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 <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", pthread_self(), (unsigned long)gettid());
 173
 174         set_affinity();
 175
 176         while (!test_go)
 177         {
 178         }
 179         cmm_smp_mb();
 180
 181         for (;;) {
 182                 struct cds_wfq_node *node = malloc(sizeof(*node));
 183                 if (!node)
 184                         goto fail;
 185                 cds_wfq_node_init(node);
 186                 cds_wfq_enqueue(&q, node);
 187                 URCU_TLS(nr_successful_enqueues)++;
 188
 189                 if (caa_unlikely(wdelay))
 190                         loop_sleep(wdelay);
 191 fail:
 192                 URCU_TLS(nr_enqueues)++;
 193                 if (caa_unlikely(!test_duration_enqueue()))
 194                         break;
 195         }
 196
 197         count[0] = URCU_TLS(nr_enqueues);
 198         count[1] = URCU_TLS(nr_successful_enqueues);
 199         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 200                        "enqueues %llu successful_enqueues %llu\n",
 201                        pthread_self(), (unsigned long)gettid(),
 202                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 203         return ((void*)1);
 204
 205 }
 206
 207 void *thr_dequeuer(void *_count)
 208 {
 209         unsigned long long *count = _count;
 210
 211         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 212                         "dequeuer", pthread_self(), (unsigned long)gettid());
 213
 214         set_affinity();
 215
 216         while (!test_go)
 217         {
 218         }
 219         cmm_smp_mb();
 220
 221         for (;;) {
 222                 struct cds_wfq_node *node = cds_wfq_dequeue_blocking(&q);
 223
 224                 if (node) {
 225                         free(node);
 226                         URCU_TLS(nr_successful_dequeues)++;
 227                 }
 228
 229                 URCU_TLS(nr_dequeues)++;
 230                 if (caa_unlikely(!test_duration_dequeue()))
 231                         break;
 232                 if (caa_unlikely(rduration))
 233                         loop_sleep(rduration);
 234         }
 235
 236         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 237                        "dequeues %llu, successful_dequeues %llu\n",
 238                        pthread_self(), (unsigned long)gettid(),
 239                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 240         count[0] = URCU_TLS(nr_dequeues);
 241         count[1] = URCU_TLS(nr_successful_dequeues);
 242         return ((void*)2);
 243 }
 244
 245 void test_end(struct cds_wfq_queue *q, unsigned long long *nr_dequeues)
 246 {
 247         struct cds_wfq_node *node;
 248
 249         do {
 250                 node = cds_wfq_dequeue_blocking(q);
 251                 if (node) {
 252                         free(node);
 253                         (*nr_dequeues)++;
 254                 }
 255         } while (node);
 256 }
 257
 258 void show_usage(int argc, char **argv)
 259 {
 260         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 261         printf(" [-d delay] (enqueuer period (in loops))");
 262         printf(" [-c duration] (dequeuer period (in loops))");
 263         printf(" [-v] (verbose output)");
 264         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 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", pthread_self(), (unsigned long)gettid());
 344
 345         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 346         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 347         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 348         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 349         cds_wfq_init(&q);
 350
 351         next_aff = 0;
 352
 353         for (i = 0; i < nr_enqueuers; i++) {
 354                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 355                                      &count_enqueuer[2 * i]);
 356                 if (err != 0)
 357                         exit(1);
 358         }
 359         for (i = 0; i < nr_dequeuers; i++) {
 360                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 361                                      &count_dequeuer[2 * i]);
 362                 if (err != 0)
 363                         exit(1);
 364         }
 365
 366         cmm_smp_mb();
 367
 368         test_go = 1;
 369
 370         for (i = 0; i < duration; i++) {
 371                 sleep(1);
 372                 if (verbose_mode)
 373                         write (1, ".", 1);
 374         }
 375
 376         test_stop = 1;
 377
 378         for (i = 0; i < nr_enqueuers; i++) {
 379                 err = pthread_join(tid_enqueuer[i], &tret);
 380                 if (err != 0)
 381                         exit(1);
 382                 tot_enqueues += count_enqueuer[2 * i];
 383                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 384         }
 385         for (i = 0; i < nr_dequeuers; i++) {
 386                 err = pthread_join(tid_dequeuer[i], &tret);
 387                 if (err != 0)
 388                         exit(1);
 389                 tot_dequeues += count_dequeuer[2 * i];
 390                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 391         }
 392
 393         test_end(&q, &end_dequeues);
 394
 395         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 396                        tot_enqueues, tot_dequeues);
 397         printf_verbose("total number of successful enqueues : %llu, "
 398                        "successful dequeues %llu\n",
 399                        tot_successful_enqueues, tot_successful_dequeues);
 400         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 401                 "nr_dequeuers %3u "
 402                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 403                 "successful enqueues %12llu successful dequeues %12llu "
 404                 "end_dequeues %llu nr_ops %12llu\n",
 405                 argv[0], duration, nr_enqueuers, wdelay,
 406                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 407                 tot_successful_enqueues,
 408                 tot_successful_dequeues, end_dequeues,
 409                 tot_enqueues + tot_dequeues);
 410         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 411                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 412                        "succ. dequeues + end dequeues %llu.\n",
 413                        tot_successful_enqueues,
 414                        tot_successful_dequeues + end_dequeues);
 415
 416         free(count_enqueuer);
 417         free(count_dequeuer);
 418         free(tid_enqueuer);
 419         free(tid_dequeuer);
 420         return 0;
 421 }