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 #ifdef 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 #ifdef 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 static
 148 void *thr_reader(void *_count)
 149 {
 150         unsigned long long *count = _count;
 151         struct test_array *local_ptr;
 152
 153         printf_verbose("thread_begin %s, tid %lu\n",
 154                         "reader", urcu_get_thread_id());
 155
 156         set_affinity();
 157
 158         rcu_register_thread();
 159
 160         while (!test_go)
 161         {
 162         }
 163         cmm_smp_mb();
 164
 165         for (;;) {
 166                 rcu_read_lock();
 167                 local_ptr = rcu_dereference(test_rcu_pointer);
 168                 rcu_debug_yield_read();
 169                 if (local_ptr)
 170                         assert(local_ptr->a == 8);
 171                 if (caa_unlikely(rduration))
 172                         loop_sleep(rduration);
 173                 rcu_read_unlock();
 174                 URCU_TLS(nr_reads)++;
 175                 if (caa_unlikely(!test_duration_read()))
 176                         break;
 177         }
 178
 179         rcu_unregister_thread();
 180
 181         *count = URCU_TLS(nr_reads);
 182         printf_verbose("thread_end %s, tid %lu\n",
 183                         "reader", urcu_get_thread_id());
 184         return ((void*)1);
 185
 186 }
 187
 188 static void rcu_gc_clear_queue(unsigned long wtidx)
 189 {
 190         void **p;
 191
 192         /* Wait for Q.S and empty queue */
 193         synchronize_rcu();
 194
 195         for (p = pending_reclaims[wtidx].queue;
 196                         p < pending_reclaims[wtidx].head; p++) {
 197                 /* poison */
 198                 if (*p)
 199                         ((struct test_array *)*p)->a = 0;
 200                 free(*p);
 201         }
 202         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 203 }
 204
 205 /* Using per-thread queue */
 206 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 207 {
 208         /* Queue pointer */
 209         *pending_reclaims[wtidx].head = old;
 210         pending_reclaims[wtidx].head++;
 211
 212         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 213                         < reclaim_batch))
 214                 return;
 215
 216         rcu_gc_clear_queue(wtidx);
 217 }
 218
 219 static
 220 void *thr_writer(void *data)
 221 {
 222         unsigned long wtidx = (unsigned long)data;
 223 #ifdef TEST_LOCAL_GC
 224         struct test_array *old = NULL;
 225 #else
 226         struct test_array *new, *old;
 227 #endif
 228
 229         printf_verbose("thread_begin %s, tid %lu\n",
 230                         "writer", urcu_get_thread_id());
 231
 232         set_affinity();
 233
 234         while (!test_go)
 235         {
 236         }
 237         cmm_smp_mb();
 238
 239         for (;;) {
 240 #ifndef TEST_LOCAL_GC
 241                 new = malloc(sizeof(*new));
 242                 new->a = 8;
 243                 old = rcu_xchg_pointer(&test_rcu_pointer, new);
 244 #endif
 245                 if (caa_unlikely(wduration))
 246                         loop_sleep(wduration);
 247                 rcu_gc_reclaim(wtidx, old);
 248                 URCU_TLS(nr_writes)++;
 249                 if (caa_unlikely(!test_duration_write()))
 250                         break;
 251                 if (caa_unlikely(wdelay))
 252                         loop_sleep(wdelay);
 253         }
 254
 255         printf_verbose("thread_end %s, tid %lu\n",
 256                         "writer", urcu_get_thread_id());
 257         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 258         return ((void*)2);
 259 }
 260
 261 static
 262 void show_usage(char **argv)
 263 {
 264         printf("Usage : %s nr_readers nr_writers duration (s) <OPTIONS>\n",
 265                 argv[0]);
 266         printf("OPTIONS:\n");
 267         printf("        [-r] [-w] (yield reader and/or writer)\n");
 268         printf("        [-d delay] (writer period (us))\n");
 269         printf("        [-c duration] (reader C.S. duration (in loops))\n");
 270         printf("        [-e duration] (writer C.S. duration (in loops))\n");
 271         printf("        [-v] (verbose output)\n");
 272         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 273         printf("\n");
 274 }
 275
 276 int main(int argc, char **argv)
 277 {
 278         int err;
 279         pthread_t *tid_reader, *tid_writer;
 280         void *tret;
 281         unsigned long long *count_reader;
 282         unsigned long long tot_reads = 0, tot_writes = 0;
 283         int i, a;
 284         unsigned int i_thr;
 285
 286         if (argc < 4) {
 287                 show_usage(argv);
 288                 return -1;
 289         }
 290
 291         err = sscanf(argv[1], "%u", &nr_readers);
 292         if (err != 1) {
 293                 show_usage(argv);
 294                 return -1;
 295         }
 296
 297         err = sscanf(argv[2], "%u", &nr_writers);
 298         if (err != 1) {
 299                 show_usage(argv);
 300                 return -1;
 301         }
 302
 303         err = sscanf(argv[3], "%lu", &duration);
 304         if (err != 1) {
 305                 show_usage(argv);
 306                 return -1;
 307         }
 308
 309         for (i = 4; i < argc; i++) {
 310                 if (argv[i][0] != '-')
 311                         continue;
 312                 switch (argv[i][1]) {
 313                 case 'r':
 314                         rcu_debug_yield_enable(RCU_YIELD_READ);
 315                         break;
 316                 case 'w':
 317                         rcu_debug_yield_enable(RCU_YIELD_WRITE);
 318                         break;
 319                 case 'a':
 320                         if (argc < i + 2) {
 321                                 show_usage(argv);
 322                                 return -1;
 323                         }
 324                         a = atoi(argv[++i]);
 325                         cpu_affinities[next_aff++] = a;
 326                         use_affinity = 1;
 327                         printf_verbose("Adding CPU %d affinity\n", a);
 328                         break;
 329                 case 'b':
 330                         if (argc < i + 2) {
 331                                 show_usage(argv);
 332                                 return -1;
 333                         }
 334                         reclaim_batch = atol(argv[++i]);
 335                         break;
 336                 case 'c':
 337                         if (argc < i + 2) {
 338                                 show_usage(argv);
 339                                 return -1;
 340                         }
 341                         rduration = atol(argv[++i]);
 342                         break;
 343                 case 'd':
 344                         if (argc < i + 2) {
 345                                 show_usage(argv);
 346                                 return -1;
 347                         }
 348                         wdelay = atol(argv[++i]);
 349                         break;
 350                 case 'e':
 351                         if (argc < i + 2) {
 352                                 show_usage(argv);
 353                                 return -1;
 354                         }
 355                         wduration = atol(argv[++i]);
 356                         break;
 357                 case 'v':
 358                         verbose_mode = 1;
 359                         break;
 360                 }
 361         }
 362
 363         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 364                 duration, nr_readers, nr_writers);
 365         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 366         printf_verbose("Reader duration : %lu loops.\n", rduration);
 367         printf_verbose("thread %-6s, tid %lu\n",
 368                         "main", urcu_get_thread_id());
 369
 370         tid_reader = calloc(nr_readers, sizeof(*tid_reader));
 371         tid_writer = calloc(nr_writers, sizeof(*tid_writer));
 372         count_reader = calloc(nr_readers, sizeof(*count_reader));
 373         tot_nr_writes = calloc(nr_writers, sizeof(*tot_nr_writes));
 374         pending_reclaims = calloc(nr_writers, sizeof(*pending_reclaims));
 375
 376         if (reclaim_batch * sizeof(*pending_reclaims[0].queue)
 377                         < CAA_CACHE_LINE_SIZE)
 378                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 379                         pending_reclaims[i_thr].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 380         else
 381                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 382                         pending_reclaims[i_thr].queue = calloc(reclaim_batch,
 383                                         sizeof(*pending_reclaims[i_thr].queue));
 384         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 385                 pending_reclaims[i_thr].head = pending_reclaims[i_thr].queue;
 386
 387         next_aff = 0;
 388
 389         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 390                 err = pthread_create(&tid_reader[i_thr], NULL, thr_reader,
 391                                      &count_reader[i_thr]);
 392                 if (err != 0)
 393                         exit(1);
 394         }
 395         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 396                 err = pthread_create(&tid_writer[i_thr], NULL, thr_writer,
 397                                      (void *)(long)i_thr);
 398                 if (err != 0)
 399                         exit(1);
 400         }
 401
 402         cmm_smp_mb();
 403
 404         test_go = 1;
 405
 406         sleep(duration);
 407
 408         test_stop = 1;
 409
 410         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 411                 err = pthread_join(tid_reader[i_thr], &tret);
 412                 if (err != 0)
 413                         exit(1);
 414                 tot_reads += count_reader[i_thr];
 415         }
 416         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 417                 err = pthread_join(tid_writer[i_thr], &tret);
 418                 if (err != 0)
 419                         exit(1);
 420                 tot_writes += tot_nr_writes[i_thr];
 421                 rcu_gc_clear_queue(i_thr);
 422         }
 423
 424         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 425                tot_writes);
 426         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 427                 "nr_writers %3u "
 428                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 429                 "batch %ld\n",
 430                 argv[0], duration, nr_readers, rduration, wduration,
 431                 nr_writers, wdelay, tot_reads, tot_writes,
 432                 tot_reads + tot_writes, reclaim_batch);
 433
 434         free(tid_reader);
 435         free(tid_writer);
 436         free(count_reader);
 437         free(tot_nr_writes);
 438
 439         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 440                 free(pending_reclaims[i_thr].queue);
 441         free(pending_reclaims);
 442
 443         return 0;
 444 }