tests/test_urcu_lfq.c

   1 /*
   2  * test_urcu_lfq.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free 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
  42 #ifdef __linux__
  43 #include <syscall.h>
  44 #endif
  45
  46 /* hardcoded number of CPUs */
  47 #define NR_CPUS 16384
  48
  49 #if defined(_syscall0)
  50 _syscall0(pid_t, gettid)
  51 #elif defined(__NR_gettid)
  52 static inline pid_t gettid(void)
  53 {
  54         return syscall(__NR_gettid);
  55 }
  56 #else
  57 #warning "use pid as tid"
  58 static inline pid_t gettid(void)
  59 {
  60         return getpid();
  61 }
  62 #endif
  63
  64 #ifndef DYNAMIC_LINK_TEST
  65 #define _LGPL_SOURCE
  66 #endif
  67 #include <urcu.h>
  68 #include <urcu/cds.h>
  69 #include <urcu-defer.h>
  70
  71 static volatile int test_go, test_stop;
  72
  73 static unsigned long rduration;
  74
  75 static unsigned long duration;
  76
  77 /* read-side C.S. duration, in loops */
  78 static unsigned long wdelay;
  79
  80 static inline void loop_sleep(unsigned long l)
  81 {
  82         while(l-- != 0)
  83                 caa_cpu_relax();
  84 }
  85
  86 static int verbose_mode;
  87
  88 #define printf_verbose(fmt, args...)            \
  89         do {                                    \
  90                 if (verbose_mode)               \
  91                         printf(fmt, args);      \
  92         } while (0)
  93
  94 static unsigned int cpu_affinities[NR_CPUS];
  95 static unsigned int next_aff = 0;
  96 static int use_affinity = 0;
  97
  98 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  99
 100 #ifndef HAVE_CPU_SET_T
 101 typedef unsigned long cpu_set_t;
 102 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 103 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 104 #endif
 105
 106 static void set_affinity(void)
 107 {
 108         cpu_set_t mask;
 109         int cpu;
 110         int ret;
 111
 112         if (!use_affinity)
 113                 return;
 114
 115 #if HAVE_SCHED_SETAFFINITY
 116         ret = pthread_mutex_lock(&affinity_mutex);
 117         if (ret) {
 118                 perror("Error in pthread mutex lock");
 119                 exit(-1);
 120         }
 121         cpu = cpu_affinities[next_aff++];
 122         ret = pthread_mutex_unlock(&affinity_mutex);
 123         if (ret) {
 124                 perror("Error in pthread mutex unlock");
 125                 exit(-1);
 126         }
 127
 128         CPU_ZERO(&mask);
 129         CPU_SET(cpu, &mask);
 130 #if SCHED_SETAFFINITY_ARGS == 2
 131         sched_setaffinity(0, &mask);
 132 #else
 133         sched_setaffinity(0, sizeof(mask), &mask);
 134 #endif
 135 #endif /* HAVE_SCHED_SETAFFINITY */
 136 }
 137
 138 /*
 139  * returns 0 if test should end.
 140  */
 141 static int test_duration_dequeue(void)
 142 {
 143         return !test_stop;
 144 }
 145
 146 static int test_duration_enqueue(void)
 147 {
 148         return !test_stop;
 149 }
 150
 151 static unsigned long long __thread nr_dequeues;
 152 static unsigned long long __thread nr_enqueues;
 153
 154 static unsigned long long __thread nr_successful_dequeues;
 155 static unsigned long long __thread nr_successful_enqueues;
 156
 157 static unsigned int nr_enqueuers;
 158 static unsigned int nr_dequeuers;
 159
 160 static struct cds_lfq_queue_rcu q;
 161
 162 void *thr_enqueuer(void *_count)
 163 {
 164         unsigned long long *count = _count;
 165
 166         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 167                         "enqueuer", pthread_self(), (unsigned long)gettid());
 168
 169         set_affinity();
 170
 171         rcu_register_thread();
 172
 173         while (!test_go)
 174         {
 175         }
 176         cmm_smp_mb();
 177
 178         for (;;) {
 179                 struct cds_lfq_node_rcu *node = malloc(sizeof(*node));
 180                 if (!node)
 181                         goto fail;
 182                 cds_lfq_node_init_rcu(node);
 183                 rcu_read_lock();
 184                 cds_lfq_enqueue_rcu(&q, node);
 185                 rcu_read_unlock();
 186                 nr_successful_enqueues++;
 187
 188                 if (unlikely(wdelay))
 189                         loop_sleep(wdelay);
 190 fail:
 191                 nr_enqueues++;
 192                 if (unlikely(!test_duration_enqueue()))
 193                         break;
 194         }
 195
 196         rcu_unregister_thread();
 197
 198         count[0] = nr_enqueues;
 199         count[1] = nr_successful_enqueues;
 200         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 201                        "enqueues %llu successful_enqueues %llu\n",
 202                        pthread_self(), (unsigned long)gettid(), nr_enqueues,
 203                        nr_successful_enqueues);
 204         return ((void*)1);
 205
 206 }
 207
 208 void *thr_dequeuer(void *_count)
 209 {
 210         unsigned long long *count = _count;
 211         int ret;
 212
 213         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 214                         "dequeuer", pthread_self(), (unsigned long)gettid());
 215
 216         set_affinity();
 217
 218         ret = rcu_defer_register_thread();
 219         if (ret) {
 220                 printf("Error in rcu_defer_register_thread\n");
 221                 exit(-1);
 222         }
 223         rcu_register_thread();
 224
 225         while (!test_go)
 226         {
 227         }
 228         cmm_smp_mb();
 229
 230         for (;;) {
 231                 struct cds_lfq_node_rcu *node;
 232
 233                 rcu_read_lock();
 234                 node = cds_lfq_dequeue_rcu(&q);
 235                 rcu_read_unlock();
 236
 237                 if (node) {
 238                         defer_rcu(free, node);
 239                         nr_successful_dequeues++;
 240                 }
 241
 242                 nr_dequeues++;
 243                 if (unlikely(!test_duration_dequeue()))
 244                         break;
 245                 if (unlikely(rduration))
 246                         loop_sleep(rduration);
 247         }
 248
 249         rcu_unregister_thread();
 250         rcu_defer_unregister_thread();
 251         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 252                        "dequeues %llu, successful_dequeues %llu\n",
 253                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 254                        nr_successful_dequeues);
 255         count[0] = nr_dequeues;
 256         count[1] = nr_successful_dequeues;
 257         return ((void*)2);
 258 }
 259
 260 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 261 {
 262         struct cds_lfq_node_rcu *node;
 263
 264         do {
 265                 rcu_read_lock();
 266                 node = cds_lfq_dequeue_rcu(q);
 267                 rcu_read_unlock();
 268                 if (node) {
 269                         free(node);     /* no more concurrent access */
 270                         (*nr_dequeues)++;
 271                 }
 272         } while (node);
 273 }
 274
 275 void show_usage(int argc, char **argv)
 276 {
 277         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 278         printf(" [-d delay] (enqueuer period (in loops))");
 279         printf(" [-c duration] (dequeuer period (in loops))");
 280         printf(" [-v] (verbose output)");
 281         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 282         printf("\n");
 283 }
 284
 285 int main(int argc, char **argv)
 286 {
 287         int err;
 288         pthread_t *tid_enqueuer, *tid_dequeuer;
 289         void *tret;
 290         unsigned long long *count_enqueuer, *count_dequeuer;
 291         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 292         unsigned long long tot_successful_enqueues = 0,
 293                            tot_successful_dequeues = 0;
 294         unsigned long long end_dequeues = 0;
 295         int i, a;
 296
 297         if (argc < 4) {
 298                 show_usage(argc, argv);
 299                 return -1;
 300         }
 301
 302         err = sscanf(argv[1], "%u", &nr_dequeuers);
 303         if (err != 1) {
 304                 show_usage(argc, argv);
 305                 return -1;
 306         }
 307
 308         err = sscanf(argv[2], "%u", &nr_enqueuers);
 309         if (err != 1) {
 310                 show_usage(argc, argv);
 311                 return -1;
 312         }
 313
 314         err = sscanf(argv[3], "%lu", &duration);
 315         if (err != 1) {
 316                 show_usage(argc, argv);
 317                 return -1;
 318         }
 319
 320         for (i = 4; i < argc; i++) {
 321                 if (argv[i][0] != '-')
 322                         continue;
 323                 switch (argv[i][1]) {
 324                 case 'a':
 325                         if (argc < i + 2) {
 326                                 show_usage(argc, argv);
 327                                 return -1;
 328                         }
 329                         a = atoi(argv[++i]);
 330                         cpu_affinities[next_aff++] = a;
 331                         use_affinity = 1;
 332                         printf_verbose("Adding CPU %d affinity\n", a);
 333                         break;
 334                 case 'c':
 335                         if (argc < i + 2) {
 336                                 show_usage(argc, argv);
 337                                 return -1;
 338                         }
 339                         rduration = atol(argv[++i]);
 340                         break;
 341                 case 'd':
 342                         if (argc < i + 2) {
 343                                 show_usage(argc, argv);
 344                                 return -1;
 345                         }
 346                         wdelay = atol(argv[++i]);
 347                         break;
 348                 case 'v':
 349                         verbose_mode = 1;
 350                         break;
 351                 }
 352         }
 353
 354         printf_verbose("running test for %lu seconds, %u enqueuers, "
 355                        "%u dequeuers.\n",
 356                        duration, nr_enqueuers, nr_dequeuers);
 357         printf_verbose("Writer delay : %lu loops.\n", rduration);
 358         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 359         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 360                         "main", pthread_self(), (unsigned long)gettid());
 361
 362         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 363         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 364         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 365         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 366         cds_lfq_init_rcu(&q, call_rcu);
 367
 368         next_aff = 0;
 369
 370         for (i = 0; i < nr_enqueuers; i++) {
 371                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 372                                      &count_enqueuer[2 * i]);
 373                 if (err != 0)
 374                         exit(1);
 375         }
 376         for (i = 0; i < nr_dequeuers; i++) {
 377                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 378                                      &count_dequeuer[2 * i]);
 379                 if (err != 0)
 380                         exit(1);
 381         }
 382
 383         cmm_smp_mb();
 384
 385         test_go = 1;
 386
 387         for (i = 0; i < duration; i++) {
 388                 sleep(1);
 389                 if (verbose_mode)
 390                         write (1, ".", 1);
 391         }
 392
 393         test_stop = 1;
 394
 395         for (i = 0; i < nr_enqueuers; i++) {
 396                 err = pthread_join(tid_enqueuer[i], &tret);
 397                 if (err != 0)
 398                         exit(1);
 399                 tot_enqueues += count_enqueuer[2 * i];
 400                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 401         }
 402         for (i = 0; i < nr_dequeuers; i++) {
 403                 err = pthread_join(tid_dequeuer[i], &tret);
 404                 if (err != 0)
 405                         exit(1);
 406                 tot_dequeues += count_dequeuer[2 * i];
 407                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 408         }
 409
 410         test_end(&q, &end_dequeues);
 411         err = cds_lfq_destroy_rcu(&q);
 412         assert(!err);
 413
 414         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 415                        tot_enqueues, tot_dequeues);
 416         printf_verbose("total number of successful enqueues : %llu, "
 417                        "successful dequeues %llu\n",
 418                        tot_successful_enqueues, tot_successful_dequeues);
 419         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 420                 "nr_dequeuers %3u "
 421                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 422                 "successful enqueues %12llu successful dequeues %12llu "
 423                 "end_dequeues %llu nr_ops %12llu\n",
 424                 argv[0], duration, nr_enqueuers, wdelay,
 425                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 426                 tot_successful_enqueues,
 427                 tot_successful_dequeues, end_dequeues,
 428                 tot_enqueues + tot_dequeues);
 429         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 430                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 431                        "succ. dequeues + end dequeues %llu.\n",
 432                        tot_successful_enqueues,
 433                        tot_successful_dequeues + end_dequeues);
 434
 435         free(count_enqueuer);
 436         free(count_dequeuer);
 437         free(tid_enqueuer);
 438         free(tid_dequeuer);
 439         return 0;
 440 }