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", pthread_self(), (unsigned long)gettid());
 186
 187         set_affinity();
 188
 189         rcu_register_thread();
 190
 191         while (!test_go)
 192         {
 193         }
 194         cmm_smp_mb();
 195
 196         for (;;) {
 197                 struct test *node = malloc(sizeof(*node));
 198                 if (!node)
 199                         goto fail;
 200                 cds_lfs_node_init(&node->list);
 201                 cds_lfs_push(&s, &node->list);
 202                 URCU_TLS(nr_successful_enqueues)++;
 203
 204                 if (caa_unlikely(wdelay))
 205                         loop_sleep(wdelay);
 206 fail:
 207                 URCU_TLS(nr_enqueues)++;
 208                 if (caa_unlikely(!test_duration_enqueue()))
 209                         break;
 210         }
 211
 212         rcu_unregister_thread();
 213
 214         count[0] = URCU_TLS(nr_enqueues);
 215         count[1] = URCU_TLS(nr_successful_enqueues);
 216         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 217                        "enqueues %llu successful_enqueues %llu\n",
 218                        pthread_self(), (unsigned long)gettid(),
 219                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 220         return ((void*)1);
 221
 222 }
 223
 224 static
 225 void free_node_cb(struct rcu_head *head)
 226 {
 227         struct test *node =
 228                 caa_container_of(head, struct test, rcu);
 229         free(node);
 230 }
 231
 232 static
 233 void do_test_pop(enum test_sync sync)
 234 {
 235         struct cds_lfs_node *snode;
 236
 237         if (sync == TEST_SYNC_RCU)
 238                 rcu_read_lock();
 239         snode = __cds_lfs_pop(&s);
 240         if (sync == TEST_SYNC_RCU)
 241                 rcu_read_unlock();
 242         if (snode) {
 243                 struct test *node;
 244
 245                 node = caa_container_of(snode,
 246                         struct test, list);
 247                 if (sync == TEST_SYNC_RCU)
 248                         call_rcu(&node->rcu, free_node_cb);
 249                 else
 250                         free(node);
 251                 URCU_TLS(nr_successful_dequeues)++;
 252         }
 253         URCU_TLS(nr_dequeues)++;
 254 }
 255
 256 static
 257 void do_test_pop_all(enum test_sync sync)
 258 {
 259         struct cds_lfs_node *snode;
 260         struct cds_lfs_head *head;
 261         struct cds_lfs_node *n;
 262
 263         head = __cds_lfs_pop_all(&s);
 264         cds_lfs_for_each_safe(head, snode, n) {
 265                 struct test *node;
 266
 267                 node = caa_container_of(snode, struct test, list);
 268                 if (sync == TEST_SYNC_RCU)
 269                         call_rcu(&node->rcu, free_node_cb);
 270                 else
 271                         free(node);
 272                 URCU_TLS(nr_successful_dequeues)++;
 273                 URCU_TLS(nr_dequeues)++;
 274         }
 275
 276 }
 277
 278 static void *thr_dequeuer(void *_count)
 279 {
 280         unsigned long long *count = _count;
 281
 282         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 283                         "dequeuer", pthread_self(), (unsigned long)gettid());
 284
 285         set_affinity();
 286
 287         rcu_register_thread();
 288
 289         while (!test_go)
 290         {
 291         }
 292         cmm_smp_mb();
 293
 294         assert(test_pop || test_pop_all);
 295
 296         for (;;) {
 297                 unsigned int counter = 0;
 298
 299                 if (test_pop && test_pop_all) {
 300                         /* both pop and pop all */
 301                         if (counter & 1)
 302                                 do_test_pop(test_sync);
 303                         else
 304                                 do_test_pop_all(test_sync);
 305                         counter++;
 306                 } else {
 307                         if (test_pop)
 308                                 do_test_pop(test_sync);
 309                         else
 310                                 do_test_pop_all(test_sync);
 311                 }
 312
 313                 if (caa_unlikely(!test_duration_dequeue()))
 314                         break;
 315                 if (caa_unlikely(rduration))
 316                         loop_sleep(rduration);
 317         }
 318
 319         rcu_unregister_thread();
 320
 321         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 322                        "dequeues %llu, successful_dequeues %llu\n",
 323                        pthread_self(), (unsigned long)gettid(),
 324                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 325         count[0] = URCU_TLS(nr_dequeues);
 326         count[1] = URCU_TLS(nr_successful_dequeues);
 327         return ((void*)2);
 328 }
 329
 330 static void test_end(struct cds_lfs_stack *s, unsigned long long *nr_dequeues)
 331 {
 332         struct cds_lfs_node *snode;
 333
 334         do {
 335                 snode = __cds_lfs_pop(s);
 336                 if (snode) {
 337                         struct test *node;
 338
 339                         node = caa_container_of(snode, struct test, list);
 340                         free(node);
 341                         (*nr_dequeues)++;
 342                 }
 343         } while (snode);
 344 }
 345
 346 static void show_usage(int argc, char **argv)
 347 {
 348         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 349         printf(" [-d delay] (enqueuer period (in loops))");
 350         printf(" [-c duration] (dequeuer period (in loops))");
 351         printf(" [-v] (verbose output)");
 352         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 353         printf(" [-p] (test pop)");
 354         printf(" [-P] (test pop_all, enabled by default)");
 355         printf(" [-R] (use RCU external synchronization)");
 356         printf("      Note: default: no external synchronization used.");
 357         printf("\n");
 358 }
 359
 360 int main(int argc, char **argv)
 361 {
 362         int err;
 363         pthread_t *tid_enqueuer, *tid_dequeuer;
 364         void *tret;
 365         unsigned long long *count_enqueuer, *count_dequeuer;
 366         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 367         unsigned long long tot_successful_enqueues = 0,
 368                            tot_successful_dequeues = 0;
 369         unsigned long long end_dequeues = 0;
 370         int i, a;
 371
 372         if (argc < 4) {
 373                 show_usage(argc, argv);
 374                 return -1;
 375         }
 376
 377         err = sscanf(argv[1], "%u", &nr_dequeuers);
 378         if (err != 1) {
 379                 show_usage(argc, argv);
 380                 return -1;
 381         }
 382
 383         err = sscanf(argv[2], "%u", &nr_enqueuers);
 384         if (err != 1) {
 385                 show_usage(argc, argv);
 386                 return -1;
 387         }
 388
 389         err = sscanf(argv[3], "%lu", &duration);
 390         if (err != 1) {
 391                 show_usage(argc, argv);
 392                 return -1;
 393         }
 394
 395         for (i = 4; i < argc; i++) {
 396                 if (argv[i][0] != '-')
 397                         continue;
 398                 switch (argv[i][1]) {
 399                 case 'a':
 400                         if (argc < i + 2) {
 401                                 show_usage(argc, argv);
 402                                 return -1;
 403                         }
 404                         a = atoi(argv[++i]);
 405                         cpu_affinities[next_aff++] = a;
 406                         use_affinity = 1;
 407                         printf_verbose("Adding CPU %d affinity\n", a);
 408                         break;
 409                 case 'c':
 410                         if (argc < i + 2) {
 411                                 show_usage(argc, argv);
 412                                 return -1;
 413                         }
 414                         rduration = atol(argv[++i]);
 415                         break;
 416                 case 'd':
 417                         if (argc < i + 2) {
 418                                 show_usage(argc, argv);
 419                                 return -1;
 420                         }
 421                         wdelay = atol(argv[++i]);
 422                         break;
 423                 case 'v':
 424                         verbose_mode = 1;
 425                         break;
 426                 case 'p':
 427                         test_pop = 1;
 428                         break;
 429                 case 'P':
 430                         test_pop_all = 1;
 431                         break;
 432                 case 'R':
 433                         test_sync = TEST_SYNC_RCU;
 434                         break;
 435                 }
 436         }
 437
 438         /* activate pop_all test by default */
 439         if (!test_pop && !test_pop_all)
 440                 test_pop_all = 1;
 441
 442         printf_verbose("running test for %lu seconds, %u enqueuers, "
 443                        "%u dequeuers.\n",
 444                        duration, nr_enqueuers, nr_dequeuers);
 445         if (test_pop)
 446                 printf_verbose("pop test activated.\n");
 447         if (test_pop_all)
 448                 printf_verbose("pop_all test activated.\n");
 449         if (test_sync == TEST_SYNC_RCU)
 450                 printf_verbose("External sync: RCU.\n");
 451         else
 452                 printf_verbose("External sync: none.\n");
 453         printf_verbose("Writer delay : %lu loops.\n", rduration);
 454         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 455         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 456                         "main", pthread_self(), (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 }