);
 372         }
 373         for (i = 0; i < nr_dequeuers; i++) {
 374                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 375                                      &count_dequeuer[2 * i]);
 376                 if (err != 0)
 377                         exit(1);
 378         }
 379
 380         cmm_smp_mb();
 381
 382         test_go = 1;
 383
 384         for (i = 0; i < duration; i++) {
 385                 sleep(1);
 386                 if (verbose_mode) {
 387                         fwrite(".", sizeof(char), 1, stdout);
 388                         fflush(stdout);
 389                 }
 390         }
 391
 392         test_stop = 1;
 393
 394         for (i = 0; i < nr_enqueuers; i++) {
 395                 err = pthread_join(tid_enqueuer[i], &tret);
 396                 if (err != 0)
 397                         exit(1);
 398                 tot_enqueues += count_enqueuer[2 * i];
 399                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 400         }
 401         for (i = 0; i < nr_dequeuers; i++) {
 402                 err = pthread_join(tid_dequeuer[i], &tret);
 403                 if (err != 0)
 404                         exit(1);
 405                 tot_dequeues += count_dequeuer[2 * i];
 406                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 407         }
 408
 409         test_end(&s, &end_dequeues);
 410
 411         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 412                        tot_enqueues, tot_dequeues);
 413         printf_verbose("total number of successful enqueues : %llu, "
 414                        "successful dequeues %llu\n",
 415                        tot_successful_enqueues, tot_successful_dequeues);
 416         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 417                 "nr_dequeuers %3u "
 418                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 419                 "successful enqueues %12llu successful dequeues %12llu "
 420                 "end_dequeues %llu nr_ops %12llu\n",
 421                 argv[0], duration, nr_enqueuers, wdelay,
 422                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 423                 tot_successful_enqueues,
 424                 tot_successful_dequeues, end_dequeues,
 425                 tot_enqueues + tot_dequeues);
 426         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 427                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 428                        "succ. dequeues + end dequeues %llu.\n",
 429                        tot_successful_enqueues,
 430                        tot_successful_dequeues + end_dequeues);
 431
 432         free_all_cpu_call_rcu_data();
 433         free(count_enqueuer);
 434         free(count_dequeuer);
 435         free(tid_enqueuer);
 436         free(tid_dequeuer);
 437         return 0;
 438 }