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 #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/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 #ifndef HAVE_CPU_SET_T
 102 typedef unsigned long cpu_set_t;
 103 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 104 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 105 #endif
 106
 107 static void set_affinity(void)
 108 {
 109         cpu_set_t mask;
 110         int cpu;
 111         int ret;
 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 struct test {
 162         struct cds_lfs_node_rcu list;
 163         struct rcu_head rcu;
 164 };
 165
 166 static struct cds_lfs_stack_rcu s;
 167
 168 void *thr_enqueuer(void *_count)
 169 {
 170         unsigned long long *count = _count;
 171
 172         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 173                         "enqueuer", (unsigned long) pthread_self(),
 174                         (unsigned long) gettid());
 175
 176         set_affinity();
 177
 178         rcu_register_thread();
 179
 180         while (!test_go)
 181         {
 182         }
 183         cmm_smp_mb();
 184
 185         for (;;) {
 186                 struct test *node = malloc(sizeof(*node));
 187                 if (!node)
 188                         goto fail;
 189                 cds_lfs_node_init_rcu(&node->list);
 190                 /* No rcu read-side is needed for push */
 191                 cds_lfs_push_rcu(&s, &node->list);
 192                 URCU_TLS(nr_successful_enqueues)++;
 193
 194                 if (caa_unlikely(wdelay))
 195                         loop_sleep(wdelay);
 196 fail:
 197                 URCU_TLS(nr_enqueues)++;
 198                 if (caa_unlikely(!test_duration_enqueue()))
 199                         break;
 200         }
 201
 202         rcu_unregister_thread();
 203
 204         count[0] = URCU_TLS(nr_enqueues);
 205         count[1] = URCU_TLS(nr_successful_enqueues);
 206         printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 207                        "enqueues %llu successful_enqueues %llu\n",
 208                        pthread_self(),
 209                         (unsigned long) gettid(),
 210                        URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 211         return ((void*)1);
 212
 213 }
 214
 215 static
 216 void free_node_cb(struct rcu_head *head)
 217 {
 218         struct test *node =
 219                 caa_container_of(head, struct test, rcu);
 220         free(node);
 221 }
 222
 223 void *thr_dequeuer(void *_count)
 224 {
 225         unsigned long long *count = _count;
 226         int ret;
 227
 228         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 229                         "dequeuer", (unsigned long) pthread_self(),
 230                         (unsigned long) gettid());
 231
 232         set_affinity();
 233
 234         ret = rcu_defer_register_thread();
 235         if (ret) {
 236                 printf("Error in rcu_defer_register_thread\n");
 237                 exit(-1);
 238         }
 239         rcu_register_thread();
 240
 241         while (!test_go)
 242         {
 243         }
 244         cmm_smp_mb();
 245
 246         for (;;) {
 247                 struct cds_lfs_node_rcu *snode;
 248                 struct test *node;
 249
 250                 rcu_read_lock();
 251                 snode = cds_lfs_pop_rcu(&s);
 252                 node = caa_container_of(snode, struct test, list);
 253                 rcu_read_unlock();
 254                 if (node) {
 255                         call_rcu(&node->rcu, free_node_cb);
 256                         URCU_TLS(nr_successful_dequeues)++;
 257                 }
 258                 URCU_TLS(nr_dequeues)++;
 259                 if (caa_unlikely(!test_duration_dequeue()))
 260                         break;
 261                 if (caa_unlikely(rduration))
 262                         loop_sleep(rduration);
 263         }
 264
 265         rcu_unregister_thread();
 266         rcu_defer_unregister_thread();
 267
 268         printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 269                        "dequeues %llu, successful_dequeues %llu\n",
 270                        pthread_self(),
 271                         (unsigned long) gettid(),
 272                        URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
 273         count[0] = URCU_TLS(nr_dequeues);
 274         count[1] = URCU_TLS(nr_successful_dequeues);
 275         return ((void*)2);
 276 }
 277
 278 void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues)
 279 {
 280         struct cds_lfs_node_rcu *snode;
 281
 282         do {
 283                 snode = cds_lfs_pop_rcu(s);
 284                 if (snode) {
 285                         struct test *node;
 286
 287                         node = caa_container_of(snode, struct test, list);
 288                         free(node);
 289                         (*nr_dequeues)++;
 290                 }
 291         } while (snode);
 292 }
 293
 294 void show_usage(int argc, char **argv)
 295 {
 296         printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]);
 297         printf(" [-d delay] (enqueuer period (in loops))");
 298         printf(" [-c duration] (dequeuer period (in loops))");
 299         printf(" [-v] (verbose output)");
 300         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 301         printf("\n");
 302 }
 303
 304 int main(int argc, char **argv)
 305 {
 306         int err;
 307         pthread_t *tid_enqueuer, *tid_dequeuer;
 308         void *tret;
 309         unsigned long long *count_enqueuer, *count_dequeuer;
 310         unsigned long long tot_enqueues = 0, tot_dequeues = 0;
 311         unsigned long long tot_successful_enqueues = 0,
 312                            tot_successful_dequeues = 0;
 313         unsigned long long end_dequeues = 0;
 314         int i, a;
 315
 316         if (argc < 4) {
 317                 show_usage(argc, argv);
 318                 return -1;
 319         }
 320
 321         err = sscanf(argv[1], "%u", &nr_dequeuers);
 322         if (err != 1) {
 323                 show_usage(argc, argv);
 324                 return -1;
 325         }
 326
 327         err = sscanf(argv[2], "%u", &nr_enqueuers);
 328         if (err != 1) {
 329                 show_usage(argc, argv);
 330                 return -1;
 331         }
 332
 333         err = sscanf(argv[3], "%lu", &duration);
 334         if (err != 1) {
 335                 show_usage(argc, argv);
 336                 return -1;
 337         }
 338
 339         for (i = 4; i < argc; i++) {
 340                 if (argv[i][0] != '-')
 341                         continue;
 342                 switch (argv[i][1]) {
 343                 case 'a':
 344                         if (argc < i + 2) {
 345                                 show_usage(argc, argv);
 346                                 return -1;
 347                         }
 348                         a = atoi(argv[++i]);
 349                         cpu_affinities[next_aff++] = a;
 350                         use_affinity = 1;
 351                         printf_verbose("Adding CPU %d affinity\n", a);
 352                         break;
 353                 case 'c':
 354                         if (argc < i + 2) {
 355                                 show_usage(argc, argv);
 356                                 return -1;
 357                         }
 358                         rduration = atol(argv[++i]);
 359                         break;
 360                 case 'd':
 361                         if (argc < i + 2) {
 362                                 show_usage(argc, argv);
 363                                 return -1;
 364                         }
 365                         wdelay = atol(argv[++i]);
 366                         break;
 367                 case 'v':
 368                         verbose_mode = 1;
 369                         break;
 370                 }
 371         }
 372
 373         printf_verbose("running test for %lu seconds, %u enqueuers, "
 374                        "%u dequeuers.\n",
 375                        duration, nr_enqueuers, nr_dequeuers);
 376         printf_verbose("Writer delay : %lu loops.\n", rduration);
 377         printf_verbose("Reader duration : %lu loops.\n", wdelay);
 378         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 379                         "main", (unsigned long) pthread_self(),
 380                         (unsigned long) gettid());
 381
 382         tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers);
 383         tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers);
 384         count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers);
 385         count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers);
 386         cds_lfs_init_rcu(&s);
 387         err = create_all_cpu_call_rcu_data(0);
 388         if (err) {
 389                 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
 390         }
 391
 392         next_aff = 0;
 393
 394         for (i = 0; i < nr_enqueuers; i++) {
 395                 err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer,
 396                                      &count_enqueuer[2 * i]);
 397                 if (err != 0)
 398                         exit(1);
 399         }
 400         for (i = 0; i < nr_dequeuers; i++) {
 401                 err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer,
 402                                      &count_dequeuer[2 * i]);
 403                 if (err != 0)
 404                         exit(1);
 405         }
 406
 407         cmm_smp_mb();
 408
 409         test_go = 1;
 410
 411         for (i = 0; i < duration; i++) {
 412                 sleep(1);
 413                 if (verbose_mode)
 414                         write (1, ".", 1);
 415         }
 416
 417         test_stop = 1;
 418
 419         for (i = 0; i < nr_enqueuers; i++) {
 420                 err = pthread_join(tid_enqueuer[i], &tret);
 421                 if (err != 0)
 422                         exit(1);
 423                 tot_enqueues += count_enqueuer[2 * i];
 424                 tot_successful_enqueues += count_enqueuer[2 * i + 1];
 425         }
 426         for (i = 0; i < nr_dequeuers; i++) {
 427                 err = pthread_join(tid_dequeuer[i], &tret);
 428                 if (err != 0)
 429                         exit(1);
 430                 tot_dequeues += count_dequeuer[2 * i];
 431                 tot_successful_dequeues += count_dequeuer[2 * i + 1];
 432         }
 433
 434         test_end(&s, &end_dequeues);
 435
 436         printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
 437                        tot_enqueues, tot_dequeues);
 438         printf_verbose("total number of successful enqueues : %llu, "
 439                        "successful dequeues %llu\n",
 440                        tot_successful_enqueues, tot_successful_dequeues);
 441         printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
 442                 "nr_dequeuers %3u "
 443                 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
 444                 "successful enqueues %12llu successful dequeues %12llu "
 445                 "end_dequeues %llu nr_ops %12llu\n",
 446                 argv[0], duration, nr_enqueuers, wdelay,
 447                 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
 448                 tot_successful_enqueues,
 449                 tot_successful_dequeues, end_dequeues,
 450                 tot_enqueues + tot_dequeues);
 451         if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
 452                 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
 453                        "succ. dequeues + end dequeues %llu.\n",
 454                        tot_successful_enqueues,
 455                        tot_successful_dequeues + end_dequeues);
 456
 457         free_all_cpu_call_rcu_data();
 458         free(count_enqueuer);
 459         free(count_dequeuer);
 460         free(tid_enqueuer);
 461         free(tid_dequeuer);
 462         return 0;
 463 }