tests/test_urcu_wfcq.c

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