tests/benchmark/test_urcu_gc.c

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