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
 150 void rcu_copy_mutex_lock(void)
 151 {
 152         int ret;
 153         ret = pthread_mutex_lock(&rcu_copy_mutex);
 154         if (ret) {
 155                 perror("Error in pthread mutex lock");
 156                 exit(-1);
 157         }
 158 }
 159
 160 void rcu_copy_mutex_unlock(void)
 161 {
 162         int ret;
 163
 164         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 165         if (ret) {
 166                 perror("Error in pthread mutex unlock");
 167                 exit(-1);
 168         }
 169 }
 170
 171 void *thr_reader(void *_count)
 172 {
 173         unsigned long long *count = _count;
 174         struct test_array *local_ptr;
 175
 176         printf_verbose("thread_begin %s, tid %lu\n",
 177                         "reader", urcu_get_thread_id());
 178
 179         set_affinity();
 180
 181         rcu_register_thread();
 182
 183         while (!test_go)
 184         {
 185         }
 186         cmm_smp_mb();
 187
 188         for (;;) {
 189                 _rcu_read_lock();
 190                 local_ptr = _rcu_dereference(test_rcu_pointer);
 191                 rcu_debug_yield_read();
 192                 if (local_ptr)
 193                         assert(local_ptr->a == 8);
 194                 if (caa_unlikely(rduration))
 195                         loop_sleep(rduration);
 196                 _rcu_read_unlock();
 197                 URCU_TLS(nr_reads)++;
 198                 /* QS each 1024 reads */
 199                 if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 200                         _rcu_quiescent_state();
 201                 if (caa_unlikely(!test_duration_read()))
 202                         break;
 203         }
 204
 205         rcu_unregister_thread();
 206
 207         *count = URCU_TLS(nr_reads);
 208         printf_verbose("thread_end %s, tid %lu\n",
 209                         "reader", urcu_get_thread_id());
 210         return ((void*)1);
 211
 212 }
 213
 214 static void rcu_gc_clear_queue(unsigned long wtidx)
 215 {
 216         void **p;
 217
 218         /* Wait for Q.S and empty queue */
 219         synchronize_rcu();
 220
 221         for (p = pending_reclaims[wtidx].queue;
 222                         p < pending_reclaims[wtidx].head; p++) {
 223                 /* poison */
 224                 if (*p)
 225                         ((struct test_array *)*p)->a = 0;
 226                 free(*p);
 227         }
 228         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 229 }
 230
 231 /* Using per-thread queue */
 232 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 233 {
 234         /* Queue pointer */
 235         *pending_reclaims[wtidx].head = old;
 236         pending_reclaims[wtidx].head++;
 237
 238         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 239                         < reclaim_batch))
 240                 return;
 241
 242         rcu_gc_clear_queue(wtidx);
 243 }
 244
 245 void *thr_writer(void *data)
 246 {
 247         unsigned long wtidx = (unsigned long)data;
 248 #ifdef TEST_LOCAL_GC
 249         struct test_array *old = NULL;
 250 #else
 251         struct test_array *new, *old;
 252 #endif
 253
 254         printf_verbose("thread_begin %s, tid %lu\n",
 255                         "writer", urcu_get_thread_id());
 256
 257         set_affinity();
 258
 259         while (!test_go)
 260         {
 261         }
 262         cmm_smp_mb();
 263
 264         for (;;) {
 265 #ifndef TEST_LOCAL_GC
 266                 new = malloc(sizeof(*new));
 267                 new->a = 8;
 268                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 269 #endif
 270                 if (caa_unlikely(wduration))
 271                         loop_sleep(wduration);
 272                 rcu_gc_reclaim(wtidx, old);
 273                 URCU_TLS(nr_writes)++;
 274                 if (caa_unlikely(!test_duration_write()))
 275                         break;
 276                 if (caa_unlikely(wdelay))
 277                         loop_sleep(wdelay);
 278         }
 279
 280         printf_verbose("thread_end %s, tid %lu\n",
 281                         "writer", urcu_get_thread_id());
 282         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 283         return ((void*)2);
 284 }
 285
 286 void show_usage(int argc, char **argv)
 287 {
 288         printf("Usage : %s nr_readers nr_writers duration (s) <OPTIONS>\n",
 289                 argv[0]);
 290         printf("OPTIONS:\n");
 291         printf("        [-r] [-w] (yield reader and/or writer)\n");
 292         printf("        [-b batch] (batch reclaim)\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         unsigned int i_thr;
 310
 311         if (argc < 4) {
 312                 show_usage(argc, argv);
 313                 return -1;
 314         }
 315
 316         err = sscanf(argv[1], "%u", &nr_readers);
 317         if (err != 1) {
 318                 show_usage(argc, argv);
 319                 return -1;
 320         }
 321
 322         err = sscanf(argv[2], "%u", &nr_writers);
 323         if (err != 1) {
 324                 show_usage(argc, argv);
 325                 return -1;
 326         }
 327
 328         err = sscanf(argv[3], "%lu", &duration);
 329         if (err != 1) {
 330                 show_usage(argc, argv);
 331                 return -1;
 332         }
 333
 334         for (i = 4; i < argc; i++) {
 335                 if (argv[i][0] != '-')
 336                         continue;
 337                 switch (argv[i][1]) {
 338                 case 'r':
 339                         rcu_debug_yield_enable(RCU_YIELD_READ);
 340                         break;
 341                 case 'w':
 342                         rcu_debug_yield_enable(RCU_YIELD_WRITE);
 343                         break;
 344                 case 'a':
 345                         if (argc < i + 2) {
 346                                 show_usage(argc, argv);
 347                                 return -1;
 348                         }
 349                         a = atoi(argv[++i]);
 350                         cpu_affinities[next_aff++] = a;
 351                         use_affinity = 1;
 352                         printf_verbose("Adding CPU %d affinity\n", a);
 353                         break;
 354                 case 'b':
 355                         if (argc < i + 2) {
 356                                 show_usage(argc, argv);
 357                                 return -1;
 358                         }
 359                         reclaim_batch = atol(argv[++i]);
 360                         break;
 361                 case 'c':
 362                         if (argc < i + 2) {
 363                                 show_usage(argc, argv);
 364                                 return -1;
 365                         }
 366                         rduration = atol(argv[++i]);
 367                         break;
 368                 case 'd':
 369                         if (argc < i + 2) {
 370                                 show_usage(argc, argv);
 371                                 return -1;
 372                         }
 373                         wdelay = atol(argv[++i]);
 374                         break;
 375                 case 'e':
 376                         if (argc < i + 2) {
 377                                 show_usage(argc, argv);
 378                                 return -1;
 379                         }
 380                         wduration = atol(argv[++i]);
 381                         break;
 382                 case 'v':
 383                         verbose_mode = 1;
 384                         break;
 385                 }
 386         }
 387
 388         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 389                 duration, nr_readers, nr_writers);
 390         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 391         printf_verbose("Reader duration : %lu loops.\n", rduration);
 392         printf_verbose("thread %-6s, tid %lu\n",
 393                         "main", urcu_get_thread_id());
 394
 395         tid_reader = calloc(nr_readers, sizeof(*tid_reader));
 396         tid_writer = calloc(nr_writers, sizeof(*tid_writer));
 397         count_reader = calloc(nr_readers, sizeof(*count_reader));
 398         tot_nr_writes = calloc(nr_writers, sizeof(*tot_nr_writes));
 399         pending_reclaims = calloc(nr_writers, sizeof(*pending_reclaims));
 400         if (reclaim_batch * sizeof(*pending_reclaims[0].queue)
 401                         < CAA_CACHE_LINE_SIZE)
 402                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 403                         pending_reclaims[i_thr].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 404         else
 405                 for (i_thr = 0; i_thr < nr_writers; i_thr++)
 406                         pending_reclaims[i_thr].queue = calloc(reclaim_batch,
 407                                         sizeof(*pending_reclaims[i_thr].queue));
 408         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 409                 pending_reclaims[i_thr].head = pending_reclaims[i_thr].queue;
 410
 411         next_aff = 0;
 412
 413         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 414                 err = pthread_create(&tid_reader[i_thr], NULL, thr_reader,
 415                                      &count_reader[i_thr]);
 416                 if (err != 0)
 417                         exit(1);
 418         }
 419         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 420                 err = pthread_create(&tid_writer[i_thr], NULL, thr_writer,
 421                                      (void *)(long)i_thr);
 422                 if (err != 0)
 423                         exit(1);
 424         }
 425
 426         cmm_smp_mb();
 427
 428         test_go = 1;
 429
 430         sleep(duration);
 431
 432         test_stop = 1;
 433
 434         for (i_thr = 0; i_thr < nr_readers; i_thr++) {
 435                 err = pthread_join(tid_reader[i_thr], &tret);
 436                 if (err != 0)
 437                         exit(1);
 438                 tot_reads += count_reader[i_thr];
 439         }
 440         for (i_thr = 0; i_thr < nr_writers; i_thr++) {
 441                 err = pthread_join(tid_writer[i_thr], &tret);
 442                 if (err != 0)
 443                         exit(1);
 444                 tot_writes += tot_nr_writes[i_thr];
 445                 rcu_gc_clear_queue(i_thr);
 446         }
 447
 448         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 449                tot_writes);
 450         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 451                 "nr_writers %3u "
 452                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 453                 "batch %ld\n",
 454                 argv[0], duration, nr_readers, rduration, wduration,
 455                 nr_writers, wdelay, tot_reads, tot_writes,
 456                 tot_reads + tot_writes, reclaim_batch);
 457
 458         free(tid_reader);
 459         free(tid_writer);
 460         free(count_reader);
 461         free(tot_nr_writes);
 462
 463         for (i_thr = 0; i_thr < nr_writers; i_thr++)
 464                 free(pending_reclaims[i_thr].queue);
 465         free(pending_reclaims);
 466
 467         return 0;
 468 }