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 #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_lfq_node_rcu list;
 163         struct rcu_head rcu;
 164 };
 165
 166 static struct cds_lfq_queue_rcu q;
 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", (unsigned long) pthread_self(),
 174                         (unsigned long) gettid());
 175
 176         set_affinity();
 177
 178         rcu_register_thread();
 179
 180         while (!test_go)
 181         {
 182         }
 183         cmm_smp_mb();
 184
 185         for (;;) {
 186                 struct test *node = malloc(sizeof(*node));
 187                 if (!node)
 188                         goto fail;
 189                 cds_lfq_node_init_rcu(&node->list);
 190                 rcu_read_lock();
 191                 cds_lfq_enqueue_rcu(&q, &node->list);
 192                 rcu_read_unlock();
 193                 URCU_TLS(nr_successful_enqueues)++;
 194
 195                 if (caa_unlikely(wdelay))
 196                         loop_sleep(wdelay);
 197 fail:
 198                 URCU_TLS(nr_enqueues)++;
 199                 if (caa_unlikely(!test_duration_enqueue()))
 200                         break;
 201         }
 202
 203         rcu_unregister_thread();
 204
 205         count[0] = URCU_TLS(nr_enqueues);
 206         count[1] = URCU_TLS(nr_successful_enqueues);
 207         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 208                        "enqueues %llu successful_enqueues %llu\n",
 209                        pthread_self(),
 210                         (unsigned long) gettid(),
 211                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 212         return ((void*)1);
 213
 214 }
 215
 216 static
 217 void free_node_cb(struct rcu_head *head)
 218 {
 219         struct test *node =
 220                 caa_container_of(head, struct test, rcu);
 221         free(node);
 222 }
 223
 224 void *thr_dequeuer(void *_count)
 225 {
 226         unsigned long long *count = _count;
 227         int ret;
 228
 229         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 230                         "dequeuer", (unsigned long) pthread_self(),
 231                         (unsigned long) gettid());
 232
 233         set_affinity();
 234
 235         ret = rcu_defer_register_thread();
 236         if (ret) {
 237                 printf("Error in rcu_defer_register_thread\n");
 238                 exit(-1);
 239         }
 240         rcu_register_thread();
 241
 242         while (!test_go)
 243         {
 244         }
 245         cmm_smp_mb();
 246
 247         for (;;) {
 248                 struct cds_lfq_node_rcu *qnode;
 249                 struct test *node;
 250
 251                 rcu_read_lock();
 252                 qnode = cds_lfq_dequeue_rcu(&q);
 253                 node = caa_container_of(qnode, struct test, list);
 254                 rcu_read_unlock();
 255
 256                 if (node) {
 257                         call_rcu(&node->rcu, free_node_cb);
 258                         URCU_TLS(nr_successful_dequeues)++;
 259                 }
 260
 261                 URCU_TLS(nr_dequeues)++;
 262                 if (caa_unlikely(!test_duration_dequeue()))
 263                         break;
 264                 if (caa_unlikely(rduration))
 265                         loop_sleep(rduration);
 266         }
 267
 268         rcu_unregister_thread();
 269         rcu_defer_unregister_thread();
 270         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 271                        "dequeues %llu, successful_dequeues %llu\n",
 272                        pthread_self(),
 273                         (unsigned long) gettid(),
 274                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 275         count[0] = URCU_TLS(nr_dequeues);
 276         count[1] = URCU_TLS(nr_successful_dequeues);
 277         return ((void*)2);
 278 }
 279
 280 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 281 {
 282         struct cds_lfq_node_rcu *snode;
 283
 284         do {
 285                 snode = cds_lfq_dequeue_rcu(q);
 286                 if (snode) {
 287                         struct test *node;
 288
 289                         node = caa_container_of(snode, struct test, list);
 290                         free(node);     /* no more concurrent access */
 291                         (*nr_dequeues)++;
 292                 }
 293         } while (snode);
 294 }
 295
 296 void show_usage(int argc, char **argv)
 297 {
 298         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 299         printf(" [-d delay] (enqueuer period (in loops))");
 300         printf(" [-c duration] (dequeuer period (in loops))");
 301         printf(" [-v] (verbose output)");
 302         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 303         printf("\n");
 304 }
 305
 306 int main(int argc, char **argv)
 307 {
 308         int err;
 309         pthread_t *tid_enqueuer, *tid_dequeuer;
 310         void *tret;
 311         unsigned long long *count_enqueuer, *count_dequeuer;
 312         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 313         unsigned long long tot_successful_enqueues = 0,
 314                            tot_successful_dequeues = 0;
 315         unsigned long long end_dequeues = 0;
 316         int i, a;
 317
 318         if (argc < 4) {
 319                 show_usage(argc, argv);
 320                 return -1;
 321         }
 322
 323         err = sscanf(argv[1], "%u", &nr_dequeuers);
 324         if (err != 1) {
 325                 show_usage(argc, argv);
 326                 return -1;
 327         }
 328
 329         err = sscanf(argv[2], "%u", &nr_enqueuers);
 330         if (err != 1) {
 331                 show_usage(argc, argv);
 332                 return -1;
 333         }
 334
 335         err = sscanf(argv[3], "%lu", &duration);
 336         if (err != 1) {
 337                 show_usage(argc, argv);
 338                 return -1;
 339         }
 340
 341         for (i = 4; i < argc; i++) {
 342                 if (argv[i][0] != '-')
 343                         continue;
 344                 switch (argv[i][1]) {
 345                 case 'a':
 346                         if (argc < i + 2) {
 347                                 show_usage(argc, argv);
 348                                 return -1;
 349                         }
 350                         a = atoi(argv[++i]);
 351                         cpu_affinities[next_aff++] = a;
 352                         use_affinity = 1;
 353                         printf_verbose("Adding CPU %d affinity\n", a);
 354                         break;
 355                 case 'c':
 356                         if (argc < i + 2) {
 357                                 show_usage(argc, argv);
 358                                 return -1;
 359                         }
 360                         rduration = atol(argv[++i]);
 361                         break;
 362                 case 'd':
 363                         if (argc < i + 2) {
 364                                 show_usage(argc, argv);
 365                                 return -1;
 366                         }
 367                         wdelay = atol(argv[++i]);
 368                         break;
 369                 case 'v':
 370                         verbose_mode = 1;
 371                         break;
 372                 }
 373         }
 374
 375         printf_verbose("running test for %lu seconds, %u enqueuers, "
 376                        "%u dequeuers.\n",
 377                        duration, nr_enqueuers, nr_dequeuers);
 378         printf_verbose("Writer delay : %lu loops.\n", rduration);
 379         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 380         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 381                         "main", (unsigned long) pthread_self(),
 382                         (unsigned long) gettid());
 383
 384         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 385         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 386         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 387         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 388         cds_lfq_init_rcu(&q, call_rcu);
 389         err = create_all_cpu_call_rcu_data(0);
 390         if (err) {
 391                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 392         }
 393
 394         next_aff = 0;
 395
 396         for (i = 0; i < nr_enqueuers; i++) {
 397                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 398                                      &count_enqueuer[2 * i]);
 399                 if (err != 0)
 400                         exit(1);
 401         }
 402         for (i = 0; i < nr_dequeuers; i++) {
 403                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 404                                      &count_dequeuer[2 * i]);
 405                 if (err != 0)
 406                         exit(1);
 407         }
 408
 409         cmm_smp_mb();
 410
 411         test_go = 1;
 412
 413         for (i = 0; i < duration; i++) {
 414                 sleep(1);
 415                 if (verbose_mode)
 416                         write (1, ".", 1);
 417         }
 418
 419         test_stop = 1;
 420
 421         for (i = 0; i < nr_enqueuers; i++) {
 422                 err = pthread_join(tid_enqueuer[i], &tret);
 423                 if (err != 0)
 424                         exit(1);
 425                 tot_enqueues += count_enqueuer[2 * i];
 426                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 427         }
 428         for (i = 0; i < nr_dequeuers; i++) {
 429                 err = pthread_join(tid_dequeuer[i], &tret);
 430                 if (err != 0)
 431                         exit(1);
 432                 tot_dequeues += count_dequeuer[2 * i];
 433                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 434         }
 435
 436         test_end(&q, &end_dequeues);
 437         err = cds_lfq_destroy_rcu(&q);
 438         assert(!err);
 439
 440         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 441                        tot_enqueues, tot_dequeues);
 442         printf_verbose("total number of successful enqueues : %llu, "
 443                        "successful dequeues %llu\n",
 444                        tot_successful_enqueues, tot_successful_dequeues);
 445         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 446                 "nr_dequeuers %3u "
 447                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 448                 "successful enqueues %12llu successful dequeues %12llu "
 449                 "end_dequeues %llu nr_ops %12llu\n",
 450                 argv[0], duration, nr_enqueuers, wdelay,
 451                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 452                 tot_successful_enqueues,
 453                 tot_successful_dequeues, end_dequeues,
 454                 tot_enqueues + tot_dequeues);
 455         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 456                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 457                        "succ. dequeues + end dequeues %llu.\n",
 458                        tot_successful_enqueues,
 459                        tot_successful_dequeues + end_dequeues);
 460
 461         free_all_cpu_call_rcu_data();
 462         free(count_enqueuer);
 463         free(count_dequeuer);
 464         free(tid_enqueuer);
 465         free(tid_dequeuer);
 466         return 0;
 467 }