tests/test_urcu_lfs.c

   1 /*
   2  * test_urcu_lfs.c
   3  *
   4  * Userspace RCU library - example lock-free stack
   5  *
   6  * Copyright 2010-2012 - 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
  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/cds.h>
  70
  71 /*
  72  * External synchronization used.
  73  */
  74 enum test_sync {
  75         TEST_SYNC_NONE = 0,
  76         TEST_SYNC_RCU,
  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 static void set_affinity(void)
 113 {
 114 #if HAVE_SCHED_SETAFFINITY
 115         cpu_set_t mask;
 116         int cpu, ret;
 117 #endif /* HAVE_SCHED_SETAFFINITY */
 118
 119         if (!use_affinity)
 120                 return;
 121
 122 #if HAVE_SCHED_SETAFFINITY
 123         ret = pthread_mutex_lock(&affinity_mutex);
 124         if (ret) {
 125                 perror("Error in pthread mutex lock");
 126                 exit(-1);
 127         }
 128         cpu = cpu_affinities[next_aff++];
 129         ret = pthread_mutex_unlock(&affinity_mutex);
 130         if (ret) {
 131                 perror("Error in pthread mutex unlock");
 132                 exit(-1);
 133         }
 134
 135         CPU_ZERO(&mask);
 136         CPU_SET(cpu, &mask);
 137 #if SCHED_SETAFFINITY_ARGS == 2
 138         sched_setaffinity(0, &mask);
 139 #else
 140         sched_setaffinity(0, sizeof(mask), &mask);
 141 #endif
 142 #endif /* HAVE_SCHED_SETAFFINITY */
 143 }
 144
 145 /*
 146  * returns 0 if test should end.
 147  */
 148 static int test_duration_dequeue(void)
 149 {
 150         return !test_stop;
 151 }
 152
 153 static int test_duration_enqueue(void)
 154 {
 155         return !test_stop;
 156 }
 157
 158 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 159 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 160
 161 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 162 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 163
 164 static unsigned int nr_enqueuers;
 165 static unsigned int nr_dequeuers;
 166
 167 struct test {
 168         struct cds_lfs_node list;
 169         struct rcu_head rcu;
 170 };
 171
 172 static struct cds_lfs_stack s;
 173
 174 static void *thr_enqueuer(void *_count)
 175 {
 176         unsigned long long *count = _count;
 177
 178         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 179                         "enqueuer", (unsigned long) pthread_self(),
 180                         (unsigned long) gettid());
 181
 182         set_affinity();
 183
 184         rcu_register_thread();
 185
 186         while (!test_go)
 187         {
 188         }
 189         cmm_smp_mb();
 190
 191         for (;;) {
 192                 struct test *node = malloc(sizeof(*node));
 193                 if (!node)
 194                         goto fail;
 195                 cds_lfs_node_init(&node->list);
 196                 cds_lfs_push(&s, &node->list);
 197                 URCU_TLS(nr_successful_enqueues)++;
 198
 199                 if (caa_unlikely(wdelay))
 200                         loop_sleep(wdelay);
 201 fail:
 202                 URCU_TLS(nr_enqueues)++;
 203                 if (caa_unlikely(!test_duration_enqueue()))
 204                         break;
 205         }
 206
 207         rcu_unregister_thread();
 208
 209         count[0] = URCU_TLS(nr_enqueues);
 210         count[1] = URCU_TLS(nr_successful_enqueues);
 211         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 212                        "enqueues %llu successful_enqueues %llu\n",
 213                        (unsigned long) pthread_self(),
 214                         (unsigned long) gettid(),
 215                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 216         return ((void*)1);
 217
 218 }
 219
 220 static
 221 void free_node_cb(struct rcu_head *head)
 222 {
 223         struct test *node =
 224                 caa_container_of(head, struct test, rcu);
 225         free(node);
 226 }
 227
 228 static
 229 void do_test_pop(enum test_sync sync)
 230 {
 231         struct cds_lfs_node *snode;
 232
 233         if (sync == TEST_SYNC_RCU)
 234                 rcu_read_lock();
 235         snode = __cds_lfs_pop(&s);
 236         if (sync == TEST_SYNC_RCU)
 237                 rcu_read_unlock();
 238         if (snode) {
 239                 struct test *node;
 240
 241                 node = caa_container_of(snode,
 242                         struct test, list);
 243                 if (sync == TEST_SYNC_RCU)
 244                         call_rcu(&node->rcu, free_node_cb);
 245                 else
 246                         free(node);
 247                 URCU_TLS(nr_successful_dequeues)++;
 248         }
 249         URCU_TLS(nr_dequeues)++;
 250 }
 251
 252 static
 253 void do_test_pop_all(enum test_sync sync)
 254 {
 255         struct cds_lfs_node *snode;
 256         struct cds_lfs_head *head;
 257         struct cds_lfs_node *n;
 258
 259         head = __cds_lfs_pop_all(&s);
 260         cds_lfs_for_each_safe(head, snode, n) {
 261                 struct test *node;
 262
 263                 node = caa_container_of(snode, struct test, list);
 264                 if (sync == TEST_SYNC_RCU)
 265                         call_rcu(&node->rcu, free_node_cb);
 266                 else
 267                         free(node);
 268                 URCU_TLS(nr_successful_dequeues)++;
 269                 URCU_TLS(nr_dequeues)++;
 270         }
 271
 272 }
 273
 274 static void *thr_dequeuer(void *_count)
 275 {
 276         unsigned long long *count = _count;
 277         unsigned int counter = 0;
 278
 279         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 280                         "dequeuer", (unsigned long) pthread_self(),
 281                         (unsigned long) gettid());
 282
 283         set_affinity();
 284
 285         rcu_register_thread();
 286
 287         while (!test_go)
 288         {
 289         }
 290         cmm_smp_mb();
 291
 292         assert(test_pop || test_pop_all);
 293
 294         for (;;) {
 295                 if (test_pop && test_pop_all) {
 296                         /* both pop and pop all */
 297                         if (counter & 1)
 298                                 do_test_pop(test_sync);
 299                         else
 300                                 do_test_pop_all(test_sync);
 301                         counter++;
 302                 } else {
 303                         if (test_pop)
 304                                 do_test_pop(test_sync);
 305                         else
 306                                 do_test_pop_all(test_sync);
 307                 }
 308
 309                 if (caa_unlikely(!test_duration_dequeue()))
 310                         break;
 311                 if (caa_unlikely(rduration))
 312                         loop_sleep(rduration);
 313         }
 314
 315         rcu_unregister_thread();
 316
 317         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 318                        "dequeues %llu, successful_dequeues %llu\n",
 319                        (unsigned long) pthread_self(),
 320                         (unsigned long) gettid(),
 321                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 322         count[0] = URCU_TLS(nr_dequeues);
 323         count[1] = URCU_TLS(nr_successful_dequeues);
 324         return ((void*)2);
 325 }
 326
 327 static void test_end(struct cds_lfs_stack *s, unsigned long long *nr_dequeues)
 328 {
 329         struct cds_lfs_node *snode;
 330
 331         do {
 332                 snode = __cds_lfs_pop(s);
 333                 if (snode) {
 334                         struct test *node;
 335
 336                         node = caa_container_of(snode, struct test, list);
 337                         free(node);
 338                         (*nr_dequeues)++;
 339                 }
 340         } while (snode);
 341 }
 342
 343 static void show_usage(int argc, char **argv)
 344 {
 345         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 346                 argv[0]);
 347         printf("OPTIONS:\n");
 348         printf("        [-d delay] (enqueuer period (in loops))\n");
 349         printf("        [-c duration] (dequeuer period (in loops))\n");
 350         printf("        [-v] (verbose output)\n");
 351         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 352         printf("        [-p] (test pop)\n");
 353         printf("        [-P] (test pop_all, enabled by default)\n");
 354         printf("        [-R] (use RCU external synchronization)\n");
 355         printf("                Note: default: no external synchronization used.\n");
 356         printf("\n");
 357 }
 358
 359 int main(int argc, char **argv)
 360 {
 361         int err;
 362         pthread_t *tid_enqueuer, *tid_dequeuer;
 363         void *tret;
 364         unsigned long long *count_enqueuer, *count_dequeuer;
 365         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 366         unsigned long long tot_successful_enqueues = 0,
 367                            tot_successful_dequeues = 0;
 368         unsigned long long end_dequeues = 0;
 369         int i, a;
 370
 371         if (argc < 4) {
 372                 show_usage(argc, argv);
 373                 return -1;
 374         }
 375
 376         err = sscanf(argv[1], "%u", &nr_dequeuers);
 377         if (err != 1) {
 378                 show_usage(argc, argv);
 379                 return -1;
 380         }
 381
 382         err = sscanf(argv[2], "%u", &nr_enqueuers);
 383         if (err != 1) {
 384                 show_usage(argc, argv);
 385                 return -1;
 386         }
 387
 388         err = sscanf(argv[3], "%lu", &duration);
 389         if (err != 1) {
 390                 show_usage(argc, argv);
 391                 return -1;
 392         }
 393
 394         for (i = 4; i < argc; i++) {
 395                 if (argv[i][0] != '-')
 396                         continue;
 397                 switch (argv[i][1]) {
 398                 case 'a':
 399                         if (argc < i + 2) {
 400                                 show_usage(argc, argv);
 401                                 return -1;
 402                         }
 403                         a = atoi(argv[++i]);
 404                         cpu_affinities[next_aff++] = a;
 405                         use_affinity = 1;
 406                         printf_verbose("Adding CPU %d affinity\n", a);
 407                         break;
 408                 case 'c':
 409                         if (argc < i + 2) {
 410                                 show_usage(argc, argv);
 411                                 return -1;
 412                         }
 413                         rduration = atol(argv[++i]);
 414                         break;
 415                 case 'd':
 416                         if (argc < i + 2) {
 417                                 show_usage(argc, argv);
 418                                 return -1;
 419                         }
 420                         wdelay = atol(argv[++i]);
 421                         break;
 422                 case 'v':
 423                         verbose_mode = 1;
 424                         break;
 425                 case 'p':
 426                         test_pop = 1;
 427                         break;
 428                 case 'P':
 429                         test_pop_all = 1;
 430                         break;
 431                 case 'R':
 432                         test_sync = TEST_SYNC_RCU;
 433                         break;
 434                 }
 435         }
 436
 437         /* activate pop_all test by default */
 438         if (!test_pop && !test_pop_all)
 439                 test_pop_all = 1;
 440
 441         printf_verbose("running test for %lu seconds, %u enqueuers, "
 442                        "%u dequeuers.\n",
 443                        duration, nr_enqueuers, nr_dequeuers);
 444         if (test_pop)
 445                 printf_verbose("pop test activated.\n");
 446         if (test_pop_all)
 447                 printf_verbose("pop_all test activated.\n");
 448         if (test_sync == TEST_SYNC_RCU)
 449                 printf_verbose("External sync: RCU.\n");
 450         else
 451                 printf_verbose("External sync: none.\n");
 452         printf_verbose("Writer delay : %lu loops.\n", rduration);
 453         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 454         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 455                         "main", (unsigned long) pthread_self(),
 456                         (unsigned long) gettid());
 457
 458         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 459         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 460         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 461         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 462         cds_lfs_init(&s);
 463         err = create_all_cpu_call_rcu_data(0);
 464         if (err) {
 465                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 466         }
 467
 468         next_aff = 0;
 469
 470         for (i = 0; i < nr_enqueuers; i++) {
 471                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 472                                      &count_enqueuer[2 * i]);
 473                 if (err != 0)
 474                         exit(1);
 475         }
 476         for (i = 0; i < nr_dequeuers; i++) {
 477                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 478                                      &count_dequeuer[2 * i]);
 479                 if (err != 0)
 480                         exit(1);
 481         }
 482
 483         cmm_smp_mb();
 484
 485         test_go = 1;
 486
 487         for (i = 0; i < duration; i++) {
 488                 sleep(1);
 489                 if (verbose_mode)
 490                         write (1, ".", 1);
 491         }
 492
 493         test_stop = 1;
 494
 495         for (i = 0; i < nr_enqueuers; i++) {
 496                 err = pthread_join(tid_enqueuer[i], &tret);
 497                 if (err != 0)
 498                         exit(1);
 499                 tot_enqueues += count_enqueuer[2 * i];
 500                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 501         }
 502         for (i = 0; i < nr_dequeuers; i++) {
 503                 err = pthread_join(tid_dequeuer[i], &tret);
 504                 if (err != 0)
 505                         exit(1);
 506                 tot_dequeues += count_dequeuer[2 * i];
 507                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 508         }
 509
 510         test_end(&s, &end_dequeues);
 511
 512         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 513                        tot_enqueues, tot_dequeues);
 514         printf_verbose("total number of successful enqueues : %llu, "
 515                        "successful dequeues %llu\n",
 516                        tot_successful_enqueues, tot_successful_dequeues);
 517         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 518                 "nr_dequeuers %3u "
 519                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 520                 "successful enqueues %12llu successful dequeues %12llu "
 521                 "end_dequeues %llu nr_ops %12llu\n",
 522                 argv[0], duration, nr_enqueuers, wdelay,
 523                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 524                 tot_successful_enqueues,
 525                 tot_successful_dequeues, end_dequeues,
 526                 tot_enqueues + tot_dequeues);
 527         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 528                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 529                        "succ. dequeues + end dequeues %llu.\n",
 530                        tot_successful_enqueues,
 531                        tot_successful_dequeues + end_dequeues);
 532
 533         free_all_cpu_call_rcu_data();
 534         free(count_enqueuer);
 535         free(count_dequeuer);
 536         free(tid_enqueuer);
 537         free(tid_dequeuer);
 538         return 0;
 539 }