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