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