tests/test_urcu_wfs.c

   1 /*
   2  * test_urcu_wfs.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/wfstack.h>
  69
  70 static volatile int test_go, test_stop;
  71
  72 static unsigned long rduration;
  73
  74 static unsigned long duration;
  75
  76 /* read-side C.S. duration, in loops */
  77 static unsigned long wdelay;
  78
  79 static inline void loop_sleep(unsigned long l)
  80 {
  81         while(l-- != 0)
  82                 caa_cpu_relax();
  83 }
  84
  85 static int verbose_mode;
  86
  87 #define printf_verbose(fmt, args...)            \
  88         do {                                    \
  89                 if (verbose_mode)               \
  90                         printf(fmt, args);      \
  91         } while (0)
  92
  93 static unsigned int cpu_affinities[NR_CPUS];
  94 static unsigned int next_aff = 0;
  95 static int use_affinity = 0;
  96
  97 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  98
  99 #ifndef HAVE_CPU_SET_T
 100 typedef unsigned long cpu_set_t;
 101 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 102 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 103 #endif
 104
 105 static void set_affinity(void)
 106 {
 107         cpu_set_t mask;
 108         int cpu;
 109         int ret;
 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 unsigned long long __thread nr_dequeues;
 151 static unsigned long long __thread nr_enqueues;
 152
 153 static unsigned long long __thread nr_successful_dequeues;
 154 static unsigned long long __thread nr_successful_enqueues;
 155
 156 static unsigned int nr_enqueuers;
 157 static unsigned int nr_dequeuers;
 158
 159 static struct cds_wfs_stack s;
 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", pthread_self(), (unsigned long)gettid());
 167
 168         set_affinity();
 169
 170         while (!test_go)
 171         {
 172         }
 173         cmm_smp_mb();
 174
 175         for (;;) {
 176                 struct cds_wfs_node *node = malloc(sizeof(*node));
 177                 if (!node)
 178                         goto fail;
 179                 cds_wfs_node_init(node);
 180                 cds_wfs_push(&s, node);
 181                 nr_successful_enqueues++;
 182
 183                 if (unlikely(wdelay))
 184                         loop_sleep(wdelay);
 185 fail:
 186                 nr_enqueues++;
 187                 if (unlikely(!test_duration_enqueue()))
 188                         break;
 189         }
 190
 191         count[0] = nr_enqueues;
 192         count[1] = nr_successful_enqueues;
 193         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 194                        "enqueues %llu successful_enqueues %llu\n",
 195                        pthread_self(), (unsigned long)gettid(), nr_enqueues,
 196                        nr_successful_enqueues);
 197         return ((void*)1);
 198
 199 }
 200
 201 void *thr_dequeuer(void *_count)
 202 {
 203         unsigned long long *count = _count;
 204
 205         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 206                         "dequeuer", pthread_self(), (unsigned long)gettid());
 207
 208         set_affinity();
 209
 210         while (!test_go)
 211         {
 212         }
 213         cmm_smp_mb();
 214
 215         for (;;) {
 216                 struct cds_wfs_node *node = cds_wfs_pop_blocking(&s);
 217
 218                 if (node) {
 219                         free(node);
 220                         nr_successful_dequeues++;
 221                 }
 222
 223                 nr_dequeues++;
 224                 if (unlikely(!test_duration_dequeue()))
 225                         break;
 226                 if (unlikely(rduration))
 227                         loop_sleep(rduration);
 228         }
 229
 230         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 231                        "dequeues %llu, successful_dequeues %llu\n",
 232                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 233                        nr_successful_dequeues);
 234         count[0] = nr_dequeues;
 235         count[1] = nr_successful_dequeues;
 236         return ((void*)2);
 237 }
 238
 239 void test_end(struct cds_wfs_stack *s, unsigned long long *nr_dequeues)
 240 {
 241         struct cds_wfs_node *node;
 242
 243         do {
 244                 node = cds_wfs_pop_blocking(s);
 245                 if (node) {
 246                         free(node);
 247                         (*nr_dequeues)++;
 248                 }
 249         } while (node);
 250 }
 251
 252 void show_usage(int argc, char **argv)
 253 {
 254         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 255         printf(" [-d delay] (enqueuer period (in loops))");
 256         printf(" [-c duration] (dequeuer period (in loops))");
 257         printf(" [-v] (verbose output)");
 258         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 259         printf("\n");
 260 }
 261
 262 int main(int argc, char **argv)
 263 {
 264         int err;
 265         pthread_t *tid_enqueuer, *tid_dequeuer;
 266         void *tret;
 267         unsigned long long *count_enqueuer, *count_dequeuer;
 268         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 269         unsigned long long tot_successful_enqueues = 0,
 270                            tot_successful_dequeues = 0;
 271         unsigned long long end_dequeues = 0;
 272         int i, a;
 273
 274         if (argc < 4) {
 275                 show_usage(argc, argv);
 276                 return -1;
 277         }
 278
 279         err = sscanf(argv[1], "%u", &nr_dequeuers);
 280         if (err != 1) {
 281                 show_usage(argc, argv);
 282                 return -1;
 283         }
 284
 285         err = sscanf(argv[2], "%u", &nr_enqueuers);
 286         if (err != 1) {
 287                 show_usage(argc, argv);
 288                 return -1;
 289         }
 290
 291         err = sscanf(argv[3], "%lu", &duration);
 292         if (err != 1) {
 293                 show_usage(argc, argv);
 294                 return -1;
 295         }
 296
 297         for (i = 4; i < argc; i++) {
 298                 if (argv[i][0] != '-')
 299                         continue;
 300                 switch (argv[i][1]) {
 301                 case 'a':
 302                         if (argc < i + 2) {
 303                                 show_usage(argc, argv);
 304                                 return -1;
 305                         }
 306                         a = atoi(argv[++i]);
 307                         cpu_affinities[next_aff++] = a;
 308                         use_affinity = 1;
 309                         printf_verbose("Adding CPU %d affinity\n", a);
 310                         break;
 311                 case 'c':
 312                         if (argc < i + 2) {
 313                                 show_usage(argc, argv);
 314                                 return -1;
 315                         }
 316                         rduration = atol(argv[++i]);
 317                         break;
 318                 case 'd':
 319                         if (argc < i + 2) {
 320                                 show_usage(argc, argv);
 321                                 return -1;
 322                         }
 323                         wdelay = atol(argv[++i]);
 324                         break;
 325                 case 'v':
 326                         verbose_mode = 1;
 327                         break;
 328                 }
 329         }
 330
 331         printf_verbose("running test for %lu seconds, %u enqueuers, "
 332                        "%u dequeuers.\n",
 333                        duration, nr_enqueuers, nr_dequeuers);
 334         printf_verbose("Writer delay : %lu loops.\n", rduration);
 335         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 336         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 337                         "main", pthread_self(), (unsigned long)gettid());
 338
 339         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 340         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 341         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 342         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 343         cds_wfs_init(&s);
 344
 345         next_aff = 0;
 346
 347         for (i = 0; i < nr_enqueuers; i++) {
 348                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 349                                      &count_enqueuer[2 * i]);
 350                 if (err != 0)
 351                         exit(1);
 352         }
 353         for (i = 0; i < nr_dequeuers; i++) {
 354                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 355                                      &count_dequeuer[2 * i]);
 356                 if (err != 0)
 357                         exit(1);
 358         }
 359
 360         cmm_smp_mb();
 361
 362         test_go = 1;
 363
 364         for (i = 0; i < duration; i++) {
 365                 sleep(1);
 366                 if (verbose_mode)
 367                         write (1, ".", 1);
 368         }
 369
 370         test_stop = 1;
 371
 372         for (i = 0; i < nr_enqueuers; i++) {
 373                 err = pthread_join(tid_enqueuer[i], &tret);
 374                 if (err != 0)
 375                         exit(1);
 376                 tot_enqueues += count_enqueuer[2 * i];
 377                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 378         }
 379         for (i = 0; i < nr_dequeuers; i++) {
 380                 err = pthread_join(tid_dequeuer[i], &tret);
 381                 if (err != 0)
 382                         exit(1);
 383                 tot_dequeues += count_dequeuer[2 * i];
 384                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 385         }
 386
 387         test_end(&s, &end_dequeues);
 388
 389         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 390                        tot_enqueues, tot_dequeues);
 391         printf_verbose("total number of successful enqueues : %llu, "
 392                        "successful dequeues %llu\n",
 393                        tot_successful_enqueues, tot_successful_dequeues);
 394         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 395                 "nr_dequeuers %3u "
 396                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 397                 "successful enqueues %12llu successful dequeues %12llu "
 398                 "end_dequeues %llu nr_ops %12llu\n",
 399                 argv[0], duration, nr_enqueuers, wdelay,
 400                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 401                 tot_successful_enqueues,
 402                 tot_successful_dequeues, end_dequeues,
 403                 tot_enqueues + tot_dequeues);
 404         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 405                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 406                        "succ. dequeues + end dequeues %llu.\n",
 407                        tot_successful_enqueues,
 408                        tot_successful_dequeues + end_dequeues);
 409
 410         free(count_enqueuer);
 411         free(count_dequeuer);
 412         free(tid_enqueuer);
 413         free(tid_dequeuer);
 414         return 0;
 415 }