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 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 105
 106 static void set_affinity(void)
 107 {
 108         cpu_set_t mask;
 109         int cpu;
 110         int ret;
 111
 112         if (!use_affinity)
 113                 return;
 114
 115         ret = pthread_mutex_lock(&affinity_mutex);
 116         if (ret) {
 117                 perror("Error in pthread mutex lock");
 118                 exit(-1);
 119         }
 120         cpu = cpu_affinities[next_aff++];
 121         ret = pthread_mutex_unlock(&affinity_mutex);
 122         if (ret) {
 123                 perror("Error in pthread mutex unlock");
 124                 exit(-1);
 125         }
 126         CPU_ZERO(&mask);
 127         CPU_SET(cpu, &mask);
 128         sched_setaffinity(0, sizeof(mask), &mask);
 129 }
 130
 131 /*
 132  * returns 0 if test should end.
 133  */
 134 static int test_duration_write(void)
 135 {
 136         return !test_stop;
 137 }
 138
 139 static int test_duration_read(void)
 140 {
 141         return !test_stop;
 142 }
 143
 144 static unsigned long long __thread nr_writes;
 145 static unsigned long long __thread nr_reads;
 146
 147 static unsigned int nr_readers;
 148 static unsigned int nr_writers;
 149
 150 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
 151 static
 152 unsigned long long __attribute__((aligned(CACHE_LINE_SIZE))) *tot_nr_writes;
 153
 154
 155 void rcu_copy_mutex_lock(void)
 156 {
 157         int ret;
 158         ret = pthread_mutex_lock(&rcu_copy_mutex);
 159         if (ret) {
 160                 perror("Error in pthread mutex lock");
 161                 exit(-1);
 162         }
 163 }
 164
 165 void rcu_copy_mutex_unlock(void)
 166 {
 167         int ret;
 168
 169         ret = pthread_mutex_unlock(&rcu_copy_mutex);
 170         if (ret) {
 171                 perror("Error in pthread mutex unlock");
 172                 exit(-1);
 173         }
 174 }
 175
 176 void *thr_reader(void *_count)
 177 {
 178         unsigned long long *count = _count;
 179         struct test_array *local_ptr;
 180
 181         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 182                         "reader", pthread_self(), (unsigned long)gettid());
 183
 184         set_affinity();
 185
 186         rcu_register_thread();
 187
 188         while (!test_go)
 189         {
 190         }
 191         smp_mb();
 192
 193         for (;;) {
 194                 _rcu_read_lock();
 195                 local_ptr = _rcu_dereference(test_rcu_pointer);
 196                 debug_yield_read();
 197                 if (local_ptr)
 198                         assert(local_ptr->a == 8);
 199                 if (unlikely(rduration))
 200                         loop_sleep(rduration);
 201                 _rcu_read_unlock();
 202                 nr_reads++;
 203                 /* QS each 1024 reads */
 204                 if (unlikely((nr_reads & ((1 << 10) - 1)) == 0))
 205                         _rcu_quiescent_state();
 206                 if (unlikely(!test_duration_read()))
 207                         break;
 208         }
 209
 210         rcu_unregister_thread();
 211
 212         *count = nr_reads;
 213         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 214                         "reader", pthread_self(), (unsigned long)gettid());
 215         return ((void*)1);
 216
 217 }
 218
 219 static void rcu_gc_clear_queue(unsigned long wtidx)
 220 {
 221         void **p;
 222
 223         /* Wait for Q.S and empty queue */
 224         synchronize_rcu();
 225
 226         for (p = pending_reclaims[wtidx].queue;
 227                         p < pending_reclaims[wtidx].head; p++) {
 228                 /* poison */
 229                 if (*p)
 230                         ((struct test_array *)*p)->a = 0;
 231                 free(*p);
 232         }
 233         pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
 234 }
 235
 236 /* Using per-thread queue */
 237 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
 238 {
 239         /* Queue pointer */
 240         *pending_reclaims[wtidx].head = old;
 241         pending_reclaims[wtidx].head++;
 242
 243         if (likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
 244                         < reclaim_batch))
 245                 return;
 246
 247         rcu_gc_clear_queue(wtidx);
 248 }
 249
 250 void *thr_writer(void *data)
 251 {
 252         unsigned long wtidx = (unsigned long)data;
 253         struct test_array *new, *old;
 254
 255         printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
 256                         "writer", pthread_self(), (unsigned long)gettid());
 257
 258         set_affinity();
 259
 260         while (!test_go)
 261         {
 262         }
 263         smp_mb();
 264
 265         for (;;) {
 266                 new = malloc(sizeof(*new));
 267                 new->a = 8;
 268                 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
 269                 rcu_gc_reclaim(wtidx, old);
 270                 nr_writes++;
 271                 if (unlikely(!test_duration_write()))
 272                         break;
 273                 if (unlikely(wdelay))
 274                         loop_sleep(wdelay);
 275         }
 276
 277         printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 278                         "writer", pthread_self(), (unsigned long)gettid());
 279         tot_nr_writes[wtidx] = nr_writes;
 280         return ((void*)2);
 281 }
 282
 283 void show_usage(int argc, char **argv)
 284 {
 285         printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
 286 #ifdef DEBUG_YIELD
 287         printf(" [-r] [-w] (yield reader and/or writer)");
 288 #endif
 289         printf(" [-d delay] (writer period (us))");
 290         printf(" [-c duration] (reader C.S. duration (in loops))");
 291         printf(" [-v] (verbose output)");
 292         printf(" [-a cpu#] [-a cpu#]... (affinity)");
 293         printf("\n");
 294 }
 295
 296 int main(int argc, char **argv)
 297 {
 298         int err;
 299         pthread_t *tid_reader, *tid_writer;
 300         void *tret;
 301         unsigned long long *count_reader;
 302         unsigned long long tot_reads = 0, tot_writes = 0;
 303         int i, a;
 304
 305         if (argc < 4) {
 306                 show_usage(argc, argv);
 307                 return -1;
 308         }
 309
 310         err = sscanf(argv[1], "%u", &nr_readers);
 311         if (err != 1) {
 312                 show_usage(argc, argv);
 313                 return -1;
 314         }
 315
 316         err = sscanf(argv[2], "%u", &nr_writers);
 317         if (err != 1) {
 318                 show_usage(argc, argv);
 319                 return -1;
 320         }
 321
 322         err = sscanf(argv[3], "%lu", &duration);
 323         if (err != 1) {
 324                 show_usage(argc, argv);
 325                 return -1;
 326         }
 327
 328         for (i = 4; i < argc; i++) {
 329                 if (argv[i][0] != '-')
 330                         continue;
 331                 switch (argv[i][1]) {
 332 #ifdef DEBUG_YIELD
 333                 case 'r':
 334                         yield_active |= YIELD_READ;
 335                         break;
 336                 case 'w':
 337                         yield_active |= YIELD_WRITE;
 338                         break;
 339 #endif
 340                 case 'a':
 341                         if (argc < i + 2) {
 342                                 show_usage(argc, argv);
 343                                 return -1;
 344                         }
 345                         a = atoi(argv[++i]);
 346                         cpu_affinities[next_aff++] = a;
 347                         use_affinity = 1;
 348                         printf_verbose("Adding CPU %d affinity\n", a);
 349                         break;
 350                 case 'b':
 351                         if (argc < i + 2) {
 352                                 show_usage(argc, argv);
 353                                 return -1;
 354                         }
 355                         reclaim_batch = atol(argv[++i]);
 356                         break;
 357                 case 'c':
 358                         if (argc < i + 2) {
 359                                 show_usage(argc, argv);
 360                                 return -1;
 361                         }
 362                         rduration = atol(argv[++i]);
 363                         break;
 364                 case 'd':
 365                         if (argc < i + 2) {
 366                                 show_usage(argc, argv);
 367                                 return -1;
 368                         }
 369                         wdelay = atol(argv[++i]);
 370                         break;
 371                 case 'v':
 372                         verbose_mode = 1;
 373                         break;
 374                 }
 375         }
 376
 377         printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
 378                 duration, nr_readers, nr_writers);
 379         printf_verbose("Writer delay : %lu loops.\n", wdelay);
 380         printf_verbose("Reader duration : %lu loops.\n", rduration);
 381         printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
 382                         "main", pthread_self(), (unsigned long)gettid());
 383
 384         tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
 385         tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
 386         count_reader = malloc(sizeof(*count_reader) * nr_readers);
 387         tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
 388         pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
 389         if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
 390                         < CACHE_LINE_SIZE)
 391                 for (i = 0; i < nr_writers; i++)
 392                         pending_reclaims[i].queue = calloc(1, CACHE_LINE_SIZE);
 393         else
 394                 for (i = 0; i < nr_writers; i++)
 395                         pending_reclaims[i].queue = calloc(reclaim_batch,
 396                                         sizeof(*pending_reclaims[i].queue));
 397         for (i = 0; i < nr_writers; i++)
 398                 pending_reclaims[i].head = pending_reclaims[i].queue;
 399
 400         next_aff = 0;
 401
 402         for (i = 0; i < nr_readers; i++) {
 403                 err = pthread_create(&tid_reader[i], NULL, thr_reader,
 404                                      &count_reader[i]);
 405                 if (err != 0)
 406                         exit(1);
 407         }
 408         for (i = 0; i < nr_writers; i++) {
 409                 err = pthread_create(&tid_writer[i], NULL, thr_writer,
 410                                      (void *)(long)i);
 411                 if (err != 0)
 412                         exit(1);
 413         }
 414
 415         smp_mb();
 416
 417         test_go = 1;
 418
 419         sleep(duration);
 420
 421         test_stop = 1;
 422
 423         for (i = 0; i < nr_readers; i++) {
 424                 err = pthread_join(tid_reader[i], &tret);
 425                 if (err != 0)
 426                         exit(1);
 427                 tot_reads += count_reader[i];
 428         }
 429         for (i = 0; i < nr_writers; i++) {
 430                 err = pthread_join(tid_writer[i], &tret);
 431                 if (err != 0)
 432                         exit(1);
 433                 tot_writes += tot_nr_writes[i];
 434                 rcu_gc_clear_queue(i);
 435         }
 436
 437         printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
 438                tot_writes);
 439         printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu "
 440                 "nr_writers %3u "
 441                 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
 442                 "batch %u\n",
 443                 argv[0], duration, nr_readers, rduration,
 444                 nr_writers, wdelay, tot_reads, tot_writes,
 445                 tot_reads + tot_writes, reclaim_batch);
 446         free(tid_reader);
 447         free(tid_writer);
 448         free(count_reader);
 449         free(tot_nr_writes);
 450         for (i = 0; i < nr_writers; i++)
 451                 free(pending_reclaims[i].queue);
 452         free(pending_reclaims);
 453
 454         return 0;
 455 }