tests/test_urcu_gc.c

   1 /*
   2  * test_urcu_gc.c
   3  *
   4  * Userspace RCU library - test program (with baatch reclamation)
   5  *
   6  * Copyright February 2009 - 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 #include "../config.h"
  25 #include <stdio.h>
  26 #include <pthread.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/types.h>
  30 #include <sys/wait.h>
  31 #include <unistd.h>
  32 #include <stdio.h>
  33 #include <assert.h>
  34 #include <sched.h>
  35 #include <errno.h>
  36
  37 #include <urcu/arch.h>
  38 #include <urcu/tls-compat.h>
  39
  40 #ifdef __linux__
  41 #include <syscall.h>
  42 #endif
  43
  44 /* hardcoded number of CPUs */
  45 #define NR_CPUS 16384
  46
  47 #if defined(_syscall0)
  48 _syscall0(pid_t, gettid)
  49 #elif defined(__NR_gettid)
  50 static inline pid_t gettid(void)
  51 {
  52         return syscall(__NR_gettid);
  53 }
  54 #else
  55 #warning "use pid as tid"
  56 static inline pid_t gettid(void)
  57 {
  58         return getpid();
  59 }
  60 #endif
  61
  62 #ifndef DYNAMIC_LINK_TEST
  63 #define _LGPL_SOURCE
  64 #else
  65 #define debug_yield_read()
  66 #endif
  67 #include <urcu.h>
  68
  69 struct test_array {
  70         int a;
  71 };
  72
  73 static volatile int test_go, test_stop;
  74
  75 static unsigned long wdelay;
  76
  77 static struct test_array *test_rcu_pointer;
  78
  79 static unsigned int reclaim_batch = 1;
  80
  81 struct reclaim_queue {
  82         void **queue;   /* Beginning of queue */
  83         void **head;    /* Insert position */
  84 };
  85
  86 static struct reclaim_queue *pending_reclaims;
  87
  88 static unsigned long duration;
  89
  90 /* read-side C.S. duration, in loops */
  91 static unsigned long rduration;
  92
  93 /* write-side C.S. duration, in loops */
  94 static unsigned long wduration;
  95
  96 static inline void loop_sleep(unsigned long l)
  97 {
  98         while(l-- != 0)
  99                 caa_cpu_relax();
 100 }
 101
 102 static int verbose_mode;
 103
 104 #define printf_verbose(fmt, args...)            \
 105         do {                                    \
 106                 if (verbose_mode)               \
 107                         printf(fmt, args);      \
 108         } while (0)
 109
 110 static unsigned int cpu_affinities[NR_CPUS];
 111 static unsigned int next_aff = 0;
 112 static int use_affinity = 0;
 113
 114 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 115
 116 #ifndef HAVE_CPU_SET_T
 117 typedef unsigned long cpu_set_t;
 118 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 119 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 120 #endif
 121
 122 static void set_affinity(void)
 123 {
 124         cpu_set_t mask;
 125         int cpu;
 126         int ret;
 127
 128         if (!use_affinity)
 129                 return;
 130
 131 #if HAVE_SCHED_SETAFFINITY
 132         ret = pthread_mutex_lock(&affinity_mutex);
 133         if (ret) {
 134                 perror("Error in pthread mutex lock");
 135                 exit(-1);
 136         }
 137         cpu = cpu_affinities[next_aff++];
 138         ret = pthread_mutex_unlock(&affinity_mutex);
 139         if (ret) {
 140                 perror("Error in pthread mutex unlock");
 141                 exit(-1);
 142         }
 143
 144         CPU_ZERO(&mask);
 145         CPU_SET(cpu, &mask);
 146 #if SCHED_SETAFFINITY_ARGS == 2
 147         sched_setaffinity(0, &mask);
 148 #else
 149         sched_setaffinity(0, sizeof(mask), &mask);
 150 #endif
 151 #endif /* HAVE_SCHED_SETAFFINITY */
 152 }
 153
 154 /*
 155  * returns 0 if test should end.
 156  */
 157 static int test_duration_write(void)
 158 {
 159         return !test_stop;
 160 }
 161
 162 static int test_duration_read(void)
 163 {
 164         return !test_stop;
 165 }
 166
 167 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
 168 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 169
 170 static
 171 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 172
 173 static unsigned int nr_readers;
 174 static unsigned int nr_writers;
 175
 176 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 177
 178 void rcu_copy_mutex_lock(void)
 179 {
 180         int ret;
 181         ret = pthread_mutex_lock(&rcu_copy_mutex);
 182         if (ret) {
 183                 perror("Error in pthread mutex lock");
 184                 exit(-1);
 185         }
 186 }
 187
 188 void rcu_copy_mutex_unlock(void)
 189 {
 190         int ret;
 191
 192         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 193         if (ret) {
 194                 perror("Error in pthread mutex unlock");
 195                 exit(-1);
 196         }
 197 }
 198
 199 void *thr_reader(void *_count)
 200 {
 201         unsigned long long *count = _count;
 202         struct test_array *local_ptr;
 203
 204         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 205                         "reader", pthread_self(), (unsigned long)gettid());
 206
 207         set_affinity();
 208
 209         rcu_register_thread();
 210
 211         while (!test_go)
 212         {
 213         }
 214         cmm_smp_mb();
 215
 216         for (;;) {
 217                 rcu_read_lock();
 218                 local_ptr = rcu_dereference(test_rcu_pointer);
 219                 debug_yield_read();
 220                 if (local_ptr)
 221                         assert(local_ptr->a == 8);
 222                 if (caa_unlikely(rduration))
 223                         loop_sleep(rduration);
 224                 rcu_read_unlock();
 225                 URCU_TLS(nr_reads)++;
 226                 if (caa_unlikely(!test_duration_read()))
 227                         break;
 228         }
 229
 230         rcu_unregister_thread();
 231
 232         *count = URCU_TLS(nr_reads);
 233         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 234                         "reader", pthread_self(), (unsigned long)gettid());
 235         return ((void*)1);
 236
 237 }
 238
 239 static void rcu_gc_clear_queue(unsigned long wtidx)
 240 {
 241         void **p;
 242
 243         /* Wait for Q.S and empty queue */
 244         synchronize_rcu();
 245
 246         for (p = pending_reclaims[wtidx].queue;
 247                         p < pending_reclaims[wtidx].head; p++) {
 248                 /* poison */
 249                 if (*p)
 250                         ((struct test_array *)*p)->a = 0;
 251                 free(*p);
 252         }
 253         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 254 }
 255
 256 /* Using per-thread queue */
 257 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 258 {
 259         /* Queue pointer */
 260         *pending_reclaims[wtidx].head = old;
 261         pending_reclaims[wtidx].head++;
 262
 263         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 264                         < reclaim_batch))
 265                 return;
 266
 267         rcu_gc_clear_queue(wtidx);
 268 }
 269
 270 void *thr_writer(void *data)
 271 {
 272         unsigned long wtidx = (unsigned long)data;
 273 #ifdef TEST_LOCAL_GC
 274         struct test_array *old = NULL;
 275 #else
 276         struct test_array *new, *old;
 277 #endif
 278
 279         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 280                         "writer", pthread_self(), (unsigned long)gettid());
 281
 282         set_affinity();
 283
 284         while (!test_go)
 285         {
 286         }
 287         cmm_smp_mb();
 288
 289         for (;;) {
 290 #ifndef TEST_LOCAL_GC
 291                 new = malloc(sizeof(*new));
 292                 new->a = 8;
 293                 old = rcu_xchg_pointer(&test_rcu_pointer, new);
 294 #endif
 295                 if (caa_unlikely(wduration))
 296                         loop_sleep(wduration);
 297                 rcu_gc_reclaim(wtidx, old);
 298                 URCU_TLS(nr_writes)++;
 299                 if (caa_unlikely(!test_duration_write()))
 300                         break;
 301                 if (caa_unlikely(wdelay))
 302                         loop_sleep(wdelay);
 303         }
 304
 305         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 306                         "writer", pthread_self(), (unsigned long)gettid());
 307         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 308         return ((void*)2);
 309 }
 310
 311 void show_usage(int argc, char **argv)
 312 {
 313         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 314 #ifdef DEBUG_YIELD
 315         printf(" [-r] [-w] (yield reader and/or writer)");
 316 #endif
 317         printf(" [-d delay] (writer period (us))");
 318         printf(" [-c duration] (reader C.S. duration (in loops))");
 319         printf(" [-e duration] (writer C.S. duration (in loops))");
 320         printf(" [-v] (verbose output)");
 321         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 322         printf("\n");
 323 }
 324
 325 int main(int argc, char **argv)
 326 {
 327         int err;
 328         pthread_t *tid_reader, *tid_writer;
 329         void *tret;
 330         unsigned long long *count_reader;
 331         unsigned long long tot_reads = 0, tot_writes = 0;
 332         int i, a;
 333
 334         if (argc < 4) {
 335                 show_usage(argc, argv);
 336                 return -1;
 337         }
 338
 339         err = sscanf(argv[1], "%u", &nr_readers);
 340         if (err != 1) {
 341                 show_usage(argc, argv);
 342                 return -1;
 343         }
 344
 345         err = sscanf(argv[2], "%u", &nr_writers);
 346         if (err != 1) {
 347                 show_usage(argc, argv);
 348                 return -1;
 349         }
 350
 351         err = sscanf(argv[3], "%lu", &duration);
 352         if (err != 1) {
 353                 show_usage(argc, argv);
 354                 return -1;
 355         }
 356
 357         for (i = 4; i < argc; i++) {
 358                 if (argv[i][0] != '-')
 359                         continue;
 360                 switch (argv[i][1]) {
 361 #ifdef DEBUG_YIELD
 362                 case 'r':
 363                         yield_active |= YIELD_READ;
 364                         break;
 365                 case 'w':
 366                         yield_active |= YIELD_WRITE;
 367                         break;
 368 #endif
 369                 case 'a':
 370                         if (argc < i + 2) {
 371                                 show_usage(argc, argv);
 372                                 return -1;
 373                         }
 374                         a = atoi(argv[++i]);
 375                         cpu_affinities[next_aff++] = a;
 376                         use_affinity = 1;
 377                         printf_verbose("Adding CPU %d affinity\n", a);
 378                         break;
 379                 case 'b':
 380                         if (argc < i + 2) {
 381                                 show_usage(argc, argv);
 382                                 return -1;
 383                         }
 384                         reclaim_batch = atol(argv[++i]);
 385                         break;
 386                 case 'c':
 387                         if (argc < i + 2) {
 388                                 show_usage(argc, argv);
 389                                 return -1;
 390                         }
 391                         rduration = atol(argv[++i]);
 392                         break;
 393                 case 'd':
 394                         if (argc < i + 2) {
 395                                 show_usage(argc, argv);
 396                                 return -1;
 397                         }
 398                         wdelay = atol(argv[++i]);
 399                         break;
 400                 case 'e':
 401                         if (argc < i + 2) {
 402                                 show_usage(argc, argv);
 403                                 return -1;
 404                         }
 405                         wduration = atol(argv[++i]);
 406                         break;
 407                 case 'v':
 408                         verbose_mode = 1;
 409                         break;
 410                 }
 411         }
 412
 413         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 414                 duration, nr_readers, nr_writers);
 415         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 416         printf_verbose("Reader duration : %lu loops.\n", rduration);
 417         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 418                         "main", pthread_self(), (unsigned long)gettid());
 419
 420         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 421         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 422         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 423         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 424         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 425         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 426                         < CAA_CACHE_LINE_SIZE)
 427                 for (i = 0; i < nr_writers; i++)
 428                         pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 429         else
 430                 for (i = 0; i < nr_writers; i++)
 431                         pending_reclaims[i].queue = calloc(reclaim_batch,
 432                                         sizeof(*pending_reclaims[i].queue));
 433         for (i = 0; i < nr_writers; i++)
 434                 pending_reclaims[i].head = pending_reclaims[i].queue;
 435
 436         next_aff = 0;
 437
 438         for (i = 0; i < nr_readers; i++) {
 439                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 440                                      &count_reader[i]);
 441                 if (err != 0)
 442                         exit(1);
 443         }
 444         for (i = 0; i < nr_writers; i++) {
 445                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 446                                      (void *)(long)i);
 447                 if (err != 0)
 448                         exit(1);
 449         }
 450
 451         cmm_smp_mb();
 452
 453         test_go = 1;
 454
 455         sleep(duration);
 456
 457         test_stop = 1;
 458
 459         for (i = 0; i < nr_readers; i++) {
 460                 err = pthread_join(tid_reader[i], &tret);
 461                 if (err != 0)
 462                         exit(1);
 463                 tot_reads += count_reader[i];
 464         }
 465         for (i = 0; i < nr_writers; i++) {
 466                 err = pthread_join(tid_writer[i], &tret);
 467                 if (err != 0)
 468                         exit(1);
 469                 tot_writes += tot_nr_writes[i];
 470                 rcu_gc_clear_queue(i);
 471         }
 472
 473         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 474                tot_writes);
 475         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 476                 "nr_writers %3u "
 477                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 478                 "batch %u\n",
 479                 argv[0], duration, nr_readers, rduration, wduration,
 480                 nr_writers, wdelay, tot_reads, tot_writes,
 481                 tot_reads + tot_writes, reclaim_batch);
 482         free(tid_reader);
 483         free(tid_writer);
 484         free(count_reader);
 485         free(tot_nr_writes);
 486         for (i = 0; i < nr_writers; i++)
 487                 free(pending_reclaims[i].queue);
 488         free(pending_reclaims);
 489
 490         return 0;
 491 }