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 <errno.h>
  38
  39 #include <urcu/arch.h>
  40 #include <urcu/tls-compat.h>
  41 #include "cpuset.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/cds.h>
  70 #include <urcu-defer.h>
  71
  72 static volatile int test_go, test_stop;
  73
  74 static unsigned long rduration;
  75
  76 static unsigned long duration;
  77
  78 /* read-side C.S. duration, in loops */
  79 static unsigned long wdelay;
  80
  81 static inline void loop_sleep(unsigned long l)
  82 {
  83         while(l-- != 0)
  84                 caa_cpu_relax();
  85 }
  86
  87 static int verbose_mode;
  88
  89 #define printf_verbose(fmt, args...)            \
  90         do {                                    \
  91                 if (verbose_mode)               \
  92                         printf(fmt, args);      \
  93         } while (0)
  94
  95 static unsigned int cpu_affinities[NR_CPUS];
  96 static unsigned int next_aff = 0;
  97 static int use_affinity = 0;
  98
  99 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 100
 101 static void set_affinity(void)
 102 {
 103         cpu_set_t mask;
 104         int cpu;
 105         int ret;
 106
 107         if (!use_affinity)
 108                 return;
 109
 110 #if HAVE_SCHED_SETAFFINITY
 111         ret = pthread_mutex_lock(&affinity_mutex);
 112         if (ret) {
 113                 perror("Error in pthread mutex lock");
 114                 exit(-1);
 115         }
 116         cpu = cpu_affinities[next_aff++];
 117         ret = pthread_mutex_unlock(&affinity_mutex);
 118         if (ret) {
 119                 perror("Error in pthread mutex unlock");
 120                 exit(-1);
 121         }
 122
 123         CPU_ZERO(&mask);
 124         CPU_SET(cpu, &mask);
 125 #if SCHED_SETAFFINITY_ARGS == 2
 126         sched_setaffinity(0, &mask);
 127 #else
 128         sched_setaffinity(0, sizeof(mask), &mask);
 129 #endif
 130 #endif /* HAVE_SCHED_SETAFFINITY */
 131 }
 132
 133 /*
 134  * returns 0 if test should end.
 135  */
 136 static int test_duration_dequeue(void)
 137 {
 138         return !test_stop;
 139 }
 140
 141 static int test_duration_enqueue(void)
 142 {
 143         return !test_stop;
 144 }
 145
 146 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
 147 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 148
 149 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
 150 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 151
 152 static unsigned int nr_enqueuers;
 153 static unsigned int nr_dequeuers;
 154
 155 struct test {
 156         struct cds_lfs_node_rcu list;
 157         struct rcu_head rcu;
 158 };
 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", (unsigned long) pthread_self(),
 168                         (unsigned long) gettid());
 169
 170         set_affinity();
 171
 172         rcu_register_thread();
 173
 174         while (!test_go)
 175         {
 176         }
 177         cmm_smp_mb();
 178
 179         for (;;) {
 180                 struct test *node = malloc(sizeof(*node));
 181                 if (!node)
 182                         goto fail;
 183                 cds_lfs_node_init_rcu(&node->list);
 184                 /* No rcu read-side is needed for push */
 185                 cds_lfs_push_rcu(&s, &node->list);
 186                 URCU_TLS(nr_successful_enqueues)++;
 187
 188                 if (caa_unlikely(wdelay))
 189                         loop_sleep(wdelay);
 190 fail:
 191                 URCU_TLS(nr_enqueues)++;
 192                 if (caa_unlikely(!test_duration_enqueue()))
 193                         break;
 194         }
 195
 196         rcu_unregister_thread();
 197
 198         count[0] = URCU_TLS(nr_enqueues);
 199         count[1] = URCU_TLS(nr_successful_enqueues);
 200         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 201                        "enqueues %llu successful_enqueues %llu\n",
 202                        pthread_self(),
 203                         (unsigned long) gettid(),
 204                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 205         return ((void*)1);
 206
 207 }
 208
 209 static
 210 void free_node_cb(struct rcu_head *head)
 211 {
 212         struct test *node =
 213                 caa_container_of(head, struct test, rcu);
 214         free(node);
 215 }
 216
 217 void *thr_dequeuer(void *_count)
 218 {
 219         unsigned long long *count = _count;
 220         int ret;
 221
 222         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 223                         "dequeuer", (unsigned long) pthread_self(),
 224                         (unsigned long) gettid());
 225
 226         set_affinity();
 227
 228         ret = rcu_defer_register_thread();
 229         if (ret) {
 230                 printf("Error in rcu_defer_register_thread\n");
 231                 exit(-1);
 232         }
 233         rcu_register_thread();
 234
 235         while (!test_go)
 236         {
 237         }
 238         cmm_smp_mb();
 239
 240         for (;;) {
 241                 struct cds_lfs_node_rcu *snode;
 242                 struct test *node;
 243
 244                 rcu_read_lock();
 245                 snode = cds_lfs_pop_rcu(&s);
 246                 node = caa_container_of(snode, struct test, list);
 247                 rcu_read_unlock();
 248                 if (node) {
 249                         call_rcu(&node->rcu, free_node_cb);
 250                         URCU_TLS(nr_successful_dequeues)++;
 251                 }
 252                 URCU_TLS(nr_dequeues)++;
 253                 if (caa_unlikely(!test_duration_dequeue()))
 254                         break;
 255                 if (caa_unlikely(rduration))
 256                         loop_sleep(rduration);
 257         }
 258
 259         rcu_unregister_thread();
 260         rcu_defer_unregister_thread();
 261
 262         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 263                        "dequeues %llu, successful_dequeues %llu\n",
 264                        pthread_self(),
 265                         (unsigned long) gettid(),
 266                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 267         count[0] = URCU_TLS(nr_dequeues);
 268         count[1] = URCU_TLS(nr_successful_dequeues);
 269         return ((void*)2);
 270 }
 271
 272 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 273 {
 274         struct cds_lfs_node_rcu *snode;
 275
 276         do {
 277                 snode = cds_lfs_pop_rcu(s);
 278                 if (snode) {
 279                         struct test *node;
 280
 281                         node = caa_container_of(snode, struct test, list);
 282                         free(node);
 283                         (*nr_dequeues)++;
 284                 }
 285         } while (snode);
 286 }
 287
 288 void show_usage(int argc, char **argv)
 289 {
 290         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 291         printf(" [-d delay] (enqueuer period (in loops))");
 292         printf(" [-c duration] (dequeuer period (in loops))");
 293         printf(" [-v] (verbose output)");
 294         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 295         printf("\n");
 296 }
 297
 298 int main(int argc, char **argv)
 299 {
 300         int err;
 301         pthread_t *tid_enqueuer, *tid_dequeuer;
 302         void *tret;
 303         unsigned long long *count_enqueuer, *count_dequeuer;
 304         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 305         unsigned long long tot_successful_enqueues = 0,
 306                            tot_successful_dequeues = 0;
 307         unsigned long long end_dequeues = 0;
 308         int i, a;
 309
 310         if (argc < 4) {
 311                 show_usage(argc, argv);
 312                 return -1;
 313         }
 314
 315         err = sscanf(argv[1], "%u", &nr_dequeuers);
 316         if (err != 1) {
 317                 show_usage(argc, argv);
 318                 return -1;
 319         }
 320
 321         err = sscanf(argv[2], "%u", &nr_enqueuers);
 322         if (err != 1) {
 323                 show_usage(argc, argv);
 324                 return -1;
 325         }
 326
 327         err = sscanf(argv[3], "%lu", &duration);
 328         if (err != 1) {
 329                 show_usage(argc, argv);
 330                 return -1;
 331         }
 332
 333         for (i = 4; i < argc; i++) {
 334                 if (argv[i][0] != '-')
 335                         continue;
 336                 switch (argv[i][1]) {
 337                 case 'a':
 338                         if (argc < i + 2) {
 339                                 show_usage(argc, argv);
 340                                 return -1;
 341                         }
 342                         a = atoi(argv[++i]);
 343                         cpu_affinities[next_aff++] = a;
 344                         use_affinity = 1;
 345                         printf_verbose("Adding CPU %d affinity\n", a);
 346                         break;
 347                 case 'c':
 348                         if (argc < i + 2) {
 349                                 show_usage(argc, argv);
 350                                 return -1;
 351                         }
 352                         rduration = atol(argv[++i]);
 353                         break;
 354                 case 'd':
 355                         if (argc < i + 2) {
 356                                 show_usage(argc, argv);
 357                                 return -1;
 358                         }
 359                         wdelay = atol(argv[++i]);
 360                         break;
 361                 case 'v':
 362                         verbose_mode = 1;
 363                         break;
 364                 }
 365         }
 366
 367         printf_verbose("running test for %lu seconds, %u enqueuers, "
 368                        "%u dequeuers.\n",
 369                        duration, nr_enqueuers, nr_dequeuers);
 370         printf_verbose("Writer delay : %lu loops.\n", rduration);
 371         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 372         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 373                         "main", (unsigned long) pthread_self(),
 374                         (unsigned long) gettid());
 375
 376         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 377         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 378         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 379         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 380         cds_lfs_init_rcu(&s);
 381         err = create_all_cpu_call_rcu_data(0);
 382         if (err) {
 383                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 384         }
 385
 386         next_aff = 0;
 387
 388         for (i = 0; i < nr_enqueuers; i++) {
 389                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 390                                      &count_enqueuer[2 * i]);
 391                 if (err != 0)
 392                         exit(1);
 393         }
 394         for (i = 0; i < nr_dequeuers; i++) {
 395                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 396                                      &count_dequeuer[2 * i]);
 397                 if (err != 0)
 398                         exit(1);
 399         }
 400
 401         cmm_smp_mb();
 402
 403         test_go = 1;
 404
 405         for (i = 0; i < duration; i++) {
 406                 sleep(1);
 407                 if (verbose_mode)
 408                         write (1, ".", 1);
 409         }
 410
 411         test_stop = 1;
 412
 413         for (i = 0; i < nr_enqueuers; i++) {
 414                 err = pthread_join(tid_enqueuer[i], &tret);
 415                 if (err != 0)
 416                         exit(1);
 417                 tot_enqueues += count_enqueuer[2 * i];
 418                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 419         }
 420         for (i = 0; i < nr_dequeuers; i++) {
 421                 err = pthread_join(tid_dequeuer[i], &tret);
 422                 if (err != 0)
 423                         exit(1);
 424                 tot_dequeues += count_dequeuer[2 * i];
 425                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 426         }
 427
 428         test_end(&s, &end_dequeues);
 429
 430         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 431                        tot_enqueues, tot_dequeues);
 432         printf_verbose("total number of successful enqueues : %llu, "
 433                        "successful dequeues %llu\n",
 434                        tot_successful_enqueues, tot_successful_dequeues);
 435         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 436                 "nr_dequeuers %3u "
 437                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 438                 "successful enqueues %12llu successful dequeues %12llu "
 439                 "end_dequeues %llu nr_ops %12llu\n",
 440                 argv[0], duration, nr_enqueuers, wdelay,
 441                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 442                 tot_successful_enqueues,
 443                 tot_successful_dequeues, end_dequeues,
 444                 tot_enqueues + tot_dequeues);
 445         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 446                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 447                        "succ. dequeues + end dequeues %llu.\n",
 448                        tot_successful_enqueues,
 449                        tot_successful_dequeues + end_dequeues);
 450
 451         free_all_cpu_call_rcu_data();
 452         free(count_enqueuer);
 453         free(count_dequeuer);
 454         free(tid_enqueuer);
 455         free(tid_dequeuer);
 456         return 0;
 457 }