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 "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
  50 /* Remove deprecation warnings from test build. */
  51 #define CDS_LFS_RCU_DEPRECATED
  52
  53 #include <urcu/cds.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         cpu_set_t mask;
  87         int cpu;
  88         int ret;
  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 #if SCHED_SETAFFINITY_ARGS == 2
 109         sched_setaffinity(0, &mask);
 110 #else
 111         sched_setaffinity(0, sizeof(mask), &mask);
 112 #endif
 113 #endif /* HAVE_SCHED_SETAFFINITY */
 114 }
 115
 116 /*
 117  * returns 0 if test should end.
 118  */
 119 static int test_duration_dequeue(void)
 120 {
 121         return !test_stop;
 122 }
 123
 124 static int test_duration_enqueue(void)
 125 {
 126         return !test_stop;
 127 }
 128
 129 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 130 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 131
 132 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 133 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 134
 135 static unsigned int nr_enqueuers;
 136 static unsigned int nr_dequeuers;
 137
 138 struct test {
 139         struct cds_lfs_node_rcu list;
 140         struct rcu_head rcu;
 141 };
 142
 143 static struct cds_lfs_stack_rcu s;
 144
 145 void *thr_enqueuer(void *_count)
 146 {
 147         unsigned long long *count = _count;
 148
 149         printf_verbose("thread_begin %s, tid %lu\n",
 150                         "enqueuer", urcu_get_thread_id());
 151
 152         set_affinity();
 153
 154         rcu_register_thread();
 155
 156         while (!test_go)
 157         {
 158         }
 159         cmm_smp_mb();
 160
 161         for (;;) {
 162                 struct test *node = malloc(sizeof(*node));
 163                 if (!node)
 164                         goto fail;
 165                 cds_lfs_node_init_rcu(&node->list);
 166                 /* No rcu read-side is needed for push */
 167                 cds_lfs_push_rcu(&s, &node->list);
 168                 URCU_TLS(nr_successful_enqueues)++;
 169
 170                 if (caa_unlikely(wdelay))
 171                         loop_sleep(wdelay);
 172 fail:
 173                 URCU_TLS(nr_enqueues)++;
 174                 if (caa_unlikely(!test_duration_enqueue()))
 175                         break;
 176         }
 177
 178         rcu_unregister_thread();
 179
 180         count[0] = URCU_TLS(nr_enqueues);
 181         count[1] = URCU_TLS(nr_successful_enqueues);
 182         printf_verbose("enqueuer thread_end, tid %lu, "
 183                         "enqueues %llu successful_enqueues %llu\n",
 184                         urcu_get_thread_id(),
 185                         URCU_TLS(nr_enqueues),
 186                         URCU_TLS(nr_successful_enqueues));
 187         return ((void*)1);
 188
 189 }
 190
 191 static
 192 void free_node_cb(struct rcu_head *head)
 193 {
 194         struct test *node =
 195                 caa_container_of(head, struct test, rcu);
 196         free(node);
 197 }
 198
 199 void *thr_dequeuer(void *_count)
 200 {
 201         unsigned long long *count = _count;
 202
 203         printf_verbose("thread_begin %s, tid %lu\n",
 204                         "dequeuer", urcu_get_thread_id());
 205
 206         set_affinity();
 207
 208         rcu_register_thread();
 209
 210         while (!test_go)
 211         {
 212         }
 213         cmm_smp_mb();
 214
 215         for (;;) {
 216                 struct cds_lfs_node_rcu *snode;
 217
 218                 rcu_read_lock();
 219                 snode = cds_lfs_pop_rcu(&s);
 220                 rcu_read_unlock();
 221                 if (snode) {
 222                         struct test *node;
 223
 224                         node = caa_container_of(snode, struct test, list);
 225                         call_rcu(&node->rcu, free_node_cb);
 226                         URCU_TLS(nr_successful_dequeues)++;
 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
 237         printf_verbose("dequeuer thread_end, tid %lu, "
 238                         "dequeues %llu, successful_dequeues %llu\n",
 239                         urcu_get_thread_id(),
 240                         URCU_TLS(nr_dequeues),
 241                         URCU_TLS(nr_successful_dequeues));
 242         count[0] = URCU_TLS(nr_dequeues);
 243         count[1] = URCU_TLS(nr_successful_dequeues);
 244         return ((void*)2);
 245 }
 246
 247 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 248 {
 249         struct cds_lfs_node_rcu *snode;
 250
 251         do {
 252                 snode = cds_lfs_pop_rcu(s);
 253                 if (snode) {
 254                         struct test *node;
 255
 256                         node = caa_container_of(snode, struct test, list);
 257                         free(node);
 258                         (*nr_dequeues)++;
 259                 }
 260         } while (snode);
 261 }
 262
 263 void show_usage(int argc, char **argv)
 264 {
 265         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 266                 argv[0]);
 267         printf("OPTIONS:\n");
 268         printf("        [-d delay] (enqueuer period (in loops))\n");
 269         printf("        [-c duration] (dequeuer period (in loops))\n");
 270         printf("        [-v] (verbose output)\n");
 271         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 272         printf("\n");
 273 }
 274
 275 int main(int argc, char **argv)
 276 {
 277         int err;
 278         pthread_t *tid_enqueuer, *tid_dequeuer;
 279         void *tret;
 280         unsigned long long *count_enqueuer, *count_dequeuer;
 281         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 282         unsigned long long tot_successful_enqueues = 0,
 283                            tot_successful_dequeues = 0;
 284         unsigned long long end_dequeues = 0;
 285         int i, a;
 286
 287         if (argc < 4) {
 288                 show_usage(argc, argv);
 289                 return -1;
 290         }
 291
 292         err = sscanf(argv[1], "%u", &nr_dequeuers);
 293         if (err != 1) {
 294                 show_usage(argc, argv);
 295                 return -1;
 296         }
 297
 298         err = sscanf(argv[2], "%u", &nr_enqueuers);
 299         if (err != 1) {
 300                 show_usage(argc, argv);
 301                 return -1;
 302         }
 303
 304         err = sscanf(argv[3], "%lu", &duration);
 305         if (err != 1) {
 306                 show_usage(argc, argv);
 307                 return -1;
 308         }
 309
 310         for (i = 4; i < argc; i++) {
 311                 if (argv[i][0] != '-')
 312                         continue;
 313                 switch (argv[i][1]) {
 314                 case 'a':
 315                         if (argc < i + 2) {
 316                                 show_usage(argc, argv);
 317                                 return -1;
 318                         }
 319                         a = atoi(argv[++i]);
 320                         cpu_affinities[next_aff++] = a;
 321                         use_affinity = 1;
 322                         printf_verbose("Adding CPU %d affinity\n", a);
 323                         break;
 324                 case 'c':
 325                         if (argc < i + 2) {
 326                                 show_usage(argc, argv);
 327                                 return -1;
 328                         }
 329                         rduration = atol(argv[++i]);
 330                         break;
 331                 case 'd':
 332                         if (argc < i + 2) {
 333                                 show_usage(argc, argv);
 334                                 return -1;
 335                         }
 336                         wdelay = atol(argv[++i]);
 337                         break;
 338                 case 'v':
 339                         verbose_mode = 1;
 340                         break;
 341                 }
 342         }
 343
 344         printf_verbose("running test for %lu seconds, %u enqueuers, "
 345                        "%u dequeuers.\n",
 346                        duration, nr_enqueuers, nr_dequeuers);
 347         printf_verbose("Writer delay : %lu loops.\n", rduration);
 348         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 349         printf_verbose("thread %-6s, tid %lu\n",
 350                         "main", urcu_get_thread_id());
 351
 352         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 353         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 354         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 355         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 356         cds_lfs_init_rcu(&s);
 357         err = create_all_cpu_call_rcu_data(0);
 358         if (err) {
 359                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 360         }
 361
 362         next_aff = 0;
 363
 364         for (i = 0; i < nr_enqueuers; i++) {
 365                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 366                                      &count_enqueuer[2 * i]);
 367                 if (err != 0)
 368                         exit(1);
 369         }
 370         for (i = 0; i < nr_dequeuers; i++) {
 371                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 372                                      &count_dequeuer[2 * i]);
 373                 if (err != 0)
 374                         exit(1);
 375         }
 376
 377         cmm_smp_mb();
 378
 379         test_go = 1;
 380
 381         for (i = 0; i < duration; i++) {
 382                 sleep(1);
 383                 if (verbose_mode) {
 384                         fwrite(".", sizeof(char), 1, stdout);
 385                         fflush(stdout);
 386                 }
 387         }
 388
 389         test_stop = 1;
 390
 391         for (i = 0; i < nr_enqueuers; i++) {
 392                 err = pthread_join(tid_enqueuer[i], &tret);
 393                 if (err != 0)
 394                         exit(1);
 395                 tot_enqueues += count_enqueuer[2 * i];
 396                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 397         }
 398         for (i = 0; i < nr_dequeuers; i++) {
 399                 err = pthread_join(tid_dequeuer[i], &tret);
 400                 if (err != 0)
 401                         exit(1);
 402                 tot_dequeues += count_dequeuer[2 * i];
 403                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 404         }
 405
 406         test_end(&s, &end_dequeues);
 407
 408         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 409                        tot_enqueues, tot_dequeues);
 410         printf_verbose("total number of successful enqueues : %llu, "
 411                        "successful dequeues %llu\n",
 412                        tot_successful_enqueues, tot_successful_dequeues);
 413         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 414                 "nr_dequeuers %3u "
 415                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 416                 "successful enqueues %12llu successful dequeues %12llu "
 417                 "end_dequeues %llu nr_ops %12llu\n",
 418                 argv[0], duration, nr_enqueuers, wdelay,
 419                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 420                 tot_successful_enqueues,
 421                 tot_successful_dequeues, end_dequeues,
 422                 tot_enqueues + tot_dequeues);
 423         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 424                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 425                        "succ. dequeues + end dequeues %llu.\n",
 426                        tot_successful_enqueues,
 427                        tot_successful_dequeues + end_dequeues);
 428
 429         free_all_cpu_call_rcu_data();
 430         free(count_enqueuer);
 431         free(count_dequeuer);
 432         free(tid_enqueuer);
 433         free(tid_dequeuer);
 434         return 0;
 435 }