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 <sched.h>
  35 #include <errno.h>
  36
  37 #include <urcu/arch.h>
  38 #include <urcu/tls-compat.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 rcu_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 loops)
  97 {
  98         while (loops-- != 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 #ifndef HAVE_CPU_SET_T
 117 typedef unsigned long cpu_set_t;
 118 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 119 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 120 #endif
 121
 122 static void set_affinity(void)
 123 {
 124 #if HAVE_SCHED_SETAFFINITY
 125         cpu_set_t mask;
 126         int cpu, ret;
 127 #endif /* HAVE_SCHED_SETAFFINITY */
 128
 129         if (!use_affinity)
 130                 return;
 131
 132 #if HAVE_SCHED_SETAFFINITY
 133         ret = pthread_mutex_lock(&affinity_mutex);
 134         if (ret) {
 135                 perror("Error in pthread mutex lock");
 136                 exit(-1);
 137         }
 138         cpu = cpu_affinities[next_aff++];
 139         ret = pthread_mutex_unlock(&affinity_mutex);
 140         if (ret) {
 141                 perror("Error in pthread mutex unlock");
 142                 exit(-1);
 143         }
 144
 145         CPU_ZERO(&mask);
 146         CPU_SET(cpu, &mask);
 147 #if SCHED_SETAFFINITY_ARGS == 2
 148         sched_setaffinity(0, &mask);
 149 #else
 150         sched_setaffinity(0, sizeof(mask), &mask);
 151 #endif
 152 #endif /* HAVE_SCHED_SETAFFINITY */
 153 }
 154
 155 /*
 156  * returns 0 if test should end.
 157  */
 158 static int test_duration_write(void)
 159 {
 160         return !test_stop;
 161 }
 162
 163 static int test_duration_read(void)
 164 {
 165         return !test_stop;
 166 }
 167
 168 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
 169 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 170
 171 static
 172 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 173
 174 static unsigned int nr_readers;
 175 static unsigned int nr_writers;
 176
 177 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 178
 179 void rcu_copy_mutex_lock(void)
 180 {
 181         int ret;
 182         ret = pthread_mutex_lock(&rcu_copy_mutex);
 183         if (ret) {
 184                 perror("Error in pthread mutex lock");
 185                 exit(-1);
 186         }
 187 }
 188
 189 void rcu_copy_mutex_unlock(void)
 190 {
 191         int ret;
 192
 193         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 194         if (ret) {
 195                 perror("Error in pthread mutex unlock");
 196                 exit(-1);
 197         }
 198 }
 199
 200 void *thr_reader(void *_count)
 201 {
 202         unsigned long long *count = _count;
 203         struct test_array *local_ptr;
 204
 205         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 206                         "reader", pthread_self(), (unsigned long)gettid());
 207
 208         set_affinity();
 209
 210         rcu_register_thread();
 211
 212         while (!test_go)
 213         {
 214         }
 215         cmm_smp_mb();
 216
 217         for (;;) {
 218                 rcu_read_lock();
 219                 local_ptr = rcu_dereference(test_rcu_pointer);
 220                 rcu_debug_yield_read();
 221                 if (local_ptr)
 222                         assert(local_ptr->a == 8);
 223                 if (caa_unlikely(rduration))
 224                         loop_sleep(rduration);
 225                 rcu_read_unlock();
 226                 URCU_TLS(nr_reads)++;
 227                 if (caa_unlikely(!test_duration_read()))
 228                         break;
 229         }
 230
 231         rcu_unregister_thread();
 232
 233         *count = URCU_TLS(nr_reads);
 234         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 235                         "reader", pthread_self(), (unsigned long)gettid());
 236         return ((void*)1);
 237
 238 }
 239
 240 static void rcu_gc_clear_queue(unsigned long wtidx)
 241 {
 242         void **p;
 243
 244         /* Wait for Q.S and empty queue */
 245         synchronize_rcu();
 246
 247         for (p = pending_reclaims[wtidx].queue;
 248                         p < pending_reclaims[wtidx].head; p++) {
 249                 /* poison */
 250                 if (*p)
 251                         ((struct test_array *)*p)->a = 0;
 252                 free(*p);
 253         }
 254         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 255 }
 256
 257 /* Using per-thread queue */
 258 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 259 {
 260         /* Queue pointer */
 261         *pending_reclaims[wtidx].head = old;
 262         pending_reclaims[wtidx].head++;
 263
 264         if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 265                         < reclaim_batch))
 266                 return;
 267
 268         rcu_gc_clear_queue(wtidx);
 269 }
 270
 271 void *thr_writer(void *data)
 272 {
 273         unsigned long wtidx = (unsigned long)data;
 274 #ifdef TEST_LOCAL_GC
 275         struct test_array *old = NULL;
 276 #else
 277         struct test_array *new, *old;
 278 #endif
 279
 280         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 281                         "writer", pthread_self(), (unsigned long)gettid());
 282
 283         set_affinity();
 284
 285         while (!test_go)
 286         {
 287         }
 288         cmm_smp_mb();
 289
 290         for (;;) {
 291 #ifndef TEST_LOCAL_GC
 292                 new = malloc(sizeof(*new));
 293                 new->a = 8;
 294                 old = rcu_xchg_pointer(&test_rcu_pointer, new);
 295 #endif
 296                 if (caa_unlikely(wduration))
 297                         loop_sleep(wduration);
 298                 rcu_gc_reclaim(wtidx, old);
 299                 URCU_TLS(nr_writes)++;
 300                 if (caa_unlikely(!test_duration_write()))
 301                         break;
 302                 if (caa_unlikely(wdelay))
 303                         loop_sleep(wdelay);
 304         }
 305
 306         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 307                         "writer", pthread_self(), (unsigned long)gettid());
 308         tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 309         return ((void*)2);
 310 }
 311
 312 void show_usage(int argc, char **argv)
 313 {
 314         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 315 #ifdef DEBUG_YIELD
 316         printf(" [-r] [-w] (yield reader and/or writer)");
 317 #endif
 318         printf(" [-d delay] (writer period (us))");
 319         printf(" [-c duration] (reader C.S. duration (in loops))");
 320         printf(" [-e duration] (writer C.S. duration (in loops))");
 321         printf(" [-v] (verbose output)");
 322         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 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", pthread_self(), (unsigned long)gettid());
 420
 421         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 422         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 423         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 424         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 425         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 426         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 427                         < CAA_CACHE_LINE_SIZE)
 428                 for (i = 0; i < nr_writers; i++)
 429                         pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
 430         else
 431                 for (i = 0; i < nr_writers; i++)
 432                         pending_reclaims[i].queue = calloc(reclaim_batch,
 433                                         sizeof(*pending_reclaims[i].queue));
 434         for (i = 0; i < nr_writers; i++)
 435                 pending_reclaims[i].head = pending_reclaims[i].queue;
 436
 437         next_aff = 0;
 438
 439         for (i = 0; i < nr_readers; i++) {
 440                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 441                                      &count_reader[i]);
 442                 if (err != 0)
 443                         exit(1);
 444         }
 445         for (i = 0; i < nr_writers; i++) {
 446                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 447                                      (void *)(long)i);
 448                 if (err != 0)
 449                         exit(1);
 450         }
 451
 452         cmm_smp_mb();
 453
 454         test_go = 1;
 455
 456         sleep(duration);
 457
 458         test_stop = 1;
 459
 460         for (i = 0; i < nr_readers; i++) {
 461                 err = pthread_join(tid_reader[i], &tret);
 462                 if (err != 0)
 463                         exit(1);
 464                 tot_reads += count_reader[i];
 465         }
 466         for (i = 0; i < nr_writers; i++) {
 467                 err = pthread_join(tid_writer[i], &tret);
 468                 if (err != 0)
 469                         exit(1);
 470                 tot_writes += tot_nr_writes[i];
 471                 rcu_gc_clear_queue(i);
 472         }
 473
 474         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 475                tot_writes);
 476         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
 477                 "nr_writers %3u "
 478                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 479                 "batch %u\n",
 480                 argv[0], duration, nr_readers, rduration, wduration,
 481                 nr_writers, wdelay, tot_reads, tot_writes,
 482                 tot_reads + tot_writes, reclaim_batch);
 483         free(tid_reader);
 484         free(tid_writer);
 485         free(count_reader);
 486         free(tot_nr_writes);
 487         for (i = 0; i < nr_writers; i++)
 488                 free(pending_reclaims[i].queue);
 489         free(pending_reclaims);
 490
 491         return 0;
 492 }