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 "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
  49 /* Remove deprecation warnings from test build. */
  50 #define CDS_WFQ_DEPRECATED
  51
  52 #include <urcu.h>
  53 #include <urcu/wfqueue.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 static struct cds_wfq_queue q;
 140
 141 void *thr_enqueuer(void *_count)
 142 {
 143         unsigned long long *count = _count;
 144
 145         printf_verbose("thread_begin %s, tid %lu\n",
 146                         "enqueuer", urcu_get_thread_id());
 147
 148         set_affinity();
 149
 150         while (!test_go)
 151         {
 152         }
 153         cmm_smp_mb();
 154
 155         for (;;) {
 156                 struct cds_wfq_node *node = malloc(sizeof(*node));
 157                 if (!node)
 158                         goto fail;
 159                 cds_wfq_node_init(node);
 160                 cds_wfq_enqueue(&q, node);
 161                 URCU_TLS(nr_successful_enqueues)++;
 162
 163                 if (caa_unlikely(wdelay))
 164                         loop_sleep(wdelay);
 165 fail:
 166                 URCU_TLS(nr_enqueues)++;
 167                 if (caa_unlikely(!test_duration_enqueue()))
 168                         break;
 169         }
 170
 171         count[0] = URCU_TLS(nr_enqueues);
 172         count[1] = URCU_TLS(nr_successful_enqueues);
 173         printf_verbose("enqueuer thread_end, tid %lu, "
 174                         "enqueues %llu successful_enqueues %llu\n",
 175                         urcu_get_thread_id(),
 176                         URCU_TLS(nr_enqueues),
 177                         URCU_TLS(nr_successful_enqueues));
 178         return ((void*)1);
 179
 180 }
 181
 182 void *thr_dequeuer(void *_count)
 183 {
 184         unsigned long long *count = _count;
 185
 186         printf_verbose("thread_begin %s, tid %lu\n",
 187                         "dequeuer", urcu_get_thread_id());
 188
 189         set_affinity();
 190
 191         while (!test_go)
 192         {
 193         }
 194         cmm_smp_mb();
 195
 196         for (;;) {
 197                 struct cds_wfq_node *node = cds_wfq_dequeue_blocking(&q);
 198
 199                 if (node) {
 200                         free(node);
 201                         URCU_TLS(nr_successful_dequeues)++;
 202                 }
 203
 204                 URCU_TLS(nr_dequeues)++;
 205                 if (caa_unlikely(!test_duration_dequeue()))
 206                         break;
 207                 if (caa_unlikely(rduration))
 208                         loop_sleep(rduration);
 209         }
 210
 211         printf_verbose("dequeuer thread_end, tid %lu, "
 212                         "dequeues %llu, successful_dequeues %llu\n",
 213                         urcu_get_thread_id(),
 214                         URCU_TLS(nr_dequeues),
 215                         URCU_TLS(nr_successful_dequeues));
 216         count[0] = URCU_TLS(nr_dequeues);
 217         count[1] = URCU_TLS(nr_successful_dequeues);
 218         return ((void*)2);
 219 }
 220
 221 void test_end(struct cds_wfq_queue *q, unsigned long long *nr_dequeues)
 222 {
 223         struct cds_wfq_node *node;
 224
 225         do {
 226                 node = cds_wfq_dequeue_blocking(q);
 227                 if (node) {
 228                         free(node);
 229                         (*nr_dequeues)++;
 230                 }
 231         } while (node);
 232 }
 233
 234 void show_usage(int argc, char **argv)
 235 {
 236         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 237                 argv[0]);
 238         printf("OPTIONS:\n");
 239         printf("        [-d delay] (enqueuer period (in loops))\n");
 240         printf("        [-c duration] (dequeuer period (in loops))\n");
 241         printf("        [-v] (verbose output)\n");
 242         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 243         printf("\n");
 244 }
 245
 246 int main(int argc, char **argv)
 247 {
 248         int err;
 249         pthread_t *tid_enqueuer, *tid_dequeuer;
 250         void *tret;
 251         unsigned long long *count_enqueuer, *count_dequeuer;
 252         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 253         unsigned long long tot_successful_enqueues = 0,
 254                            tot_successful_dequeues = 0;
 255         unsigned long long end_dequeues = 0;
 256         int i, a;
 257         unsigned int i_thr;
 258
 259         if (argc < 4) {
 260                 show_usage(argc, argv);
 261                 return -1;
 262         }
 263
 264         err = sscanf(argv[1], "%u", &nr_dequeuers);
 265         if (err != 1) {
 266                 show_usage(argc, argv);
 267                 return -1;
 268         }
 269
 270         err = sscanf(argv[2], "%u", &nr_enqueuers);
 271         if (err != 1) {
 272                 show_usage(argc, argv);
 273                 return -1;
 274         }
 275
 276         err = sscanf(argv[3], "%lu", &duration);
 277         if (err != 1) {
 278                 show_usage(argc, argv);
 279                 return -1;
 280         }
 281
 282         for (i = 4; i < argc; i++) {
 283                 if (argv[i][0] != '-')
 284                         continue;
 285                 switch (argv[i][1]) {
 286                 case 'a':
 287                         if (argc < i + 2) {
 288                                 show_usage(argc, argv);
 289                                 return -1;
 290                         }
 291                         a = atoi(argv[++i]);
 292                         cpu_affinities[next_aff++] = a;
 293                         use_affinity = 1;
 294                         printf_verbose("Adding CPU %d affinity\n", a);
 295                         break;
 296                 case 'c':
 297                         if (argc < i + 2) {
 298                                 show_usage(argc, argv);
 299                                 return -1;
 300                         }
 301                         rduration = atol(argv[++i]);
 302                         break;
 303                 case 'd':
 304                         if (argc < i + 2) {
 305                                 show_usage(argc, argv);
 306                                 return -1;
 307                         }
 308                         wdelay = atol(argv[++i]);
 309                         break;
 310                 case 'v':
 311                         verbose_mode = 1;
 312                         break;
 313                 }
 314         }
 315
 316         printf_verbose("running test for %lu seconds, %u enqueuers, "
 317                        "%u dequeuers.\n",
 318                        duration, nr_enqueuers, nr_dequeuers);
 319         printf_verbose("Writer delay : %lu loops.\n", rduration);
 320         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 321         printf_verbose("thread %-6s, tid %lu\n",
 322                         "main", urcu_get_thread_id());
 323
 324         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 325         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 326         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 327         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 328         cds_wfq_init(&q);
 329
 330         next_aff = 0;
 331
 332         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 333                 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
 334                                      &count_enqueuer[2 * i_thr]);
 335                 if (err != 0)
 336                         exit(1);
 337         }
 338         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 339                 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
 340                                      &count_dequeuer[2 * i_thr]);
 341                 if (err != 0)
 342                         exit(1);
 343         }
 344
 345         cmm_smp_mb();
 346
 347         test_go = 1;
 348
 349         for (i_thr = 0; i_thr < duration; i_thr++) {
 350                 sleep(1);
 351                 if (verbose_mode) {
 352                         fwrite(".", sizeof(char), 1, stdout);
 353                         fflush(stdout);
 354                 }
 355         }
 356
 357         test_stop = 1;
 358
 359         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 360                 err = pthread_join(tid_enqueuer[i_thr], &tret);
 361                 if (err != 0)
 362                         exit(1);
 363                 tot_enqueues += count_enqueuer[2 * i_thr];
 364                 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
 365         }
 366         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 367                 err = pthread_join(tid_dequeuer[i_thr], &tret);
 368                 if (err != 0)
 369                         exit(1);
 370                 tot_dequeues += count_dequeuer[2 * i_thr];
 371                 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
 372         }
 373
 374         test_end(&q, &end_dequeues);
 375
 376         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 377                        tot_enqueues, tot_dequeues);
 378         printf_verbose("total number of successful enqueues : %llu, "
 379                        "successful dequeues %llu\n",
 380                        tot_successful_enqueues, tot_successful_dequeues);
 381         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 382                 "nr_dequeuers %3u "
 383                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 384                 "successful enqueues %12llu successful dequeues %12llu "
 385                 "end_dequeues %llu nr_ops %12llu\n",
 386                 argv[0], duration, nr_enqueuers, wdelay,
 387                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 388                 tot_successful_enqueues,
 389                 tot_successful_dequeues, end_dequeues,
 390                 tot_enqueues + tot_dequeues);
 391         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 392                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 393                        "succ. dequeues + end dequeues %llu.\n",
 394                        tot_successful_enqueues,
 395                        tot_successful_dequeues + end_dequeues);
 396
 397         cds_wfq_destroy(&q);
 398         free(count_enqueuer);
 399         free(count_dequeuer);
 400         free(tid_enqueuer);
 401         free(tid_dequeuer);
 402
 403         return 0;
 404 }