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 <sys/syscall.h>
  38 #include <sched.h>
  39 #include <errno.h>
  40
  41 #include <urcu/arch.h>
  42
  43 /* hardcoded number of CPUs */
  44 #define NR_CPUS 16384
  45
  46 #if defined(_syscall0)
  47 _syscall0(pid_t, gettid)
  48 #elif defined(__NR_gettid)
  49 static inline pid_t gettid(void)
  50 {
  51         return syscall(__NR_gettid);
  52 }
  53 #else
  54 #warning "use pid as tid"
  55 static inline pid_t gettid(void)
  56 {
  57         return getpid();
  58 }
  59 #endif
  60
  61 #ifndef DYNAMIC_LINK_TEST
  62 #define _LGPL_SOURCE
  63 #endif
  64 #include <urcu.h>
  65 #include <urcu/wfstack.h>
  66
  67 static volatile int test_go, test_stop;
  68
  69 static unsigned long rduration;
  70
  71 static unsigned long duration;
  72
  73 /* read-side C.S. duration, in loops */
  74 static unsigned long wdelay;
  75
  76 static inline void loop_sleep(unsigned long l)
  77 {
  78         while(l-- != 0)
  79                 cpu_relax();
  80 }
  81
  82 static int verbose_mode;
  83
  84 #define printf_verbose(fmt, args...)            \
  85         do {                                    \
  86                 if (verbose_mode)               \
  87                         printf(fmt, args);      \
  88         } while (0)
  89
  90 static unsigned int cpu_affinities[NR_CPUS];
  91 static unsigned int next_aff = 0;
  92 static int use_affinity = 0;
  93
  94 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  95
  96 #ifndef HAVE_CPU_SET_T
  97 typedef unsigned long cpu_set_t;
  98 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
  99 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 100 #endif
 101
 102 static void set_affinity(void)
 103 {
 104         cpu_set_t mask;
 105         int cpu;
 106         int ret;
 107
 108         if (!use_affinity)
 109                 return;
 110
 111 #if HAVE_SCHED_SETAFFINITY
 112         ret = pthread_mutex_lock(&affinity_mutex);
 113         if (ret) {
 114                 perror("Error in pthread mutex lock");
 115                 exit(-1);
 116         }
 117         cpu = cpu_affinities[next_aff++];
 118         ret = pthread_mutex_unlock(&affinity_mutex);
 119         if (ret) {
 120                 perror("Error in pthread mutex unlock");
 121                 exit(-1);
 122         }
 123
 124         CPU_ZERO(&mask);
 125         CPU_SET(cpu, &mask);
 126 #if SCHED_SETAFFINITY_ARGS == 2
 127         sched_setaffinity(0, &mask);
 128 #else
 129         sched_setaffinity(0, sizeof(mask), &mask);
 130 #endif
 131 #endif /* HAVE_SCHED_SETAFFINITY */
 132 }
 133
 134 /*
 135  * returns 0 if test should end.
 136  */
 137 static int test_duration_dequeue(void)
 138 {
 139         return !test_stop;
 140 }
 141
 142 static int test_duration_enqueue(void)
 143 {
 144         return !test_stop;
 145 }
 146
 147 static unsigned long long __thread nr_dequeues;
 148 static unsigned long long __thread nr_enqueues;
 149
 150 static unsigned long long __thread nr_successful_dequeues;
 151 static unsigned long long __thread nr_successful_enqueues;
 152
 153 static unsigned int nr_enqueuers;
 154 static unsigned int nr_dequeuers;
 155
 156 static struct wfs_stack s;
 157
 158 void *thr_enqueuer(void *_count)
 159 {
 160         unsigned long long *count = _count;
 161
 162         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 163                         "enqueuer", pthread_self(), (unsigned long)gettid());
 164
 165         set_affinity();
 166
 167         while (!test_go)
 168         {
 169         }
 170         cmm_smp_mb();
 171
 172         for (;;) {
 173                 struct wfs_node *node = malloc(sizeof(*node));
 174                 if (!node)
 175                         goto fail;
 176                 wfs_node_init(node);
 177                 wfs_push(&s, node);
 178                 nr_successful_enqueues++;
 179
 180                 if (unlikely(wdelay))
 181                         loop_sleep(wdelay);
 182 fail:
 183                 nr_enqueues++;
 184                 if (unlikely(!test_duration_enqueue()))
 185                         break;
 186         }
 187
 188         count[0] = nr_enqueues;
 189         count[1] = nr_successful_enqueues;
 190         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 191                        "enqueues %llu successful_enqueues %llu\n",
 192                        pthread_self(), (unsigned long)gettid(), nr_enqueues,
 193                        nr_successful_enqueues);
 194         return ((void*)1);
 195
 196 }
 197
 198 void *thr_dequeuer(void *_count)
 199 {
 200         unsigned long long *count = _count;
 201
 202         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 203                         "dequeuer", pthread_self(), (unsigned long)gettid());
 204
 205         set_affinity();
 206
 207         while (!test_go)
 208         {
 209         }
 210         cmm_smp_mb();
 211
 212         for (;;) {
 213                 struct wfs_node *node = wfs_pop_blocking(&s);
 214
 215                 if (node) {
 216                         free(node);
 217                         nr_successful_dequeues++;
 218                 }
 219
 220                 nr_dequeues++;
 221                 if (unlikely(!test_duration_dequeue()))
 222                         break;
 223                 if (unlikely(rduration))
 224                         loop_sleep(rduration);
 225         }
 226
 227         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 228                        "dequeues %llu, successful_dequeues %llu\n",
 229                        pthread_self(), (unsigned long)gettid(), nr_dequeues,
 230                        nr_successful_dequeues);
 231         count[0] = nr_dequeues;
 232         count[1] = nr_successful_dequeues;
 233         return ((void*)2);
 234 }
 235
 236 void test_end(struct wfs_stack *s, unsigned long long *nr_dequeues)
 237 {
 238         struct wfs_node *node;
 239
 240         do {
 241                 node = wfs_pop_blocking(s);
 242                 if (node) {
 243                         free(node);
 244                         (*nr_dequeues)++;
 245                 }
 246         } while (node);
 247 }
 248
 249 void show_usage(int argc, char **argv)
 250 {
 251         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 252         printf(" [-d delay] (enqueuer period (in loops))");
 253         printf(" [-c duration] (dequeuer period (in loops))");
 254         printf(" [-v] (verbose output)");
 255         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 256         printf("\n");
 257 }
 258
 259 int main(int argc, char **argv)
 260 {
 261         int err;
 262         pthread_t *tid_enqueuer, *tid_dequeuer;
 263         void *tret;
 264         unsigned long long *count_enqueuer, *count_dequeuer;
 265         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 266         unsigned long long tot_successful_enqueues = 0,
 267                            tot_successful_dequeues = 0;
 268         unsigned long long end_dequeues = 0;
 269         int i, a;
 270
 271         if (argc < 4) {
 272                 show_usage(argc, argv);
 273                 return -1;
 274         }
 275
 276         err = sscanf(argv[1], "%u", &nr_dequeuers);
 277         if (err != 1) {
 278                 show_usage(argc, argv);
 279                 return -1;
 280         }
 281
 282         err = sscanf(argv[2], "%u", &nr_enqueuers);
 283         if (err != 1) {
 284                 show_usage(argc, argv);
 285                 return -1;
 286         }
 287
 288         err = sscanf(argv[3], "%lu", &duration);
 289         if (err != 1) {
 290                 show_usage(argc, argv);
 291                 return -1;
 292         }
 293
 294         for (i = 4; i < argc; i++) {
 295                 if (argv[i][0] != '-')
 296                         continue;
 297                 switch (argv[i][1]) {
 298                 case 'a':
 299                         if (argc < i + 2) {
 300                                 show_usage(argc, argv);
 301                                 return -1;
 302                         }
 303                         a = atoi(argv[++i]);
 304                         cpu_affinities[next_aff++] = a;
 305                         use_affinity = 1;
 306                         printf_verbose("Adding CPU %d affinity\n", a);
 307                         break;
 308                 case 'c':
 309                         if (argc < i + 2) {
 310                                 show_usage(argc, argv);
 311                                 return -1;
 312                         }
 313                         rduration = atol(argv[++i]);
 314                         break;
 315                 case 'd':
 316                         if (argc < i + 2) {
 317                                 show_usage(argc, argv);
 318                                 return -1;
 319                         }
 320                         wdelay = atol(argv[++i]);
 321                         break;
 322                 case 'v':
 323                         verbose_mode = 1;
 324                         break;
 325                 }
 326         }
 327
 328         printf_verbose("running test for %lu seconds, %u enqueuers, "
 329                        "%u dequeuers.\n",
 330                        duration, nr_enqueuers, nr_dequeuers);
 331         printf_verbose("Writer delay : %lu loops.\n", rduration);
 332         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 333         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 334                         "main", pthread_self(), (unsigned long)gettid());
 335
 336         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 337         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 338         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 339         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 340         wfs_init(&s);
 341
 342         next_aff = 0;
 343
 344         for (i = 0; i < nr_enqueuers; i++) {
 345                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 346                                      &count_enqueuer[2 * i]);
 347                 if (err != 0)
 348                         exit(1);
 349         }
 350         for (i = 0; i < nr_dequeuers; i++) {
 351                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 352                                      &count_dequeuer[2 * i]);
 353                 if (err != 0)
 354                         exit(1);
 355         }
 356
 357         cmm_smp_mb();
 358
 359         test_go = 1;
 360
 361         for (i = 0; i < duration; i++) {
 362                 sleep(1);
 363                 if (verbose_mode)
 364                         write (1, ".", 1);
 365         }
 366
 367         test_stop = 1;
 368
 369         for (i = 0; i < nr_enqueuers; i++) {
 370                 err = pthread_join(tid_enqueuer[i], &tret);
 371                 if (err != 0)
 372                         exit(1);
 373                 tot_enqueues += count_enqueuer[2 * i];
 374                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 375         }
 376         for (i = 0; i < nr_dequeuers; i++) {
 377                 err = pthread_join(tid_dequeuer[i], &tret);
 378                 if (err != 0)
 379                         exit(1);
 380                 tot_dequeues += count_dequeuer[2 * i];
 381                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 382         }
 383
 384         test_end(&s, &end_dequeues);
 385
 386         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 387                        tot_enqueues, tot_dequeues);
 388         printf_verbose("total number of successful enqueues : %llu, "
 389                        "successful dequeues %llu\n",
 390                        tot_successful_enqueues, tot_successful_dequeues);
 391         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 392                 "nr_dequeuers %3u "
 393                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 394                 "successful enqueues %12llu successful dequeues %12llu "
 395                 "end_dequeues %llu nr_ops %12llu\n",
 396                 argv[0], duration, nr_enqueuers, wdelay,
 397                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 398                 tot_successful_enqueues,
 399                 tot_successful_dequeues, end_dequeues,
 400                 tot_enqueues + tot_dequeues);
 401         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 402                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 403                        "succ. dequeues + end dequeues %llu.\n",
 404                        tot_successful_enqueues,
 405                        tot_successful_dequeues + end_dequeues);
 406
 407         free(count_enqueuer);
 408         free(count_dequeuer);
 409         free(tid_enqueuer);
 410         free(tid_dequeuer);
 411         return 0;
 412 }