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 void *thr_enqueuer(void *_count)
 137 {
 138         unsigned long long *count = _count;
 139
 140         printf_verbose("thread_begin %s, tid %lu\n",
 141                         "enqueuer", urcu_get_thread_id());
 142
 143         set_affinity();
 144
 145         while (!test_go)
 146         {
 147         }
 148         cmm_smp_mb();
 149
 150         for (;;) {
 151                 struct cds_wfq_node *node = malloc(sizeof(*node));
 152                 if (!node)
 153                         goto fail;
 154                 cds_wfq_node_init(node);
 155                 cds_wfq_enqueue(&q, node);
 156                 URCU_TLS(nr_successful_enqueues)++;
 157
 158                 if (caa_unlikely(wdelay))
 159                         loop_sleep(wdelay);
 160 fail:
 161                 URCU_TLS(nr_enqueues)++;
 162                 if (caa_unlikely(!test_duration_enqueue()))
 163                         break;
 164         }
 165
 166         count[0] = URCU_TLS(nr_enqueues);
 167         count[1] = URCU_TLS(nr_successful_enqueues);
 168         printf_verbose("enqueuer thread_end, tid %lu, "
 169                         "enqueues %llu successful_enqueues %llu\n",
 170                         urcu_get_thread_id(),
 171                         URCU_TLS(nr_enqueues),
 172                         URCU_TLS(nr_successful_enqueues));
 173         return ((void*)1);
 174
 175 }
 176
 177 void *thr_dequeuer(void *_count)
 178 {
 179         unsigned long long *count = _count;
 180
 181         printf_verbose("thread_begin %s, tid %lu\n",
 182                         "dequeuer", urcu_get_thread_id());
 183
 184         set_affinity();
 185
 186         while (!test_go)
 187         {
 188         }
 189         cmm_smp_mb();
 190
 191         for (;;) {
 192                 struct cds_wfq_node *node = cds_wfq_dequeue_blocking(&q);
 193
 194                 if (node) {
 195                         free(node);
 196                         URCU_TLS(nr_successful_dequeues)++;
 197                 }
 198
 199                 URCU_TLS(nr_dequeues)++;
 200                 if (caa_unlikely(!test_duration_dequeue()))
 201                         break;
 202                 if (caa_unlikely(rduration))
 203                         loop_sleep(rduration);
 204         }
 205
 206         printf_verbose("dequeuer thread_end, tid %lu, "
 207                         "dequeues %llu, successful_dequeues %llu\n",
 208                         urcu_get_thread_id(),
 209                         URCU_TLS(nr_dequeues),
 210                         URCU_TLS(nr_successful_dequeues));
 211         count[0] = URCU_TLS(nr_dequeues);
 212         count[1] = URCU_TLS(nr_successful_dequeues);
 213         return ((void*)2);
 214 }
 215
 216 void test_end(struct cds_wfq_queue *q, unsigned long long *nr_dequeues)
 217 {
 218         struct cds_wfq_node *node;
 219
 220         do {
 221                 node = cds_wfq_dequeue_blocking(q);
 222                 if (node) {
 223                         free(node);
 224                         (*nr_dequeues)++;
 225                 }
 226         } while (node);
 227 }
 228
 229 void show_usage(int argc, char **argv)
 230 {
 231         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 232                 argv[0]);
 233         printf("OPTIONS:\n");
 234         printf("        [-d delay] (enqueuer period (in loops))\n");
 235         printf("        [-c duration] (dequeuer period (in loops))\n");
 236         printf("        [-v] (verbose output)\n");
 237         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 238         printf("\n");
 239 }
 240
 241 int main(int argc, char **argv)
 242 {
 243         int err;
 244         pthread_t *tid_enqueuer, *tid_dequeuer;
 245         void *tret;
 246         unsigned long long *count_enqueuer, *count_dequeuer;
 247         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 248         unsigned long long tot_successful_enqueues = 0,
 249                            tot_successful_dequeues = 0;
 250         unsigned long long end_dequeues = 0;
 251         int i, a;
 252         unsigned int i_thr;
 253
 254         if (argc < 4) {
 255                 show_usage(argc, argv);
 256                 return -1;
 257         }
 258
 259         err = sscanf(argv[1], "%u", &nr_dequeuers);
 260         if (err != 1) {
 261                 show_usage(argc, argv);
 262                 return -1;
 263         }
 264
 265         err = sscanf(argv[2], "%u", &nr_enqueuers);
 266         if (err != 1) {
 267                 show_usage(argc, argv);
 268                 return -1;
 269         }
 270
 271         err = sscanf(argv[3], "%lu", &duration);
 272         if (err != 1) {
 273                 show_usage(argc, argv);
 274                 return -1;
 275         }
 276
 277         for (i = 4; i < argc; i++) {
 278                 if (argv[i][0] != '-')
 279                         continue;
 280                 switch (argv[i][1]) {
 281                 case 'a':
 282                         if (argc < i + 2) {
 283                                 show_usage(argc, argv);
 284                                 return -1;
 285                         }
 286                         a = atoi(argv[++i]);
 287                         cpu_affinities[next_aff++] = a;
 288                         use_affinity = 1;
 289                         printf_verbose("Adding CPU %d affinity\n", a);
 290                         break;
 291                 case 'c':
 292                         if (argc < i + 2) {
 293                                 show_usage(argc, argv);
 294                                 return -1;
 295                         }
 296                         rduration = atol(argv[++i]);
 297                         break;
 298                 case 'd':
 299                         if (argc < i + 2) {
 300                                 show_usage(argc, argv);
 301                                 return -1;
 302                         }
 303                         wdelay = atol(argv[++i]);
 304                         break;
 305                 case 'v':
 306                         verbose_mode = 1;
 307                         break;
 308                 }
 309         }
 310
 311         printf_verbose("running test for %lu seconds, %u enqueuers, "
 312                        "%u dequeuers.\n",
 313                        duration, nr_enqueuers, nr_dequeuers);
 314         printf_verbose("Writer delay : %lu loops.\n", rduration);
 315         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 316         printf_verbose("thread %-6s, tid %lu\n",
 317                         "main", urcu_get_thread_id());
 318
 319         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 320         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 321         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 322         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 323         cds_wfq_init(&q);
 324
 325         next_aff = 0;
 326
 327         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 328                 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
 329                                      &count_enqueuer[2 * i_thr]);
 330                 if (err != 0)
 331                         exit(1);
 332         }
 333         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 334                 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
 335                                      &count_dequeuer[2 * i_thr]);
 336                 if (err != 0)
 337                         exit(1);
 338         }
 339
 340         cmm_smp_mb();
 341
 342         test_go = 1;
 343
 344         for (i_thr = 0; i_thr < duration; i_thr++) {
 345                 sleep(1);
 346                 if (verbose_mode) {
 347                         fwrite(".", sizeof(char), 1, stdout);
 348                         fflush(stdout);
 349                 }
 350         }
 351
 352         test_stop = 1;
 353
 354         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 355                 err = pthread_join(tid_enqueuer[i_thr], &tret);
 356                 if (err != 0)
 357                         exit(1);
 358                 tot_enqueues += count_enqueuer[2 * i_thr];
 359                 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
 360         }
 361         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 362                 err = pthread_join(tid_dequeuer[i_thr], &tret);
 363                 if (err != 0)
 364                         exit(1);
 365                 tot_dequeues += count_dequeuer[2 * i_thr];
 366                 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
 367         }
 368
 369         test_end(&q, &end_dequeues);
 370
 371         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 372                        tot_enqueues, tot_dequeues);
 373         printf_verbose("total number of successful enqueues : %llu, "
 374                        "successful dequeues %llu\n",
 375                        tot_successful_enqueues, tot_successful_dequeues);
 376         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 377                 "nr_dequeuers %3u "
 378                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 379                 "successful enqueues %12llu successful dequeues %12llu "
 380                 "end_dequeues %llu nr_ops %12llu\n",
 381                 argv[0], duration, nr_enqueuers, wdelay,
 382                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 383                 tot_successful_enqueues,
 384                 tot_successful_dequeues, end_dequeues,
 385                 tot_enqueues + tot_dequeues);
 386         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 387                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 388                        "succ. dequeues + end dequeues %llu.\n",
 389                        tot_successful_enqueues,
 390                        tot_successful_dequeues + end_dequeues);
 391
 392         cds_wfq_destroy(&q);
 393         free(count_enqueuer);
 394         free(count_dequeuer);
 395         free(tid_enqueuer);
 396         free(tid_dequeuer);
 397
 398         return 0;
 399 }