tests/benchmark/test_urcu_wfq.c

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