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 #include <urcu-defer.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 l)
  81 {
  82         while(l-- != 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         cpu_set_t mask;
 109         int cpu;
 110         int ret;
 111
 112         if (!use_affinity)
 113                 return;
 114
 115 #if HAVE_SCHED_SETAFFINITY
 116         ret = pthread_mutex_lock(&affinity_mutex);
 117         if (ret) {
 118                 perror("Error in pthread mutex lock");
 119                 exit(-1);
 120         }
 121         cpu = cpu_affinities[next_aff++];
 122         ret = pthread_mutex_unlock(&affinity_mutex);
 123         if (ret) {
 124                 perror("Error in pthread mutex unlock");
 125                 exit(-1);
 126         }
 127
 128         CPU_ZERO(&mask);
 129         CPU_SET(cpu, &mask);
 130 #if SCHED_SETAFFINITY_ARGS == 2
 131         sched_setaffinity(0, &mask);
 132 #else
 133         sched_setaffinity(0, sizeof(mask), &mask);
 134 #endif
 135 #endif /* HAVE_SCHED_SETAFFINITY */
 136 }
 137
 138 /*
 139  * returns 0 if test should end.
 140  */
 141 static int test_duration_dequeue(void)
 142 {
 143         return !test_stop;
 144 }
 145
 146 static int test_duration_enqueue(void)
 147 {
 148         return !test_stop;
 149 }
 150
 151 static unsigned long long __thread nr_dequeues;
 152 static unsigned long long __thread nr_enqueues;
 153
 154 static unsigned long long __thread nr_successful_dequeues;
 155 static unsigned long long __thread nr_successful_enqueues;
 156
 157 static unsigned int nr_enqueuers;
 158 static unsigned int nr_dequeuers;
 159
 160 struct test {
 161         struct cds_lfq_node_rcu list;
 162         struct rcu_head rcu;
 163 };
 164
 165 static struct cds_lfq_queue_rcu q;
 166
 167 void *thr_enqueuer(void *_count)
 168 {
 169         unsigned long long *count = _count;
 170
 171         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 172                         "enqueuer", pthread_self(), (unsigned long)gettid());
 173
 174         set_affinity();
 175
 176         rcu_register_thread();
 177
 178         while (!test_go)
 179         {
 180         }
 181         cmm_smp_mb();
 182
 183         for (;;) {
 184                 struct test *node = malloc(sizeof(*node));
 185                 if (!node)
 186                         goto fail;
 187                 cds_lfq_node_init_rcu(&node->list);
 188                 rcu_read_lock();
 189                 cds_lfq_enqueue_rcu(&q, &node->list);
 190                 rcu_read_unlock();
 191                 nr_successful_enqueues++;
 192
 193                 if (unlikely(wdelay))
 194                         loop_sleep(wdelay);
 195 fail:
 196                 nr_enqueues++;
 197                 if (unlikely(!test_duration_enqueue()))
 198                         break;
 199         }
 200
 201         rcu_unregister_thread();
 202
 203         count[0] = nr_enqueues;
 204         count[1] = 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(), nr_enqueues,
 208                        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_lfq_node_rcu *qnode;
 245                 struct test *node;
 246
 247                 rcu_read_lock();
 248                 qnode = cds_lfq_dequeue_rcu(&q);
 249                 node = caa_container_of(qnode, struct test, list);
 250                 rcu_read_unlock();
 251
 252                 if (node) {
 253                         call_rcu(&node->rcu, free_node_cb);
 254                         nr_successful_dequeues++;
 255                 }
 256
 257                 nr_dequeues++;
 258                 if (unlikely(!test_duration_dequeue()))
 259                         break;
 260                 if (unlikely(rduration))
 261                         loop_sleep(rduration);
 262         }
 263
 264         rcu_unregister_thread();
 265         rcu_defer_unregister_thread();
 266         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 267                        "dequeues %llu, successful_dequeues %llu\n",
 268                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 269                        nr_successful_dequeues);
 270         count[0] = nr_dequeues;
 271         count[1] = nr_successful_dequeues;
 272         return ((void*)2);
 273 }
 274
 275 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 276 {
 277         struct cds_lfq_node_rcu *snode;
 278
 279         do {
 280                 snode = cds_lfq_dequeue_rcu(q);
 281                 if (snode) {
 282                         struct test *node;
 283
 284                         node = caa_container_of(snode, struct test, list);
 285                         free(node);     /* no more concurrent access */
 286                         (*nr_dequeues)++;
 287                 }
 288         } while (snode);
 289 }
 290
 291 void show_usage(int argc, char **argv)
 292 {
 293         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 294         printf(" [-d delay] (enqueuer period (in loops))");
 295         printf(" [-c duration] (dequeuer period (in loops))");
 296         printf(" [-v] (verbose output)");
 297         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 298         printf("\n");
 299 }
 300
 301 int main(int argc, char **argv)
 302 {
 303         int err;
 304         pthread_t *tid_enqueuer, *tid_dequeuer;
 305         void *tret;
 306         unsigned long long *count_enqueuer, *count_dequeuer;
 307         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 308         unsigned long long tot_successful_enqueues = 0,
 309                            tot_successful_dequeues = 0;
 310         unsigned long long end_dequeues = 0;
 311         int i, a;
 312
 313         if (argc < 4) {
 314                 show_usage(argc, argv);
 315                 return -1;
 316         }
 317
 318         err = sscanf(argv[1], "%u", &nr_dequeuers);
 319         if (err != 1) {
 320                 show_usage(argc, argv);
 321                 return -1;
 322         }
 323
 324         err = sscanf(argv[2], "%u", &nr_enqueuers);
 325         if (err != 1) {
 326                 show_usage(argc, argv);
 327                 return -1;
 328         }
 329
 330         err = sscanf(argv[3], "%lu", &duration);
 331         if (err != 1) {
 332                 show_usage(argc, argv);
 333                 return -1;
 334         }
 335
 336         for (i = 4; i < argc; i++) {
 337                 if (argv[i][0] != '-')
 338                         continue;
 339                 switch (argv[i][1]) {
 340                 case 'a':
 341                         if (argc < i + 2) {
 342                                 show_usage(argc, argv);
 343                                 return -1;
 344                         }
 345                         a = atoi(argv[++i]);
 346                         cpu_affinities[next_aff++] = a;
 347                         use_affinity = 1;
 348                         printf_verbose("Adding CPU %d affinity\n", a);
 349                         break;
 350                 case 'c':
 351                         if (argc < i + 2) {
 352                                 show_usage(argc, argv);
 353                                 return -1;
 354                         }
 355                         rduration = atol(argv[++i]);
 356                         break;
 357                 case 'd':
 358                         if (argc < i + 2) {
 359                                 show_usage(argc, argv);
 360                                 return -1;
 361                         }
 362                         wdelay = atol(argv[++i]);
 363                         break;
 364                 case 'v':
 365                         verbose_mode = 1;
 366                         break;
 367                 }
 368         }
 369
 370         printf_verbose("running test for %lu seconds, %u enqueuers, "
 371                        "%u dequeuers.\n",
 372                        duration, nr_enqueuers, nr_dequeuers);
 373         printf_verbose("Writer delay : %lu loops.\n", rduration);
 374         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 375         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 376                         "main", pthread_self(), (unsigned long)gettid());
 377
 378         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 379         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 380         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 381         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 382         cds_lfq_init_rcu(&q, call_rcu);
 383         err = create_all_cpu_call_rcu_data(0);
 384         assert(!err);
 385
 386         next_aff = 0;
 387
 388         for (i = 0; i < nr_enqueuers; i++) {
 389                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 390                                      &count_enqueuer[2 * i]);
 391                 if (err != 0)
 392                         exit(1);
 393         }
 394         for (i = 0; i < nr_dequeuers; i++) {
 395                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 396                                      &count_dequeuer[2 * i]);
 397                 if (err != 0)
 398                         exit(1);
 399         }
 400
 401         cmm_smp_mb();
 402
 403         test_go = 1;
 404
 405         for (i = 0; i < duration; i++) {
 406                 sleep(1);
 407                 if (verbose_mode)
 408                         write (1, ".", 1);
 409         }
 410
 411         test_stop = 1;
 412
 413         for (i = 0; i < nr_enqueuers; i++) {
 414                 err = pthread_join(tid_enqueuer[i], &tret);
 415                 if (err != 0)
 416                         exit(1);
 417                 tot_enqueues += count_enqueuer[2 * i];
 418                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 419         }
 420         for (i = 0; i < nr_dequeuers; i++) {
 421                 err = pthread_join(tid_dequeuer[i], &tret);
 422                 if (err != 0)
 423                         exit(1);
 424                 tot_dequeues += count_dequeuer[2 * i];
 425                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 426         }
 427
 428         test_end(&q, &end_dequeues);
 429         err = cds_lfq_destroy_rcu(&q);
 430         assert(!err);
 431
 432         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 433                        tot_enqueues, tot_dequeues);
 434         printf_verbose("total number of successful enqueues : %llu, "
 435                        "successful dequeues %llu\n",
 436                        tot_successful_enqueues, tot_successful_dequeues);
 437         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 438                 "nr_dequeuers %3u "
 439                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 440                 "successful enqueues %12llu successful dequeues %12llu "
 441                 "end_dequeues %llu nr_ops %12llu\n",
 442                 argv[0], duration, nr_enqueuers, wdelay,
 443                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 444                 tot_successful_enqueues,
 445                 tot_successful_dequeues, end_dequeues,
 446                 tot_enqueues + tot_dequeues);
 447         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 448                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 449                        "succ. dequeues + end dequeues %llu.\n",
 450                        tot_successful_enqueues,
 451                        tot_successful_dequeues + end_dequeues);
 452
 453         free_all_cpu_call_rcu_data();
 454         free(count_enqueuer);
 455         free(count_dequeuer);
 456         free(tid_enqueuer);
 457         free(tid_dequeuer);
 458         return 0;
 459 }