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 #define _GNU_SOURCE
  25 #include "config.h"
  26 #include <stdio.h>
  27 #include <pthread.h>
  28 #include <stdlib.h>
  29 #include <stdint.h>
  30 #include <stdbool.h>
  31 #include <string.h>
  32 #include <sys/types.h>
  33 #include <sys/wait.h>
  34 #include <unistd.h>
  35 #include <stdio.h>
  36 #include <assert.h>
  37 #include <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include "cpuset.h"
  42 #include "thread-id.h"
  43
  44 /* hardcoded number of CPUs */
  45 #define NR_CPUS 16384
  46
  47 #ifndef DYNAMIC_LINK_TEST
  48 #define _LGPL_SOURCE
  49 #endif
  50 #include <urcu.h>
  51
  52 /* Remove deprecation warnings from test build. */
  53 #define CDS_LFS_RCU_DEPRECATED
  54
  55 #include <urcu/cds.h>
  56
  57 static volatile int test_go, test_stop;
  58
  59 static unsigned long rduration;
  60
  61 static unsigned long duration;
  62
  63 /* read-side C.S. duration, in loops */
  64 static unsigned long wdelay;
  65
  66 static inline void loop_sleep(unsigned long loops)
  67 {
  68         while (loops-- != 0)
  69                 caa_cpu_relax();
  70 }
  71
  72 static int verbose_mode;
  73
  74 #define printf_verbose(fmt, args...)            \
  75         do {                                    \
  76                 if (verbose_mode)               \
  77                         printf(fmt, args);      \
  78         } while (0)
  79
  80 static unsigned int cpu_affinities[NR_CPUS];
  81 static unsigned int next_aff = 0;
  82 static int use_affinity = 0;
  83
  84 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  85
  86 static void set_affinity(void)
  87 {
  88         cpu_set_t mask;
  89         int cpu;
  90         int ret;
  91
  92         if (!use_affinity)
  93                 return;
  94
  95 #if HAVE_SCHED_SETAFFINITY
  96         ret = pthread_mutex_lock(&affinity_mutex);
  97         if (ret) {
  98                 perror("Error in pthread mutex lock");
  99                 exit(-1);
 100         }
 101         cpu = cpu_affinities[next_aff++];
 102         ret = pthread_mutex_unlock(&affinity_mutex);
 103         if (ret) {
 104                 perror("Error in pthread mutex unlock");
 105                 exit(-1);
 106         }
 107
 108         CPU_ZERO(&mask);
 109         CPU_SET(cpu, &mask);
 110 #if SCHED_SETAFFINITY_ARGS == 2
 111         sched_setaffinity(0, &mask);
 112 #else
 113         sched_setaffinity(0, sizeof(mask), &mask);
 114 #endif
 115 #endif /* HAVE_SCHED_SETAFFINITY */
 116 }
 117
 118 /*
 119  * returns 0 if test should end.
 120  */
 121 static int test_duration_dequeue(void)
 122 {
 123         return !test_stop;
 124 }
 125
 126 static int test_duration_enqueue(void)
 127 {
 128         return !test_stop;
 129 }
 130
 131 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 132 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 133
 134 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 135 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 136
 137 static unsigned int nr_enqueuers;
 138 static unsigned int nr_dequeuers;
 139
 140 struct test {
 141         struct cds_lfs_node_rcu list;
 142         struct rcu_head rcu;
 143 };
 144
 145 static struct cds_lfs_stack_rcu s;
 146
 147 void *thr_enqueuer(void *_count)
 148 {
 149         unsigned long long *count = _count;
 150
 151         printf_verbose("thread_begin %s, tid %lu\n",
 152                         "enqueuer", urcu_get_thread_id());
 153
 154         set_affinity();
 155
 156         rcu_register_thread();
 157
 158         while (!test_go)
 159         {
 160         }
 161         cmm_smp_mb();
 162
 163         for (;;) {
 164                 struct test *node = malloc(sizeof(*node));
 165                 if (!node)
 166                         goto fail;
 167                 cds_lfs_node_init_rcu(&node->list);
 168                 /* No rcu read-side is needed for push */
 169                 cds_lfs_push_rcu(&s, &node->list);
 170                 URCU_TLS(nr_successful_enqueues)++;
 171
 172                 if (caa_unlikely(wdelay))
 173                         loop_sleep(wdelay);
 174 fail:
 175                 URCU_TLS(nr_enqueues)++;
 176                 if (caa_unlikely(!test_duration_enqueue()))
 177                         break;
 178         }
 179
 180         rcu_unregister_thread();
 181
 182         count[0] = URCU_TLS(nr_enqueues);
 183         count[1] = URCU_TLS(nr_successful_enqueues);
 184         printf_verbose("enqueuer thread_end, tid %lu, "
 185                         "enqueues %llu successful_enqueues %llu\n",
 186                         urcu_get_thread_id(),
 187                         URCU_TLS(nr_enqueues),
 188                         URCU_TLS(nr_successful_enqueues));
 189         return ((void*)1);
 190
 191 }
 192
 193 static
 194 void free_node_cb(struct rcu_head *head)
 195 {
 196         struct test *node =
 197                 caa_container_of(head, struct test, rcu);
 198         free(node);
 199 }
 200
 201 void *thr_dequeuer(void *_count)
 202 {
 203         unsigned long long *count = _count;
 204
 205         printf_verbose("thread_begin %s, tid %lu\n",
 206                         "dequeuer", urcu_get_thread_id());
 207
 208         set_affinity();
 209
 210         rcu_register_thread();
 211
 212         while (!test_go)
 213         {
 214         }
 215         cmm_smp_mb();
 216
 217         for (;;) {
 218                 struct cds_lfs_node_rcu *snode;
 219
 220                 rcu_read_lock();
 221                 snode = cds_lfs_pop_rcu(&s);
 222                 rcu_read_unlock();
 223                 if (snode) {
 224                         struct test *node;
 225
 226                         node = caa_container_of(snode, struct test, list);
 227                         call_rcu(&node->rcu, free_node_cb);
 228                         URCU_TLS(nr_successful_dequeues)++;
 229                 }
 230                 URCU_TLS(nr_dequeues)++;
 231                 if (caa_unlikely(!test_duration_dequeue()))
 232                         break;
 233                 if (caa_unlikely(rduration))
 234                         loop_sleep(rduration);
 235         }
 236
 237         rcu_unregister_thread();
 238
 239         printf_verbose("dequeuer thread_end, tid %lu, "
 240                         "dequeues %llu, successful_dequeues %llu\n",
 241                         urcu_get_thread_id(),
 242                         URCU_TLS(nr_dequeues),
 243                         URCU_TLS(nr_successful_dequeues));
 244         count[0] = URCU_TLS(nr_dequeues);
 245         count[1] = URCU_TLS(nr_successful_dequeues);
 246         return ((void*)2);
 247 }
 248
 249 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 250 {
 251         struct cds_lfs_node_rcu *snode;
 252
 253         do {
 254                 snode = cds_lfs_pop_rcu(s);
 255                 if (snode) {
 256                         struct test *node;
 257
 258                         node = caa_container_of(snode, struct test, list);
 259                         free(node);
 260                         (*nr_dequeues)++;
 261                 }
 262         } while (snode);
 263 }
 264
 265 void show_usage(int argc, char **argv)
 266 {
 267         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 268                 argv[0]);
 269         printf("OPTIONS:\n");
 270         printf("        [-d delay] (enqueuer period (in loops))\n");
 271         printf("        [-c duration] (dequeuer period (in loops))\n");
 272         printf("        [-v] (verbose output)\n");
 273         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 274         printf("\n");
 275 }
 276
 277 int main(int argc, char **argv)
 278 {
 279         int err;
 280         pthread_t *tid_enqueuer, *tid_dequeuer;
 281         void *tret;
 282         unsigned long long *count_enqueuer, *count_dequeuer;
 283         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 284         unsigned long long tot_successful_enqueues = 0,
 285                            tot_successful_dequeues = 0;
 286         unsigned long long end_dequeues = 0;
 287         int i, a;
 288
 289         if (argc < 4) {
 290                 show_usage(argc, argv);
 291                 return -1;
 292         }
 293
 294         err = sscanf(argv[1], "%u", &nr_dequeuers);
 295         if (err != 1) {
 296                 show_usage(argc, argv);
 297                 return -1;
 298         }
 299
 300         err = sscanf(argv[2], "%u", &nr_enqueuers);
 301         if (err != 1) {
 302                 show_usage(argc, argv);
 303                 return -1;
 304         }
 305
 306         err = sscanf(argv[3], "%lu", &duration);
 307         if (err != 1) {
 308                 show_usage(argc, argv);
 309                 return -1;
 310         }
 311
 312         for (i = 4; i < argc; i++) {
 313                 if (argv[i][0] != '-')
 314                         continue;
 315                 switch (argv[i][1]) {
 316                 case 'a':
 317                         if (argc < i + 2) {
 318                                 show_usage(argc, argv);
 319                                 return -1;
 320                         }
 321                         a = atoi(argv[++i]);
 322                         cpu_affinities[next_aff++] = a;
 323                         use_affinity = 1;
 324                         printf_verbose("Adding CPU %d affinity\n", a);
 325                         break;
 326                 case 'c':
 327                         if (argc < i + 2) {
 328                                 show_usage(argc, argv);
 329                                 return -1;
 330                         }
 331                         rduration = atol(argv[++i]);
 332                         break;
 333                 case 'd':
 334                         if (argc < i + 2) {
 335                                 show_usage(argc, argv);
 336                                 return -1;
 337                         }
 338                         wdelay = atol(argv[++i]);
 339                         break;
 340                 case 'v':
 341                         verbose_mode = 1;
 342                         break;
 343                 }
 344         }
 345
 346         printf_verbose("running test for %lu seconds, %u enqueuers, "
 347                        "%u dequeuers.\n",
 348                        duration, nr_enqueuers, nr_dequeuers);
 349         printf_verbose("Writer delay : %lu loops.\n", rduration);
 350         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 351         printf_verbose("thread %-6s, tid %lu\n",
 352                         "main", urcu_get_thread_id());
 353
 354         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 355         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 356         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 357         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 358         cds_lfs_init_rcu(&s);
 359         err = create_all_cpu_call_rcu_data(0);
 360         if (err) {
 361                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 362         }
 363
 364         next_aff = 0;
 365
 366         for (i = 0; i < nr_enqueuers; i++) {
 367                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 368                                      &count_enqueuer[2 * i]);
 369                 if (err != 0)
 370                         exit(1);
 371         }
 372         for (i = 0; i < nr_dequeuers; i++) {
 373                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 374                                      &count_dequeuer[2 * i]);
 375                 if (err != 0)
 376                         exit(1);
 377         }
 378
 379         cmm_smp_mb();
 380
 381         test_go = 1;
 382
 383         for (i = 0; i < duration; i++) {
 384                 sleep(1);
 385                 if (verbose_mode)
 386                         (void) write(1, ".", 1);
 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 }