tests/test_urcu_rbtree.c

   1 /*
   2  * test_urcu_rbtree.c
   3  *
   4  * Userspace RCU library - test program for RB tree
   5  *
   6  * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with this program; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  21  */
  22
  23 #define _GNU_SOURCE
  24 #ifndef DYNAMIC_LINK_TEST
  25 #define _LGPL_SOURCE
  26 #else
  27 #define debug_yield_read()
  28 #endif
  29 #include "../config.h"
  30 #include <stdio.h>
  31 #include <pthread.h>
  32 #include <stdlib.h>
  33 #include <string.h>
  34 #include <sys/types.h>
  35 #include <sys/wait.h>
  36 #include <unistd.h>
  37 #include <stdio.h>
  38 #include <assert.h>
  39 #include <sys/syscall.h>
  40 #include <sched.h>
  41 #include <errno.h>
  42 #include <time.h>
  43
  44 #include <urcu/arch.h>
  45
  46 extern int __thread disable_debug;
  47
  48 /* hardcoded number of CPUs */
  49 #define NR_CPUS 16384
  50
  51 /* Default number of insert/delete */
  52 #define DEFAULT_NR_RAND 6
  53
  54 /* Default number of global items (used by readers for lookups) */
  55 #define DEFAULT_NR_GLOBAL       10
  56
  57 #if defined(_syscall0)
  58 _syscall0(pid_t, gettid)
  59 #elif defined(__NR_gettid)
  60 static inline pid_t gettid(void)
  61 {
  62         return syscall(__NR_gettid);
  63 }
  64 #else
  65 #warning "use pid as tid"
  66 static inline pid_t gettid(void)
  67 {
  68         return getpid();
  69 }
  70 #endif
  71
  72 #include <urcu.h>
  73 #include <urcu/rcurbtree.h>
  74 #include <urcu-defer.h>
  75
  76 int tree_comp(void *a, void *b)
  77 {
  78         if ((unsigned long)a < (unsigned long)b)
  79                 return -1;
  80         else if ((unsigned long)a > (unsigned long)b)
  81                 return 1;
  82         else
  83                 return 0;
  84 }
  85
  86 static DEFINE_RCU_RBTREE(rbtree, tree_comp, malloc, free, call_rcu);
  87
  88 static volatile int test_go, test_stop;
  89
  90 static unsigned long wdelay;
  91
  92 static unsigned long duration;
  93
  94 /* read-side C.S. duration, in loops */
  95 static unsigned long rduration;
  96
  97 /* write-side C.S. duration, in loops */
  98 static unsigned long wduration;
  99
 100 static unsigned long nr_rand_items = DEFAULT_NR_RAND;
 101
 102 static int opt_search_begin,
 103         opt_search_bottom,
 104         opt_search_end,
 105         opt_search_mid,
 106         opt_iter_min_max,
 107         opt_iter_max_min,
 108         opt_benchmark;
 109
 110 static inline void loop_sleep(unsigned long l)
 111 {
 112         while (l-- != 0)
 113                 caa_cpu_relax();
 114 }
 115
 116 static int verbose_mode;
 117
 118 #define printf_verbose(fmt, args...)            \
 119         do {                                    \
 120                 if (verbose_mode)               \
 121                         printf(fmt, args);      \
 122         } while (0)
 123
 124 static unsigned int cpu_affinities[NR_CPUS];
 125 static unsigned int next_aff = 0;
 126 static int use_affinity = 0;
 127
 128 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 129
 130 #ifndef HAVE_CPU_SET_T
 131 typedef unsigned long cpu_set_t;
 132 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 133 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 134 #endif
 135
 136 static void set_affinity(void)
 137 {
 138         cpu_set_t mask;
 139         int cpu;
 140         int ret;
 141
 142         if (!use_affinity)
 143                 return;
 144
 145 #if HAVE_SCHED_SETAFFINITY
 146         ret = pthread_mutex_lock(&affinity_mutex);
 147         if (ret) {
 148                 perror("Error in pthread mutex lock");
 149                 exit(-1);
 150         }
 151         cpu = cpu_affinities[next_aff++];
 152         ret = pthread_mutex_unlock(&affinity_mutex);
 153         if (ret) {
 154                 perror("Error in pthread mutex unlock");
 155                 exit(-1);
 156         }
 157
 158         CPU_ZERO(&mask);
 159         CPU_SET(cpu, &mask);
 160 #if SCHED_SETAFFINITY_ARGS == 2
 161         sched_setaffinity(0, &mask);
 162 #else
 163         sched_setaffinity(0, sizeof(mask), &mask);
 164 #endif
 165 #endif /* HAVE_SCHED_SETAFFINITY */
 166 }
 167
 168 /*
 169  * returns 0 if test should end.
 170  */
 171 static int test_duration_write(void)
 172 {
 173         return !test_stop;
 174 }
 175
 176 static int test_duration_read(void)
 177 {
 178         return !test_stop;
 179 }
 180
 181 static unsigned long long __thread nr_writes;
 182 static unsigned long long __thread nr_reads;
 183
 184 static unsigned int nr_readers;
 185 static unsigned int nr_writers;
 186
 187 static unsigned long global_items = DEFAULT_NR_GLOBAL;
 188 static void **global_key = NULL;
 189
 190 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 191
 192 void rcu_copy_mutex_lock(void)
 193 {
 194         int ret;
 195         ret = pthread_mutex_lock(&rcu_copy_mutex);
 196         if (ret) {
 197                 perror("Error in pthread mutex lock");
 198                 exit(-1);
 199         }
 200 }
 201
 202 void rcu_copy_mutex_unlock(void)
 203 {
 204         int ret;
 205
 206         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 207         if (ret) {
 208                 perror("Error in pthread mutex unlock");
 209                 exit(-1);
 210         }
 211 }
 212
 213 static
 214 void set_lookup_index(struct rcu_rbtree_node *node,
 215                 char *lookup_hit)
 216 {
 217         int i;
 218
 219         for (i = 0; i < global_items; i++) {
 220                 if (node->begin == global_key[i]
 221                     && !lookup_hit[i]) {
 222                         lookup_hit[i] = 1;
 223                         break;
 224                 }
 225         }
 226 }
 227
 228 void *thr_reader(void *_count)
 229 {
 230         unsigned long long *count = _count;
 231         struct rcu_rbtree_node *node;
 232         int i, index;
 233         char *lookup_hit;
 234
 235         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 236                         "reader", pthread_self(), (unsigned long)gettid());
 237
 238         set_affinity();
 239
 240         rcu_register_thread();
 241
 242         lookup_hit = malloc(sizeof(*lookup_hit) * global_items);
 243
 244         while (!test_go)
 245         {
 246         }
 247         cmm_smp_mb();
 248
 249         for (;;) {
 250                 /* search begin key */
 251                 if (opt_search_begin) {
 252                         for (i = 0; i < global_items; i++) {
 253                                 rcu_read_lock();
 254                                 node = rcu_rbtree_search_begin_key(&rbtree,
 255                                                          rcu_dereference(rbtree.root),
 256                                                          global_key[i]);
 257                                 assert(!rcu_rbtree_is_nil(&rbtree, node));
 258                                 rcu_read_unlock();
 259                                 nr_reads++;
 260                         }
 261                 }
 262
 263                 /* search bottom of range */
 264                 if (opt_search_bottom) {
 265                         for (i = 0; i < global_items; i++) {
 266                                 rcu_read_lock();
 267                                 node = rcu_rbtree_search(&rbtree,
 268                                                          rcu_dereference(rbtree.root),
 269                                                          global_key[i]);
 270                                 assert(!rcu_rbtree_is_nil(&rbtree, node));
 271                                 rcu_read_unlock();
 272                                 nr_reads++;
 273                         }
 274                 }
 275
 276                 /* search end of range */
 277                 if (opt_search_end) {
 278                         for (i = 0; i < global_items; i++) {
 279                                 rcu_read_lock();
 280                                 node = rcu_rbtree_search(&rbtree,
 281                                                          rcu_dereference(rbtree.root),
 282                                                          (void*) ((unsigned long) global_key[i] + 3));
 283                                 assert(!rcu_rbtree_is_nil(&rbtree, node));
 284                                 rcu_read_unlock();
 285                                 nr_reads++;
 286                         }
 287                 }
 288
 289                 /* search range (middle) */
 290                 if (opt_search_mid) {
 291                         for (i = 0; i < global_items; i++) {
 292                                 rcu_read_lock();
 293                                 node = rcu_rbtree_search_range(&rbtree,
 294                                                          rcu_dereference(rbtree.root),
 295                                                          (void*) ((unsigned long) global_key[i] + 1),
 296                                                          (void*) ((unsigned long) global_key[i] + 2));
 297                                 assert(!rcu_rbtree_is_nil(&rbtree, node));
 298                                 rcu_read_unlock();
 299                                 nr_reads++;
 300                         }
 301                 }
 302
 303                 /* min + next */
 304                 if (opt_iter_min_max) {
 305                         memset(lookup_hit, 0, sizeof(*lookup_hit) * global_items);
 306
 307                         rcu_read_lock();
 308                         node = rcu_rbtree_min(&rbtree,
 309                                               rcu_dereference(rbtree.root));
 310                         while (!rcu_rbtree_is_nil(&rbtree, node)) {
 311                                 if (!opt_benchmark)
 312                                         set_lookup_index(node, lookup_hit);
 313                                 node = rcu_rbtree_next(&rbtree, node);
 314                                 nr_reads++;
 315                         }
 316                         rcu_read_unlock();
 317
 318                         if (!opt_benchmark) {
 319                                 for (i = 0; i < global_items; i++)
 320                                         assert(lookup_hit[i]);
 321                         }
 322                 }
 323
 324                 /* max + prev */
 325                 if (opt_iter_max_min) {
 326                         memset(lookup_hit, 0, sizeof(*lookup_hit) * global_items);
 327
 328                         rcu_read_lock();
 329                         node = rcu_rbtree_max(&rbtree,
 330                                               rcu_dereference(rbtree.root));
 331                         while (!rcu_rbtree_is_nil(&rbtree, node)) {
 332                                 if (!opt_benchmark)
 333                                         set_lookup_index(node, lookup_hit);
 334                                 node = rcu_rbtree_prev(&rbtree, node);
 335                                 nr_reads++;
 336                         }
 337                         rcu_read_unlock();
 338
 339                         if (!opt_benchmark) {
 340                                 for (i = 0; i < global_items; i++)
 341                                         assert(lookup_hit[i]);
 342                         }
 343                 }
 344
 345                 debug_yield_read();
 346                 if (unlikely(rduration))
 347                         loop_sleep(rduration);
 348                 if (unlikely(!test_duration_read()))
 349                         break;
 350         }
 351
 352         rcu_unregister_thread();
 353
 354         /* test extra thread registration */
 355         rcu_register_thread();
 356         rcu_unregister_thread();
 357
 358         free(lookup_hit);
 359
 360         *count = nr_reads;
 361         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 362                         "reader", pthread_self(), (unsigned long)gettid());
 363         return ((void*)1);
 364
 365 }
 366
 367 void *thr_writer(void *_count)
 368 {
 369         unsigned long long *count = _count;
 370         struct rcu_rbtree_node *node;
 371         void **key;
 372         int i;
 373
 374         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 375                         "writer", pthread_self(), (unsigned long)gettid());
 376
 377         set_affinity();
 378
 379         key = malloc(sizeof(*key) * nr_rand_items);
 380         assert(key);
 381         //disable_debug = 1;
 382
 383         rcu_register_thread();
 384
 385         while (!test_go)
 386         {
 387         }
 388         cmm_smp_mb();
 389
 390         for (;;) {
 391                 rcu_copy_mutex_lock();
 392
 393                 for (i = 0; i < nr_rand_items; i++) {
 394                         //key[i] = (void *)(unsigned long)(rand() % 2048);
 395                         key[i] = (void *)(unsigned long)(((unsigned long) rand() * 4) % 2048);
 396                         //For more collisions
 397                         //key[i] = (void *)(unsigned long)(rand() % 6);
 398                         //node->begin = key[i];
 399                         //node->end = (void *)((unsigned long) key[i] + 1);
 400                         //node->end = (void *)((unsigned long) key[i] + 4);
 401                         rcu_read_lock();
 402                         rcu_rbtree_insert(&rbtree, key[i],
 403                                           (void *)((unsigned long) key[i] + 4));
 404                         rcu_read_unlock();
 405                         nr_writes++;
 406                 }
 407                 rcu_copy_mutex_unlock();
 408
 409                 if (unlikely(wduration))
 410                         loop_sleep(wduration);
 411
 412                 rcu_copy_mutex_lock();
 413                 for (i = 0; i < nr_rand_items; i++) {
 414 #if 0
 415                         node = rcu_rbtree_min(rbtree, rbtree->root);
 416                         while (!rcu_rbtree_is_nil(&rbtree, node)) {
 417                                 printf("{ 0x%lX p:%lX r:%lX l:%lX %s %s %s} ",
 418                                         (unsigned long)node->key,
 419                                         node->p->key,
 420                                         node->right->key,
 421                                         node->left->key,
 422                                         node->color ? "red" : "black",
 423                                         node->pos ? "right" : "left",
 424                                         node->nil ? "nil" : "");
 425                                 node = rcu_rbtree_next(rbtree, node);
 426                         }
 427                         printf("\n");
 428 #endif
 429                         rcu_read_lock();
 430                         node = rcu_rbtree_search(&rbtree, rbtree.root, key[i]);
 431                         assert(!rcu_rbtree_is_nil(&rbtree, node));
 432                         rcu_rbtree_remove(&rbtree, node);
 433                         rcu_read_unlock();
 434                         nr_writes++;
 435                 }
 436
 437                 rcu_copy_mutex_unlock();
 438                 if (unlikely(!test_duration_write()))
 439                         break;
 440                 if (unlikely(wdelay))
 441                         loop_sleep(wdelay);
 442         }
 443
 444         rcu_unregister_thread();
 445
 446         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 447                         "writer", pthread_self(), (unsigned long)gettid());
 448         *count = nr_writes;
 449         free(key);
 450         return ((void*)2);
 451 }
 452
 453 void show_usage(int argc, char **argv)
 454 {
 455         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 456 #ifdef DEBUG_YIELD
 457         printf(" [-r] [-w] (yield reader and/or writer)");
 458 #endif
 459         printf(" [-d delay] (writer period (us))");
 460         printf(" [-c duration] (reader C.S. duration (in loops))");
 461         printf(" [-e duration] (writer C.S. duration (in loops))");
 462         printf(" [-v] (verbose output)");
 463         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 464         printf(" [-g items] Initially populate n items (for reader lookups)");
 465         printf(" [-f items] Writers add n random items and then remove these items");
 466         printf(" [-b] Reader: search begin key");
 467         printf(" [-n] Reader: search bottom of range");
 468         printf(" [-N] Reader: search end of range");
 469         printf(" [-m] Reader: search range (middle)");
 470         printf(" [-i] Reader: iterate from min to max");
 471         printf(" [-I] Reader: iterate from max to min");
 472         printf(" [-B] Benchmark mode (no validation)");
 473         printf("\n");
 474 }
 475
 476 int main(int argc, char **argv)
 477 {
 478         int err;
 479         pthread_t *tid_reader, *tid_writer;
 480         void *tret;
 481         unsigned long long *count_reader, *count_writer;
 482         unsigned long long tot_reads = 0, tot_writes = 0;
 483         int i, a;
 484         struct rcu_rbtree_node *node;
 485
 486         if (argc < 4) {
 487                 show_usage(argc, argv);
 488                 return -1;
 489         }
 490
 491         err = sscanf(argv[1], "%u", &nr_readers);
 492         if (err != 1) {
 493                 show_usage(argc, argv);
 494                 return -1;
 495         }
 496
 497         err = sscanf(argv[2], "%u", &nr_writers);
 498         if (err != 1) {
 499                 show_usage(argc, argv);
 500                 return -1;
 501         }
 502
 503         err = sscanf(argv[3], "%lu", &duration);
 504         if (err != 1) {
 505                 show_usage(argc, argv);
 506                 return -1;
 507         }
 508
 509         for (i = 4; i < argc; i++) {
 510                 if (argv[i][0] != '-')
 511                         continue;
 512                 switch (argv[i][1]) {
 513 #ifdef DEBUG_YIELD
 514                 case 'r':
 515                         yield_active |= YIELD_READ;
 516                         break;
 517                 case 'w':
 518                         yield_active |= YIELD_WRITE;
 519                         break;
 520 #endif
 521                 case 'a':
 522                         if (argc < i + 2) {
 523                                 show_usage(argc, argv);
 524                                 return -1;
 525                         }
 526                         a = atoi(argv[++i]);
 527                         cpu_affinities[next_aff++] = a;
 528                         use_affinity = 1;
 529                         printf_verbose("Adding CPU %d affinity\n", a);
 530                         break;
 531                 case 'c':
 532                         if (argc < i + 2) {
 533                                 show_usage(argc, argv);
 534                                 return -1;
 535                         }
 536                         rduration = atol(argv[++i]);
 537                         break;
 538                 case 'd':
 539                         if (argc < i + 2) {
 540                                 show_usage(argc, argv);
 541                                 return -1;
 542                         }
 543                         wdelay = atol(argv[++i]);
 544                         break;
 545                 case 'e':
 546                         if (argc < i + 2) {
 547                                 show_usage(argc, argv);
 548                                 return -1;
 549                         }
 550                         wduration = atol(argv[++i]);
 551                         break;
 552                 case 'v':
 553                         verbose_mode = 1;
 554                         break;
 555                 case 'g':
 556                         if (argc < i + 2) {
 557                                 show_usage(argc, argv);
 558                                 return -1;
 559                         }
 560                         global_items = atol(argv[++i]);
 561                         break;
 562                 case 'b':
 563                         opt_search_begin = 1;
 564                         break;
 565                 case 'n':
 566                         opt_search_bottom = 1;
 567                         break;
 568                 case 'N':
 569                         opt_search_end = 1;
 570                         break;
 571                 case 'm':
 572                         opt_search_mid = 1;
 573                         break;
 574                 case 'i':
 575                         opt_iter_min_max = 1;
 576                         break;
 577                 case 'I':
 578                         opt_iter_max_min = 1;
 579                         break;
 580                 case 'B':
 581                         opt_benchmark = 1;
 582                         break;
 583                 case 'f':
 584                         if (argc < i + 2) {
 585                                 show_usage(argc, argv);
 586                                 return -1;
 587                         }
 588                         nr_rand_items = atol(argv[++i]);
 589                         break;
 590                 }
 591         }
 592
 593         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 594                 duration, nr_readers, nr_writers);
 595         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 596         printf_verbose("Reader duration : %lu loops.\n", rduration);
 597         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 598                         "main", pthread_self(), (unsigned long)gettid());
 599
 600         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 601         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 602         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 603         count_writer = malloc(sizeof(*count_writer) * nr_writers);
 604         global_key = malloc(sizeof(*global_key) * global_items);
 605
 606         srand(time(NULL));
 607
 608         err = create_all_cpu_call_rcu_data(0);
 609         assert(!err);
 610
 611         next_aff = 0;
 612
 613         for (i = 0; i < nr_readers; i++) {
 614                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 615                                      &count_reader[i]);
 616                 if (err != 0)
 617                         exit(1);
 618         }
 619         for (i = 0; i < nr_writers; i++) {
 620                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 621                                      &count_writer[i]);
 622                 if (err != 0)
 623                         exit(1);
 624         }
 625
 626         rcu_register_thread();
 627         rcu_read_lock();
 628         /* Insert items looked up by readers */
 629         for (i = 0; i < global_items; i++) {
 630                 global_key[i] = (void *)(unsigned long)(((unsigned long) rand() * 4) % 2048);
 631                 //global_key[i] = (void *)(unsigned long)(rand() % 2048);
 632                 //For more collisions
 633                 //global_key[i] = (void *)(unsigned long)(rand() % 6);
 634                 //node->begin = global_key[i];
 635                 //node->end = (void *)((unsigned long) global_key[i] + 1);
 636                 //node->end = (void *)((unsigned long) global_key[i] + 4);
 637                 rcu_rbtree_insert(&rbtree, global_key[i],
 638                                   (void *)((unsigned long) global_key[i] + 4));
 639         }
 640         rcu_read_unlock();
 641
 642         cmm_smp_mb();
 643
 644         test_go = 1;
 645
 646         sleep(duration);
 647
 648         test_stop = 1;
 649
 650         for (i = 0; i < nr_readers; i++) {
 651                 err = pthread_join(tid_reader[i], &tret);
 652                 if (err != 0)
 653                         exit(1);
 654                 tot_reads += count_reader[i];
 655         }
 656         for (i = 0; i < nr_writers; i++) {
 657                 err = pthread_join(tid_writer[i], &tret);
 658                 if (err != 0)
 659                         exit(1);
 660                 tot_writes += count_writer[i];
 661         }
 662
 663         rcu_read_lock();
 664         for (i = 0; i < global_items; i++) {
 665                 node = rcu_rbtree_search(&rbtree, rbtree.root, global_key[i]);
 666                 assert(!rcu_rbtree_is_nil(&rbtree, node));
 667                 rcu_rbtree_remove(&rbtree, node);
 668         }
 669         rcu_read_unlock();
 670         rcu_unregister_thread();
 671
 672         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 673                tot_writes);
 674         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 675                 "nr_writers %3u "
 676                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 677                 "global_items %6lu rand_items %6lu\n",
 678                 argv[0], duration, nr_readers, rduration, wduration,
 679                 nr_writers, wdelay, tot_reads, tot_writes,
 680                 tot_reads + tot_writes, global_items, nr_rand_items);
 681         free_all_cpu_call_rcu_data();
 682         free(tid_reader);
 683         free(tid_writer);
 684         free(count_reader);
 685         free(count_writer);
 686         free(global_key);
 687         return 0;
 688 }