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