tests/test_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 <sched.h>
  35 #include <errno.h>
  36
  37 #include <urcu/arch.h>
  38
  39 #ifdef __linux__
  40 #include <syscall.h>
  41 #endif
  42
  43 /* hardcoded number of CPUs */
  44 #define NR_CPUS 16384
  45
  46 #if defined(_syscall0)
  47 _syscall0(pid_t, gettid)
  48 #elif defined(__NR_gettid)
  49 static inline pid_t gettid(void)
  50 {
  51         return syscall(__NR_gettid);
  52 }
  53 #else
  54 #warning "use pid as tid"
  55 static inline pid_t gettid(void)
  56 {
  57         return getpid();
  58 }
  59 #endif
  60
  61 #define _LGPL_SOURCE
  62 #include <urcu-qsbr.h>
  63
  64 struct test_array {
  65         int a;
  66 };
  67
  68 static volatile int test_go, test_stop;
  69
  70 static unsigned long wdelay;
  71
  72 static struct test_array *test_rcu_pointer;
  73
  74 static unsigned long duration;
  75
  76 /* read-side C.S. duration, in loops */
  77 static unsigned long rduration;
  78 static unsigned int reclaim_batch = 1;
  79
  80 struct reclaim_queue {
  81         void **queue;   /* Beginning of queue */
  82         void **head;    /* Insert position */
  83 };
  84
  85 static struct reclaim_queue *pending_reclaims;
  86
  87
  88 /* write-side C.S. duration, in loops */
  89 static unsigned long wduration;
  90
  91 static inline void loop_sleep(unsigned long l)
  92 {
  93         while(l-- != 0)
  94                 caa_cpu_relax();
  95 }
  96
  97 static int verbose_mode;
  98
  99 #define printf_verbose(fmt, args...)            \
 100         do {                                    \
 101                 if (verbose_mode)               \
 102                         printf(fmt, args);      \
 103         } while (0)
 104
 105 static unsigned int cpu_affinities[NR_CPUS];
 106 static unsigned int next_aff = 0;
 107 static int use_affinity = 0;
 108
 109 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 110
 111 #ifndef HAVE_CPU_SET_T
 112 typedef unsigned long cpu_set_t;
 113 # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0)
 114 # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0)
 115 #endif
 116
 117 static void set_affinity(void)
 118 {
 119         cpu_set_t mask;
 120         int cpu;
 121         int ret;
 122
 123         if (!use_affinity)
 124                 return;
 125
 126 #if HAVE_SCHED_SETAFFINITY
 127         ret = pthread_mutex_lock(&affinity_mutex);
 128         if (ret) {
 129                 perror("Error in pthread mutex lock");
 130                 exit(-1);
 131         }
 132         cpu = cpu_affinities[next_aff++];
 133         ret = pthread_mutex_unlock(&affinity_mutex);
 134         if (ret) {
 135                 perror("Error in pthread mutex unlock");
 136                 exit(-1);
 137         }
 138
 139         CPU_ZERO(&mask);
 140         CPU_SET(cpu, &mask);
 141 #if SCHED_SETAFFINITY_ARGS == 2
 142         sched_setaffinity(0, &mask);
 143 #else
 144         sched_setaffinity(0, sizeof(mask), &mask);
 145 #endif
 146 #endif /* HAVE_SCHED_SETAFFINITY */
 147 }
 148
 149 /*
 150  * returns 0 if test should end.
 151  */
 152 static int test_duration_write(void)
 153 {
 154         return !test_stop;
 155 }
 156
 157 static int test_duration_read(void)
 158 {
 159         return !test_stop;
 160 }
 161
 162 static unsigned long long __thread nr_writes;
 163 static unsigned long long __thread nr_reads;
 164
 165 static unsigned int nr_readers;
 166 static unsigned int nr_writers;
 167
 168 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 169 static
 170 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
 171
 172
 173 void rcu_copy_mutex_lock(void)
 174 {
 175         int ret;
 176         ret = pthread_mutex_lock(&rcu_copy_mutex);
 177         if (ret) {
 178                 perror("Error in pthread mutex lock");
 179                 exit(-1);
 180         }
 181 }
 182
 183 void rcu_copy_mutex_unlock(void)
 184 {
 185         int ret;
 186
 187         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 188         if (ret) {
 189                 perror("Error in pthread mutex unlock");
 190                 exit(-1);
 191         }
 192 }
 193
 194 void *thr_reader(void *_count)
 195 {
 196         unsigned long long *count = _count;
 197         struct test_array *local_ptr;
 198
 199         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 200                         "reader", pthread_self(), (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 (unlikely(rduration))
 218                         loop_sleep(rduration);
 219                 _rcu_read_unlock();
 220                 nr_reads++;
 221                 /* QS each 1024 reads */
 222                 if (unlikely((nr_reads & ((1 << 10) - 1)) == 0))
 223                         _rcu_quiescent_state();
 224                 if (unlikely(!test_duration_read()))
 225                         break;
 226         }
 227
 228         rcu_unregister_thread();
 229
 230         *count = nr_reads;
 231         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 232                         "reader", pthread_self(), (unsigned long)gettid());
 233         return ((void*)1);
 234
 235 }
 236
 237 static void rcu_gc_clear_queue(unsigned long wtidx)
 238 {
 239         void **p;
 240
 241         /* Wait for Q.S and empty queue */
 242         synchronize_rcu();
 243
 244         for (p = pending_reclaims[wtidx].queue;
 245                         p < pending_reclaims[wtidx].head; p++) {
 246                 /* poison */
 247                 if (*p)
 248                         ((struct test_array *)*p)->a = 0;
 249                 free(*p);
 250         }
 251         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 252 }
 253
 254 /* Using per-thread queue */
 255 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 256 {
 257         /* Queue pointer */
 258         *pending_reclaims[wtidx].head = old;
 259         pending_reclaims[wtidx].head++;
 260
 261         if (likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 262                         < reclaim_batch))
 263                 return;
 264
 265         rcu_gc_clear_queue(wtidx);
 266 }
 267
 268 void *thr_writer(void *data)
 269 {
 270         unsigned long wtidx = (unsigned long)data;
 271 #ifdef TEST_LOCAL_GC
 272         struct test_array *old = NULL;
 273 #else
 274         struct test_array *new, *old;
 275 #endif
 276
 277         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 278                         "writer", pthread_self(), (unsigned long)gettid());
 279
 280         set_affinity();
 281
 282         while (!test_go)
 283         {
 284         }
 285         cmm_smp_mb();
 286
 287         for (;;) {
 288 #ifndef TEST_LOCAL_GC
 289                 new = malloc(sizeof(*new));
 290                 new->a = 8;
 291                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 292 #endif
 293                 if (unlikely(wduration))
 294                         loop_sleep(wduration);
 295                 rcu_gc_reclaim(wtidx, old);
 296                 nr_writes++;
 297                 if (unlikely(!test_duration_write()))
 298                         break;
 299                 if (unlikely(wdelay))
 300                         loop_sleep(wdelay);
 301         }
 302
 303         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 304                         "writer", pthread_self(), (unsigned long)gettid());
 305         tot_nr_writes[wtidx] = 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(" [-b batch] (batch reclaim)");
 316         printf(" [-d delay] (writer period (us))");
 317         printf(" [-c duration] (reader C.S. duration (in loops))");
 318         printf(" [-e duration] (writer C.S. duration (in loops))");
 319         printf(" [-v] (verbose output)");
 320         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 321         printf("\n");
 322 }
 323
 324 int main(int argc, char **argv)
 325 {
 326         int err;
 327         pthread_t *tid_reader, *tid_writer;
 328         void *tret;
 329         unsigned long long *count_reader;
 330         unsigned long long tot_reads = 0, tot_writes = 0;
 331         int i, a;
 332
 333         if (argc < 4) {
 334                 show_usage(argc, argv);
 335                 return -1;
 336         }
 337
 338         err = sscanf(argv[1], "%u", &nr_readers);
 339         if (err != 1) {
 340                 show_usage(argc, argv);
 341                 return -1;
 342         }
 343
 344         err = sscanf(argv[2], "%u", &nr_writers);
 345         if (err != 1) {
 346                 show_usage(argc, argv);
 347                 return -1;
 348         }
 349
 350         err = sscanf(argv[3], "%lu", &duration);
 351         if (err != 1) {
 352                 show_usage(argc, argv);
 353                 return -1;
 354         }
 355
 356         for (i = 4; i < argc; i++) {
 357                 if (argv[i][0] != '-')
 358                         continue;
 359                 switch (argv[i][1]) {
 360 #ifdef DEBUG_YIELD
 361                 case 'r':
 362                         yield_active |= YIELD_READ;
 363                         break;
 364                 case 'w':
 365                         yield_active |= YIELD_WRITE;
 366                         break;
 367 #endif
 368                 case 'a':
 369                         if (argc < i + 2) {
 370                                 show_usage(argc, argv);
 371                                 return -1;
 372                         }
 373                         a = atoi(argv[++i]);
 374                         cpu_affinities[next_aff++] = a;
 375                         use_affinity = 1;
 376                         printf_verbose("Adding CPU %d affinity\n", a);
 377                         break;
 378                 case 'b':
 379                         if (argc < i + 2) {
 380                                 show_usage(argc, argv);
 381                                 return -1;
 382                         }
 383                         reclaim_batch = atol(argv[++i]);
 384                         break;
 385                 case 'c':
 386                         if (argc < i + 2) {
 387                                 show_usage(argc, argv);
 388                                 return -1;
 389                         }
 390                         rduration = atol(argv[++i]);
 391                         break;
 392                 case 'd':
 393                         if (argc < i + 2) {
 394                                 show_usage(argc, argv);
 395                                 return -1;
 396                         }
 397                         wdelay = atol(argv[++i]);
 398                         break;
 399                 case 'e':
 400                         if (argc < i + 2) {
 401                                 show_usage(argc, argv);
 402                                 return -1;
 403                         }
 404                         wduration = atol(argv[++i]);
 405                         break;
 406                 case 'v':
 407                         verbose_mode = 1;
 408                         break;
 409                 }
 410         }
 411
 412         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 413                 duration, nr_readers, nr_writers);
 414         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 415         printf_verbose("Reader duration : %lu loops.\n", rduration);
 416         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 417                         "main", pthread_self(), (unsigned long)gettid());
 418
 419         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 420         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 421         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 422         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 423         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 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 }