tests/test_urcu_wfcq.c

   1 /*
   2  * test_urcu_wfcq.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free concurrent queue
   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/wfcqueue.h>
  70
  71 enum test_sync {
  72         TEST_SYNC_MUTEX = 0,
  73         TEST_SYNC_NONE,
  74 };
  75
  76 static enum test_sync test_sync;
  77
  78 static volatile int test_go, test_stop;
  79
  80 static unsigned long rduration;
  81
  82 static unsigned long duration;
  83
  84 /* read-side C.S. duration, in loops */
  85 static unsigned long wdelay;
  86
  87 static inline void loop_sleep(unsigned long loops)
  88 {
  89         while (loops-- != 0)
  90                 caa_cpu_relax();
  91 }
  92
  93 static int verbose_mode;
  94
  95 static int test_dequeue, test_splice;
  96
  97 #define printf_verbose(fmt, args...)            \
  98         do {                                    \
  99                 if (verbose_mode)               \
 100                         printf(fmt, ## args);   \
 101         } while (0)
 102
 103 static unsigned int cpu_affinities[NR_CPUS];
 104 static unsigned int next_aff = 0;
 105 static int use_affinity = 0;
 106
 107 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 108
 109 #ifndef HAVE_CPU_SET_T
 110 typedef unsigned long cpu_set_t;
 111 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 112 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 113 #endif
 114
 115 static void set_affinity(void)
 116 {
 117 #if HAVE_SCHED_SETAFFINITY
 118         cpu_set_t mask;
 119         int cpu, ret;
 120 #endif /* HAVE_SCHED_SETAFFINITY */
 121
 122         if (!use_affinity)
 123                 return;
 124
 125 #if HAVE_SCHED_SETAFFINITY
 126         ret = pthread_mutex_lock(&affinity_mutex);
 127         if (ret) {
 128                 perror("Error in pthread mutex lock");
 129                 exit(-1);
 130         }
 131         cpu = cpu_affinities[next_aff++];
 132         ret = pthread_mutex_unlock(&affinity_mutex);
 133         if (ret) {
 134                 perror("Error in pthread mutex unlock");
 135                 exit(-1);
 136         }
 137
 138         CPU_ZERO(&mask);
 139         CPU_SET(cpu, &mask);
 140 #if SCHED_SETAFFINITY_ARGS == 2
 141         sched_setaffinity(0, &mask);
 142 #else
 143         sched_setaffinity(0, sizeof(mask), &mask);
 144 #endif
 145 #endif /* HAVE_SCHED_SETAFFINITY */
 146 }
 147
 148 /*
 149  * returns 0 if test should end.
 150  */
 151 static int test_duration_dequeue(void)
 152 {
 153         return !test_stop;
 154 }
 155
 156 static int test_duration_enqueue(void)
 157 {
 158         return !test_stop;
 159 }
 160
 161 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 162 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 163
 164 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 165 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 166
 167 static unsigned int nr_enqueuers;
 168 static unsigned int nr_dequeuers;
 169
 170 static struct cds_wfcq_head __attribute__((aligned(CAA_CACHE_LINE_SIZE))) head;
 171 static struct cds_wfcq_tail __attribute__((aligned(CAA_CACHE_LINE_SIZE))) tail;
 172
 173 static void *thr_enqueuer(void *_count)
 174 {
 175         unsigned long long *count = _count;
 176
 177         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 178                         "enqueuer", (unsigned long) pthread_self(),
 179                         (unsigned long) gettid());
 180
 181         set_affinity();
 182
 183         while (!test_go)
 184         {
 185         }
 186         cmm_smp_mb();
 187
 188         for (;;) {
 189                 struct cds_wfcq_node *node = malloc(sizeof(*node));
 190                 if (!node)
 191                         goto fail;
 192                 cds_wfcq_node_init(node);
 193                 cds_wfcq_enqueue(&head, &tail, node);
 194                 URCU_TLS(nr_successful_enqueues)++;
 195
 196                 if (caa_unlikely(wdelay))
 197                         loop_sleep(wdelay);
 198 fail:
 199                 URCU_TLS(nr_enqueues)++;
 200                 if (caa_unlikely(!test_duration_enqueue()))
 201                         break;
 202         }
 203
 204         count[0] = URCU_TLS(nr_enqueues);
 205         count[1] = URCU_TLS(nr_successful_enqueues);
 206         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 207                        "enqueues %llu successful_enqueues %llu\n",
 208                        pthread_self(),
 209                         (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_dequeue(enum test_sync sync)
 216 {
 217         struct cds_wfcq_node *node;
 218
 219         if (sync == TEST_SYNC_MUTEX)
 220                 node = cds_wfcq_dequeue_blocking(&head, &tail);
 221         else
 222                 node = __cds_wfcq_dequeue_blocking(&head, &tail);
 223
 224         if (node) {
 225                 free(node);
 226                 URCU_TLS(nr_successful_dequeues)++;
 227         }
 228         URCU_TLS(nr_dequeues)++;
 229 }
 230
 231 static void do_test_splice(enum test_sync sync)
 232 {
 233         struct cds_wfcq_head tmp_head;
 234         struct cds_wfcq_tail tmp_tail;
 235         struct cds_wfcq_node *node, *n;
 236
 237         cds_wfcq_init(&tmp_head, &tmp_tail);
 238
 239         if (sync == TEST_SYNC_MUTEX)
 240                 cds_wfcq_splice_blocking(&tmp_head, &tmp_tail,
 241                         &head, &tail);
 242         else
 243                 __cds_wfcq_splice_blocking(&tmp_head, &tmp_tail,
 244                         &head, &tail);
 245
 246         __cds_wfcq_for_each_blocking_safe(&tmp_head, &tmp_tail, node, n) {
 247                 free(node);
 248                 URCU_TLS(nr_successful_dequeues)++;
 249                 URCU_TLS(nr_dequeues)++;
 250         }
 251 }
 252
 253 static void *thr_dequeuer(void *_count)
 254 {
 255         unsigned long long *count = _count;
 256         unsigned int counter;
 257
 258         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 259                         "dequeuer", (unsigned long) pthread_self(),
 260                         (unsigned long) gettid());
 261
 262         set_affinity();
 263
 264         while (!test_go)
 265         {
 266         }
 267         cmm_smp_mb();
 268
 269         for (;;) {
 270                 if (test_dequeue && test_splice) {
 271                         if (counter & 1)
 272                                 do_test_dequeue(test_sync);
 273                         else
 274                                 do_test_splice(test_sync);
 275                         counter++;
 276                 } else {
 277                         if (test_dequeue)
 278                                 do_test_dequeue(test_sync);
 279                         else
 280                                 do_test_splice(test_sync);
 281                 }
 282                 if (caa_unlikely(!test_duration_dequeue()))
 283                         break;
 284                 if (caa_unlikely(rduration))
 285                         loop_sleep(rduration);
 286         }
 287
 288         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 289                        "dequeues %llu, successful_dequeues %llu\n",
 290                        pthread_self(),
 291                         (unsigned long) gettid(),
 292                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 293         count[0] = URCU_TLS(nr_dequeues);
 294         count[1] = URCU_TLS(nr_successful_dequeues);
 295         return ((void*)2);
 296 }
 297
 298 static void test_end(unsigned long long *nr_dequeues)
 299 {
 300         struct cds_wfcq_node *node;
 301
 302         do {
 303                 node = cds_wfcq_dequeue_blocking(&head, &tail);
 304                 if (node) {
 305                         free(node);
 306                         (*nr_dequeues)++;
 307                 }
 308         } while (node);
 309 }
 310
 311 static void show_usage(int argc, char **argv)
 312 {
 313         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 314         printf(" [-d delay] (enqueuer period (in loops))");
 315         printf(" [-c duration] (dequeuer period (in loops))");
 316         printf(" [-v] (verbose output)");
 317         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 318         printf(" [-q] (test dequeue)");
 319         printf(" [-s] (test splice, enabled by default)");
 320         printf(" [-M] (use mutex external synchronization)");
 321         printf(" [-0] (use no external synchronization)");
 322         printf("      Note: default: mutex external synchronization used.");
 323         printf("\n");
 324 }
 325
 326 int main(int argc, char **argv)
 327 {
 328         int err;
 329         pthread_t *tid_enqueuer, *tid_dequeuer;
 330         void *tret;
 331         unsigned long long *count_enqueuer, *count_dequeuer;
 332         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 333         unsigned long long tot_successful_enqueues = 0,
 334                            tot_successful_dequeues = 0;
 335         unsigned long long end_dequeues = 0;
 336         int i, a;
 337
 338         if (argc < 4) {
 339                 show_usage(argc, argv);
 340                 return -1;
 341         }
 342
 343         err = sscanf(argv[1], "%u", &nr_dequeuers);
 344         if (err != 1) {
 345                 show_usage(argc, argv);
 346                 return -1;
 347         }
 348
 349         err = sscanf(argv[2], "%u", &nr_enqueuers);
 350         if (err != 1) {
 351                 show_usage(argc, argv);
 352                 return -1;
 353         }
 354
 355         err = sscanf(argv[3], "%lu", &duration);
 356         if (err != 1) {
 357                 show_usage(argc, argv);
 358                 return -1;
 359         }
 360
 361         for (i = 4; i < argc; i++) {
 362                 if (argv[i][0] != '-')
 363                         continue;
 364                 switch (argv[i][1]) {
 365                 case 'a':
 366                         if (argc < i + 2) {
 367                                 show_usage(argc, argv);
 368                                 return -1;
 369                         }
 370                         a = atoi(argv[++i]);
 371                         cpu_affinities[next_aff++] = a;
 372                         use_affinity = 1;
 373                         printf_verbose("Adding CPU %d affinity\n", a);
 374                         break;
 375                 case 'c':
 376                         if (argc < i + 2) {
 377                                 show_usage(argc, argv);
 378                                 return -1;
 379                         }
 380                         rduration = atol(argv[++i]);
 381                         break;
 382                 case 'd':
 383                         if (argc < i + 2) {
 384                                 show_usage(argc, argv);
 385                                 return -1;
 386                         }
 387                         wdelay = atol(argv[++i]);
 388                         break;
 389                 case 'v':
 390                         verbose_mode = 1;
 391                         break;
 392                 case 'q':
 393                         test_dequeue = 1;
 394                         break;
 395                 case 's':
 396                         test_splice = 1;
 397                         break;
 398                 case 'M':
 399                         test_sync = TEST_SYNC_MUTEX;
 400                         break;
 401                 case '0':
 402                         test_sync = TEST_SYNC_NONE;
 403                         break;
 404                 }
 405         }
 406
 407         /* activate splice test by default */
 408         if (!test_dequeue && !test_splice)
 409                 test_splice = 1;
 410
 411         printf_verbose("running test for %lu seconds, %u enqueuers, "
 412                        "%u dequeuers.\n",
 413                        duration, nr_enqueuers, nr_dequeuers);
 414         if (test_dequeue)
 415                 printf_verbose("dequeue test activated.\n");
 416         else
 417                 printf_verbose("splice test activated.\n");
 418         if (test_sync == TEST_SYNC_MUTEX)
 419                 printf_verbose("External sync: mutex.\n");
 420         else
 421                 printf_verbose("External sync: none.\n");
 422         printf_verbose("Writer delay : %lu loops.\n", rduration);
 423         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 424         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 425                         "main", (unsigned long) pthread_self(),
 426                         (unsigned long) gettid());
 427
 428         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 429         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 430         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 431         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 432         cds_wfcq_init(&head, &tail);
 433
 434         next_aff = 0;
 435
 436         for (i = 0; i < nr_enqueuers; i++) {
 437                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 438                                      &count_enqueuer[2 * i]);
 439                 if (err != 0)
 440                         exit(1);
 441         }
 442         for (i = 0; i < nr_dequeuers; i++) {
 443                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 444                                      &count_dequeuer[2 * i]);
 445                 if (err != 0)
 446                         exit(1);
 447         }
 448
 449         cmm_smp_mb();
 450
 451         test_go = 1;
 452
 453         for (i = 0; i < duration; i++) {
 454                 sleep(1);
 455                 if (verbose_mode)
 456                         write (1, ".", 1);
 457         }
 458
 459         test_stop = 1;
 460
 461         for (i = 0; i < nr_enqueuers; i++) {
 462                 err = pthread_join(tid_enqueuer[i], &tret);
 463                 if (err != 0)
 464                         exit(1);
 465                 tot_enqueues += count_enqueuer[2 * i];
 466                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 467         }
 468         for (i = 0; i < nr_dequeuers; i++) {
 469                 err = pthread_join(tid_dequeuer[i], &tret);
 470                 if (err != 0)
 471                         exit(1);
 472                 tot_dequeues += count_dequeuer[2 * i];
 473                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 474         }
 475
 476         test_end(&end_dequeues);
 477
 478         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 479                        tot_enqueues, tot_dequeues);
 480         printf_verbose("total number of successful enqueues : %llu, "
 481                        "successful dequeues %llu\n",
 482                        tot_successful_enqueues, tot_successful_dequeues);
 483         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 484                 "nr_dequeuers %3u "
 485                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 486                 "successful enqueues %12llu successful dequeues %12llu "
 487                 "end_dequeues %llu nr_ops %12llu\n",
 488                 argv[0], duration, nr_enqueuers, wdelay,
 489                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 490                 tot_successful_enqueues,
 491                 tot_successful_dequeues, end_dequeues,
 492                 tot_enqueues + tot_dequeues);
 493         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 494                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 495                        "succ. dequeues + end dequeues %llu.\n",
 496                        tot_successful_enqueues,
 497                        tot_successful_dequeues + end_dequeues);
 498
 499         free(count_enqueuer);
 500         free(count_dequeuer);
 501         free(tid_enqueuer);
 502         free(tid_dequeuer);
 503         return 0;
 504 }