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 "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 #include <urcu.h>
  48
  49 /* Remove deprecation warnings from test build. */
  50 #define CDS_LFS_RCU_DEPRECATED
  51
  52 #include <urcu/cds.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 struct test {
 135         struct cds_lfs_node_rcu list;
 136         struct rcu_head rcu;
 137 };
 138
 139 static struct cds_lfs_stack_rcu s;
 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         rcu_register_thread();
 151
 152         while (!test_go)
 153         {
 154         }
 155         cmm_smp_mb();
 156
 157         for (;;) {
 158                 struct test *node = malloc(sizeof(*node));
 159                 if (!node)
 160                         goto fail;
 161                 cds_lfs_node_init_rcu(&node->list);
 162                 /* No rcu read-side is needed for push */
 163                 cds_lfs_push_rcu(&s, &node->list);
 164                 URCU_TLS(nr_successful_enqueues)++;
 165
 166                 if (caa_unlikely(wdelay))
 167                         loop_sleep(wdelay);
 168 fail:
 169                 URCU_TLS(nr_enqueues)++;
 170                 if (caa_unlikely(!test_duration_enqueue()))
 171                         break;
 172         }
 173
 174         rcu_unregister_thread();
 175
 176         count[0] = URCU_TLS(nr_enqueues);
 177         count[1] = URCU_TLS(nr_successful_enqueues);
 178         printf_verbose("enqueuer thread_end, tid %lu, "
 179                         "enqueues %llu successful_enqueues %llu\n",
 180                         urcu_get_thread_id(),
 181                         URCU_TLS(nr_enqueues),
 182                         URCU_TLS(nr_successful_enqueues));
 183         return ((void*)1);
 184
 185 }
 186
 187 static
 188 void free_node_cb(struct rcu_head *head)
 189 {
 190         struct test *node =
 191                 caa_container_of(head, struct test, rcu);
 192         free(node);
 193 }
 194
 195 void *thr_dequeuer(void *_count)
 196 {
 197         unsigned long long *count = _count;
 198
 199         printf_verbose("thread_begin %s, tid %lu\n",
 200                         "dequeuer", urcu_get_thread_id());
 201
 202         set_affinity();
 203
 204         rcu_register_thread();
 205
 206         while (!test_go)
 207         {
 208         }
 209         cmm_smp_mb();
 210
 211         for (;;) {
 212                 struct cds_lfs_node_rcu *snode;
 213
 214                 rcu_read_lock();
 215                 snode = cds_lfs_pop_rcu(&s);
 216                 rcu_read_unlock();
 217                 if (snode) {
 218                         struct test *node;
 219
 220                         node = caa_container_of(snode, struct test, list);
 221                         call_rcu(&node->rcu, free_node_cb);
 222                         URCU_TLS(nr_successful_dequeues)++;
 223                 }
 224                 URCU_TLS(nr_dequeues)++;
 225                 if (caa_unlikely(!test_duration_dequeue()))
 226                         break;
 227                 if (caa_unlikely(rduration))
 228                         loop_sleep(rduration);
 229         }
 230
 231         rcu_unregister_thread();
 232
 233         printf_verbose("dequeuer thread_end, tid %lu, "
 234                         "dequeues %llu, successful_dequeues %llu\n",
 235                         urcu_get_thread_id(),
 236                         URCU_TLS(nr_dequeues),
 237                         URCU_TLS(nr_successful_dequeues));
 238         count[0] = URCU_TLS(nr_dequeues);
 239         count[1] = URCU_TLS(nr_successful_dequeues);
 240         return ((void*)2);
 241 }
 242
 243 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 244 {
 245         struct cds_lfs_node_rcu *snode;
 246
 247         do {
 248                 snode = cds_lfs_pop_rcu(s);
 249                 if (snode) {
 250                         struct test *node;
 251
 252                         node = caa_container_of(snode, struct test, list);
 253                         free(node);
 254                         (*nr_dequeues)++;
 255                 }
 256         } while (snode);
 257 }
 258
 259 void show_usage(int argc, char **argv)
 260 {
 261         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 262                 argv[0]);
 263         printf("OPTIONS:\n");
 264         printf("        [-d delay] (enqueuer period (in loops))\n");
 265         printf("        [-c duration] (dequeuer period (in loops))\n");
 266         printf("        [-v] (verbose output)\n");
 267         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 268         printf("\n");
 269 }
 270
 271 int main(int argc, char **argv)
 272 {
 273         int err;
 274         pthread_t *tid_enqueuer, *tid_dequeuer;
 275         void *tret;
 276         unsigned long long *count_enqueuer, *count_dequeuer;
 277         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 278         unsigned long long tot_successful_enqueues = 0,
 279                            tot_successful_dequeues = 0;
 280         unsigned long long end_dequeues = 0;
 281         int i, a;
 282         unsigned int i_thr;
 283
 284         if (argc < 4) {
 285                 show_usage(argc, argv);
 286                 return -1;
 287         }
 288
 289         err = sscanf(argv[1], "%u", &nr_dequeuers);
 290         if (err != 1) {
 291                 show_usage(argc, argv);
 292                 return -1;
 293         }
 294
 295         err = sscanf(argv[2], "%u", &nr_enqueuers);
 296         if (err != 1) {
 297                 show_usage(argc, argv);
 298                 return -1;
 299         }
 300
 301         err = sscanf(argv[3], "%lu", &duration);
 302         if (err != 1) {
 303                 show_usage(argc, argv);
 304                 return -1;
 305         }
 306
 307         for (i = 4; i < argc; i++) {
 308                 if (argv[i][0] != '-')
 309                         continue;
 310                 switch (argv[i][1]) {
 311                 case 'a':
 312                         if (argc < i + 2) {
 313                                 show_usage(argc, argv);
 314                                 return -1;
 315                         }
 316                         a = atoi(argv[++i]);
 317                         cpu_affinities[next_aff++] = a;
 318                         use_affinity = 1;
 319                         printf_verbose("Adding CPU %d affinity\n", a);
 320                         break;
 321                 case 'c':
 322                         if (argc < i + 2) {
 323                                 show_usage(argc, argv);
 324                                 return -1;
 325                         }
 326                         rduration = atol(argv[++i]);
 327                         break;
 328                 case 'd':
 329                         if (argc < i + 2) {
 330                                 show_usage(argc, argv);
 331                                 return -1;
 332                         }
 333                         wdelay = atol(argv[++i]);
 334                         break;
 335                 case 'v':
 336                         verbose_mode = 1;
 337                         break;
 338                 }
 339         }
 340
 341         printf_verbose("running test for %lu seconds, %u enqueuers, "
 342                        "%u dequeuers.\n",
 343                        duration, nr_enqueuers, nr_dequeuers);
 344         printf_verbose("Writer delay : %lu loops.\n", rduration);
 345         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 346         printf_verbose("thread %-6s, tid %lu\n",
 347                         "main", urcu_get_thread_id());
 348
 349         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 350         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 351         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 352         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 353         cds_lfs_init_rcu(&s);
 354         err = create_all_cpu_call_rcu_data(0);
 355         if (err) {
 356                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 357         }
 358
 359         next_aff = 0;
 360
 361         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 362                 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
 363                                      &count_enqueuer[2 * i_thr]);
 364                 if (err != 0)
 365                         exit(1);
 366         }
 367         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 368                 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
 369                                      &count_dequeuer[2 * i_thr]);
 370                 if (err != 0)
 371                         exit(1);
 372         }
 373
 374         cmm_smp_mb();
 375
 376         test_go = 1;
 377
 378         for (i_thr = 0; i_thr < duration; i_thr++) {
 379                 sleep(1);
 380                 if (verbose_mode) {
 381                         fwrite(".", sizeof(char), 1, stdout);
 382                         fflush(stdout);
 383                 }
 384         }
 385
 386         test_stop = 1;
 387
 388         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 389                 err = pthread_join(tid_enqueuer[i_thr], &tret);
 390                 if (err != 0)
 391                         exit(1);
 392                 tot_enqueues += count_enqueuer[2 * i_thr];
 393                 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
 394         }
 395         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 396                 err = pthread_join(tid_dequeuer[i_thr], &tret);
 397                 if (err != 0)
 398                         exit(1);
 399                 tot_dequeues += count_dequeuer[2 * i_thr];
 400                 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
 401         }
 402
 403         test_end(&s, &end_dequeues);
 404
 405         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 406                        tot_enqueues, tot_dequeues);
 407         printf_verbose("total number of successful enqueues : %llu, "
 408                        "successful dequeues %llu\n",
 409                        tot_successful_enqueues, tot_successful_dequeues);
 410         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 411                 "nr_dequeuers %3u "
 412                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 413                 "successful enqueues %12llu successful dequeues %12llu "
 414                 "end_dequeues %llu nr_ops %12llu\n",
 415                 argv[0], duration, nr_enqueuers, wdelay,
 416                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 417                 tot_successful_enqueues,
 418                 tot_successful_dequeues, end_dequeues,
 419                 tot_enqueues + tot_dequeues);
 420         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 421                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 422                        "succ. dequeues + end dequeues %llu.\n",
 423                        tot_successful_enqueues,
 424                        tot_successful_dequeues + end_dequeues);
 425
 426         free_all_cpu_call_rcu_data();
 427         free(count_enqueuer);
 428         free(count_dequeuer);
 429         free(tid_enqueuer);
 430         free(tid_dequeuer);
 431         return 0;
 432 }