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 static 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", (unsigned long) pthread_self(),
 181                         (unsigned long) gettid());
 182
 183         set_affinity();
 184
 185         while (!test_go)
 186         {
 187         }
 188         cmm_smp_mb();
 189
 190         for (;;) {
 191                 struct cds_wfs_node *node = malloc(sizeof(*node));
 192                 if (!node)
 193                         goto fail;
 194                 cds_wfs_node_init(node);
 195                 cds_wfs_push(&s, node);
 196                 URCU_TLS(nr_successful_enqueues)++;
 197
 198                 if (caa_unlikely(wdelay))
 199                         loop_sleep(wdelay);
 200 fail:
 201                 URCU_TLS(nr_enqueues)++;
 202                 if (caa_unlikely(!test_duration_enqueue()))
 203                         break;
 204         }
 205
 206         count[0] = URCU_TLS(nr_enqueues);
 207         count[1] = URCU_TLS(nr_successful_enqueues);
 208         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 209                        "enqueues %llu successful_enqueues %llu\n",
 210                        pthread_self(),
 211                         (unsigned long) gettid(),
 212                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 213         return ((void*)1);
 214
 215 }
 216
 217 static void do_test_pop(enum test_sync sync)
 218 {
 219         struct cds_wfs_node *node;
 220
 221         if (sync == TEST_SYNC_MUTEX)
 222                 cds_wfs_pop_lock(&s);
 223         node = __cds_wfs_pop_blocking(&s);
 224         if (sync == TEST_SYNC_MUTEX)
 225                 cds_wfs_pop_unlock(&s);
 226
 227         if (node) {
 228                 free(node);
 229                 URCU_TLS(nr_successful_dequeues)++;
 230         }
 231         URCU_TLS(nr_dequeues)++;
 232 }
 233
 234 static void do_test_pop_all(enum test_sync sync)
 235 {
 236         struct cds_wfs_head *head;
 237         struct cds_wfs_node *node, *n;
 238
 239         if (sync == TEST_SYNC_MUTEX)
 240                 cds_wfs_pop_lock(&s);
 241         head = __cds_wfs_pop_all(&s);
 242         if (sync == TEST_SYNC_MUTEX)
 243                 cds_wfs_pop_unlock(&s);
 244
 245         cds_wfs_for_each_blocking_safe(head, node, n) {
 246                 free(node);
 247                 URCU_TLS(nr_successful_dequeues)++;
 248                 URCU_TLS(nr_dequeues)++;
 249         }
 250 }
 251
 252 static void *thr_dequeuer(void *_count)
 253 {
 254         unsigned long long *count = _count;
 255         unsigned int counter;
 256
 257         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 258                         "dequeuer", (unsigned long) pthread_self(),
 259                         (unsigned long) gettid());
 260
 261         set_affinity();
 262
 263         while (!test_go)
 264         {
 265         }
 266         cmm_smp_mb();
 267
 268         assert(test_pop || test_pop_all);
 269
 270         for (;;) {
 271                 if (test_pop && test_pop_all) {
 272                         if (counter & 1)
 273                                 do_test_pop(test_sync);
 274                         else
 275                                 do_test_pop_all(test_sync);
 276                         counter++;
 277                 } else {
 278                         if (test_pop)
 279                                 do_test_pop(test_sync);
 280                         else
 281                                 do_test_pop_all(test_sync);
 282                 }
 283
 284                 if (caa_unlikely(!test_duration_dequeue()))
 285                         break;
 286                 if (caa_unlikely(rduration))
 287                         loop_sleep(rduration);
 288         }
 289
 290         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 291                        "dequeues %llu, successful_dequeues %llu\n",
 292                        pthread_self(),
 293                         (unsigned long) gettid(),
 294                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 295         count[0] = URCU_TLS(nr_dequeues);
 296         count[1] = URCU_TLS(nr_successful_dequeues);
 297         return ((void*)2);
 298 }
 299
 300 static void test_end(struct cds_wfs_stack *s, unsigned long long *nr_dequeues)
 301 {
 302         struct cds_wfs_node *node;
 303
 304         do {
 305                 node = cds_wfs_pop_blocking(s);
 306                 if (node) {
 307                         free(node);
 308                         (*nr_dequeues)++;
 309                 }
 310         } while (node);
 311 }
 312
 313 static void show_usage(int argc, char **argv)
 314 {
 315         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 316         printf(" [-d delay] (enqueuer period (in loops))");
 317         printf(" [-c duration] (dequeuer period (in loops))");
 318         printf(" [-v] (verbose output)");
 319         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 320         printf(" [-p] (test pop)");
 321         printf(" [-P] (test pop_all, enabled by default)");
 322         printf(" [-M] (use mutex external synchronization)");
 323         printf("      Note: default: no external synchronization used.");
 324         printf("\n");
 325 }
 326
 327 int main(int argc, char **argv)
 328 {
 329         int err;
 330         pthread_t *tid_enqueuer, *tid_dequeuer;
 331         void *tret;
 332         unsigned long long *count_enqueuer, *count_dequeuer;
 333         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 334         unsigned long long tot_successful_enqueues = 0,
 335                            tot_successful_dequeues = 0;
 336         unsigned long long end_dequeues = 0;
 337         int i, a;
 338
 339         if (argc < 4) {
 340                 show_usage(argc, argv);
 341                 return -1;
 342         }
 343
 344         err = sscanf(argv[1], "%u", &nr_dequeuers);
 345         if (err != 1) {
 346                 show_usage(argc, argv);
 347                 return -1;
 348         }
 349
 350         err = sscanf(argv[2], "%u", &nr_enqueuers);
 351         if (err != 1) {
 352                 show_usage(argc, argv);
 353                 return -1;
 354         }
 355
 356         err = sscanf(argv[3], "%lu", &duration);
 357         if (err != 1) {
 358                 show_usage(argc, argv);
 359                 return -1;
 360         }
 361
 362         for (i = 4; i < argc; i++) {
 363                 if (argv[i][0] != '-')
 364                         continue;
 365                 switch (argv[i][1]) {
 366                 case 'a':
 367                         if (argc < i + 2) {
 368                                 show_usage(argc, argv);
 369                                 return -1;
 370                         }
 371                         a = atoi(argv[++i]);
 372                         cpu_affinities[next_aff++] = a;
 373                         use_affinity = 1;
 374                         printf_verbose("Adding CPU %d affinity\n", a);
 375                         break;
 376                 case 'c':
 377                         if (argc < i + 2) {
 378                                 show_usage(argc, argv);
 379                                 return -1;
 380                         }
 381                         rduration = atol(argv[++i]);
 382                         break;
 383                 case 'd':
 384                         if (argc < i + 2) {
 385                                 show_usage(argc, argv);
 386                                 return -1;
 387                         }
 388                         wdelay = atol(argv[++i]);
 389                         break;
 390                 case 'v':
 391                         verbose_mode = 1;
 392                         break;
 393                 case 'p':
 394                         test_pop = 1;
 395                         break;
 396                 case 'P':
 397                         test_pop_all = 1;
 398                         break;
 399                 case 'M':
 400                         test_sync = TEST_SYNC_MUTEX;
 401                         break;
 402                 }
 403         }
 404
 405         /* activate pop_all test by default */
 406         if (!test_pop && !test_pop_all)
 407                 test_pop_all = 1;
 408
 409         printf_verbose("running test for %lu seconds, %u enqueuers, "
 410                        "%u dequeuers.\n",
 411                        duration, nr_enqueuers, nr_dequeuers);
 412         if (test_pop)
 413                 printf_verbose("pop test activated.\n");
 414         if (test_pop_all)
 415                 printf_verbose("pop_all test activated.\n");
 416         if (test_sync == TEST_SYNC_MUTEX)
 417                 printf_verbose("External sync: mutex.\n");
 418         else
 419                 printf_verbose("External sync: none.\n");
 420         printf_verbose("Writer delay : %lu loops.\n", rduration);
 421         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 422         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 423                         "main", (unsigned long) pthread_self(),
 424                         (unsigned long) gettid());
 425
 426         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 427         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 428         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 429         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 430         cds_wfs_init(&s);
 431
 432         next_aff = 0;
 433
 434         for (i = 0; i < nr_enqueuers; i++) {
 435                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 436                                      &count_enqueuer[2 * i]);
 437                 if (err != 0)
 438                         exit(1);
 439         }
 440         for (i = 0; i < nr_dequeuers; i++) {
 441                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 442                                      &count_dequeuer[2 * i]);
 443                 if (err != 0)
 444                         exit(1);
 445         }
 446
 447         cmm_smp_mb();
 448
 449         test_go = 1;
 450
 451         for (i = 0; i < duration; i++) {
 452                 sleep(1);
 453                 if (verbose_mode)
 454                         write (1, ".", 1);
 455         }
 456
 457         test_stop = 1;
 458
 459         for (i = 0; i < nr_enqueuers; i++) {
 460                 err = pthread_join(tid_enqueuer[i], &tret);
 461                 if (err != 0)
 462                         exit(1);
 463                 tot_enqueues += count_enqueuer[2 * i];
 464                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 465         }
 466         for (i = 0; i < nr_dequeuers; i++) {
 467                 err = pthread_join(tid_dequeuer[i], &tret);
 468                 if (err != 0)
 469                         exit(1);
 470                 tot_dequeues += count_dequeuer[2 * i];
 471                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 472         }
 473
 474         test_end(&s, &end_dequeues);
 475
 476         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 477                        tot_enqueues, tot_dequeues);
 478         printf_verbose("total number of successful enqueues : %llu, "
 479                        "successful dequeues %llu\n",
 480                        tot_successful_enqueues, tot_successful_dequeues);
 481         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 482                 "nr_dequeuers %3u "
 483                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 484                 "successful enqueues %12llu successful dequeues %12llu "
 485                 "end_dequeues %llu nr_ops %12llu\n",
 486                 argv[0], duration, nr_enqueuers, wdelay,
 487                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 488                 tot_successful_enqueues,
 489                 tot_successful_dequeues, end_dequeues,
 490                 tot_enqueues + tot_dequeues);
 491         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 492                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 493                        "succ. dequeues + end dequeues %llu.\n",
 494                        tot_successful_enqueues,
 495                        tot_successful_dequeues + end_dequeues);
 496
 497         free(count_enqueuer);
 498         free(count_dequeuer);
 499         free(tid_enqueuer);
 500         free(tid_dequeuer);
 501         return 0;
 502 }