tests/benchmark/test_urcu_lfq.c

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