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
  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 loops)
  81 {
  82         while (loops-- != 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 #if HAVE_SCHED_SETAFFINITY
 109         cpu_set_t mask;
 110         int cpu, ret;
 111 #endif /* HAVE_SCHED_SETAFFINITY */
 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", 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_lfq_node_init_rcu(&node->list);
 189                 rcu_read_lock();
 190                 cds_lfq_enqueue_rcu(&q, &node->list);
 191                 rcu_read_unlock();
 192                 URCU_TLS(nr_successful_enqueues)++;
 193
 194                 if (caa_unlikely(wdelay))
 195                         loop_sleep(wdelay);
 196 fail:
 197                 URCU_TLS(nr_enqueues)++;
 198                 if (caa_unlikely(!test_duration_enqueue()))
 199                         break;
 200         }
 201
 202         rcu_unregister_thread();
 203
 204         count[0] = URCU_TLS(nr_enqueues);
 205         count[1] = URCU_TLS(nr_successful_enqueues);
 206         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 207                        "enqueues %llu successful_enqueues %llu\n",
 208                        pthread_self(), (unsigned long)gettid(),
 209                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 210         return ((void*)1);
 211
 212 }
 213
 214 static
 215 void free_node_cb(struct rcu_head *head)
 216 {
 217         struct test *node =
 218                 caa_container_of(head, struct test, rcu);
 219         free(node);
 220 }
 221
 222 void *thr_dequeuer(void *_count)
 223 {
 224         unsigned long long *count = _count;
 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 *qnode;
 240
 241                 rcu_read_lock();
 242                 qnode = cds_lfq_dequeue_rcu(&q);
 243                 rcu_read_unlock();
 244
 245                 if (qnode) {
 246                         struct test *node;
 247
 248                         node = caa_container_of(qnode, struct test, list);
 249                         call_rcu(&node->rcu, free_node_cb);
 250                         URCU_TLS(nr_successful_dequeues)++;
 251                 }
 252
 253                 URCU_TLS(nr_dequeues)++;
 254                 if (caa_unlikely(!test_duration_dequeue()))
 255                         break;
 256                 if (caa_unlikely(rduration))
 257                         loop_sleep(rduration);
 258         }
 259
 260         rcu_unregister_thread();
 261         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 262                        "dequeues %llu, successful_dequeues %llu\n",
 263                        pthread_self(), (unsigned long)gettid(),
 264                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 265         count[0] = URCU_TLS(nr_dequeues);
 266         count[1] = URCU_TLS(nr_successful_dequeues);
 267         return ((void*)2);
 268 }
 269
 270 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 271 {
 272         struct cds_lfq_node_rcu *snode;
 273
 274         do {
 275                 snode = cds_lfq_dequeue_rcu(q);
 276                 if (snode) {
 277                         struct test *node;
 278
 279                         node = caa_container_of(snode, struct test, list);
 280                         free(node);     /* no more concurrent access */
 281                         (*nr_dequeues)++;
 282                 }
 283         } while (snode);
 284 }
 285
 286 void show_usage(int argc, char **argv)
 287 {
 288         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 289         printf(" [-d delay] (enqueuer period (in loops))");
 290         printf(" [-c duration] (dequeuer period (in loops))");
 291         printf(" [-v] (verbose output)");
 292         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 293         printf("\n");
 294 }
 295
 296 int main(int argc, char **argv)
 297 {
 298         int err;
 299         pthread_t *tid_enqueuer, *tid_dequeuer;
 300         void *tret;
 301         unsigned long long *count_enqueuer, *count_dequeuer;
 302         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 303         unsigned long long tot_successful_enqueues = 0,
 304                            tot_successful_dequeues = 0;
 305         unsigned long long end_dequeues = 0;
 306         int i, a;
 307
 308         if (argc < 4) {
 309                 show_usage(argc, argv);
 310                 return -1;
 311         }
 312
 313         err = sscanf(argv[1], "%u", &nr_dequeuers);
 314         if (err != 1) {
 315                 show_usage(argc, argv);
 316                 return -1;
 317         }
 318
 319         err = sscanf(argv[2], "%u", &nr_enqueuers);
 320         if (err != 1) {
 321                 show_usage(argc, argv);
 322                 return -1;
 323         }
 324
 325         err = sscanf(argv[3], "%lu", &duration);
 326         if (err != 1) {
 327                 show_usage(argc, argv);
 328                 return -1;
 329         }
 330
 331         for (i = 4; i < argc; i++) {
 332                 if (argv[i][0] != '-')
 333                         continue;
 334                 switch (argv[i][1]) {
 335                 case 'a':
 336                         if (argc < i + 2) {
 337                                 show_usage(argc, argv);
 338                                 return -1;
 339                         }
 340                         a = atoi(argv[++i]);
 341                         cpu_affinities[next_aff++] = a;
 342                         use_affinity = 1;
 343                         printf_verbose("Adding CPU %d affinity\n", a);
 344                         break;
 345                 case 'c':
 346                         if (argc < i + 2) {
 347                                 show_usage(argc, argv);
 348                                 return -1;
 349                         }
 350                         rduration = atol(argv[++i]);
 351                         break;
 352                 case 'd':
 353                         if (argc < i + 2) {
 354                                 show_usage(argc, argv);
 355                                 return -1;
 356                         }
 357                         wdelay = atol(argv[++i]);
 358                         break;
 359                 case 'v':
 360                         verbose_mode = 1;
 361                         break;
 362                 }
 363         }
 364
 365         printf_verbose("running test for %lu seconds, %u enqueuers, "
 366                        "%u dequeuers.\n",
 367                        duration, nr_enqueuers, nr_dequeuers);
 368         printf_verbose("Writer delay : %lu loops.\n", rduration);
 369         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 370         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 371                         "main", pthread_self(), (unsigned long)gettid());
 372
 373         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 374         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 375         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 376         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 377         cds_lfq_init_rcu(&q, call_rcu);
 378         err = create_all_cpu_call_rcu_data(0);
 379         if (err) {
 380                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 381         }
 382
 383         next_aff = 0;
 384
 385         for (i = 0; i < nr_enqueuers; i++) {
 386                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 387                                      &count_enqueuer[2 * i]);
 388                 if (err != 0)
 389                         exit(1);
 390         }
 391         for (i = 0; i < nr_dequeuers; i++) {
 392                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 393                                      &count_dequeuer[2 * i]);
 394                 if (err != 0)
 395                         exit(1);
 396         }
 397
 398         cmm_smp_mb();
 399
 400         test_go = 1;
 401
 402         for (i = 0; i < duration; i++) {
 403                 sleep(1);
 404                 if (verbose_mode)
 405                         write (1, ".", 1);
 406         }
 407
 408         test_stop = 1;
 409
 410         for (i = 0; i < nr_enqueuers; i++) {
 411                 err = pthread_join(tid_enqueuer[i], &tret);
 412                 if (err != 0)
 413                         exit(1);
 414                 tot_enqueues += count_enqueuer[2 * i];
 415                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 416         }
 417         for (i = 0; i < nr_dequeuers; i++) {
 418                 err = pthread_join(tid_dequeuer[i], &tret);
 419                 if (err != 0)
 420                         exit(1);
 421                 tot_dequeues += count_dequeuer[2 * i];
 422                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 423         }
 424
 425         test_end(&q, &end_dequeues);
 426         err = cds_lfq_destroy_rcu(&q);
 427         assert(!err);
 428
 429         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 430                        tot_enqueues, tot_dequeues);
 431         printf_verbose("total number of successful enqueues : %llu, "
 432                        "successful dequeues %llu\n",
 433                        tot_successful_enqueues, tot_successful_dequeues);
 434         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 435                 "nr_dequeuers %3u "
 436                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 437                 "successful enqueues %12llu successful dequeues %12llu "
 438                 "end_dequeues %llu nr_ops %12llu\n",
 439                 argv[0], duration, nr_enqueuers, wdelay,
 440                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 441                 tot_successful_enqueues,
 442                 tot_successful_dequeues, end_dequeues,
 443                 tot_enqueues + tot_dequeues);
 444         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 445                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 446                        "succ. dequeues + end dequeues %llu.\n",
 447                        tot_successful_enqueues,
 448                        tot_successful_dequeues + end_dequeues);
 449
 450         free_all_cpu_call_rcu_data();
 451         free(count_enqueuer);
 452         free(count_dequeuer);
 453         free(tid_enqueuer);
 454         free(tid_dequeuer);
 455         return 0;
 456 }