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