tests/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 #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
  40 #ifdef __linux__
  41 #include <syscall.h>
  42 #endif
  43
  44 /* hardcoded number of CPUs */
  45 #define NR_CPUS 16384
  46
  47 #if defined(_syscall0)
  48 _syscall0(pid_t, gettid)
  49 #elif defined(__NR_gettid)
  50 static inline pid_t gettid(void)
  51 {
  52         return syscall(__NR_gettid);
  53 }
  54 #else
  55 #warning "use pid as tid"
  56 static inline pid_t gettid(void)
  57 {
  58         return getpid();
  59 }
  60 #endif
  61
  62 #define _LGPL_SOURCE
  63 #include <urcu-qsbr.h>
  64
  65 struct test_array {
  66         int a;
  67 };
  68
  69 static volatile int test_go, test_stop;
  70
  71 static unsigned long wdelay;
  72
  73 static struct test_array *test_rcu_pointer;
  74
  75 static unsigned long duration;
  76
  77 /* read-side C.S. duration, in loops */
  78 static unsigned long rduration;
  79 static unsigned int reclaim_batch = 1;
  80
  81 struct reclaim_queue {
  82         void **queue;   /* Beginning of queue */
  83         void **head;    /* Insert position */
  84 };
  85
  86 static struct reclaim_queue *pending_reclaims;
  87
  88
  89 /* write-side C.S. duration, in loops */
  90 static unsigned long wduration;
  91
  92 static inline void loop_sleep(unsigned long loops)
  93 {
  94         while (loops-- != 0)
  95                 caa_cpu_relax();
  96 }
  97
  98 static int verbose_mode;
  99
 100 #define printf_verbose(fmt, args...)            \
 101         do {                                    \
 102                 if (verbose_mode)               \
 103                         printf(fmt, args);      \
 104         } while (0)
 105
 106 static unsigned int cpu_affinities[NR_CPUS];
 107 static unsigned int next_aff = 0;
 108 static int use_affinity = 0;
 109
 110 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 111
 112 static void set_affinity(void)
 113 {
 114 #if HAVE_SCHED_SETAFFINITY
 115         cpu_set_t mask;
 116         int cpu, ret;
 117 #endif /* HAVE_SCHED_SETAFFINITY */
 118
 119         if (!use_affinity)
 120                 return;
 121
 122 #if HAVE_SCHED_SETAFFINITY
 123         ret = pthread_mutex_lock(&affinity_mutex);
 124         if (ret) {
 125                 perror("Error in pthread mutex lock");
 126                 exit(-1);
 127         }
 128         cpu = cpu_affinities[next_aff++];
 129         ret = pthread_mutex_unlock(&affinity_mutex);
 130         if (ret) {
 131                 perror("Error in pthread mutex unlock");
 132                 exit(-1);
 133         }
 134
 135         CPU_ZERO(&mask);
 136         CPU_SET(cpu, &mask);
 137 #if SCHED_SETAFFINITY_ARGS == 2
 138         sched_setaffinity(0, &mask);
 139 #else
 140         sched_setaffinity(0, sizeof(mask), &mask);
 141 #endif
 142 #endif /* HAVE_SCHED_SETAFFINITY */
 143 }
 144
 145 /*
 146  * returns 0 if test should end.
 147  */
 148 static int test_duration_write(void)
 149 {
 150         return !test_stop;
 151 }
 152
 153 static int test_duration_read(void)
 154 {
 155         return !test_stop;
 156 }
 157
 158 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
 159 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 160
 161 static unsigned int nr_readers;
 162 static unsigned int nr_writers;
 163
 164 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 165 static
 166 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 167
 168
 169 void rcu_copy_mutex_lock(void)
 170 {
 171         int ret;
 172         ret = pthread_mutex_lock(&rcu_copy_mutex);
 173         if (ret) {
 174                 perror("Error in pthread mutex lock");
 175                 exit(-1);
 176         }
 177 }
 178
 179 void rcu_copy_mutex_unlock(void)
 180 {
 181         int ret;
 182
 183         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 184         if (ret) {
 185                 perror("Error in pthread mutex unlock");
 186                 exit(-1);
 187         }
 188 }
 189
 190 void *thr_reader(void *_count)
 191 {
 192         unsigned long long *count = _count;
 193         struct test_array *local_ptr;
 194
 195         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 196                         "reader", (unsigned long) pthread_self(),
 197                         (unsigned long) gettid());
 198
 199         set_affinity();
 200
 201         rcu_register_thread();
 202
 203         while (!test_go)
 204         {
 205         }
 206         cmm_smp_mb();
 207
 208         for (;;) {
 209                 _rcu_read_lock();
 210                 local_ptr = _rcu_dereference(test_rcu_pointer);
 211                 rcu_debug_yield_read();
 212                 if (local_ptr)
 213                         assert(local_ptr->a == 8);
 214                 if (caa_unlikely(rduration))
 215                         loop_sleep(rduration);
 216                 _rcu_read_unlock();
 217                 URCU_TLS(nr_reads)++;
 218                 /* QS each 1024 reads */
 219                 if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 220                         _rcu_quiescent_state();
 221                 if (caa_unlikely(!test_duration_read()))
 222                         break;
 223         }
 224
 225         rcu_unregister_thread();
 226
 227         *count = URCU_TLS(nr_reads);
 228         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 229                         "reader", (unsigned long) pthread_self(),
 230                         (unsigned long) gettid());
 231         return ((void*)1);
 232
 233 }
 234
 235 static void rcu_gc_clear_queue(unsigned long wtidx)
 236 {
 237         void **p;
 238
 239         /* Wait for Q.S and empty queue */
 240         synchronize_rcu();
 241
 242         for (p = pending_reclaims[wtidx].queue;
 243                         p < pending_reclaims[wtidx].head; p++) {
 244                 /* poison */
 245                 if (*p)
 246                         ((struct test_array *)*p)->a = 0;
 247                 free(*p);
 248         }
 249         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 250 }
 251
 252 /* Using per-thread queue */
 253 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 254 {
 255         /* Queue pointer */
 256         *pending_reclaims[wtidx].head = old;
 257         pending_reclaims[wtidx].head++;
 258
 259         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 260                         < reclaim_batch))
 261                 return;
 262
 263         rcu_gc_clear_queue(wtidx);
 264 }
 265
 266 void *thr_writer(void *data)
 267 {
 268         unsigned long wtidx = (unsigned long)data;
 269 #ifdef TEST_LOCAL_GC
 270         struct test_array *old = NULL;
 271 #else
 272         struct test_array *new, *old;
 273 #endif
 274
 275         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 276                         "writer", (unsigned long) pthread_self(),
 277                         (unsigned long) gettid());
 278
 279         set_affinity();
 280
 281         while (!test_go)
 282         {
 283         }
 284         cmm_smp_mb();
 285
 286         for (;;) {
 287 #ifndef TEST_LOCAL_GC
 288                 new = malloc(sizeof(*new));
 289                 new->a = 8;
 290                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 291 #endif
 292                 if (caa_unlikely(wduration))
 293                         loop_sleep(wduration);
 294                 rcu_gc_reclaim(wtidx, old);
 295                 URCU_TLS(nr_writes)++;
 296                 if (caa_unlikely(!test_duration_write()))
 297                         break;
 298                 if (caa_unlikely(wdelay))
 299                         loop_sleep(wdelay);
 300         }
 301
 302         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 303                         "writer", (unsigned long) pthread_self(),
 304                         (unsigned long) gettid());
 305         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 306         return ((void*)2);
 307 }
 308
 309 void show_usage(int argc, char **argv)
 310 {
 311         printf("Usage : %s nr_readers nr_writers duration (s) <OPTIONS>\n",
 312                 argv[0]);
 313         printf("OPTIONS:\n");
 314 #ifdef DEBUG_YIELD
 315         printf("        [-r] [-w] (yield reader and/or writer)\n");
 316 #endif
 317         printf("        [-b batch] (batch reclaim)\n");
 318         printf("        [-d delay] (writer period (us))\n");
 319         printf("        [-c duration] (reader C.S. duration (in loops))\n");
 320         printf("        [-e duration] (writer C.S. duration (in loops))\n");
 321         printf("        [-v] (verbose output)\n");
 322         printf("        [-a cpu#] [-a cpu#]... (affinity)\n");
 323         printf("\n");
 324 }
 325
 326 int main(int argc, char **argv)
 327 {
 328         int err;
 329         pthread_t *tid_reader, *tid_writer;
 330         void *tret;
 331         unsigned long long *count_reader;
 332         unsigned long long tot_reads = 0, tot_writes = 0;
 333         int i, a;
 334
 335         if (argc < 4) {
 336                 show_usage(argc, argv);
 337                 return -1;
 338         }
 339
 340         err = sscanf(argv[1], "%u", &nr_readers);
 341         if (err != 1) {
 342                 show_usage(argc, argv);
 343                 return -1;
 344         }
 345
 346         err = sscanf(argv[2], "%u", &nr_writers);
 347         if (err != 1) {
 348                 show_usage(argc, argv);
 349                 return -1;
 350         }
 351
 352         err = sscanf(argv[3], "%lu", &duration);
 353         if (err != 1) {
 354                 show_usage(argc, argv);
 355                 return -1;
 356         }
 357
 358         for (i = 4; i < argc; i++) {
 359                 if (argv[i][0] != '-')
 360                         continue;
 361                 switch (argv[i][1]) {
 362 #ifdef DEBUG_YIELD
 363                 case 'r':
 364                         rcu_yield_active |= RCU_YIELD_READ;
 365                         break;
 366                 case 'w':
 367                         rcu_yield_active |= RCU_YIELD_WRITE;
 368                         break;
 369 #endif
 370                 case 'a':
 371                         if (argc < i + 2) {
 372                                 show_usage(argc, argv);
 373                                 return -1;
 374                         }
 375                         a = atoi(argv[++i]);
 376                         cpu_affinities[next_aff++] = a;
 377                         use_affinity = 1;
 378                         printf_verbose("Adding CPU %d affinity\n", a);
 379                         break;
 380                 case 'b':
 381                         if (argc < i + 2) {
 382                                 show_usage(argc, argv);
 383                                 return -1;
 384                         }
 385                         reclaim_batch = atol(argv[++i]);
 386                         break;
 387                 case 'c':
 388                         if (argc < i + 2) {
 389                                 show_usage(argc, argv);
 390                                 return -1;
 391                         }
 392                         rduration = atol(argv[++i]);
 393                         break;
 394                 case 'd':
 395                         if (argc < i + 2) {
 396                                 show_usage(argc, argv);
 397                                 return -1;
 398                         }
 399                         wdelay = atol(argv[++i]);
 400                         break;
 401                 case 'e':
 402                         if (argc < i + 2) {
 403                                 show_usage(argc, argv);
 404                                 return -1;
 405                         }
 406                         wduration = atol(argv[++i]);
 407                         break;
 408                 case 'v':
 409                         verbose_mode = 1;
 410                         break;
 411                 }
 412         }
 413
 414         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 415                 duration, nr_readers, nr_writers);
 416         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 417         printf_verbose("Reader duration : %lu loops.\n", rduration);
 418         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 419                         "main", (unsigned long) pthread_self(),
 420                         (unsigned long) gettid());
 421
 422         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 423         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 424         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 425         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 426         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 427         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 428                         < CAA_CACHE_LINE_SIZE)
 429                 for (i = 0; i < nr_writers; i++)
 430                         pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 431         else
 432                 for (i = 0; i < nr_writers; i++)
 433                         pending_reclaims[i].queue = calloc(reclaim_batch,
 434                                         sizeof(*pending_reclaims[i].queue));
 435         for (i = 0; i < nr_writers; i++)
 436                 pending_reclaims[i].head = pending_reclaims[i].queue;
 437
 438         next_aff = 0;
 439
 440         for (i = 0; i < nr_readers; i++) {
 441                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 442                                      &count_reader[i]);
 443                 if (err != 0)
 444                         exit(1);
 445         }
 446         for (i = 0; i < nr_writers; i++) {
 447                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 448                                      (void *)(long)i);
 449                 if (err != 0)
 450                         exit(1);
 451         }
 452
 453         cmm_smp_mb();
 454
 455         test_go = 1;
 456
 457         sleep(duration);
 458
 459         test_stop = 1;
 460
 461         for (i = 0; i < nr_readers; i++) {
 462                 err = pthread_join(tid_reader[i], &tret);
 463                 if (err != 0)
 464                         exit(1);
 465                 tot_reads += count_reader[i];
 466         }
 467         for (i = 0; i < nr_writers; i++) {
 468                 err = pthread_join(tid_writer[i], &tret);
 469                 if (err != 0)
 470                         exit(1);
 471                 tot_writes += tot_nr_writes[i];
 472                 rcu_gc_clear_queue(i);
 473         }
 474
 475         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 476                tot_writes);
 477         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 478                 "nr_writers %3u "
 479                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 480                 "batch %u\n",
 481                 argv[0], duration, nr_readers, rduration, wduration,
 482                 nr_writers, wdelay, tot_reads, tot_writes,
 483                 tot_reads + tot_writes, reclaim_batch);
 484         free(tid_reader);
 485         free(tid_writer);
 486         free(count_reader);
 487         free(tot_nr_writes);
 488         for (i = 0; i < nr_writers; i++)
 489                 free(pending_reclaims[i].queue);
 490         free(pending_reclaims);
 491
 492         return 0;
 493 }