tests/benchmark/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 #include <stdio.h>
  25 #include <pthread.h>
  26 #include <stdlib.h>
  27 #include <stdint.h>
  28 #include <stdbool.h>
  29 #include <string.h>
  30 #include <sys/types.h>
  31 #include <sys/wait.h>
  32 #include <unistd.h>
  33 #include <stdio.h>
  34 #include <assert.h>
  35 #include <errno.h>
  36
  37 #include <urcu/arch.h>
  38 #include <urcu/tls-compat.h>
  39 #include "thread-id.h"
  40
  41 /* hardcoded number of CPUs */
  42 #define NR_CPUS 16384
  43
  44 #ifndef DYNAMIC_LINK_TEST
  45 #define _LGPL_SOURCE
  46 #endif
  47 #include <urcu.h>
  48 #include <urcu/cds.h>
  49
  50 static volatile int test_go, test_stop;
  51
  52 static unsigned long rduration;
  53
  54 static unsigned long duration;
  55
  56 /* read-side C.S. duration, in loops */
  57 static unsigned long wdelay;
  58
  59 static inline void loop_sleep(unsigned long loops)
  60 {
  61         while (loops-- != 0)
  62                 caa_cpu_relax();
  63 }
  64
  65 static int verbose_mode;
  66
  67 #define printf_verbose(fmt, args...)            \
  68         do {                                    \
  69                 if (verbose_mode)               \
  70                         printf(fmt, args);      \
  71         } while (0)
  72
  73 static unsigned int cpu_affinities[NR_CPUS];
  74 static unsigned int next_aff = 0;
  75 static int use_affinity = 0;
  76
  77 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  78
  79 static void set_affinity(void)
  80 {
  81 #if HAVE_SCHED_SETAFFINITY
  82         cpu_set_t mask;
  83         int cpu, ret;
  84 #endif /* HAVE_SCHED_SETAFFINITY */
  85
  86         if (!use_affinity)
  87                 return;
  88
  89 #if HAVE_SCHED_SETAFFINITY
  90         ret = pthread_mutex_lock(&affinity_mutex);
  91         if (ret) {
  92                 perror("Error in pthread mutex lock");
  93                 exit(-1);
  94         }
  95         cpu = cpu_affinities[next_aff++];
  96         ret = pthread_mutex_unlock(&affinity_mutex);
  97         if (ret) {
  98                 perror("Error in pthread mutex unlock");
  99                 exit(-1);
 100         }
 101
 102         CPU_ZERO(&mask);
 103         CPU_SET(cpu, &mask);
 104         sched_setaffinity(0, sizeof(mask), &mask);
 105 #endif /* HAVE_SCHED_SETAFFINITY */
 106 }
 107
 108 /*
 109  * returns 0 if test should end.
 110  */
 111 static int test_duration_dequeue(void)
 112 {
 113         return !test_stop;
 114 }
 115
 116 static int test_duration_enqueue(void)
 117 {
 118         return !test_stop;
 119 }
 120
 121 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 122 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 123
 124 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 125 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 126
 127 static unsigned int nr_enqueuers;
 128 static unsigned int nr_dequeuers;
 129
 130 struct test {
 131         struct cds_lfq_node_rcu list;
 132         struct rcu_head rcu;
 133 };
 134
 135 static struct cds_lfq_queue_rcu q;
 136
 137 void *thr_enqueuer(void *_count)
 138 {
 139         unsigned long long *count = _count;
 140
 141         printf_verbose("thread_begin %s, tid %lu\n",
 142                         "enqueuer", urcu_get_thread_id());
 143
 144         set_affinity();
 145
 146         rcu_register_thread();
 147
 148         while (!test_go)
 149         {
 150         }
 151         cmm_smp_mb();
 152
 153         for (;;) {
 154                 struct test *node = malloc(sizeof(*node));
 155                 if (!node)
 156                         goto fail;
 157                 cds_lfq_node_init_rcu(&node->list);
 158                 rcu_read_lock();
 159                 cds_lfq_enqueue_rcu(&q, &node->list);
 160                 rcu_read_unlock();
 161                 URCU_TLS(nr_successful_enqueues)++;
 162
 163                 if (caa_unlikely(wdelay))
 164                         loop_sleep(wdelay);
 165 fail:
 166                 URCU_TLS(nr_enqueues)++;
 167                 if (caa_unlikely(!test_duration_enqueue()))
 168                         break;
 169         }
 170
 171         rcu_unregister_thread();
 172
 173         count[0] = URCU_TLS(nr_enqueues);
 174         count[1] = URCU_TLS(nr_successful_enqueues);
 175         printf_verbose("enqueuer thread_end, tid %lu, "
 176                         "enqueues %llu successful_enqueues %llu\n",
 177                         urcu_get_thread_id(),
 178                         URCU_TLS(nr_enqueues),
 179                         URCU_TLS(nr_successful_enqueues));
 180         return ((void*)1);
 181
 182 }
 183
 184 static
 185 void free_node_cb(struct rcu_head *head)
 186 {
 187         struct test *node =
 188                 caa_container_of(head, struct test, rcu);
 189         free(node);
 190 }
 191
 192 void *thr_dequeuer(void *_count)
 193 {
 194         unsigned long long *count = _count;
 195
 196         printf_verbose("thread_begin %s, tid %lu\n",
 197                         "dequeuer", urcu_get_thread_id());
 198
 199         set_affinity();
 200
 201         rcu_register_thread();
 202
 203         while (!test_go)
 204         {
 205         }
 206         cmm_smp_mb();
 207
 208         for (;;) {
 209                 struct cds_lfq_node_rcu *qnode;
 210
 211                 rcu_read_lock();
 212                 qnode = cds_lfq_dequeue_rcu(&q);
 213                 rcu_read_unlock();
 214
 215                 if (qnode) {
 216                         struct test *node;
 217
 218                         node = caa_container_of(qnode, struct test, list);
 219                         call_rcu(&node->rcu, free_node_cb);
 220                         URCU_TLS(nr_successful_dequeues)++;
 221                 }
 222
 223                 URCU_TLS(nr_dequeues)++;
 224                 if (caa_unlikely(!test_duration_dequeue()))
 225                         break;
 226                 if (caa_unlikely(rduration))
 227                         loop_sleep(rduration);
 228         }
 229
 230         rcu_unregister_thread();
 231         printf_verbose("dequeuer thread_end, tid %lu, "
 232                         "dequeues %llu, successful_dequeues %llu\n",
 233                         urcu_get_thread_id(),
 234                         URCU_TLS(nr_dequeues),
 235                         URCU_TLS(nr_successful_dequeues));
 236         count[0] = URCU_TLS(nr_dequeues);
 237         count[1] = URCU_TLS(nr_successful_dequeues);
 238         return ((void*)2);
 239 }
 240
 241 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 242 {
 243         struct cds_lfq_node_rcu *snode;
 244
 245         do {
 246                 snode = cds_lfq_dequeue_rcu(q);
 247                 if (snode) {
 248                         struct test *node;
 249
 250                         node = caa_container_of(snode, struct test, list);
 251                         free(node);     /* no more concurrent access */
 252                         (*nr_dequeues)++;
 253                 }
 254         } while (snode);
 255 }
 256
 257 void show_usage(int argc, char **argv)
 258 {
 259         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 260                 argv[0]);
 261         printf("OPTIONS:\n");
 262         printf("        [-d delay] (enqueuer period (in loops))\n");
 263         printf("        [-c duration] (dequeuer period (in loops))\n");
 264         printf("        [-v] (verbose output)\n");
 265         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 266         printf("\n");
 267 }
 268
 269 int main(int argc, char **argv)
 270 {
 271         int err;
 272         pthread_t *tid_enqueuer, *tid_dequeuer;
 273         void *tret;
 274         unsigned long long *count_enqueuer, *count_dequeuer;
 275         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 276         unsigned long long tot_successful_enqueues = 0,
 277                            tot_successful_dequeues = 0;
 278         unsigned long long end_dequeues = 0;
 279         int i, a;
 280         unsigned int i_thr;
 281
 282         if (argc < 4) {
 283                 show_usage(argc, argv);
 284                 return -1;
 285         }
 286
 287         err = sscanf(argv[1], "%u", &nr_dequeuers);
 288         if (err != 1) {
 289                 show_usage(argc, argv);
 290                 return -1;
 291         }
 292
 293         err = sscanf(argv[2], "%u", &nr_enqueuers);
 294         if (err != 1) {
 295                 show_usage(argc, argv);
 296                 return -1;
 297         }
 298
 299         err = sscanf(argv[3], "%lu", &duration);
 300         if (err != 1) {
 301                 show_usage(argc, argv);
 302                 return -1;
 303         }
 304
 305         for (i = 4; i < argc; i++) {
 306                 if (argv[i][0] != '-')
 307                         continue;
 308                 switch (argv[i][1]) {
 309                 case 'a':
 310                         if (argc < i + 2) {
 311                                 show_usage(argc, argv);
 312                                 return -1;
 313                         }
 314                         a = atoi(argv[++i]);
 315                         cpu_affinities[next_aff++] = a;
 316                         use_affinity = 1;
 317                         printf_verbose("Adding CPU %d affinity\n", a);
 318                         break;
 319                 case 'c':
 320                         if (argc < i + 2) {
 321                                 show_usage(argc, argv);
 322                                 return -1;
 323                         }
 324                         rduration = atol(argv[++i]);
 325                         break;
 326                 case 'd':
 327                         if (argc < i + 2) {
 328                                 show_usage(argc, argv);
 329                                 return -1;
 330                         }
 331                         wdelay = atol(argv[++i]);
 332                         break;
 333                 case 'v':
 334                         verbose_mode = 1;
 335                         break;
 336                 }
 337         }
 338
 339         printf_verbose("running test for %lu seconds, %u enqueuers, "
 340                        "%u dequeuers.\n",
 341                        duration, nr_enqueuers, nr_dequeuers);
 342         printf_verbose("Writer delay : %lu loops.\n", rduration);
 343         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 344         printf_verbose("thread %-6s, tid %lu\n",
 345                         "main", urcu_get_thread_id());
 346
 347         tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
 348         tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
 349         count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
 350         count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
 351         cds_lfq_init_rcu(&q, call_rcu);
 352         err = create_all_cpu_call_rcu_data(0);
 353         if (err) {
 354                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 355         }
 356
 357         next_aff = 0;
 358
 359         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 360                 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
 361                                      &count_enqueuer[2 * i_thr]);
 362                 if (err != 0)
 363                         exit(1);
 364         }
 365         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 366                 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
 367                                      &count_dequeuer[2 * i_thr]);
 368                 if (err != 0)
 369                         exit(1);
 370         }
 371
 372         cmm_smp_mb();
 373
 374         test_go = 1;
 375
 376         for (i_thr = 0; i_thr < duration; i_thr++) {
 377                 sleep(1);
 378                 if (verbose_mode) {
 379                         fwrite(".", sizeof(char), 1, stdout);
 380                         fflush(stdout);
 381                 }
 382         }
 383
 384         test_stop = 1;
 385
 386         for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
 387                 err = pthread_join(tid_enqueuer[i_thr], &tret);
 388                 if (err != 0)
 389                         exit(1);
 390                 tot_enqueues += count_enqueuer[2 * i_thr];
 391                 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
 392         }
 393         for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
 394                 err = pthread_join(tid_dequeuer[i_thr], &tret);
 395                 if (err != 0)
 396                         exit(1);
 397                 tot_dequeues += count_dequeuer[2 * i_thr];
 398                 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
 399         }
 400
 401         test_end(&q, &end_dequeues);
 402         err = cds_lfq_destroy_rcu(&q);
 403         assert(!err);
 404
 405         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 406                        tot_enqueues, tot_dequeues);
 407         printf_verbose("total number of successful enqueues : %llu, "
 408                        "successful dequeues %llu\n",
 409                        tot_successful_enqueues, tot_successful_dequeues);
 410         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 411                 "nr_dequeuers %3u "
 412                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 413                 "successful enqueues %12llu successful dequeues %12llu "
 414                 "end_dequeues %llu nr_ops %12llu\n",
 415                 argv[0], duration, nr_enqueuers, wdelay,
 416                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 417                 tot_successful_enqueues,
 418                 tot_successful_dequeues, end_dequeues,
 419                 tot_enqueues + tot_dequeues);
 420         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 421                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 422                        "succ. dequeues + end dequeues %llu.\n",
 423                        tot_successful_enqueues,
 424                        tot_successful_dequeues + end_dequeues);
 425
 426         free_all_cpu_call_rcu_data();
 427         free(count_enqueuer);
 428         free(count_dequeuer);
 429         free(tid_enqueuer);
 430         free(tid_dequeuer);
 431
 432         return 0;
 433 }