tests/test_urcu_lfs.c

   1 /*
   2  * test_urcu_lfs.c
   3  *
   4  * Userspace RCU library - example RCU-based lock-free stack
   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
  42 #ifdef __linux__
  43 #include <syscall.h>
  44 #endif
  45
  46 /* hardcoded number of CPUs */
  47 #define NR_CPUS 16384
  48
  49 #if defined(_syscall0)
  50 _syscall0(pid_t, gettid)
  51 #elif defined(__NR_gettid)
  52 static inline pid_t gettid(void)
  53 {
  54         return syscall(__NR_gettid);
  55 }
  56 #else
  57 #warning "use pid as tid"
  58 static inline pid_t gettid(void)
  59 {
  60         return getpid();
  61 }
  62 #endif
  63
  64 #ifndef DYNAMIC_LINK_TEST
  65 #define _LGPL_SOURCE
  66 #endif
  67 #include <urcu.h>
  68 #include <urcu/cds.h>
  69 #include <urcu-defer.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 l)
  81 {
  82         while(l-- != 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         cpu_set_t mask;
 109         int cpu;
 110         int ret;
 111
 112         if (!use_affinity)
 113                 return;
 114
 115 #if HAVE_SCHED_SETAFFINITY
 116         ret = pthread_mutex_lock(&affinity_mutex);
 117         if (ret) {
 118                 perror("Error in pthread mutex lock");
 119                 exit(-1);
 120         }
 121         cpu = cpu_affinities[next_aff++];
 122         ret = pthread_mutex_unlock(&affinity_mutex);
 123         if (ret) {
 124                 perror("Error in pthread mutex unlock");
 125                 exit(-1);
 126         }
 127
 128         CPU_ZERO(&mask);
 129         CPU_SET(cpu, &mask);
 130 #if SCHED_SETAFFINITY_ARGS == 2
 131         sched_setaffinity(0, &mask);
 132 #else
 133         sched_setaffinity(0, sizeof(mask), &mask);
 134 #endif
 135 #endif /* HAVE_SCHED_SETAFFINITY */
 136 }
 137
 138 /*
 139  * returns 0 if test should end.
 140  */
 141 static int test_duration_dequeue(void)
 142 {
 143         return !test_stop;
 144 }
 145
 146 static int test_duration_enqueue(void)
 147 {
 148         return !test_stop;
 149 }
 150
 151 static unsigned long long __thread nr_dequeues;
 152 static unsigned long long __thread nr_enqueues;
 153
 154 static unsigned long long __thread nr_successful_dequeues;
 155 static unsigned long long __thread nr_successful_enqueues;
 156
 157 static unsigned int nr_enqueuers;
 158 static unsigned int nr_dequeuers;
 159
 160 static struct cds_lfs_stack_rcu s;
 161
 162 void *thr_enqueuer(void *_count)
 163 {
 164         unsigned long long *count = _count;
 165
 166         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 167                         "enqueuer", pthread_self(), (unsigned long)gettid());
 168
 169         set_affinity();
 170
 171         rcu_register_thread();
 172
 173         while (!test_go)
 174         {
 175         }
 176         cmm_smp_mb();
 177
 178         for (;;) {
 179                 struct cds_lfs_node_rcu *node = malloc(sizeof(*node));
 180                 if (!node)
 181                         goto fail;
 182                 cds_lfs_node_init_rcu(node);
 183                 /* No rcu read-side is needed for push */
 184                 cds_lfs_push_rcu(&s, node);
 185                 nr_successful_enqueues++;
 186
 187                 if (unlikely(wdelay))
 188                         loop_sleep(wdelay);
 189 fail:
 190                 nr_enqueues++;
 191                 if (unlikely(!test_duration_enqueue()))
 192                         break;
 193         }
 194
 195         rcu_unregister_thread();
 196
 197         count[0] = nr_enqueues;
 198         count[1] = 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(), nr_enqueues,
 202                        nr_successful_enqueues);
 203         return ((void*)1);
 204
 205 }
 206
 207 void *thr_dequeuer(void *_count)
 208 {
 209         unsigned long long *count = _count;
 210         int ret;
 211
 212         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 213                         "dequeuer", pthread_self(), (unsigned long)gettid());
 214
 215         set_affinity();
 216
 217         ret = rcu_defer_register_thread();
 218         if (ret) {
 219                 printf("Error in rcu_defer_register_thread\n");
 220                 exit(-1);
 221         }
 222         rcu_register_thread();
 223
 224         while (!test_go)
 225         {
 226         }
 227         cmm_smp_mb();
 228
 229         for (;;) {
 230                 struct cds_lfs_node_rcu *node;
 231
 232                 rcu_read_lock();
 233                 node = cds_lfs_pop_rcu(&s);
 234                 rcu_read_unlock();
 235                 if (node) {
 236                         defer_rcu(free, node);
 237                         nr_successful_dequeues++;
 238                 }
 239                 nr_dequeues++;
 240                 if (unlikely(!test_duration_dequeue()))
 241                         break;
 242                 if (unlikely(rduration))
 243                         loop_sleep(rduration);
 244         }
 245
 246         rcu_unregister_thread();
 247         rcu_defer_unregister_thread();
 248
 249         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 250                        "dequeues %llu, successful_dequeues %llu\n",
 251                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 252                        nr_successful_dequeues);
 253         count[0] = nr_dequeues;
 254         count[1] = nr_successful_dequeues;
 255         return ((void*)2);
 256 }
 257
 258 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 259 {
 260         struct cds_lfs_node_rcu *node;
 261
 262         do {
 263                 node = cds_lfs_pop_rcu(s);
 264                 if (node) {
 265                         free(node);
 266                         (*nr_dequeues)++;
 267                 }
 268         } while (node);
 269 }
 270
 271 void show_usage(int argc, char **argv)
 272 {
 273         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 274         printf(" [-d delay] (enqueuer period (in loops))");
 275         printf(" [-c duration] (dequeuer period (in loops))");
 276         printf(" [-v] (verbose output)");
 277         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 278         printf("\n");
 279 }
 280
 281 int main(int argc, char **argv)
 282 {
 283         int err;
 284         pthread_t *tid_enqueuer, *tid_dequeuer;
 285         void *tret;
 286         unsigned long long *count_enqueuer, *count_dequeuer;
 287         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 288         unsigned long long tot_successful_enqueues = 0,
 289                            tot_successful_dequeues = 0;
 290         unsigned long long end_dequeues = 0;
 291         int i, a;
 292
 293         if (argc < 4) {
 294                 show_usage(argc, argv);
 295                 return -1;
 296         }
 297
 298         err = sscanf(argv[1], "%u", &nr_dequeuers);
 299         if (err != 1) {
 300                 show_usage(argc, argv);
 301                 return -1;
 302         }
 303
 304         err = sscanf(argv[2], "%u", &nr_enqueuers);
 305         if (err != 1) {
 306                 show_usage(argc, argv);
 307                 return -1;
 308         }
 309
 310         err = sscanf(argv[3], "%lu", &duration);
 311         if (err != 1) {
 312                 show_usage(argc, argv);
 313                 return -1;
 314         }
 315
 316         for (i = 4; i < argc; i++) {
 317                 if (argv[i][0] != '-')
 318                         continue;
 319                 switch (argv[i][1]) {
 320                 case 'a':
 321                         if (argc < i + 2) {
 322                                 show_usage(argc, argv);
 323                                 return -1;
 324                         }
 325                         a = atoi(argv[++i]);
 326                         cpu_affinities[next_aff++] = a;
 327                         use_affinity = 1;
 328                         printf_verbose("Adding CPU %d affinity\n", a);
 329                         break;
 330                 case 'c':
 331                         if (argc < i + 2) {
 332                                 show_usage(argc, argv);
 333                                 return -1;
 334                         }
 335                         rduration = atol(argv[++i]);
 336                         break;
 337                 case 'd':
 338                         if (argc < i + 2) {
 339                                 show_usage(argc, argv);
 340                                 return -1;
 341                         }
 342                         wdelay = atol(argv[++i]);
 343                         break;
 344                 case 'v':
 345                         verbose_mode = 1;
 346                         break;
 347                 }
 348         }
 349
 350         printf_verbose("running test for %lu seconds, %u enqueuers, "
 351                        "%u dequeuers.\n",
 352                        duration, nr_enqueuers, nr_dequeuers);
 353         printf_verbose("Writer delay : %lu loops.\n", rduration);
 354         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 355         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 356                         "main", pthread_self(), (unsigned long)gettid());
 357
 358         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 359         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 360         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 361         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 362         cds_lfs_init_rcu(&s);
 363
 364         next_aff = 0;
 365
 366         for (i = 0; i < nr_enqueuers; i++) {
 367                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 368                                      &count_enqueuer[2 * i]);
 369                 if (err != 0)
 370                         exit(1);
 371         }
 372         for (i = 0; i < nr_dequeuers; i++) {
 373                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 374                                      &count_dequeuer[2 * i]);
 375                 if (err != 0)
 376                         exit(1);
 377         }
 378
 379         cmm_smp_mb();
 380
 381         test_go = 1;
 382
 383         for (i = 0; i < duration; i++) {
 384                 sleep(1);
 385                 if (verbose_mode)
 386                         write (1, ".", 1);
 387         }
 388
 389         test_stop = 1;
 390
 391         for (i = 0; i < nr_enqueuers; i++) {
 392                 err = pthread_join(tid_enqueuer[i], &tret);
 393                 if (err != 0)
 394                         exit(1);
 395                 tot_enqueues += count_enqueuer[2 * i];
 396                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 397         }
 398         for (i = 0; i < nr_dequeuers; i++) {
 399                 err = pthread_join(tid_dequeuer[i], &tret);
 400                 if (err != 0)
 401                         exit(1);
 402                 tot_dequeues += count_dequeuer[2 * i];
 403                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 404         }
 405
 406         test_end(&s, &end_dequeues);
 407
 408         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 409                        tot_enqueues, tot_dequeues);
 410         printf_verbose("total number of successful enqueues : %llu, "
 411                        "successful dequeues %llu\n",
 412                        tot_successful_enqueues, tot_successful_dequeues);
 413         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 414                 "nr_dequeuers %3u "
 415                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 416                 "successful enqueues %12llu successful dequeues %12llu "
 417                 "end_dequeues %llu nr_ops %12llu\n",
 418                 argv[0], duration, nr_enqueuers, wdelay,
 419                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 420                 tot_successful_enqueues,
 421                 tot_successful_dequeues, end_dequeues,
 422                 tot_enqueues + tot_dequeues);
 423         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 424                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 425                        "succ. dequeues + end dequeues %llu.\n",
 426                        tot_successful_enqueues,
 427                        tot_successful_dequeues + end_dequeues);
 428
 429         free(count_enqueuer);
 430         free(count_dequeuer);
 431         free(tid_enqueuer);
 432         free(tid_dequeuer);
 433         return 0;
 434 }