test_urcu_wfq.c

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