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