tests/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 <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include "cpuset.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 static void set_affinity(void)
 105 {
 106 #if HAVE_SCHED_SETAFFINITY
 107         cpu_set_t mask;
 108         int cpu, ret;
 109 #endif /* HAVE_SCHED_SETAFFINITY */
 110
 111         if (!use_affinity)
 112                 return;
 113
 114 #if HAVE_SCHED_SETAFFINITY
 115         ret = pthread_mutex_lock(&affinity_mutex);
 116         if (ret) {
 117                 perror("Error in pthread mutex lock");
 118                 exit(-1);
 119         }
 120         cpu = cpu_affinities[next_aff++];
 121         ret = pthread_mutex_unlock(&affinity_mutex);
 122         if (ret) {
 123                 perror("Error in pthread mutex unlock");
 124                 exit(-1);
 125         }
 126
 127         CPU_ZERO(&mask);
 128         CPU_SET(cpu, &mask);
 129 #if SCHED_SETAFFINITY_ARGS == 2
 130         sched_setaffinity(0, &mask);
 131 #else
 132         sched_setaffinity(0, sizeof(mask), &mask);
 133 #endif
 134 #endif /* HAVE_SCHED_SETAFFINITY */
 135 }
 136
 137 /*
 138  * returns 0 if test should end.
 139  */
 140 static int test_duration_dequeue(void)
 141 {
 142         return !test_stop;
 143 }
 144
 145 static int test_duration_enqueue(void)
 146 {
 147         return !test_stop;
 148 }
 149
 150 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 151 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 152
 153 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 154 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 155
 156 static unsigned int nr_enqueuers;
 157 static unsigned int nr_dequeuers;
 158
 159 static struct cds_wfq_queue q;
 160
 161 void *thr_enqueuer(void *_count)
 162 {
 163         unsigned long long *count = _count;
 164
 165         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 166                         "enqueuer", (unsigned long) pthread_self(),
 167                         (unsigned long) gettid());
 168
 169         set_affinity();
 170
 171         while (!test_go)
 172         {
 173         }
 174         cmm_smp_mb();
 175
 176         for (;;) {
 177                 struct cds_wfq_node *node = malloc(sizeof(*node));
 178                 if (!node)
 179                         goto fail;
 180                 cds_wfq_node_init(node);
 181                 cds_wfq_enqueue(&q, node);
 182                 URCU_TLS(nr_successful_enqueues)++;
 183
 184                 if (caa_unlikely(wdelay))
 185                         loop_sleep(wdelay);
 186 fail:
 187                 URCU_TLS(nr_enqueues)++;
 188                 if (caa_unlikely(!test_duration_enqueue()))
 189                         break;
 190         }
 191
 192         count[0] = URCU_TLS(nr_enqueues);
 193         count[1] = URCU_TLS(nr_successful_enqueues);
 194         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 195                        "enqueues %llu successful_enqueues %llu\n",
 196                        (unsigned long) pthread_self(),
 197                         (unsigned long) gettid(),
 198                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 199         return ((void*)1);
 200
 201 }
 202
 203 void *thr_dequeuer(void *_count)
 204 {
 205         unsigned long long *count = _count;
 206
 207         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 208                         "dequeuer", (unsigned long) pthread_self(),
 209                         (unsigned long) gettid());
 210
 211         set_affinity();
 212
 213         while (!test_go)
 214         {
 215         }
 216         cmm_smp_mb();
 217
 218         for (;;) {
 219                 struct cds_wfq_node *node = cds_wfq_dequeue_blocking(&q);
 220
 221                 if (node) {
 222                         free(node);
 223                         URCU_TLS(nr_successful_dequeues)++;
 224                 }
 225
 226                 URCU_TLS(nr_dequeues)++;
 227                 if (caa_unlikely(!test_duration_dequeue()))
 228                         break;
 229                 if (caa_unlikely(rduration))
 230                         loop_sleep(rduration);
 231         }
 232
 233         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 234                        "dequeues %llu, successful_dequeues %llu\n",
 235                        (unsigned long) pthread_self(),
 236                         (unsigned long) gettid(),
 237                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 238         count[0] = URCU_TLS(nr_dequeues);
 239         count[1] = URCU_TLS(nr_successful_dequeues);
 240         return ((void*)2);
 241 }
 242
 243 void test_end(struct cds_wfq_queue *q, unsigned long long *nr_dequeues)
 244 {
 245         struct cds_wfq_node *node;
 246
 247         do {
 248                 node = cds_wfq_dequeue_blocking(q);
 249                 if (node) {
 250                         free(node);
 251                         (*nr_dequeues)++;
 252                 }
 253         } while (node);
 254 }
 255
 256 void show_usage(int argc, char **argv)
 257 {
 258         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
 259                 argv[0]);
 260         printf("OPTIONS:\n");
 261         printf("        [-d delay] (enqueuer period (in loops))\n");
 262         printf("        [-c duration] (dequeuer period (in loops))\n");
 263         printf("        [-v] (verbose output)\n");
 264         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 265         printf("\n");
 266 }
 267
 268 int main(int argc, char **argv)
 269 {
 270         int err;
 271         pthread_t *tid_enqueuer, *tid_dequeuer;
 272         void *tret;
 273         unsigned long long *count_enqueuer, *count_dequeuer;
 274         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 275         unsigned long long tot_successful_enqueues = 0,
 276                            tot_successful_dequeues = 0;
 277         unsigned long long end_dequeues = 0;
 278         int i, a;
 279
 280         if (argc < 4) {
 281                 show_usage(argc, argv);
 282                 return -1;
 283         }
 284
 285         err = sscanf(argv[1], "%u", &nr_dequeuers);
 286         if (err != 1) {
 287                 show_usage(argc, argv);
 288                 return -1;
 289         }
 290
 291         err = sscanf(argv[2], "%u", &nr_enqueuers);
 292         if (err != 1) {
 293                 show_usage(argc, argv);
 294                 return -1;
 295         }
 296
 297         err = sscanf(argv[3], "%lu", &duration);
 298         if (err != 1) {
 299                 show_usage(argc, argv);
 300                 return -1;
 301         }
 302
 303         for (i = 4; i < argc; i++) {
 304                 if (argv[i][0] != '-')
 305                         continue;
 306                 switch (argv[i][1]) {
 307                 case 'a':
 308                         if (argc < i + 2) {
 309                                 show_usage(argc, argv);
 310                                 return -1;
 311                         }
 312                         a = atoi(argv[++i]);
 313                         cpu_affinities[next_aff++] = a;
 314                         use_affinity = 1;
 315                         printf_verbose("Adding CPU %d affinity\n", a);
 316                         break;
 317                 case 'c':
 318                         if (argc < i + 2) {
 319                                 show_usage(argc, argv);
 320                                 return -1;
 321                         }
 322                         rduration = atol(argv[++i]);
 323                         break;
 324                 case 'd':
 325                         if (argc < i + 2) {
 326                                 show_usage(argc, argv);
 327                                 return -1;
 328                         }
 329                         wdelay = atol(argv[++i]);
 330                         break;
 331                 case 'v':
 332                         verbose_mode = 1;
 333                         break;
 334                 }
 335         }
 336
 337         printf_verbose("running test for %lu seconds, %u enqueuers, "
 338                        "%u dequeuers.\n",
 339                        duration, nr_enqueuers, nr_dequeuers);
 340         printf_verbose("Writer delay : %lu loops.\n", rduration);
 341         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 342         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 343                         "main", (unsigned long) pthread_self(),
 344                         (unsigned long) gettid());
 345
 346         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 347         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 348         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 349         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 350         cds_wfq_init(&q);
 351
 352         next_aff = 0;
 353
 354         for (i = 0; i < nr_enqueuers; i++) {
 355                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 356                                      &count_enqueuer[2 * i]);
 357                 if (err != 0)
 358                         exit(1);
 359         }
 360         for (i = 0; i < nr_dequeuers; i++) {
 361                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 362                                      &count_dequeuer[2 * i]);
 363                 if (err != 0)
 364                         exit(1);
 365         }
 366
 367         cmm_smp_mb();
 368
 369         test_go = 1;
 370
 371         for (i = 0; i < duration; i++) {
 372                 sleep(1);
 373                 if (verbose_mode)
 374                         write (1, ".", 1);
 375         }
 376
 377         test_stop = 1;
 378
 379         for (i = 0; i < nr_enqueuers; i++) {
 380                 err = pthread_join(tid_enqueuer[i], &tret);
 381                 if (err != 0)
 382                         exit(1);
 383                 tot_enqueues += count_enqueuer[2 * i];
 384                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 385         }
 386         for (i = 0; i < nr_dequeuers; i++) {
 387                 err = pthread_join(tid_dequeuer[i], &tret);
 388                 if (err != 0)
 389                         exit(1);
 390                 tot_dequeues += count_dequeuer[2 * i];
 391                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 392         }
 393
 394         test_end(&q, &end_dequeues);
 395
 396         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 397                        tot_enqueues, tot_dequeues);
 398         printf_verbose("total number of successful enqueues : %llu, "
 399                        "successful dequeues %llu\n",
 400                        tot_successful_enqueues, tot_successful_dequeues);
 401         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 402                 "nr_dequeuers %3u "
 403                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 404                 "successful enqueues %12llu successful dequeues %12llu "
 405                 "end_dequeues %llu nr_ops %12llu\n",
 406                 argv[0], duration, nr_enqueuers, wdelay,
 407                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 408                 tot_successful_enqueues,
 409                 tot_successful_dequeues, end_dequeues,
 410                 tot_enqueues + tot_dequeues);
 411         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 412                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 413                        "succ. dequeues + end dequeues %llu.\n",
 414                        tot_successful_enqueues,
 415                        tot_successful_dequeues + end_dequeues);
 416
 417         free(count_enqueuer);
 418         free(count_dequeuer);
 419         free(tid_enqueuer);
 420         free(tid_dequeuer);
 421         return 0;
 422 }