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 <sys/syscall.h>
  38 #include <sched.h>
  39 #include <errno.h>
  40
  41 #include <urcu/arch.h>
  42
  43 /* hardcoded number of CPUs */
  44 #define NR_CPUS 16384
  45
  46 #if defined(_syscall0)
  47 _syscall0(pid_t, gettid)
  48 #elif defined(__NR_gettid)
  49 static inline pid_t gettid(void)
  50 {
  51         return syscall(__NR_gettid);
  52 }
  53 #else
  54 #warning "use pid as tid"
  55 static inline pid_t gettid(void)
  56 {
  57         return getpid();
  58 }
  59 #endif
  60
  61 #ifndef DYNAMIC_LINK_TEST
  62 #define _LGPL_SOURCE
  63 #endif
  64 #include <urcu.h>
  65 #include <urcu/rculfqueue.h>
  66
  67 static volatile int test_go, test_stop;
  68
  69 static unsigned long rduration;
  70
  71 static unsigned long duration;
  72
  73 /* read-side C.S. duration, in loops */
  74 static unsigned long wdelay;
  75
  76 static inline void loop_sleep(unsigned long l)
  77 {
  78         while(l-- != 0)
  79                 caa_cpu_relax();
  80 }
  81
  82 static int verbose_mode;
  83
  84 #define printf_verbose(fmt, args...)            \
  85         do {                                    \
  86                 if (verbose_mode)               \
  87                         printf(fmt, args);      \
  88         } while (0)
  89
  90 static unsigned int cpu_affinities[NR_CPUS];
  91 static unsigned int next_aff = 0;
  92 static int use_affinity = 0;
  93
  94 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  95
  96 #ifndef HAVE_CPU_SET_T
  97 typedef unsigned long cpu_set_t;
  98 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
  99 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 100 #endif
 101
 102 static void set_affinity(void)
 103 {
 104         cpu_set_t mask;
 105         int cpu;
 106         int ret;
 107
 108         if (!use_affinity)
 109                 return;
 110
 111 #if HAVE_SCHED_SETAFFINITY
 112         ret = pthread_mutex_lock(&affinity_mutex);
 113         if (ret) {
 114                 perror("Error in pthread mutex lock");
 115                 exit(-1);
 116         }
 117         cpu = cpu_affinities[next_aff++];
 118         ret = pthread_mutex_unlock(&affinity_mutex);
 119         if (ret) {
 120                 perror("Error in pthread mutex unlock");
 121                 exit(-1);
 122         }
 123
 124         CPU_ZERO(&mask);
 125         CPU_SET(cpu, &mask);
 126 #if SCHED_SETAFFINITY_ARGS == 2
 127         sched_setaffinity(0, &mask);
 128 #else
 129         sched_setaffinity(0, sizeof(mask), &mask);
 130 #endif
 131 #endif /* HAVE_SCHED_SETAFFINITY */
 132 }
 133
 134 /*
 135  * returns 0 if test should end.
 136  */
 137 static int test_duration_dequeue(void)
 138 {
 139         return !test_stop;
 140 }
 141
 142 static int test_duration_enqueue(void)
 143 {
 144         return !test_stop;
 145 }
 146
 147 static unsigned long long __thread nr_dequeues;
 148 static unsigned long long __thread nr_enqueues;
 149
 150 static unsigned long long __thread nr_successful_dequeues;
 151 static unsigned long long __thread nr_successful_enqueues;
 152
 153 static unsigned int nr_enqueuers;
 154 static unsigned int nr_dequeuers;
 155
 156 static struct cds_lfq_queue_rcu q;
 157
 158 void *thr_enqueuer(void *_count)
 159 {
 160         unsigned long long *count = _count;
 161
 162         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 163                         "enqueuer", pthread_self(), (unsigned long)gettid());
 164
 165         set_affinity();
 166
 167         rcu_register_thread();
 168
 169         while (!test_go)
 170         {
 171         }
 172         cmm_smp_mb();
 173
 174         for (;;) {
 175                 struct cds_lfq_node_rcu *node = malloc(sizeof(*node));
 176                 if (!node)
 177                         goto fail;
 178                 cds_lfq_node_init_rcu(node);
 179                 rcu_read_lock();
 180                 cds_lfq_enqueue_rcu(&q, node);
 181                 rcu_read_unlock();
 182                 nr_successful_enqueues++;
 183
 184                 if (unlikely(wdelay))
 185                         loop_sleep(wdelay);
 186 fail:
 187                 nr_enqueues++;
 188                 if (unlikely(!test_duration_enqueue()))
 189                         break;
 190         }
 191
 192         rcu_unregister_thread();
 193
 194         count[0] = nr_enqueues;
 195         count[1] = nr_successful_enqueues;
 196         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 197                        "enqueues %llu successful_enqueues %llu\n",
 198                        pthread_self(), (unsigned long)gettid(), nr_enqueues,
 199                        nr_successful_enqueues);
 200         return ((void*)1);
 201
 202 }
 203
 204 static void rcu_free_node(struct rcu_head *head)
 205 {
 206         struct cds_lfq_node_rcu *node =
 207                 caa_container_of(head, struct cds_lfq_node_rcu, rcu_head);
 208         free(node);
 209 }
 210
 211 static void ref_release_node(struct urcu_ref *ref)
 212 {
 213         struct cds_lfq_node_rcu *node =
 214                 caa_container_of(ref, struct cds_lfq_node_rcu, ref);
 215         call_rcu(&node->rcu_head, rcu_free_node);
 216 }
 217
 218 void *thr_dequeuer(void *_count)
 219 {
 220         unsigned long long *count = _count;
 221         int ret;
 222
 223         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 224                         "dequeuer", pthread_self(), (unsigned long)gettid());
 225
 226         set_affinity();
 227
 228         rcu_register_thread();
 229
 230         while (!test_go)
 231         {
 232         }
 233         cmm_smp_mb();
 234
 235         for (;;) {
 236                 struct cds_lfq_node_rcu *node;
 237
 238                 rcu_read_lock();
 239                 node = cds_lfq_dequeue_rcu(&q);
 240                 rcu_read_unlock();
 241
 242                 if (node) {
 243                         urcu_ref_put(&node->ref, ref_release_node);
 244                         nr_successful_dequeues++;
 245                 }
 246
 247                 nr_dequeues++;
 248                 if (unlikely(!test_duration_dequeue()))
 249                         break;
 250                 if (unlikely(rduration))
 251                         loop_sleep(rduration);
 252         }
 253
 254         rcu_unregister_thread();
 255         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 256                        "dequeues %llu, successful_dequeues %llu\n",
 257                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 258                        nr_successful_dequeues);
 259         count[0] = nr_dequeues;
 260         count[1] = nr_successful_dequeues;
 261         return ((void*)2);
 262 }
 263
 264 static void release_node(struct urcu_ref *ref)
 265 {
 266         struct cds_lfq_node_rcu *node = caa_container_of(ref, struct cds_lfq_node_rcu, ref);
 267         free(node);
 268 }
 269
 270 void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues)
 271 {
 272         struct cds_lfq_node_rcu *node;
 273
 274         do {
 275                 rcu_read_lock();
 276                 node = cds_lfq_dequeue_rcu(q);
 277                 rcu_read_unlock();
 278                 if (node) {
 279                         urcu_ref_put(&node->ref, release_node);
 280                         (*nr_dequeues)++;
 281                 }
 282         } while (node);
 283 }
 284
 285 void show_usage(int argc, char **argv)
 286 {
 287         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 288         printf(" [-d delay] (enqueuer period (in loops))");
 289         printf(" [-c duration] (dequeuer period (in loops))");
 290         printf(" [-v] (verbose output)");
 291         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 292         printf("\n");
 293 }
 294
 295 int main(int argc, char **argv)
 296 {
 297         int err;
 298         pthread_t *tid_enqueuer, *tid_dequeuer;
 299         void *tret;
 300         unsigned long long *count_enqueuer, *count_dequeuer;
 301         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 302         unsigned long long tot_successful_enqueues = 0,
 303                            tot_successful_dequeues = 0;
 304         unsigned long long end_dequeues = 0;
 305         int i, a;
 306
 307         if (argc < 4) {
 308                 show_usage(argc, argv);
 309                 return -1;
 310         }
 311
 312         err = sscanf(argv[1], "%u", &nr_dequeuers);
 313         if (err != 1) {
 314                 show_usage(argc, argv);
 315                 return -1;
 316         }
 317
 318         err = sscanf(argv[2], "%u", &nr_enqueuers);
 319         if (err != 1) {
 320                 show_usage(argc, argv);
 321                 return -1;
 322         }
 323
 324         err = sscanf(argv[3], "%lu", &duration);
 325         if (err != 1) {
 326                 show_usage(argc, argv);
 327                 return -1;
 328         }
 329
 330         for (i = 4; i < argc; i++) {
 331                 if (argv[i][0] != '-')
 332                         continue;
 333                 switch (argv[i][1]) {
 334                 case 'a':
 335                         if (argc < i + 2) {
 336                                 show_usage(argc, argv);
 337                                 return -1;
 338                         }
 339                         a = atoi(argv[++i]);
 340                         cpu_affinities[next_aff++] = a;
 341                         use_affinity = 1;
 342                         printf_verbose("Adding CPU %d affinity\n", a);
 343                         break;
 344                 case 'c':
 345                         if (argc < i + 2) {
 346                                 show_usage(argc, argv);
 347                                 return -1;
 348                         }
 349                         rduration = atol(argv[++i]);
 350                         break;
 351                 case 'd':
 352                         if (argc < i + 2) {
 353                                 show_usage(argc, argv);
 354                                 return -1;
 355                         }
 356                         wdelay = atol(argv[++i]);
 357                         break;
 358                 case 'v':
 359                         verbose_mode = 1;
 360                         break;
 361                 }
 362         }
 363
 364         printf_verbose("running test for %lu seconds, %u enqueuers, "
 365                        "%u dequeuers.\n",
 366                        duration, nr_enqueuers, nr_dequeuers);
 367         printf_verbose("Writer delay : %lu loops.\n", rduration);
 368         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 369         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 370                         "main", pthread_self(), (unsigned long)gettid());
 371
 372         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 373         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 374         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 375         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 376         cds_lfq_init_rcu(&q, ref_release_node);
 377
 378         next_aff = 0;
 379
 380         for (i = 0; i < nr_enqueuers; i++) {
 381                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 382                                      &count_enqueuer[2 * i]);
 383                 if (err != 0)
 384                         exit(1);
 385         }
 386         for (i = 0; i < nr_dequeuers; i++) {
 387                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 388                                      &count_dequeuer[2 * i]);
 389                 if (err != 0)
 390                         exit(1);
 391         }
 392
 393         cmm_smp_mb();
 394
 395         test_go = 1;
 396
 397         for (i = 0; i < duration; i++) {
 398                 sleep(1);
 399                 if (verbose_mode)
 400                         write (1, ".", 1);
 401         }
 402
 403         test_stop = 1;
 404
 405         for (i = 0; i < nr_enqueuers; i++) {
 406                 err = pthread_join(tid_enqueuer[i], &tret);
 407                 if (err != 0)
 408                         exit(1);
 409                 tot_enqueues += count_enqueuer[2 * i];
 410                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 411         }
 412         for (i = 0; i < nr_dequeuers; i++) {
 413                 err = pthread_join(tid_dequeuer[i], &tret);
 414                 if (err != 0)
 415                         exit(1);
 416                 tot_dequeues += count_dequeuer[2 * i];
 417                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 418         }
 419
 420         test_end(&q, &end_dequeues);
 421
 422         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 423                        tot_enqueues, tot_dequeues);
 424         printf_verbose("total number of successful enqueues : %llu, "
 425                        "successful dequeues %llu\n",
 426                        tot_successful_enqueues, tot_successful_dequeues);
 427         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 428                 "nr_dequeuers %3u "
 429                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 430                 "successful enqueues %12llu successful dequeues %12llu "
 431                 "end_dequeues %llu nr_ops %12llu\n",
 432                 argv[0], duration, nr_enqueuers, wdelay,
 433                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 434                 tot_successful_enqueues,
 435                 tot_successful_dequeues, end_dequeues,
 436                 tot_enqueues + tot_dequeues);
 437         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 438                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 439                        "succ. dequeues + end dequeues %llu.\n",
 440                        tot_successful_enqueues,
 441                        tot_successful_dequeues + end_dequeues);
 442
 443         free(count_enqueuer);
 444         free(count_dequeuer);
 445         free(tid_enqueuer);
 446         free(tid_dequeuer);
 447         return 0;
 448 }