tests/test_urcu_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 #ifndef DYNAMIC_LINK_TEST
  63 #define _LGPL_SOURCE
  64 #else
  65 #define debug_yield_read()
  66 #endif
  67 #include <urcu.h>
  68
  69 struct test_array {
  70         int a;
  71 };
  72
  73 static volatile int test_go, test_stop;
  74
  75 static unsigned long wdelay;
  76
  77 static struct test_array *test_rcu_pointer;
  78
  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 static unsigned long duration;
  89
  90 /* read-side C.S. duration, in loops */
  91 static unsigned long rduration;
  92
  93 /* write-side C.S. duration, in loops */
  94 static unsigned long wduration;
  95
  96 static inline void loop_sleep(unsigned long l)
  97 {
  98         while(l-- != 0)
  99                 caa_cpu_relax();
 100 }
 101
 102 static int verbose_mode;
 103
 104 #define printf_verbose(fmt, args...)            \
 105         do {                                    \
 106                 if (verbose_mode)               \
 107                         printf(fmt, args);      \
 108         } while (0)
 109
 110 static unsigned int cpu_affinities[NR_CPUS];
 111 static unsigned int next_aff = 0;
 112 static int use_affinity = 0;
 113
 114 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 115
 116 static void set_affinity(void)
 117 {
 118         cpu_set_t mask;
 119         int cpu;
 120         int ret;
 121
 122         if (!use_affinity)
 123                 return;
 124
 125 #if HAVE_SCHED_SETAFFINITY
 126         ret = pthread_mutex_lock(&affinity_mutex);
 127         if (ret) {
 128                 perror("Error in pthread mutex lock");
 129                 exit(-1);
 130         }
 131         cpu = cpu_affinities[next_aff++];
 132         ret = pthread_mutex_unlock(&affinity_mutex);
 133         if (ret) {
 134                 perror("Error in pthread mutex unlock");
 135                 exit(-1);
 136         }
 137
 138         CPU_ZERO(&mask);
 139         CPU_SET(cpu, &mask);
 140 #if SCHED_SETAFFINITY_ARGS == 2
 141         sched_setaffinity(0, &mask);
 142 #else
 143         sched_setaffinity(0, sizeof(mask), &mask);
 144 #endif
 145 #endif /* HAVE_SCHED_SETAFFINITY */
 146 }
 147
 148 /*
 149  * returns 0 if test should end.
 150  */
 151 static int test_duration_write(void)
 152 {
 153         return !test_stop;
 154 }
 155
 156 static int test_duration_read(void)
 157 {
 158         return !test_stop;
 159 }
 160
 161 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
 162 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 163
 164 static
 165 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 166
 167 static unsigned int nr_readers;
 168 static unsigned int nr_writers;
 169
 170 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 171
 172 void rcu_copy_mutex_lock(void)
 173 {
 174         int ret;
 175         ret = pthread_mutex_lock(&rcu_copy_mutex);
 176         if (ret) {
 177                 perror("Error in pthread mutex lock");
 178                 exit(-1);
 179         }
 180 }
 181
 182 void rcu_copy_mutex_unlock(void)
 183 {
 184         int ret;
 185
 186         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 187         if (ret) {
 188                 perror("Error in pthread mutex unlock");
 189                 exit(-1);
 190         }
 191 }
 192
 193 void *thr_reader(void *_count)
 194 {
 195         unsigned long long *count = _count;
 196         struct test_array *local_ptr;
 197
 198         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 199                         "reader", (unsigned long) pthread_self(),
 200                         (unsigned long) gettid());
 201
 202         set_affinity();
 203
 204         rcu_register_thread();
 205
 206         while (!test_go)
 207         {
 208         }
 209         cmm_smp_mb();
 210
 211         for (;;) {
 212                 rcu_read_lock();
 213                 local_ptr = rcu_dereference(test_rcu_pointer);
 214                 debug_yield_read();
 215                 if (local_ptr)
 216                         assert(local_ptr->a == 8);
 217                 if (caa_unlikely(rduration))
 218                         loop_sleep(rduration);
 219                 rcu_read_unlock();
 220                 URCU_TLS(nr_reads)++;
 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)", argv[0]);
 312 #ifdef DEBUG_YIELD
 313         printf(" [-r] [-w] (yield reader and/or writer)");
 314 #endif
 315         printf(" [-d delay] (writer period (us))");
 316         printf(" [-c duration] (reader C.S. duration (in loops))");
 317         printf(" [-e duration] (writer C.S. duration (in loops))");
 318         printf(" [-v] (verbose output)");
 319         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 320         printf("\n");
 321 }
 322
 323 int main(int argc, char **argv)
 324 {
 325         int err;
 326         pthread_t *tid_reader, *tid_writer;
 327         void *tret;
 328         unsigned long long *count_reader;
 329         unsigned long long tot_reads = 0, tot_writes = 0;
 330         int i, a;
 331
 332         if (argc < 4) {
 333                 show_usage(argc, argv);
 334                 return -1;
 335         }
 336
 337         err = sscanf(argv[1], "%u", &nr_readers);
 338         if (err != 1) {
 339                 show_usage(argc, argv);
 340                 return -1;
 341         }
 342
 343         err = sscanf(argv[2], "%u", &nr_writers);
 344         if (err != 1) {
 345                 show_usage(argc, argv);
 346                 return -1;
 347         }
 348
 349         err = sscanf(argv[3], "%lu", &duration);
 350         if (err != 1) {
 351                 show_usage(argc, argv);
 352                 return -1;
 353         }
 354
 355         for (i = 4; i < argc; i++) {
 356                 if (argv[i][0] != '-')
 357                         continue;
 358                 switch (argv[i][1]) {
 359 #ifdef DEBUG_YIELD
 360                 case 'r':
 361                         yield_active |= YIELD_READ;
 362                         break;
 363                 case 'w':
 364                         yield_active |= YIELD_WRITE;
 365                         break;
 366 #endif
 367                 case 'a':
 368                         if (argc < i + 2) {
 369                                 show_usage(argc, argv);
 370                                 return -1;
 371                         }
 372                         a = atoi(argv[++i]);
 373                         cpu_affinities[next_aff++] = a;
 374                         use_affinity = 1;
 375                         printf_verbose("Adding CPU %d affinity\n", a);
 376                         break;
 377                 case 'b':
 378                         if (argc < i + 2) {
 379                                 show_usage(argc, argv);
 380                                 return -1;
 381                         }
 382                         reclaim_batch = atol(argv[++i]);
 383                         break;
 384                 case 'c':
 385                         if (argc < i + 2) {
 386                                 show_usage(argc, argv);
 387                                 return -1;
 388                         }
 389                         rduration = atol(argv[++i]);
 390                         break;
 391                 case 'd':
 392                         if (argc < i + 2) {
 393                                 show_usage(argc, argv);
 394                                 return -1;
 395                         }
 396                         wdelay = atol(argv[++i]);
 397                         break;
 398                 case 'e':
 399                         if (argc < i + 2) {
 400                                 show_usage(argc, argv);
 401                                 return -1;
 402                         }
 403                         wduration = atol(argv[++i]);
 404                         break;
 405                 case 'v':
 406                         verbose_mode = 1;
 407                         break;
 408                 }
 409         }
 410
 411         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 412                 duration, nr_readers, nr_writers);
 413         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 414         printf_verbose("Reader duration : %lu loops.\n", rduration);
 415         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 416                         "main", (unsigned long) pthread_self(),
 417                         (unsigned long) gettid());
 418
 419         tid_reader = calloc(nr_readers, sizeof(*tid_reader));
 420         tid_writer = calloc(nr_writers, sizeof(*tid_writer));
 421         count_reader = calloc(nr_readers, sizeof(*count_reader));
 422         tot_nr_writes = calloc(nr_writers, sizeof(*tot_nr_writes));
 423         pending_reclaims = calloc(nr_writers, sizeof(*pending_reclaims));
 424         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 425                         < CAA_CACHE_LINE_SIZE)
 426                 for (i = 0; i < nr_writers; i++)
 427                         pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 428         else
 429                 for (i = 0; i < nr_writers; i++)
 430                         pending_reclaims[i].queue = calloc(reclaim_batch,
 431                                         sizeof(*pending_reclaims[i].queue));
 432         for (i = 0; i < nr_writers; i++)
 433                 pending_reclaims[i].head = pending_reclaims[i].queue;
 434
 435         next_aff = 0;
 436
 437         for (i = 0; i < nr_readers; i++) {
 438                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 439                                      &count_reader[i]);
 440                 if (err != 0)
 441                         exit(1);
 442         }
 443         for (i = 0; i < nr_writers; i++) {
 444                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 445                                      (void *)(long)i);
 446                 if (err != 0)
 447                         exit(1);
 448         }
 449
 450         cmm_smp_mb();
 451
 452         test_go = 1;
 453
 454         sleep(duration);
 455
 456         test_stop = 1;
 457
 458         for (i = 0; i < nr_readers; i++) {
 459                 err = pthread_join(tid_reader[i], &tret);
 460                 if (err != 0)
 461                         exit(1);
 462                 tot_reads += count_reader[i];
 463         }
 464         for (i = 0; i < nr_writers; i++) {
 465                 err = pthread_join(tid_writer[i], &tret);
 466                 if (err != 0)
 467                         exit(1);
 468                 tot_writes += tot_nr_writes[i];
 469                 rcu_gc_clear_queue(i);
 470         }
 471
 472         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 473                tot_writes);
 474         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 475                 "nr_writers %3u "
 476                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 477                 "batch %u\n",
 478                 argv[0], duration, nr_readers, rduration, wduration,
 479                 nr_writers, wdelay, tot_reads, tot_writes,
 480                 tot_reads + tot_writes, reclaim_batch);
 481         free(tid_reader);
 482         free(tid_writer);
 483         free(count_reader);
 484         free(tot_nr_writes);
 485         for (i = 0; i < nr_writers; i++)
 486                 free(pending_reclaims[i].queue);
 487         free(pending_reclaims);
 488
 489         return 0;
 490 }