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@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 #define _LGPL_SOURCE
  60 #include "urcu-qsbr.h"
  61
  62 struct test_array {
  63         int a;
  64 };
  65
  66 static volatile int test_go, test_stop;
  67
  68 static unsigned long wdelay;
  69
  70 static struct test_array *test_rcu_pointer;
  71
  72 static unsigned long duration;
  73
  74 /* read-side C.S. duration, in loops */
  75 static unsigned long rduration;
  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
  86 static inline void loop_sleep(unsigned long l)
  87 {
  88         while(l-- != 0)
  89                 cpu_relax();
  90 }
  91
  92 static int verbose_mode;
  93
  94 #define printf_verbose(fmt, args...)            \
  95         do {                                    \
  96                 if (verbose_mode)               \
  97                         printf(fmt, args);      \
  98         } while (0)
  99
 100 static unsigned int cpu_affinities[NR_CPUS];
 101 static unsigned int next_aff = 0;
 102 static int use_affinity = 0;
 103
 104 static void set_affinity(void)
 105 {
 106         cpu_set_t mask;
 107         int cpu;
 108
 109         if (!use_affinity)
 110                 return;
 111
 112         cpu = cpu_affinities[next_aff++];
 113         CPU_ZERO(&mask);
 114         CPU_SET(cpu, &mask);
 115         sched_setaffinity(0, sizeof(mask), &mask);
 116 }
 117
 118
 119
 120 /*
 121  * returns 0 if test should end.
 122  */
 123 static int test_duration_write(void)
 124 {
 125         return !test_stop;
 126 }
 127
 128 static int test_duration_read(void)
 129 {
 130         return !test_stop;
 131 }
 132
 133 static unsigned long long __thread nr_writes;
 134 static unsigned long long __thread nr_reads;
 135
 136 static unsigned int nr_readers;
 137 static unsigned int nr_writers;
 138
 139 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 140 static
 141 unsigned long long __attribute__((aligned(CACHE_LINE_SIZE))) *tot_nr_writes;
 142
 143
 144 void rcu_copy_mutex_lock(void)
 145 {
 146         int ret;
 147         ret = pthread_mutex_lock(&rcu_copy_mutex);
 148         if (ret) {
 149                 perror("Error in pthread mutex lock");
 150                 exit(-1);
 151         }
 152 }
 153
 154 void rcu_copy_mutex_unlock(void)
 155 {
 156         int ret;
 157
 158         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 159         if (ret) {
 160                 perror("Error in pthread mutex unlock");
 161                 exit(-1);
 162         }
 163 }
 164
 165 void *thr_reader(void *_count)
 166 {
 167         unsigned long long *count = _count;
 168         struct test_array *local_ptr;
 169
 170         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 171                         "reader", pthread_self(), (unsigned long)gettid());
 172
 173         set_affinity();
 174
 175         rcu_register_thread();
 176
 177         while (!test_go)
 178         {
 179         }
 180         smp_mb();
 181
 182         for (;;) {
 183                 _rcu_read_lock();
 184                 local_ptr = _rcu_dereference(test_rcu_pointer);
 185                 debug_yield_read();
 186                 if (local_ptr)
 187                         assert(local_ptr->a == 8);
 188                 if (unlikely(rduration))
 189                         loop_sleep(rduration);
 190                 _rcu_read_unlock();
 191                 nr_reads++;
 192                 /* QS each 1024 reads */
 193                 if (unlikely((nr_reads & ((1 << 10) - 1)) == 0))
 194                         _rcu_quiescent_state();
 195                 if (unlikely(!test_duration_read()))
 196                         break;
 197         }
 198
 199         rcu_unregister_thread();
 200
 201         *count = nr_reads;
 202         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 203                         "reader", pthread_self(), (unsigned long)gettid());
 204         return ((void*)1);
 205
 206 }
 207
 208 static void rcu_gc_clear_queue(unsigned long wtidx)
 209 {
 210         void **p;
 211
 212         /* Wait for Q.S and empty queue */
 213         synchronize_rcu();
 214
 215         for (p = pending_reclaims[wtidx].queue;
 216                         p < pending_reclaims[wtidx].head; p++) {
 217                 /* poison */
 218                 if (*p)
 219                         ((struct test_array *)*p)->a = 0;
 220                 free(*p);
 221         }
 222         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 223 }
 224
 225 /* Using per-thread queue */
 226 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 227 {
 228         /* Queue pointer */
 229         *pending_reclaims[wtidx].head = old;
 230         pending_reclaims[wtidx].head++;
 231
 232         if (likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 233                         < reclaim_batch))
 234                 return;
 235
 236         rcu_gc_clear_queue(wtidx);
 237 }
 238
 239 void *thr_writer(void *data)
 240 {
 241         unsigned long wtidx = (unsigned long)data;
 242         struct test_array *new, *old;
 243
 244         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 245                         "writer", pthread_self(), (unsigned long)gettid());
 246
 247         set_affinity();
 248
 249         while (!test_go)
 250         {
 251         }
 252         smp_mb();
 253
 254         for (;;) {
 255                 new = malloc(sizeof(*new));
 256                 rcu_copy_mutex_lock();
 257                 old = test_rcu_pointer;
 258                 if (old)
 259                         assert(old->a == 8);
 260                 new->a = 8;
 261                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 262                 rcu_copy_mutex_unlock();
 263                 rcu_gc_reclaim(wtidx, old);
 264                 nr_writes++;
 265                 if (unlikely(!test_duration_write()))
 266                         break;
 267                 if (unlikely(wdelay))
 268                         loop_sleep(wdelay);
 269         }
 270
 271         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 272                         "writer", pthread_self(), (unsigned long)gettid());
 273         tot_nr_writes[wtidx] = nr_writes;
 274         return ((void*)2);
 275 }
 276
 277 void show_usage(int argc, char **argv)
 278 {
 279         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 280 #ifdef DEBUG_YIELD
 281         printf(" [-r] [-w] (yield reader and/or writer)");
 282 #endif
 283         printf(" [-d delay] (writer period (us))");
 284         printf(" [-c duration] (reader C.S. duration (in loops))");
 285         printf(" [-v] (verbose output)");
 286         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 287         printf("\n");
 288 }
 289
 290 int main(int argc, char **argv)
 291 {
 292         int err;
 293         pthread_t *tid_reader, *tid_writer;
 294         void *tret;
 295         unsigned long long *count_reader;
 296         unsigned long long tot_reads = 0, tot_writes = 0;
 297         int i, a;
 298
 299         if (argc < 4) {
 300                 show_usage(argc, argv);
 301                 return -1;
 302         }
 303
 304         err = sscanf(argv[1], "%u", &nr_readers);
 305         if (err != 1) {
 306                 show_usage(argc, argv);
 307                 return -1;
 308         }
 309
 310         err = sscanf(argv[2], "%u", &nr_writers);
 311         if (err != 1) {
 312                 show_usage(argc, argv);
 313                 return -1;
 314         }
 315
 316         err = sscanf(argv[3], "%lu", &duration);
 317         if (err != 1) {
 318                 show_usage(argc, argv);
 319                 return -1;
 320         }
 321
 322         for (i = 4; i < argc; i++) {
 323                 if (argv[i][0] != '-')
 324                         continue;
 325                 switch (argv[i][1]) {
 326 #ifdef DEBUG_YIELD
 327                 case 'r':
 328                         yield_active |= YIELD_READ;
 329                         break;
 330                 case 'w':
 331                         yield_active |= YIELD_WRITE;
 332                         break;
 333 #endif
 334                 case 'a':
 335                         if (argc < i + 2) {
 336                                 show_usage(argc, argv);
 337                                 return -1;
 338                         }
 339                         a = atoi(argv[++i]);
 340                         cpu_affinities[next_aff++] = a;
 341                         use_affinity = 1;
 342                         printf_verbose("Adding CPU %d affinity\n", a);
 343                         break;
 344                 case 'b':
 345                         if (argc < i + 2) {
 346                                 show_usage(argc, argv);
 347                                 return -1;
 348                         }
 349                         reclaim_batch = atol(argv[++i]);
 350                         break;
 351                 case 'c':
 352                         if (argc < i + 2) {
 353                                 show_usage(argc, argv);
 354                                 return -1;
 355                         }
 356                         rduration = atol(argv[++i]);
 357                         break;
 358                 case 'd':
 359                         if (argc < i + 2) {
 360                                 show_usage(argc, argv);
 361                                 return -1;
 362                         }
 363                         wdelay = atol(argv[++i]);
 364                         break;
 365                 case 'v':
 366                         verbose_mode = 1;
 367                         break;
 368                 }
 369         }
 370
 371         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 372                 duration, nr_readers, nr_writers);
 373         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 374         printf_verbose("Reader duration : %lu loops.\n", rduration);
 375         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 376                         "main", pthread_self(), (unsigned long)gettid());
 377
 378         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 379         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 380         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 381         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 382         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 383         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 384                         < CACHE_LINE_SIZE)
 385                 for (i = 0; i < nr_writers; i++)
 386                         pending_reclaims[i].queue = calloc(1, CACHE_LINE_SIZE);
 387         else
 388                 for (i = 0; i < nr_writers; i++)
 389                         pending_reclaims[i].queue = calloc(reclaim_batch,
 390                                         sizeof(*pending_reclaims[i].queue));
 391         for (i = 0; i < nr_writers; i++)
 392                 pending_reclaims[i].head = pending_reclaims[i].queue;
 393
 394         next_aff = 0;
 395
 396         for (i = 0; i < nr_readers; i++) {
 397                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 398                                      &count_reader[i]);
 399                 if (err != 0)
 400                         exit(1);
 401         }
 402         for (i = 0; i < nr_writers; i++) {
 403                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 404                                      (void *)(long)i);
 405                 if (err != 0)
 406                         exit(1);
 407         }
 408
 409         smp_mb();
 410
 411         test_go = 1;
 412
 413         sleep(duration);
 414
 415         test_stop = 1;
 416
 417         for (i = 0; i < nr_readers; i++) {
 418                 err = pthread_join(tid_reader[i], &tret);
 419                 if (err != 0)
 420                         exit(1);
 421                 tot_reads += count_reader[i];
 422         }
 423         for (i = 0; i < nr_writers; i++) {
 424                 err = pthread_join(tid_writer[i], &tret);
 425                 if (err != 0)
 426                         exit(1);
 427                 tot_writes += tot_nr_writes[i];
 428                 rcu_gc_clear_queue(i);
 429         }
 430
 431         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 432                tot_writes);
 433         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu "
 434                 "nr_writers %3u "
 435                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 436                 "batch %u\n",
 437                 argv[0], duration, nr_readers, rduration,
 438                 nr_writers, wdelay, tot_reads, tot_writes,
 439                 tot_reads + tot_writes, reclaim_batch);
 440         free(tid_reader);
 441         free(tid_writer);
 442         free(count_reader);
 443         free(tot_nr_writes);
 444         for (i = 0; i < nr_writers; i++)
 445                 free(pending_reclaims[i].queue);
 446         free(pending_reclaims);
 447
 448         return 0;
 449 }