tests/test_urcu_lfs.c

   1 /*
   2  * test_urcu_lfs.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free stack
   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/cds.h>
  70 #include <urcu-defer.h>
  71
  72 static volatile int test_go, test_stop;
  73
  74 static unsigned long rduration;
  75
  76 static unsigned long duration;
  77
  78 /* read-side C.S. duration, in loops */
  79 static unsigned long wdelay;
  80
  81 static inline void loop_sleep(unsigned long l)
  82 {
  83         while(l-- != 0)
  84                 caa_cpu_relax();
  85 }
  86
  87 static int verbose_mode;
  88
  89 #define printf_verbose(fmt, args...)            \
  90         do {                                    \
  91                 if (verbose_mode)               \
  92                         printf(fmt, args);      \
  93         } while (0)
  94
  95 static unsigned int cpu_affinities[NR_CPUS];
  96 static unsigned int next_aff = 0;
  97 static int use_affinity = 0;
  98
  99 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 100
 101 #ifndef HAVE_CPU_SET_T
 102 typedef unsigned long cpu_set_t;
 103 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 104 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 105 #endif
 106
 107 static void set_affinity(void)
 108 {
 109         cpu_set_t mask;
 110         int cpu;
 111         int ret;
 112
 113         if (!use_affinity)
 114                 return;
 115
 116 #if HAVE_SCHED_SETAFFINITY
 117         ret = pthread_mutex_lock(&affinity_mutex);
 118         if (ret) {
 119                 perror("Error in pthread mutex lock");
 120                 exit(-1);
 121         }
 122         cpu = cpu_affinities[next_aff++];
 123         ret = pthread_mutex_unlock(&affinity_mutex);
 124         if (ret) {
 125                 perror("Error in pthread mutex unlock");
 126                 exit(-1);
 127         }
 128
 129         CPU_ZERO(&mask);
 130         CPU_SET(cpu, &mask);
 131 #if SCHED_SETAFFINITY_ARGS == 2
 132         sched_setaffinity(0, &mask);
 133 #else
 134         sched_setaffinity(0, sizeof(mask), &mask);
 135 #endif
 136 #endif /* HAVE_SCHED_SETAFFINITY */
 137 }
 138
 139 /*
 140  * returns 0 if test should end.
 141  */
 142 static int test_duration_dequeue(void)
 143 {
 144         return !test_stop;
 145 }
 146
 147 static int test_duration_enqueue(void)
 148 {
 149         return !test_stop;
 150 }
 151
 152 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 153 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 154
 155 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 156 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 157
 158 static unsigned int nr_enqueuers;
 159 static unsigned int nr_dequeuers;
 160
 161 struct test {
 162         struct cds_lfs_node_rcu list;
 163         struct rcu_head rcu;
 164 };
 165
 166 static struct cds_lfs_stack_rcu s;
 167
 168 void *thr_enqueuer(void *_count)
 169 {
 170         unsigned long long *count = _count;
 171
 172         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 173                         "enqueuer", pthread_self(), (unsigned long)gettid());
 174
 175         set_affinity();
 176
 177         rcu_register_thread();
 178
 179         while (!test_go)
 180         {
 181         }
 182         cmm_smp_mb();
 183
 184         for (;;) {
 185                 struct test *node = malloc(sizeof(*node));
 186                 if (!node)
 187                         goto fail;
 188                 cds_lfs_node_init_rcu(&node->list);
 189                 /* No rcu read-side is needed for push */
 190                 cds_lfs_push_rcu(&s, &node->list);
 191                 URCU_TLS(nr_successful_enqueues)++;
 192
 193                 if (caa_unlikely(wdelay))
 194                         loop_sleep(wdelay);
 195 fail:
 196                 URCU_TLS(nr_enqueues)++;
 197                 if (caa_unlikely(!test_duration_enqueue()))
 198                         break;
 199         }
 200
 201         rcu_unregister_thread();
 202
 203         count[0] = URCU_TLS(nr_enqueues);
 204         count[1] = URCU_TLS(nr_successful_enqueues);
 205         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 206                        "enqueues %llu successful_enqueues %llu\n",
 207                        pthread_self(), (unsigned long)gettid(),
 208                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 209         return ((void*)1);
 210
 211 }
 212
 213 static
 214 void free_node_cb(struct rcu_head *head)
 215 {
 216         struct test *node =
 217                 caa_container_of(head, struct test, rcu);
 218         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         ret = rcu_defer_register_thread();
 232         if (ret) {
 233                 printf("Error in rcu_defer_register_thread\n");
 234                 exit(-1);
 235         }
 236         rcu_register_thread();
 237
 238         while (!test_go)
 239         {
 240         }
 241         cmm_smp_mb();
 242
 243         for (;;) {
 244                 struct cds_lfs_node_rcu *snode;
 245                 struct test *node;
 246
 247                 rcu_read_lock();
 248                 snode = cds_lfs_pop_rcu(&s);
 249                 node = caa_container_of(snode, struct test, list);
 250                 rcu_read_unlock();
 251                 if (node) {
 252                         call_rcu(&node->rcu, free_node_cb);
 253                         URCU_TLS(nr_successful_dequeues)++;
 254                 }
 255                 URCU_TLS(nr_dequeues)++;
 256                 if (caa_unlikely(!test_duration_dequeue()))
 257                         break;
 258                 if (caa_unlikely(rduration))
 259                         loop_sleep(rduration);
 260         }
 261
 262         rcu_unregister_thread();
 263         rcu_defer_unregister_thread();
 264
 265         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 266                        "dequeues %llu, successful_dequeues %llu\n",
 267                        pthread_self(), (unsigned long)gettid(),
 268                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 269         count[0] = URCU_TLS(nr_dequeues);
 270         count[1] = URCU_TLS(nr_successful_dequeues);
 271         return ((void*)2);
 272 }
 273
 274 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 275 {
 276         struct cds_lfs_node_rcu *snode;
 277
 278         do {
 279                 snode = cds_lfs_pop_rcu(s);
 280                 if (snode) {
 281                         struct test *node;
 282
 283                         node = caa_container_of(snode, struct test, list);
 284                         free(node);
 285                         (*nr_dequeues)++;
 286                 }
 287         } while (snode);
 288 }
 289
 290 void show_usage(int argc, char **argv)
 291 {
 292         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 293         printf(" [-d delay] (enqueuer period (in loops))");
 294         printf(" [-c duration] (dequeuer period (in loops))");
 295         printf(" [-v] (verbose output)");
 296         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 297         printf("\n");
 298 }
 299
 300 int main(int argc, char **argv)
 301 {
 302         int err;
 303         pthread_t *tid_enqueuer, *tid_dequeuer;
 304         void *tret;
 305         unsigned long long *count_enqueuer, *count_dequeuer;
 306         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 307         unsigned long long tot_successful_enqueues = 0,
 308                            tot_successful_dequeues = 0;
 309         unsigned long long end_dequeues = 0;
 310         int i, a;
 311
 312         if (argc < 4) {
 313                 show_usage(argc, argv);
 314                 return -1;
 315         }
 316
 317         err = sscanf(argv[1], "%u", &nr_dequeuers);
 318         if (err != 1) {
 319                 show_usage(argc, argv);
 320                 return -1;
 321         }
 322
 323         err = sscanf(argv[2], "%u", &nr_enqueuers);
 324         if (err != 1) {
 325                 show_usage(argc, argv);
 326                 return -1;
 327         }
 328
 329         err = sscanf(argv[3], "%lu", &duration);
 330         if (err != 1) {
 331                 show_usage(argc, argv);
 332                 return -1;
 333         }
 334
 335         for (i = 4; i < argc; i++) {
 336                 if (argv[i][0] != '-')
 337                         continue;
 338                 switch (argv[i][1]) {
 339                 case 'a':
 340                         if (argc < i + 2) {
 341                                 show_usage(argc, argv);
 342                                 return -1;
 343                         }
 344                         a = atoi(argv[++i]);
 345                         cpu_affinities[next_aff++] = a;
 346                         use_affinity = 1;
 347                         printf_verbose("Adding CPU %d affinity\n", a);
 348                         break;
 349                 case 'c':
 350                         if (argc < i + 2) {
 351                                 show_usage(argc, argv);
 352                                 return -1;
 353                         }
 354                         rduration = atol(argv[++i]);
 355                         break;
 356                 case 'd':
 357                         if (argc < i + 2) {
 358                                 show_usage(argc, argv);
 359                                 return -1;
 360                         }
 361                         wdelay = atol(argv[++i]);
 362                         break;
 363                 case 'v':
 364                         verbose_mode = 1;
 365                         break;
 366                 }
 367         }
 368
 369         printf_verbose("running test for %lu seconds, %u enqueuers, "
 370                        "%u dequeuers.\n",
 371                        duration, nr_enqueuers, nr_dequeuers);
 372         printf_verbose("Writer delay : %lu loops.\n", rduration);
 373         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 374         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 375                         "main", pthread_self(), (unsigned long)gettid());
 376
 377         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 378         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 379         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 380         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 381         cds_lfs_init_rcu(&s);
 382         err = create_all_cpu_call_rcu_data(0);
 383         if (err) {
 384                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 385         }
 386
 387         next_aff = 0;
 388
 389         for (i = 0; i < nr_enqueuers; i++) {
 390                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 391                                      &count_enqueuer[2 * i]);
 392                 if (err != 0)
 393                         exit(1);
 394         }
 395         for (i = 0; i < nr_dequeuers; i++) {
 396                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 397                                      &count_dequeuer[2 * i]);
 398                 if (err != 0)
 399                         exit(1);
 400         }
 401
 402         cmm_smp_mb();
 403
 404         test_go = 1;
 405
 406         for (i = 0; i < duration; i++) {
 407                 sleep(1);
 408                 if (verbose_mode)
 409                         write (1, ".", 1);
 410         }
 411
 412         test_stop = 1;
 413
 414         for (i = 0; i < nr_enqueuers; i++) {
 415                 err = pthread_join(tid_enqueuer[i], &tret);
 416                 if (err != 0)
 417                         exit(1);
 418                 tot_enqueues += count_enqueuer[2 * i];
 419                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 420         }
 421         for (i = 0; i < nr_dequeuers; i++) {
 422                 err = pthread_join(tid_dequeuer[i], &tret);
 423                 if (err != 0)
 424                         exit(1);
 425                 tot_dequeues += count_dequeuer[2 * i];
 426                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 427         }
 428
 429         test_end(&s, &end_dequeues);
 430
 431         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 432                        tot_enqueues, tot_dequeues);
 433         printf_verbose("total number of successful enqueues : %llu, "
 434                        "successful dequeues %llu\n",
 435                        tot_successful_enqueues, tot_successful_dequeues);
 436         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 437                 "nr_dequeuers %3u "
 438                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 439                 "successful enqueues %12llu successful dequeues %12llu "
 440                 "end_dequeues %llu nr_ops %12llu\n",
 441                 argv[0], duration, nr_enqueuers, wdelay,
 442                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 443                 tot_successful_enqueues,
 444                 tot_successful_dequeues, end_dequeues,
 445                 tot_enqueues + tot_dequeues);
 446         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 447                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 448                        "succ. dequeues + end dequeues %llu.\n",
 449                        tot_successful_enqueues,
 450                        tot_successful_dequeues + end_dequeues);
 451
 452         free_all_cpu_call_rcu_data();
 453         free(count_enqueuer);
 454         free(count_dequeuer);
 455         free(tid_enqueuer);
 456         free(tid_dequeuer);
 457         return 0;
 458 }