tests/test_urcu_wfs.c

   1 /*
   2  * test_urcu_wfs.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 <sched.h>
  38 #include <errno.h>
  39
  40 #include <urcu/arch.h>
  41 #include <urcu/tls-compat.h>
  42
  43 #ifdef __linux__
  44 #include <syscall.h>
  45 #endif
  46
  47 /* hardcoded number of CPUs */
  48 #define NR_CPUS 16384
  49
  50 #if defined(_syscall0)
  51 _syscall0(pid_t, gettid)
  52 #elif defined(__NR_gettid)
  53 static inline pid_t gettid(void)
  54 {
  55         return syscall(__NR_gettid);
  56 }
  57 #else
  58 #warning "use pid as tid"
  59 static inline pid_t gettid(void)
  60 {
  61         return getpid();
  62 }
  63 #endif
  64
  65 #ifndef DYNAMIC_LINK_TEST
  66 #define _LGPL_SOURCE
  67 #endif
  68 #include <urcu.h>
  69 #include <urcu/wfstack.h>
  70
  71 /*
  72  * External synchronization used.
  73  */
  74 enum test_sync {
  75         TEST_SYNC_NONE = 0,
  76         TEST_SYNC_MUTEX,
  77 };
  78
  79 static enum test_sync test_sync;
  80
  81 static volatile int test_go, test_stop;
  82
  83 static unsigned long rduration;
  84
  85 static unsigned long duration;
  86
  87 /* read-side C.S. duration, in loops */
  88 static unsigned long wdelay;
  89
  90 static inline void loop_sleep(unsigned long loops)
  91 {
  92         while (loops-- != 0)
  93                 caa_cpu_relax();
  94 }
  95
  96 static int verbose_mode;
  97
  98 static int test_pop, test_pop_all;
  99
 100 #define printf_verbose(fmt, args...)            \
 101         do {                                    \
 102                 if (verbose_mode)               \
 103                         printf(fmt, ## args);   \
 104         } while (0)
 105
 106 static unsigned int cpu_affinities[NR_CPUS];
 107 static unsigned int next_aff = 0;
 108 static int use_affinity = 0;
 109
 110 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 111
 112 #ifndef HAVE_CPU_SET_T
 113 typedef unsigned long cpu_set_t;
 114 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 115 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 116 #endif
 117
 118 static void set_affinity(void)
 119 {
 120 #if HAVE_SCHED_SETAFFINITY
 121         cpu_set_t mask;
 122         int cpu, ret;
 123 #endif /* HAVE_SCHED_SETAFFINITY */
 124
 125         if (!use_affinity)
 126                 return;
 127
 128 #if HAVE_SCHED_SETAFFINITY
 129         ret = pthread_mutex_lock(&affinity_mutex);
 130         if (ret) {
 131                 perror("Error in pthread mutex lock");
 132                 exit(-1);
 133         }
 134         cpu = cpu_affinities[next_aff++];
 135         ret = pthread_mutex_unlock(&affinity_mutex);
 136         if (ret) {
 137                 perror("Error in pthread mutex unlock");
 138                 exit(-1);
 139         }
 140
 141         CPU_ZERO(&mask);
 142         CPU_SET(cpu, &mask);
 143 #if SCHED_SETAFFINITY_ARGS == 2
 144         sched_setaffinity(0, &mask);
 145 #else
 146         sched_setaffinity(0, sizeof(mask), &mask);
 147 #endif
 148 #endif /* HAVE_SCHED_SETAFFINITY */
 149 }
 150
 151 /*
 152  * returns 0 if test should end.
 153  */
 154 static int test_duration_dequeue(void)
 155 {
 156         return !test_stop;
 157 }
 158
 159 static int test_duration_enqueue(void)
 160 {
 161         return !test_stop;
 162 }
 163
 164 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 165 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 166
 167 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 168 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 169
 170 static unsigned int nr_enqueuers;
 171 static unsigned int nr_dequeuers;
 172
 173 static struct cds_wfs_stack s;
 174
 175 void *thr_enqueuer(void *_count)
 176 {
 177         unsigned long long *count = _count;
 178
 179         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 180                         "enqueuer", pthread_self(), (unsigned long)gettid());
 181
 182         set_affinity();
 183
 184         while (!test_go)
 185         {
 186         }
 187         cmm_smp_mb();
 188
 189         for (;;) {
 190                 struct cds_wfs_node *node = malloc(sizeof(*node));
 191                 if (!node)
 192                         goto fail;
 193                 cds_wfs_node_init(node);
 194                 cds_wfs_push(&s, node);
 195                 URCU_TLS(nr_successful_enqueues)++;
 196
 197                 if (caa_unlikely(wdelay))
 198                         loop_sleep(wdelay);
 199 fail:
 200                 URCU_TLS(nr_enqueues)++;
 201                 if (caa_unlikely(!test_duration_enqueue()))
 202                         break;
 203         }
 204
 205         count[0] = URCU_TLS(nr_enqueues);
 206         count[1] = URCU_TLS(nr_successful_enqueues);
 207         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 208                        "enqueues %llu successful_enqueues %llu\n",
 209                        pthread_self(), (unsigned long)gettid(),
 210                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 211         return ((void*)1);
 212
 213 }
 214
 215 static void do_test_pop(enum test_sync sync)
 216 {
 217         struct cds_wfs_node *node;
 218
 219         if (sync == TEST_SYNC_MUTEX)
 220                 cds_wfs_pop_lock(&s);
 221         node = __cds_wfs_pop_blocking(&s);
 222         if (sync == TEST_SYNC_MUTEX)
 223                 cds_wfs_pop_unlock(&s);
 224
 225         if (node) {
 226                 free(node);
 227                 URCU_TLS(nr_successful_dequeues)++;
 228         }
 229         URCU_TLS(nr_dequeues)++;
 230 }
 231
 232 static void do_test_pop_all(enum test_sync sync)
 233 {
 234         struct cds_wfs_head *head;
 235         struct cds_wfs_node *node, *n;
 236
 237         if (sync == TEST_SYNC_MUTEX)
 238                 cds_wfs_pop_lock(&s);
 239         head = __cds_wfs_pop_all(&s);
 240         if (sync == TEST_SYNC_MUTEX)
 241                 cds_wfs_pop_unlock(&s);
 242
 243         cds_wfs_for_each_blocking_safe(head, node, n) {
 244                 free(node);
 245                 URCU_TLS(nr_successful_dequeues)++;
 246                 URCU_TLS(nr_dequeues)++;
 247         }
 248 }
 249
 250 void *thr_dequeuer(void *_count)
 251 {
 252         unsigned long long *count = _count;
 253         unsigned int counter;
 254
 255         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 256                         "dequeuer", pthread_self(), (unsigned long)gettid());
 257
 258         set_affinity();
 259
 260         while (!test_go)
 261         {
 262         }
 263         cmm_smp_mb();
 264
 265         assert(test_pop || test_pop_all);
 266
 267         for (;;) {
 268                 if (test_pop && test_pop_all) {
 269                         if (counter & 1)
 270                                 do_test_pop(test_sync);
 271                         else
 272                                 do_test_pop_all(test_sync);
 273                         counter++;
 274                 } else {
 275                         if (test_pop)
 276                                 do_test_pop(test_sync);
 277                         else
 278                                 do_test_pop_all(test_sync);
 279                 }
 280
 281                 if (caa_unlikely(!test_duration_dequeue()))
 282                         break;
 283                 if (caa_unlikely(rduration))
 284                         loop_sleep(rduration);
 285         }
 286
 287         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 288                        "dequeues %llu, successful_dequeues %llu\n",
 289                        pthread_self(), (unsigned long)gettid(),
 290                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 291         count[0] = URCU_TLS(nr_dequeues);
 292         count[1] = URCU_TLS(nr_successful_dequeues);
 293         return ((void*)2);
 294 }
 295
 296 void test_end(struct cds_wfs_stack *s, unsigned long long *nr_dequeues)
 297 {
 298         struct cds_wfs_node *node;
 299
 300         do {
 301                 node = cds_wfs_pop_blocking(s);
 302                 if (node) {
 303                         free(node);
 304                         (*nr_dequeues)++;
 305                 }
 306         } while (node);
 307 }
 308
 309 void show_usage(int argc, char **argv)
 310 {
 311         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 312         printf(" [-d delay] (enqueuer period (in loops))");
 313         printf(" [-c duration] (dequeuer period (in loops))");
 314         printf(" [-v] (verbose output)");
 315         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 316         printf(" [-p] (test pop)");
 317         printf(" [-P] (test pop_all, enabled by default)");
 318         printf(" [-M] (use mutex external synchronization)");
 319         printf("      Note: default: no external synchronization used.");
 320         printf("\n");
 321 }
 322
 323 int main(int argc, char **argv)
 324 {
 325         int err;
 326         pthread_t *tid_enqueuer, *tid_dequeuer;
 327         void *tret;
 328         unsigned long long *count_enqueuer, *count_dequeuer;
 329         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 330         unsigned long long tot_successful_enqueues = 0,
 331                            tot_successful_dequeues = 0;
 332         unsigned long long end_dequeues = 0;
 333         int i, a;
 334
 335         if (argc < 4) {
 336                 show_usage(argc, argv);
 337                 return -1;
 338         }
 339
 340         err = sscanf(argv[1], "%u", &nr_dequeuers);
 341         if (err != 1) {
 342                 show_usage(argc, argv);
 343                 return -1;
 344         }
 345
 346         err = sscanf(argv[2], "%u", &nr_enqueuers);
 347         if (err != 1) {
 348                 show_usage(argc, argv);
 349                 return -1;
 350         }
 351
 352         err = sscanf(argv[3], "%lu", &duration);
 353         if (err != 1) {
 354                 show_usage(argc, argv);
 355                 return -1;
 356         }
 357
 358         for (i = 4; i < argc; i++) {
 359                 if (argv[i][0] != '-')
 360                         continue;
 361                 switch (argv[i][1]) {
 362                 case 'a':
 363                         if (argc < i + 2) {
 364                                 show_usage(argc, argv);
 365                                 return -1;
 366                         }
 367                         a = atoi(argv[++i]);
 368                         cpu_affinities[next_aff++] = a;
 369                         use_affinity = 1;
 370                         printf_verbose("Adding CPU %d affinity\n", a);
 371                         break;
 372                 case 'c':
 373                         if (argc < i + 2) {
 374                                 show_usage(argc, argv);
 375                                 return -1;
 376                         }
 377                         rduration = atol(argv[++i]);
 378                         break;
 379                 case 'd':
 380                         if (argc < i + 2) {
 381                                 show_usage(argc, argv);
 382                                 return -1;
 383                         }
 384                         wdelay = atol(argv[++i]);
 385                         break;
 386                 case 'v':
 387                         verbose_mode = 1;
 388                         break;
 389                 case 'p':
 390                         test_pop = 1;
 391                         break;
 392                 case 'P':
 393                         test_pop_all = 1;
 394                         break;
 395                 case 'M':
 396                         test_sync = TEST_SYNC_MUTEX;
 397                         break;
 398                 }
 399         }
 400
 401         /* activate pop_all test by default */
 402         if (!test_pop && !test_pop_all)
 403                 test_pop_all = 1;
 404
 405         printf_verbose("running test for %lu seconds, %u enqueuers, "
 406                        "%u dequeuers.\n",
 407                        duration, nr_enqueuers, nr_dequeuers);
 408         if (test_pop)
 409                 printf_verbose("pop test activated.\n");
 410         if (test_pop_all)
 411                 printf_verbose("pop_all test activated.\n");
 412         printf_verbose("Writer delay : %lu loops.\n", rduration);
 413         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 414         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 415                         "main", pthread_self(), (unsigned long)gettid());
 416
 417         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 418         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 419         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 420         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 421         cds_wfs_init(&s);
 422
 423         next_aff = 0;
 424
 425         for (i = 0; i < nr_enqueuers; i++) {
 426                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 427                                      &count_enqueuer[2 * i]);
 428                 if (err != 0)
 429                         exit(1);
 430         }
 431         for (i = 0; i < nr_dequeuers; i++) {
 432                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 433                                      &count_dequeuer[2 * i]);
 434                 if (err != 0)
 435                         exit(1);
 436         }
 437
 438         cmm_smp_mb();
 439
 440         test_go = 1;
 441
 442         for (i = 0; i < duration; i++) {
 443                 sleep(1);
 444                 if (verbose_mode)
 445                         write (1, ".", 1);
 446         }
 447
 448         test_stop = 1;
 449
 450         for (i = 0; i < nr_enqueuers; i++) {
 451                 err = pthread_join(tid_enqueuer[i], &tret);
 452                 if (err != 0)
 453                         exit(1);
 454                 tot_enqueues += count_enqueuer[2 * i];
 455                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 456         }
 457         for (i = 0; i < nr_dequeuers; i++) {
 458                 err = pthread_join(tid_dequeuer[i], &tret);
 459                 if (err != 0)
 460                         exit(1);
 461                 tot_dequeues += count_dequeuer[2 * i];
 462                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 463         }
 464
 465         test_end(&s, &end_dequeues);
 466
 467         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 468                        tot_enqueues, tot_dequeues);
 469         printf_verbose("total number of successful enqueues : %llu, "
 470                        "successful dequeues %llu\n",
 471                        tot_successful_enqueues, tot_successful_dequeues);
 472         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 473                 "nr_dequeuers %3u "
 474                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 475                 "successful enqueues %12llu successful dequeues %12llu "
 476                 "end_dequeues %llu nr_ops %12llu\n",
 477                 argv[0], duration, nr_enqueuers, wdelay,
 478                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 479                 tot_successful_enqueues,
 480                 tot_successful_dequeues, end_dequeues,
 481                 tot_enqueues + tot_dequeues);
 482         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 483                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 484                        "succ. dequeues + end dequeues %llu.\n",
 485                        tot_successful_enqueues,
 486                        tot_successful_dequeues + end_dequeues);
 487
 488         free(count_enqueuer);
 489         free(count_dequeuer);
 490         free(tid_enqueuer);
 491         free(tid_dequeuer);
 492         return 0;
 493 }