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