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 #include <urcu.h>
  69 #include <urcu/wfqueue.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_wfq_queue q;
 162
 163 void *thr_enqueuer(void *_count)
 164 {
 165         unsigned long long *count = _count;
 166
 167         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 168                         "enqueuer", pthread_self(), (unsigned long)gettid());
 169
 170         set_affinity();
 171
 172         while (!test_go)
 173         {
 174         }
 175         cmm_smp_mb();
 176
 177         for (;;) {
 178                 struct cds_wfq_node *node = malloc(sizeof(*node));
 179                 if (!node)
 180                         goto fail;
 181                 cds_wfq_node_init(node);
 182                 cds_wfq_enqueue(&q, node);
 183                 URCU_TLS(nr_successful_enqueues)++;
 184
 185                 if (caa_unlikely(wdelay))
 186                         loop_sleep(wdelay);
 187 fail:
 188                 URCU_TLS(nr_enqueues)++;
 189                 if (caa_unlikely(!test_duration_enqueue()))
 190                         break;
 191         }
 192
 193         count[0] = URCU_TLS(nr_enqueues);
 194         count[1] = URCU_TLS(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(),
 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", pthread_self(), (unsigned long)gettid());
 209
 210         set_affinity();
 211
 212         while (!test_go)
 213         {
 214         }
 215         cmm_smp_mb();
 216
 217         for (;;) {
 218                 struct cds_wfq_node *node = cds_wfq_dequeue_blocking(&q);
 219
 220                 if (node) {
 221                         free(node);
 222                         URCU_TLS(nr_successful_dequeues)++;
 223                 }
 224
 225                 URCU_TLS(nr_dequeues)++;
 226                 if (caa_unlikely(!test_duration_dequeue()))
 227                         break;
 228                 if (caa_unlikely(rduration))
 229                         loop_sleep(rduration);
 230         }
 231
 232         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 233                        "dequeues %llu, successful_dequeues %llu\n",
 234                        pthread_self(), (unsigned long)gettid(),
 235                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 236         count[0] = URCU_TLS(nr_dequeues);
 237         count[1] = URCU_TLS(nr_successful_dequeues);
 238         return ((void*)2);
 239 }
 240
 241 void test_end(struct cds_wfq_queue *q, unsigned long long *nr_dequeues)
 242 {
 243         struct cds_wfq_node *node;
 244
 245         do {
 246                 node = cds_wfq_dequeue_blocking(q);
 247                 if (node) {
 248                         free(node);
 249                         (*nr_dequeues)++;
 250                 }
 251         } while (node);
 252 }
 253
 254 void show_usage(int argc, char **argv)
 255 {
 256         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 257         printf(" [-d delay] (enqueuer period (in loops))");
 258         printf(" [-c duration] (dequeuer period (in loops))");
 259         printf(" [-v] (verbose output)");
 260         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 261         printf("\n");
 262 }
 263
 264 int main(int argc, char **argv)
 265 {
 266         int err;
 267         pthread_t *tid_enqueuer, *tid_dequeuer;
 268         void *tret;
 269         unsigned long long *count_enqueuer, *count_dequeuer;
 270         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 271         unsigned long long tot_successful_enqueues = 0,
 272                            tot_successful_dequeues = 0;
 273         unsigned long long end_dequeues = 0;
 274         int i, a;
 275
 276         if (argc < 4) {
 277                 show_usage(argc, argv);
 278                 return -1;
 279         }
 280
 281         err = sscanf(argv[1], "%u", &nr_dequeuers);
 282         if (err != 1) {
 283                 show_usage(argc, argv);
 284                 return -1;
 285         }
 286
 287         err = sscanf(argv[2], "%u", &nr_enqueuers);
 288         if (err != 1) {
 289                 show_usage(argc, argv);
 290                 return -1;
 291         }
 292
 293         err = sscanf(argv[3], "%lu", &duration);
 294         if (err != 1) {
 295                 show_usage(argc, argv);
 296                 return -1;
 297         }
 298
 299         for (i = 4; i < argc; i++) {
 300                 if (argv[i][0] != '-')
 301                         continue;
 302                 switch (argv[i][1]) {
 303                 case 'a':
 304                         if (argc < i + 2) {
 305                                 show_usage(argc, argv);
 306                                 return -1;
 307                         }
 308                         a = atoi(argv[++i]);
 309                         cpu_affinities[next_aff++] = a;
 310                         use_affinity = 1;
 311                         printf_verbose("Adding CPU %d affinity\n", a);
 312                         break;
 313                 case 'c':
 314                         if (argc < i + 2) {
 315                                 show_usage(argc, argv);
 316                                 return -1;
 317                         }
 318                         rduration = atol(argv[++i]);
 319                         break;
 320                 case 'd':
 321                         if (argc < i + 2) {
 322                                 show_usage(argc, argv);
 323                                 return -1;
 324                         }
 325                         wdelay = atol(argv[++i]);
 326                         break;
 327                 case 'v':
 328                         verbose_mode = 1;
 329                         break;
 330                 }
 331         }
 332
 333         printf_verbose("running test for %lu seconds, %u enqueuers, "
 334                        "%u dequeuers.\n",
 335                        duration, nr_enqueuers, nr_dequeuers);
 336         printf_verbose("Writer delay : %lu loops.\n", rduration);
 337         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 338         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 339                         "main", pthread_self(), (unsigned long)gettid());
 340
 341         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 342         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 343         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 344         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 345         cds_wfq_init(&q);
 346
 347         next_aff = 0;
 348
 349         for (i = 0; i < nr_enqueuers; i++) {
 350                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 351                                      &count_enqueuer[2 * i]);
 352                 if (err != 0)
 353                         exit(1);
 354         }
 355         for (i = 0; i < nr_dequeuers; i++) {
 356                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 357                                      &count_dequeuer[2 * i]);
 358                 if (err != 0)
 359                         exit(1);
 360         }
 361
 362         cmm_smp_mb();
 363
 364         test_go = 1;
 365
 366         for (i = 0; i < duration; i++) {
 367                 sleep(1);
 368                 if (verbose_mode)
 369                         write (1, ".", 1);
 370         }
 371
 372         test_stop = 1;
 373
 374         for (i = 0; i < nr_enqueuers; i++) {
 375                 err = pthread_join(tid_enqueuer[i], &tret);
 376                 if (err != 0)
 377                         exit(1);
 378                 tot_enqueues += count_enqueuer[2 * i];
 379                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 380         }
 381         for (i = 0; i < nr_dequeuers; i++) {
 382                 err = pthread_join(tid_dequeuer[i], &tret);
 383                 if (err != 0)
 384                         exit(1);
 385                 tot_dequeues += count_dequeuer[2 * i];
 386                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 387         }
 388
 389         test_end(&q, &end_dequeues);
 390
 391         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 392                        tot_enqueues, tot_dequeues);
 393         printf_verbose("total number of successful enqueues : %llu, "
 394                        "successful dequeues %llu\n",
 395                        tot_successful_enqueues, tot_successful_dequeues);
 396         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 397                 "nr_dequeuers %3u "
 398                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 399                 "successful enqueues %12llu successful dequeues %12llu "
 400                 "end_dequeues %llu nr_ops %12llu\n",
 401                 argv[0], duration, nr_enqueuers, wdelay,
 402                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 403                 tot_successful_enqueues,
 404                 tot_successful_dequeues, end_dequeues,
 405                 tot_enqueues + tot_dequeues);
 406         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 407                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 408                        "succ. dequeues + end dequeues %llu.\n",
 409                        tot_successful_enqueues,
 410                        tot_successful_dequeues + end_dequeues);
 411
 412         free(count_enqueuer);
 413         free(count_dequeuer);
 414         free(tid_enqueuer);
 415         free(tid_dequeuer);
 416         return 0;
 417 }