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@polymtl.ca>
   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 <stdio.h>
  25 #include <pthread.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/types.h>
  29 #include <sys/wait.h>
  30 #include <unistd.h>
  31 #include <stdio.h>
  32 #include <assert.h>
  33 #include <sys/syscall.h>
  34 #include <sched.h>
  35
  36 #include "arch.h"
  37
  38 /* Make this big enough to include the POWER5+ L3 cacheline size of 256B */
  39 #define CACHE_LINE_SIZE 4096
  40
  41 #if defined(_syscall0)
  42 _syscall0(pid_t, gettid)
  43 #elif defined(__NR_gettid)
  44 static inline pid_t gettid(void)
  45 {
  46         return syscall(__NR_gettid);
  47 }
  48 #else
  49 #warning "use pid as tid"
  50 static inline pid_t gettid(void)
  51 {
  52         return getpid();
  53 }
  54 #endif
  55
  56 #ifndef DYNAMIC_LINK_TEST
  57 #define _LGPL_SOURCE
  58 #else
  59 #define debug_yield_read()
  60 #endif
  61 #include "urcu.h"
  62
  63 struct test_array {
  64         int a;
  65 };
  66
  67 static volatile int test_go, test_stop;
  68
  69 static unsigned long wdelay;
  70
  71 static struct test_array *test_rcu_pointer;
  72
  73 static int reclaim_batch = 1;
  74
  75 struct reclaim_queue {
  76         void **queue;   /* Beginning of queue */
  77         void **head;    /* Insert position */
  78 };
  79
  80 static struct reclaim_queue *pending_reclaims;
  81
  82 static unsigned long duration;
  83
  84 /* read-side C.S. duration, in loops */
  85 static unsigned long rduration;
  86
  87 static inline void loop_sleep(unsigned long l)
  88 {
  89         while(l-- != 0)
  90                 cpu_relax();
  91 }
  92
  93 static int verbose_mode;
  94
  95 #define printf_verbose(fmt, args...)            \
  96         do {                                    \
  97                 if (verbose_mode)               \
  98                         printf(fmt, args);      \
  99         } while (0)
 100
 101 /*
 102  * returns 0 if test should end.
 103  */
 104 static int test_duration_write(void)
 105 {
 106         return !test_stop;
 107 }
 108
 109 static int test_duration_read(void)
 110 {
 111         return !test_stop;
 112 }
 113
 114 static unsigned long long __thread nr_writes;
 115 static unsigned long long __thread nr_reads;
 116
 117 static
 118 unsigned long long __attribute__((aligned(CACHE_LINE_SIZE))) *tot_nr_writes;
 119
 120 static unsigned int nr_readers;
 121 static unsigned int nr_writers;
 122
 123 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 124
 125 void rcu_copy_mutex_lock(void)
 126 {
 127         int ret;
 128         ret = pthread_mutex_lock(&rcu_copy_mutex);
 129         if (ret) {
 130                 perror("Error in pthread mutex lock");
 131                 exit(-1);
 132         }
 133 }
 134
 135 void rcu_copy_mutex_unlock(void)
 136 {
 137         int ret;
 138
 139         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 140         if (ret) {
 141                 perror("Error in pthread mutex unlock");
 142                 exit(-1);
 143         }
 144 }
 145
 146 void *thr_reader(void *_count)
 147 {
 148         unsigned long long *count = _count;
 149         struct test_array *local_ptr;
 150
 151         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 152                         "reader", pthread_self(), (unsigned long)gettid());
 153
 154         rcu_register_thread();
 155
 156         while (!test_go)
 157         {
 158         }
 159         smp_mb();
 160
 161         for (;;) {
 162                 rcu_read_lock();
 163                 local_ptr = rcu_dereference(test_rcu_pointer);
 164                 debug_yield_read();
 165                 if (local_ptr)
 166                         assert(local_ptr->a == 8);
 167                 if (unlikely(rduration))
 168                         loop_sleep(rduration);
 169                 rcu_read_unlock();
 170                 nr_reads++;
 171                 if (unlikely(!test_duration_read()))
 172                         break;
 173         }
 174
 175         rcu_unregister_thread();
 176
 177         *count = nr_reads;
 178         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 179                         "reader", pthread_self(), (unsigned long)gettid());
 180         return ((void*)1);
 181
 182 }
 183
 184 /* Using per-thread queue */
 185 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 186 {
 187         void **p;
 188
 189         /* Queue pointer */
 190         *pending_reclaims[wtidx].head = old;
 191         pending_reclaims[wtidx].head++;
 192
 193         if (likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 194                         < reclaim_batch))
 195                 return;
 196
 197         /* If queue is full, wait for Q.S and empty queue */
 198         synchronize_rcu();
 199
 200         for (p = pending_reclaims[wtidx].queue;
 201                         p < pending_reclaims[wtidx].head; p++) {
 202                 /* poison */
 203                 if (*p)
 204                         ((struct test_array *)*p)->a = 0;
 205                 free(*p);
 206         }
 207         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 208 }
 209
 210 static void rcu_gc_cleanup(unsigned long wtidx)
 211 {
 212         void **p;
 213
 214         /* Wait for Q.S and empty queue */
 215         synchronize_rcu();
 216
 217         for (p = pending_reclaims[wtidx].queue;
 218                         p < pending_reclaims[wtidx].head; p++) {
 219                 /* poison */
 220                 if (*p)
 221                         ((struct test_array *)*p)->a = 0;
 222                 free(*p);
 223         }
 224         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 225 }
 226
 227 void *thr_writer(void *data)
 228 {
 229         unsigned long wtidx = (unsigned long)data;
 230         struct test_array *new, *old;
 231
 232         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 233                         "writer", pthread_self(), (unsigned long)gettid());
 234
 235         while (!test_go)
 236         {
 237         }
 238         smp_mb();
 239
 240         for (;;) {
 241                 new = malloc(sizeof(*new));
 242                 rcu_copy_mutex_lock();
 243                 old = test_rcu_pointer;
 244                 if (old)
 245                         assert(old->a == 8);
 246                 new->a = 8;
 247                 old = rcu_xchg_pointer(&test_rcu_pointer, new);
 248                 rcu_copy_mutex_unlock();
 249                 rcu_gc_reclaim(wtidx, old);
 250                 nr_writes++;
 251                 if (unlikely(!test_duration_write()))
 252                         break;
 253                 if (unlikely(wdelay))
 254                         loop_sleep(wdelay);
 255         }
 256
 257         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 258                         "writer", pthread_self(), (unsigned long)gettid());
 259         tot_nr_writes[wtidx] = nr_writes;
 260         return ((void*)2);
 261 }
 262
 263 void show_usage(int argc, char **argv)
 264 {
 265         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 266 #ifdef DEBUG_YIELD
 267         printf(" [-r] [-w] (yield reader and/or writer)");
 268 #endif
 269         printf(" [-d delay] (writer period (us))");
 270         printf(" [-c duration] (reader C.S. duration (in loops))");
 271         printf(" [-v] (verbose output)");
 272         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 273         printf("\n");
 274 }
 275
 276 cpu_set_t affinity;
 277
 278 int main(int argc, char **argv)
 279 {
 280         int err;
 281         pthread_t *tid_reader, *tid_writer;
 282         void *tret;
 283         unsigned long long *count_reader;
 284         unsigned long long tot_reads = 0, tot_writes = 0;
 285         int i, a;
 286         int use_affinity = 0;
 287
 288         if (argc < 4) {
 289                 show_usage(argc, argv);
 290                 return -1;
 291         }
 292
 293         err = sscanf(argv[1], "%u", &nr_readers);
 294         if (err != 1) {
 295                 show_usage(argc, argv);
 296                 return -1;
 297         }
 298
 299         err = sscanf(argv[2], "%u", &nr_writers);
 300         if (err != 1) {
 301                 show_usage(argc, argv);
 302                 return -1;
 303         }
 304
 305         err = sscanf(argv[3], "%lu", &duration);
 306         if (err != 1) {
 307                 show_usage(argc, argv);
 308                 return -1;
 309         }
 310
 311         CPU_ZERO(&affinity);
 312
 313         for (i = 4; i < argc; i++) {
 314                 if (argv[i][0] != '-')
 315                         continue;
 316                 switch (argv[i][1]) {
 317 #ifdef DEBUG_YIELD
 318                 case 'r':
 319                         yield_active |= YIELD_READ;
 320                         break;
 321                 case 'w':
 322                         yield_active |= YIELD_WRITE;
 323                         break;
 324 #endif
 325                 case 'a':
 326                         if (argc < i + 2) {
 327                                 show_usage(argc, argv);
 328                                 return -1;
 329                         }
 330                         a = atoi(argv[++i]);
 331                         CPU_SET(a, &affinity);
 332                         use_affinity = 1;
 333                         printf_verbose("Adding CPU %d affinity\n", a);
 334                         break;
 335                 case 'b':
 336                         if (argc < i + 2) {
 337                                 show_usage(argc, argv);
 338                                 return -1;
 339                         }
 340                         reclaim_batch = atol(argv[++i]);
 341                         break;
 342                 case 'c':
 343                         if (argc < i + 2) {
 344                                 show_usage(argc, argv);
 345                                 return -1;
 346                         }
 347                         rduration = atol(argv[++i]);
 348                         break;
 349                 case 'd':
 350                         if (argc < i + 2) {
 351                                 show_usage(argc, argv);
 352                                 return -1;
 353                         }
 354                         wdelay = atol(argv[++i]);
 355                         break;
 356                 case 'v':
 357                         verbose_mode = 1;
 358                         break;
 359                 }
 360         }
 361
 362         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 363                 duration, nr_readers, nr_writers);
 364         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 365         printf_verbose("Reader duration : %lu loops.\n", rduration);
 366         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 367                         "main", pthread_self(), (unsigned long)gettid());
 368
 369         if (use_affinity
 370             && sched_setaffinity(0, sizeof(affinity), &affinity) < 0) {
 371                 perror("sched_setaffinity");
 372                 exit(-1);
 373         }
 374
 375         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 376         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 377         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 378         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 379         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 380         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 381                         < CACHE_LINE_SIZE)
 382                 for (i = 0; i < nr_writers; i++)
 383                         pending_reclaims[i].queue = calloc(1, CACHE_LINE_SIZE);
 384         else
 385                 for (i = 0; i < nr_writers; i++)
 386                         pending_reclaims[i].queue = calloc(reclaim_batch,
 387                                         sizeof(*pending_reclaims[i].queue));
 388         for (i = 0; i < nr_writers; i++)
 389                 pending_reclaims[i].head = pending_reclaims[i].queue;
 390
 391         for (i = 0; i < nr_readers; i++) {
 392                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 393                                      &count_reader[i]);
 394                 if (err != 0)
 395                         exit(1);
 396         }
 397         for (i = 0; i < nr_writers; i++) {
 398                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 399                                      (void *)(long)i);
 400                 if (err != 0)
 401                         exit(1);
 402         }
 403
 404         smp_mb();
 405
 406         test_go = 1;
 407
 408         sleep(duration);
 409
 410         test_stop = 1;
 411
 412         for (i = 0; i < nr_readers; i++) {
 413                 err = pthread_join(tid_reader[i], &tret);
 414                 if (err != 0)
 415                         exit(1);
 416                 tot_reads += count_reader[i];
 417         }
 418         for (i = 0; i < nr_writers; i++) {
 419                 err = pthread_join(tid_writer[i], &tret);
 420                 if (err != 0)
 421                         exit(1);
 422                 tot_writes += tot_nr_writes[i];
 423                 rcu_gc_cleanup(i);
 424         }
 425
 426         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 427                tot_writes);
 428         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu "
 429                 "nr_writers %3u "
 430                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu\n",
 431                 argv[0], duration, nr_readers, rduration,
 432                 nr_writers, wdelay, tot_reads, tot_writes,
 433                 tot_reads + tot_writes);
 434         free(tid_reader);
 435         free(tid_writer);
 436         free(count_reader);
 437         free(tot_nr_writes);
 438         for (i = 0; i < nr_writers; i++)
 439                 free(pending_reclaims[i].queue);
 440         free(pending_reclaims);
 441
 442         return 0;
 443 }