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 <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/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 #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 struct test {
 174         struct cds_lfs_node list;
 175         struct rcu_head rcu;
 176 };
 177
 178 static struct cds_lfs_stack s;
 179
 180 static void *thr_enqueuer(void *_count)
 181 {
 182         unsigned long long *count = _count;
 183
 184         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 185                         "enqueuer", (unsigned long) pthread_self(),
 186                         (unsigned long) gettid());
 187
 188         set_affinity();
 189
 190         rcu_register_thread();
 191
 192         while (!test_go)
 193         {
 194         }
 195         cmm_smp_mb();
 196
 197         for (;;) {
 198                 struct test *node = malloc(sizeof(*node));
 199                 if (!node)
 200                         goto fail;
 201                 cds_lfs_node_init(&node->list);
 202                 cds_lfs_push(&s, &node->list);
 203                 URCU_TLS(nr_successful_enqueues)++;
 204
 205                 if (caa_unlikely(wdelay))
 206                         loop_sleep(wdelay);
 207 fail:
 208                 URCU_TLS(nr_enqueues)++;
 209                 if (caa_unlikely(!test_duration_enqueue()))
 210                         break;
 211         }
 212
 213         rcu_unregister_thread();
 214
 215         count[0] = URCU_TLS(nr_enqueues);
 216         count[1] = URCU_TLS(nr_successful_enqueues);
 217         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 218                        "enqueues %llu successful_enqueues %llu\n",
 219                        pthread_self(),
 220                         (unsigned long) gettid(),
 221                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 222         return ((void*)1);
 223
 224 }
 225
 226 static
 227 void free_node_cb(struct rcu_head *head)
 228 {
 229         struct test *node =
 230                 caa_container_of(head, struct test, rcu);
 231         free(node);
 232 }
 233
 234 static
 235 void do_test_pop(enum test_sync sync)
 236 {
 237         struct cds_lfs_node *snode;
 238
 239         if (sync == TEST_SYNC_RCU)
 240                 rcu_read_lock();
 241         snode = __cds_lfs_pop(&s);
 242         if (sync == TEST_SYNC_RCU)
 243                 rcu_read_unlock();
 244         if (snode) {
 245                 struct test *node;
 246
 247                 node = caa_container_of(snode,
 248                         struct test, list);
 249                 if (sync == TEST_SYNC_RCU)
 250                         call_rcu(&node->rcu, free_node_cb);
 251                 else
 252                         free(node);
 253                 URCU_TLS(nr_successful_dequeues)++;
 254         }
 255         URCU_TLS(nr_dequeues)++;
 256 }
 257
 258 static
 259 void do_test_pop_all(enum test_sync sync)
 260 {
 261         struct cds_lfs_node *snode;
 262         struct cds_lfs_head *head;
 263         struct cds_lfs_node *n;
 264
 265         head = __cds_lfs_pop_all(&s);
 266         cds_lfs_for_each_safe(head, snode, n) {
 267                 struct test *node;
 268
 269                 node = caa_container_of(snode, struct test, list);
 270                 if (sync == TEST_SYNC_RCU)
 271                         call_rcu(&node->rcu, free_node_cb);
 272                 else
 273                         free(node);
 274                 URCU_TLS(nr_successful_dequeues)++;
 275                 URCU_TLS(nr_dequeues)++;
 276         }
 277
 278 }
 279
 280 static void *thr_dequeuer(void *_count)
 281 {
 282         unsigned long long *count = _count;
 283
 284         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 285                         "dequeuer", (unsigned long) pthread_self(),
 286                         (unsigned long) gettid());
 287
 288         set_affinity();
 289
 290         rcu_register_thread();
 291
 292         while (!test_go)
 293         {
 294         }
 295         cmm_smp_mb();
 296
 297         assert(test_pop || test_pop_all);
 298
 299         for (;;) {
 300                 unsigned int counter = 0;
 301
 302                 if (test_pop && test_pop_all) {
 303                         /* both pop and pop all */
 304                         if (counter & 1)
 305                                 do_test_pop(test_sync);
 306                         else
 307                                 do_test_pop_all(test_sync);
 308                         counter++;
 309                 } else {
 310                         if (test_pop)
 311                                 do_test_pop(test_sync);
 312                         else
 313                                 do_test_pop_all(test_sync);
 314                 }
 315
 316                 if (caa_unlikely(!test_duration_dequeue()))
 317                         break;
 318                 if (caa_unlikely(rduration))
 319                         loop_sleep(rduration);
 320         }
 321
 322         rcu_unregister_thread();
 323
 324         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 325                        "dequeues %llu, successful_dequeues %llu\n",
 326                        pthread_self(),
 327                         (unsigned long) gettid(),
 328                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 329         count[0] = URCU_TLS(nr_dequeues);
 330         count[1] = URCU_TLS(nr_successful_dequeues);
 331         return ((void*)2);
 332 }
 333
 334 static void test_end(struct cds_lfs_stack *s, unsigned long long *nr_dequeues)
 335 {
 336         struct cds_lfs_node *snode;
 337
 338         do {
 339                 snode = __cds_lfs_pop(s);
 340                 if (snode) {
 341                         struct test *node;
 342
 343                         node = caa_container_of(snode, struct test, list);
 344                         free(node);
 345                         (*nr_dequeues)++;
 346                 }
 347         } while (snode);
 348 }
 349
 350 static void show_usage(int argc, char **argv)
 351 {
 352         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 353         printf(" [-d delay] (enqueuer period (in loops))");
 354         printf(" [-c duration] (dequeuer period (in loops))");
 355         printf(" [-v] (verbose output)");
 356         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 357         printf(" [-p] (test pop)");
 358         printf(" [-P] (test pop_all, enabled by default)");
 359         printf(" [-R] (use RCU external synchronization)");
 360         printf("      Note: default: no external synchronization used.");
 361         printf("\n");
 362 }
 363
 364 int main(int argc, char **argv)
 365 {
 366         int err;
 367         pthread_t *tid_enqueuer, *tid_dequeuer;
 368         void *tret;
 369         unsigned long long *count_enqueuer, *count_dequeuer;
 370         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 371         unsigned long long tot_successful_enqueues = 0,
 372                            tot_successful_dequeues = 0;
 373         unsigned long long end_dequeues = 0;
 374         int i, a;
 375
 376         if (argc < 4) {
 377                 show_usage(argc, argv);
 378                 return -1;
 379         }
 380
 381         err = sscanf(argv[1], "%u", &nr_dequeuers);
 382         if (err != 1) {
 383                 show_usage(argc, argv);
 384                 return -1;
 385         }
 386
 387         err = sscanf(argv[2], "%u", &nr_enqueuers);
 388         if (err != 1) {
 389                 show_usage(argc, argv);
 390                 return -1;
 391         }
 392
 393         err = sscanf(argv[3], "%lu", &duration);
 394         if (err != 1) {
 395                 show_usage(argc, argv);
 396                 return -1;
 397         }
 398
 399         for (i = 4; i < argc; i++) {
 400                 if (argv[i][0] != '-')
 401                         continue;
 402                 switch (argv[i][1]) {
 403                 case 'a':
 404                         if (argc < i + 2) {
 405                                 show_usage(argc, argv);
 406                                 return -1;
 407                         }
 408                         a = atoi(argv[++i]);
 409                         cpu_affinities[next_aff++] = a;
 410                         use_affinity = 1;
 411                         printf_verbose("Adding CPU %d affinity\n", a);
 412                         break;
 413                 case 'c':
 414                         if (argc < i + 2) {
 415                                 show_usage(argc, argv);
 416                                 return -1;
 417                         }
 418                         rduration = atol(argv[++i]);
 419                         break;
 420                 case 'd':
 421                         if (argc < i + 2) {
 422                                 show_usage(argc, argv);
 423                                 return -1;
 424                         }
 425                         wdelay = atol(argv[++i]);
 426                         break;
 427                 case 'v':
 428                         verbose_mode = 1;
 429                         break;
 430                 case 'p':
 431                         test_pop = 1;
 432                         break;
 433                 case 'P':
 434                         test_pop_all = 1;
 435                         break;
 436                 case 'R':
 437                         test_sync = TEST_SYNC_RCU;
 438                         break;
 439                 }
 440         }
 441
 442         /* activate pop_all test by default */
 443         if (!test_pop && !test_pop_all)
 444                 test_pop_all = 1;
 445
 446         printf_verbose("running test for %lu seconds, %u enqueuers, "
 447                        "%u dequeuers.\n",
 448                        duration, nr_enqueuers, nr_dequeuers);
 449         if (test_pop)
 450                 printf_verbose("pop test activated.\n");
 451         if (test_pop_all)
 452                 printf_verbose("pop_all test activated.\n");
 453         if (test_sync == TEST_SYNC_RCU)
 454                 printf_verbose("External sync: RCU.\n");
 455         else
 456                 printf_verbose("External sync: none.\n");
 457         printf_verbose("Writer delay : %lu loops.\n", rduration);
 458         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 459         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 460                         "main", (unsigned long) pthread_self(),
 461                         (unsigned long) gettid());
 462
 463         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 464         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 465         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 466         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 467         cds_lfs_init(&s);
 468         err = create_all_cpu_call_rcu_data(0);
 469         if (err) {
 470                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 471         }
 472
 473         next_aff = 0;
 474
 475         for (i = 0; i < nr_enqueuers; i++) {
 476                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 477                                      &count_enqueuer[2 * i]);
 478                 if (err != 0)
 479                         exit(1);
 480         }
 481         for (i = 0; i < nr_dequeuers; i++) {
 482                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 483                                      &count_dequeuer[2 * i]);
 484                 if (err != 0)
 485                         exit(1);
 486         }
 487
 488         cmm_smp_mb();
 489
 490         test_go = 1;
 491
 492         for (i = 0; i < duration; i++) {
 493                 sleep(1);
 494                 if (verbose_mode)
 495                         write (1, ".", 1);
 496         }
 497
 498         test_stop = 1;
 499
 500         for (i = 0; i < nr_enqueuers; i++) {
 501                 err = pthread_join(tid_enqueuer[i], &tret);
 502                 if (err != 0)
 503                         exit(1);
 504                 tot_enqueues += count_enqueuer[2 * i];
 505                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 506         }
 507         for (i = 0; i < nr_dequeuers; i++) {
 508                 err = pthread_join(tid_dequeuer[i], &tret);
 509                 if (err != 0)
 510                         exit(1);
 511                 tot_dequeues += count_dequeuer[2 * i];
 512                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 513         }
 514
 515         test_end(&s, &end_dequeues);
 516
 517         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 518                        tot_enqueues, tot_dequeues);
 519         printf_verbose("total number of successful enqueues : %llu, "
 520                        "successful dequeues %llu\n",
 521                        tot_successful_enqueues, tot_successful_dequeues);
 522         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 523                 "nr_dequeuers %3u "
 524                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 525                 "successful enqueues %12llu successful dequeues %12llu "
 526                 "end_dequeues %llu nr_ops %12llu\n",
 527                 argv[0], duration, nr_enqueuers, wdelay,
 528                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 529                 tot_successful_enqueues,
 530                 tot_successful_dequeues, end_dequeues,
 531                 tot_enqueues + tot_dequeues);
 532         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 533                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 534                        "succ. dequeues + end dequeues %llu.\n",
 535                        tot_successful_enqueues,
 536                        tot_successful_dequeues + end_dequeues);
 537
 538         free_all_cpu_call_rcu_data();
 539         free(count_enqueuer);
 540         free(count_dequeuer);
 541         free(tid_enqueuer);
 542         free(tid_dequeuer);
 543         return 0;
 544 }