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