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