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
  70 static volatile int test_go, test_stop;
  71
  72 static unsigned long rduration;
  73
  74 static unsigned long duration;
  75
  76 /* read-side C.S. duration, in loops */
  77 static unsigned long wdelay;
  78
  79 static inline void loop_sleep(unsigned long l)
  80 {
  81         while(l-- != 0)
  82                 caa_cpu_relax();
  83 }
  84
  85 static int verbose_mode;
  86
  87 #define printf_verbose(fmt, args...)            \
  88         do {                                    \
  89                 if (verbose_mode)               \
  90                         printf(fmt, args);      \
  91         } while (0)
  92
  93 static unsigned int cpu_affinities[NR_CPUS];
  94 static unsigned int next_aff = 0;
  95 static int use_affinity = 0;
  96
  97 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  98
  99 #ifndef HAVE_CPU_SET_T
 100 typedef unsigned long cpu_set_t;
 101 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 102 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 103 #endif
 104
 105 static void set_affinity(void)
 106 {
 107         cpu_set_t mask;
 108         int cpu;
 109         int ret;
 110
 111         if (!use_affinity)
 112                 return;
 113
 114 #if HAVE_SCHED_SETAFFINITY
 115         ret = pthread_mutex_lock(&affinity_mutex);
 116         if (ret) {
 117                 perror("Error in pthread mutex lock");
 118                 exit(-1);
 119         }
 120         cpu = cpu_affinities[next_aff++];
 121         ret = pthread_mutex_unlock(&affinity_mutex);
 122         if (ret) {
 123                 perror("Error in pthread mutex unlock");
 124                 exit(-1);
 125         }
 126
 127         CPU_ZERO(&mask);
 128         CPU_SET(cpu, &mask);
 129 #if SCHED_SETAFFINITY_ARGS == 2
 130         sched_setaffinity(0, &mask);
 131 #else
 132         sched_setaffinity(0, sizeof(mask), &mask);
 133 #endif
 134 #endif /* HAVE_SCHED_SETAFFINITY */
 135 }
 136
 137 /*
 138  * returns 0 if test should end.
 139  */
 140 static int test_duration_dequeue(void)
 141 {
 142         return !test_stop;
 143 }
 144
 145 static int test_duration_enqueue(void)
 146 {
 147         return !test_stop;
 148 }
 149
 150 static unsigned long long __thread nr_dequeues;
 151 static unsigned long long __thread nr_enqueues;
 152
 153 static unsigned long long __thread nr_successful_dequeues;
 154 static unsigned long long __thread nr_successful_enqueues;
 155
 156 static unsigned int nr_enqueuers;
 157 static unsigned int nr_dequeuers;
 158
 159 static struct cds_lfq_queue_rcu q;
 160
 161 void *thr_enqueuer(void *_count)
 162 {
 163         unsigned long long *count = _count;
 164
 165         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 166                         "enqueuer", pthread_self(), (unsigned long)gettid());
 167
 168         set_affinity();
 169
 170         rcu_register_thread();
 171
 172         while (!test_go)
 173         {
 174         }
 175         cmm_smp_mb();
 176
 177         for (;;) {
 178                 struct cds_lfq_node_rcu *node = malloc(sizeof(*node));
 179                 if (!node)
 180                         goto fail;
 181                 cds_lfq_node_init_rcu(node);
 182                 rcu_read_lock();
 183                 cds_lfq_enqueue_rcu(&q, node);
 184                 rcu_read_unlock();
 185                 nr_successful_enqueues++;
 186
 187                 if (unlikely(wdelay))
 188                         loop_sleep(wdelay);
 189 fail:
 190                 nr_enqueues++;
 191                 if (unlikely(!test_duration_enqueue()))
 192                         break;
 193         }
 194
 195         rcu_unregister_thread();
 196
 197         count[0] = nr_enqueues;
 198         count[1] = nr_successful_enqueues;
 199         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 200                        "enqueues %llu successful_enqueues %llu\n",
 201                        pthread_self(), (unsigned long)gettid(), nr_enqueues,
 202                        nr_successful_enqueues);
 203         return ((void*)1);
 204
 205 }
 206
 207 static void rcu_free_node(struct rcu_head *head)
 208 {
 209         struct cds_lfq_node_rcu *node =
 210                 caa_container_of(head, struct cds_lfq_node_rcu, rcu_head);
 211         free(node);
 212 }
 213
 214 static void ref_release_node(struct urcu_ref *ref)
 215 {
 216         struct cds_lfq_node_rcu *node =
 217                 caa_container_of(ref, struct cds_lfq_node_rcu, ref);
 218         call_rcu(&node->rcu_head, rcu_free_node);
 219 }
 220
 221 void *thr_dequeuer(void *_count)
 222 {
 223         unsigned long long *count = _count;
 224         int ret;
 225
 226         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 227                         "dequeuer", pthread_self(), (unsigned long)gettid());
 228
 229         set_affinity();
 230
 231         rcu_register_thread();
 232
 233         while (!test_go)
 234         {
 235         }
 236         cmm_smp_mb();
 237
 238         for (;;) {
 239                 struct cds_lfq_node_rcu *node;
 240
 241                 rcu_read_lock();
 242                 node = cds_lfq_dequeue_rcu(&q);
 243                 rcu_read_unlock();
 244
 245                 if (node) {
 246                         urcu_ref_put(&node->ref, ref_release_node);
 247                         nr_successful_dequeues++;
 248                 }
 249
 250                 nr_dequeues++;
 251                 if (unlikely(!test_duration_dequeue()))
 252                         break;
 253                 if (unlikely(rduration))
 254                         loop_sleep(rduration);
 255         }
 256
 257         rcu_unregister_thread();
 258         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 259                        "dequeues %llu, successful_dequeues %llu\n",
 260                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 261                        nr_successful_dequeues);
 262         count[0] = nr_dequeues;
 263         count[1] = nr_successful_dequeues;
 264         return ((void*)2);
 265 }
 266
 267 static void release_node(struct urcu_ref *ref)
 268 {
 269         struct cds_lfq_node_rcu *node = caa_container_of(ref, struct cds_lfq_node_rcu, ref);
 270         free(node);
 271 }
 272
 273 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 274 {
 275         struct cds_lfq_node_rcu *node;
 276
 277         do {
 278                 rcu_read_lock();
 279                 node = cds_lfq_dequeue_rcu(q);
 280                 rcu_read_unlock();
 281                 if (node) {
 282                         urcu_ref_put(&node->ref, release_node);
 283                         (*nr_dequeues)++;
 284                 }
 285         } while (node);
 286 }
 287
 288 void show_usage(int argc, char **argv)
 289 {
 290         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 291         printf(" [-d delay] (enqueuer period (in loops))");
 292         printf(" [-c duration] (dequeuer period (in loops))");
 293         printf(" [-v] (verbose output)");
 294         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 295         printf("\n");
 296 }
 297
 298 int main(int argc, char **argv)
 299 {
 300         int err;
 301         pthread_t *tid_enqueuer, *tid_dequeuer;
 302         void *tret;
 303         unsigned long long *count_enqueuer, *count_dequeuer;
 304         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 305         unsigned long long tot_successful_enqueues = 0,
 306                            tot_successful_dequeues = 0;
 307         unsigned long long end_dequeues = 0;
 308         int i, a;
 309
 310         if (argc < 4) {
 311                 show_usage(argc, argv);
 312                 return -1;
 313         }
 314
 315         err = sscanf(argv[1], "%u", &nr_dequeuers);
 316         if (err != 1) {
 317                 show_usage(argc, argv);
 318                 return -1;
 319         }
 320
 321         err = sscanf(argv[2], "%u", &nr_enqueuers);
 322         if (err != 1) {
 323                 show_usage(argc, argv);
 324                 return -1;
 325         }
 326
 327         err = sscanf(argv[3], "%lu", &duration);
 328         if (err != 1) {
 329                 show_usage(argc, argv);
 330                 return -1;
 331         }
 332
 333         for (i = 4; i < argc; i++) {
 334                 if (argv[i][0] != '-')
 335                         continue;
 336                 switch (argv[i][1]) {
 337                 case 'a':
 338                         if (argc < i + 2) {
 339                                 show_usage(argc, argv);
 340                                 return -1;
 341                         }
 342                         a = atoi(argv[++i]);
 343                         cpu_affinities[next_aff++] = a;
 344                         use_affinity = 1;
 345                         printf_verbose("Adding CPU %d affinity\n", a);
 346                         break;
 347                 case 'c':
 348                         if (argc < i + 2) {
 349                                 show_usage(argc, argv);
 350                                 return -1;
 351                         }
 352                         rduration = atol(argv[++i]);
 353                         break;
 354                 case 'd':
 355                         if (argc < i + 2) {
 356                                 show_usage(argc, argv);
 357                                 return -1;
 358                         }
 359                         wdelay = atol(argv[++i]);
 360                         break;
 361                 case 'v':
 362                         verbose_mode = 1;
 363                         break;
 364                 }
 365         }
 366
 367         printf_verbose("running test for %lu seconds, %u enqueuers, "
 368                        "%u dequeuers.\n",
 369                        duration, nr_enqueuers, nr_dequeuers);
 370         printf_verbose("Writer delay : %lu loops.\n", rduration);
 371         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 372         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 373                         "main", pthread_self(), (unsigned long)gettid());
 374
 375         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 376         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 377         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 378         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 379         cds_lfq_init_rcu(&q, ref_release_node);
 380
 381         next_aff = 0;
 382
 383         for (i = 0; i < nr_enqueuers; i++) {
 384                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 385                                      &count_enqueuer[2 * i]);
 386                 if (err != 0)
 387                         exit(1);
 388         }
 389         for (i = 0; i < nr_dequeuers; i++) {
 390                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 391                                      &count_dequeuer[2 * i]);
 392                 if (err != 0)
 393                         exit(1);
 394         }
 395
 396         cmm_smp_mb();
 397
 398         test_go = 1;
 399
 400         for (i = 0; i < duration; i++) {
 401                 sleep(1);
 402                 if (verbose_mode)
 403                         write (1, ".", 1);
 404         }
 405
 406         test_stop = 1;
 407
 408         for (i = 0; i < nr_enqueuers; i++) {
 409                 err = pthread_join(tid_enqueuer[i], &tret);
 410                 if (err != 0)
 411                         exit(1);
 412                 tot_enqueues += count_enqueuer[2 * i];
 413                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 414         }
 415         for (i = 0; i < nr_dequeuers; i++) {
 416                 err = pthread_join(tid_dequeuer[i], &tret);
 417                 if (err != 0)
 418                         exit(1);
 419                 tot_dequeues += count_dequeuer[2 * i];
 420                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 421         }
 422
 423         test_end(&q, &end_dequeues);
 424
 425         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 426                        tot_enqueues, tot_dequeues);
 427         printf_verbose("total number of successful enqueues : %llu, "
 428                        "successful dequeues %llu\n",
 429                        tot_successful_enqueues, tot_successful_dequeues);
 430         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 431                 "nr_dequeuers %3u "
 432                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 433                 "successful enqueues %12llu successful dequeues %12llu "
 434                 "end_dequeues %llu nr_ops %12llu\n",
 435                 argv[0], duration, nr_enqueuers, wdelay,
 436                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 437                 tot_successful_enqueues,
 438                 tot_successful_dequeues, end_dequeues,
 439                 tot_enqueues + tot_dequeues);
 440         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 441                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 442                        "succ. dequeues + end dequeues %llu.\n",
 443                        tot_successful_enqueues,
 444                        tot_successful_dequeues + end_dequeues);
 445
 446         free(count_enqueuer);
 447         free(count_dequeuer);
 448         free(tid_enqueuer);
 449         free(tid_dequeuer);
 450         return 0;
 451 }