tests/benchmark/test_urcu_qsbr_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 #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 #define _LGPL_SOURCE
  43 #include <urcu-qsbr.h>
  44
  45 struct test_array {
  46         int a;
  47 };
  48
  49 static volatile int test_go, test_stop;
  50
  51 static unsigned long wdelay;
  52
  53 static struct test_array *test_rcu_pointer;
  54
  55 static unsigned long duration;
  56
  57 /* read-side C.S. duration, in loops */
  58 static unsigned long rduration;
  59 static long reclaim_batch = 1;
  60
  61 struct reclaim_queue {
  62         void **queue;   /* Beginning of queue */
  63         void **head;    /* Insert position */
  64 };
  65
  66 static struct reclaim_queue *pending_reclaims;
  67
  68
  69 /* write-side C.S. duration, in loops */
  70 static unsigned long wduration;
  71
  72 static inline void loop_sleep(unsigned long loops)
  73 {
  74         while (loops-- != 0)
  75                 caa_cpu_relax();
  76 }
  77
  78 static int verbose_mode;
  79
  80 #define printf_verbose(fmt, args...)            \
  81         do {                                    \
  82                 if (verbose_mode)               \
  83                         printf(fmt, args);      \
  84         } while (0)
  85
  86 static unsigned int cpu_affinities[NR_CPUS];
  87 static unsigned int next_aff = 0;
  88 static int use_affinity = 0;
  89
  90 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
  91
  92 static void set_affinity(void)
  93 {
  94 #if HAVE_SCHED_SETAFFINITY
  95         cpu_set_t mask;
  96         int cpu, ret;
  97 #endif /* HAVE_SCHED_SETAFFINITY */
  98
  99         if (!use_affinity)
 100                 return;
 101
 102 #if HAVE_SCHED_SETAFFINITY
 103         ret = pthread_mutex_lock(&affinity_mutex);
 104         if (ret) {
 105                 perror("Error in pthread mutex lock");
 106                 exit(-1);
 107         }
 108         cpu = cpu_affinities[next_aff++];
 109         ret = pthread_mutex_unlock(&affinity_mutex);
 110         if (ret) {
 111                 perror("Error in pthread mutex unlock");
 112                 exit(-1);
 113         }
 114
 115         CPU_ZERO(&mask);
 116         CPU_SET(cpu, &mask);
 117         sched_setaffinity(0, sizeof(mask), &mask);
 118 #endif /* HAVE_SCHED_SETAFFINITY */
 119 }
 120
 121 /*
 122  * returns 0 if test should end.
 123  */
 124 static int test_duration_write(void)
 125 {
 126         return !test_stop;
 127 }
 128
 129 static int test_duration_read(void)
 130 {
 131         return !test_stop;
 132 }
 133
 134 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
 135 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 136
 137 static unsigned int nr_readers;
 138 static unsigned int nr_writers;
 139
 140 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 141 static
 142 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 143
 144
 145 void rcu_copy_mutex_lock(void)
 146 {
 147         int ret;
 148         ret = pthread_mutex_lock(&rcu_copy_mutex);
 149         if (ret) {
 150                 perror("Error in pthread mutex lock");
 151                 exit(-1);
 152         }
 153 }
 154
 155 void rcu_copy_mutex_unlock(void)
 156 {
 157         int ret;
 158
 159         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 160         if (ret) {
 161                 perror("Error in pthread mutex unlock");
 162                 exit(-1);
 163         }
 164 }
 165
 166 void *thr_reader(void *_count)
 167 {
 168         unsigned long long *count = _count;
 169         struct test_array *local_ptr;
 170
 171         printf_verbose("thread_begin %s, tid %lu\n",
 172                         "reader", urcu_get_thread_id());
 173
 174         set_affinity();
 175
 176         rcu_register_thread();
 177
 178         while (!test_go)
 179         {
 180         }
 181         cmm_smp_mb();
 182
 183         for (;;) {
 184                 _rcu_read_lock();
 185                 local_ptr = _rcu_dereference(test_rcu_pointer);
 186                 rcu_debug_yield_read();
 187                 if (local_ptr)
 188                         assert(local_ptr->a == 8);
 189                 if (caa_unlikely(rduration))
 190                         loop_sleep(rduration);
 191                 _rcu_read_unlock();
 192                 URCU_TLS(nr_reads)++;
 193                 /* QS each 1024 reads */
 194                 if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 195                         _rcu_quiescent_state();
 196                 if (caa_unlikely(!test_duration_read()))
 197                         break;
 198         }
 199
 200         rcu_unregister_thread();
 201
 202         *count = URCU_TLS(nr_reads);
 203         printf_verbose("thread_end %s, tid %lu\n",
 204                         "reader", urcu_get_thread_id());
 205         return ((void*)1);
 206
 207 }
 208
 209 static void rcu_gc_clear_queue(unsigned long wtidx)
 210 {
 211         void **p;
 212
 213         /* Wait for Q.S and empty queue */
 214         synchronize_rcu();
 215
 216         for (p = pending_reclaims[wtidx].queue;
 217                         p < pending_reclaims[wtidx].head; p++) {
 218                 /* poison */
 219                 if (*p)
 220                         ((struct test_array *)*p)->a = 0;
 221                 free(*p);
 222         }
 223         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 224 }
 225
 226 /* Using per-thread queue */
 227 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 228 {
 229         /* Queue pointer */
 230         *pending_reclaims[wtidx].head = old;
 231         pending_reclaims[wtidx].head++;
 232
 233         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 234                         < reclaim_batch))
 235                 return;
 236
 237         rcu_gc_clear_queue(wtidx);
 238 }
 239
 240 void *thr_writer(void *data)
 241 {
 242         unsigned long wtidx = (unsigned long)data;
 243 #ifdef TEST_LOCAL_GC
 244         struct test_array *old = NULL;
 245 #else
 246         struct test_array *new, *old;
 247 #endif
 248
 249         printf_verbose("thread_begin %s, tid %lu\n",
 250                         "writer", urcu_get_thread_id());
 251
 252         set_affinity();
 253
 254         while (!test_go)
 255         {
 256         }
 257         cmm_smp_mb();
 258
 259         for (;;) {
 260 #ifndef TEST_LOCAL_GC
 261                 new = malloc(sizeof(*new));
 262                 new->a = 8;
 263                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 264 #endif
 265                 if (caa_unlikely(wduration))
 266                         loop_sleep(wduration);
 267                 rcu_gc_reclaim(wtidx, old);
 268                 URCU_TLS(nr_writes)++;
 269                 if (caa_unlikely(!test_duration_write()))
 270                         break;
 271                 if (caa_unlikely(wdelay))
 272                         loop_sleep(wdelay);
 273         }
 274
 275         printf_verbose("thread_end %s, tid %lu\n",
 276                         "writer", urcu_get_thread_id());
 277         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 278         return ((void*)2);
 279 }
 280
 281 void show_usage(int argc, char **argv)
 282 {
 283         printf("Usage : %s nr_readers nr_writers duration (s) <OPTIONS>\n",
 284                 argv[0]);
 285         printf("OPTIONS:\n");
 286         printf("        [-r] [-w] (yield reader and/or writer)\n");
 287         printf("        [-b batch] (batch reclaim)\n");
 288         printf("        [-d delay] (writer period (us))\n");
 289         printf("        [-c duration] (reader C.S. duration (in loops))\n");
 290         printf("        [-e duration] (writer C.S. duration (in loops))\n");
 291         printf("        [-v] (verbose output)\n");
 292         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 293         printf("\n");
 294 }
 295
 296 int main(int argc, char **argv)
 297 {
 298         int err;
 299         pthread_t *tid_reader, *tid_writer;
 300         void *tret;
 301         unsigned long long *count_reader;
 302         unsigned long long tot_reads = 0, tot_writes = 0;
 303         int i, a;
 304         unsigned int i_thr;
 305
 306         if (argc < 4) {
 307                 show_usage(argc, argv);
 308                 return -1;
 309         }
 310
 311         err = sscanf(argv[1], "%u", &nr_readers);
 312         if (err != 1) {
 313                 show_usage(argc, argv);
 314                 return -1;
 315         }
 316
 317         err = sscanf(argv[2], "%u", &nr_writers);
 318         if (err != 1) {
 319                 show_usage(argc, argv);
 320                 return -1;
 321         }
 322
 323         err = sscanf(argv[3], "%lu", &duration);
 324         if (err != 1) {
 325                 show_usage(argc, argv);
 326                 return -1;
 327         }
 328
 329         for (i = 4; i < argc; i++) {
 330                 if (argv[i][0] != '-')
 331                         continue;
 332                 switch (argv[i][1]) {
 333                 case 'r':
 334                         rcu_debug_yield_enable(RCU_YIELD_READ);
 335                         break;
 336                 case 'w':
 337                         rcu_debug_yield_enable(RCU_YIELD_WRITE);
 338                         break;
 339                 case 'a':
 340                         if (argc < i + 2) {
 341                                 show_usage(argc, argv);
 342                                 return -1;
 343                         }
 344                         a = atoi(argv[++i]);
 345                         cpu_affinities[next_aff++] = a;
 346                         use_affinity = 1;
 347                         printf_verbose("Adding CPU %d affinity\n", a);
 348                         break;
 349                 case 'b':
 350                         if (argc < i + 2) {
 351                                 show_usage(argc, argv);
 352                                 return -1;
 353                         }
 354                         reclaim_batch = atol(argv[++i]);
 355                         break;
 356                 case 'c':
 357                         if (argc < i + 2) {
 358                                 show_usage(argc, argv);
 359                                 return -1;
 360                         }
 361                         rduration = atol(argv[++i]);
 362                         break;
 363                 case 'd':
 364                         if (argc < i + 2) {
 365                                 show_usage(argc, argv);
 366                                 return -1;
 367                         }
 368                         wdelay = atol(argv[++i]);
 369                         break;
 370                 case 'e':
 371                         if (argc < i + 2) {
 372                                 show_usage(argc, argv);
 373                                 return -1;
 374                         }
 375                         wduration = atol(argv[++i]);
 376                         break;
 377                 case 'v':
 378                         verbose_mode = 1;
 379                         break;
 380                 }
 381         }
 382
 383         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 384                 duration, nr_readers, nr_writers);
 385         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 386         printf_verbose("Reader duration : %lu loops.\n", rduration);
 387         printf_verbose("thread %-6s, tid %lu\n",
 388                         "main", urcu_get_thread_id());
 389
 390         tid_reader = calloc(nr_readers, sizeof(*tid_reader));
 391         tid_writer = calloc(nr_writers, sizeof(*tid_writer));
 392         count_reader = calloc(nr_readers, sizeof(*count_reader));
 393         tot_nr_writes = calloc(nr_writers, sizeof(*tot_nr_writes));
 394         pending_reclaims = calloc(nr_writers, sizeof(*pending_reclaims));
 395         if (reclaim_batch * sizeof(*pending_reclaims[0].queue)
 396                         < CAA_CACHE_LINE_SIZE)
 397                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 398                         pending_reclaims[i_thr].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 399         else
 400                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 401                         pending_reclaims[i_thr].queue = calloc(reclaim_batch,
 402                                         sizeof(*pending_reclaims[i_thr].queue));
 403         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 404                 pending_reclaims[i_thr].head = pending_reclaims[i_thr].queue;
 405
 406         next_aff = 0;
 407
 408         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 409                 err = pthread_create(&tid_reader[i_thr], NULL, thr_reader,
 410                                      &count_reader[i_thr]);
 411                 if (err != 0)
 412                         exit(1);
 413         }
 414         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 415                 err = pthread_create(&tid_writer[i_thr], NULL, thr_writer,
 416                                      (void *)(long)i_thr);
 417                 if (err != 0)
 418                         exit(1);
 419         }
 420
 421         cmm_smp_mb();
 422
 423         test_go = 1;
 424
 425         sleep(duration);
 426
 427         test_stop = 1;
 428
 429         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 430                 err = pthread_join(tid_reader[i_thr], &tret);
 431                 if (err != 0)
 432                         exit(1);
 433                 tot_reads += count_reader[i_thr];
 434         }
 435         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 436                 err = pthread_join(tid_writer[i_thr], &tret);
 437                 if (err != 0)
 438                         exit(1);
 439                 tot_writes += tot_nr_writes[i_thr];
 440                 rcu_gc_clear_queue(i_thr);
 441         }
 442
 443         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 444                tot_writes);
 445         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 446                 "nr_writers %3u "
 447                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 448                 "batch %ld\n",
 449                 argv[0], duration, nr_readers, rduration, wduration,
 450                 nr_writers, wdelay, tot_reads, tot_writes,
 451                 tot_reads + tot_writes, reclaim_batch);
 452
 453         free(tid_reader);
 454         free(tid_writer);
 455         free(count_reader);
 456         free(tot_nr_writes);
 457
 458         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 459                 free(pending_reclaims[i_thr].queue);
 460         free(pending_reclaims);
 461
 462         return 0;
 463 }