tests/rcutorture.h

   1 /*
   2  * rcutorture.h: simple user-level performance/stress test of RCU.
   3  *
   4  * Usage:
   5  *      ./rcu <nreaders> rperf [ <cpustride> ]
   6  *              Run a read-side performance test with the specified
   7  *              number of readers spaced by <cpustride>.
   8  *              Thus "./rcu 16 rperf 2" would run 16 readers on even-numbered
   9  *              CPUs from 0 to 30.
  10  *      ./rcu <nupdaters> uperf [ <cpustride> ]
  11  *              Run an update-side performance test with the specified
  12  *              number of updaters and specified CPU spacing.
  13  *      ./rcu <nreaders> perf [ <cpustride> ]
  14  *              Run a combined read/update performance test with the specified
  15  *              number of readers and one updater and specified CPU spacing.
  16  *              The readers run on the low-numbered CPUs and the updater
  17  *              of the highest-numbered CPU.
  18  *
  19  * The above tests produce output as follows:
  20  *
  21  * n_reads: 46008000  n_updates: 146026  nreaders: 2  nupdaters: 1 duration: 1
  22  * ns/read: 43.4707  ns/update: 6848.1
  23  *
  24  * The first line lists the total number of RCU reads and updates executed
  25  * during the test, the number of reader threads, the number of updater
  26  * threads, and the duration of the test in seconds.  The second line
  27  * lists the average duration of each type of operation in nanoseconds,
  28  * or "nan" if the corresponding type of operation was not performed.
  29  *
  30  *      ./rcu <nreaders> stress
  31  *              Run a stress test with the specified number of readers and
  32  *              one updater.  None of the threads are affinitied to any
  33  *              particular CPU.
  34  *
  35  * This test produces output as follows:
  36  *
  37  * n_reads: 114633217  n_updates: 3903415  n_mberror: 0
  38  * rcu_stress_count: 114618391 14826 0 0 0 0 0 0 0 0 0
  39  *
  40  * The first line lists the number of RCU read and update operations
  41  * executed, followed by the number of memory-ordering violations
  42  * (which will be zero in a correct RCU implementation).  The second
  43  * line lists the number of readers observing progressively more stale
  44  * data.  A correct RCU implementation will have all but the first two
  45  * numbers non-zero.
  46  *
  47  * This program is free software; you can redistribute it and/or modify
  48  * it under the terms of the GNU General Public License as published by
  49  * the Free Software Foundation; either version 2 of the License, or
  50  * (at your option) any later version.
  51  *
  52  * This program is distributed in the hope that it will be useful,
  53  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  54  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  55  * GNU General Public License for more details.
  56  *
  57  * You should have received a copy of the GNU General Public License
  58  * along with this program; if not, write to the Free Software
  59  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  60  *
  61  * Copyright (c) 2008 Paul E. McKenney, IBM Corporation.
  62  */
  63
  64 /*
  65  * Test variables.
  66  */
  67
  68 DEFINE_PER_THREAD(long long, n_reads_pt);
  69 DEFINE_PER_THREAD(long long, n_updates_pt);
  70
  71 long long n_reads = 0LL;
  72 long n_updates = 0L;
  73 int nthreadsrunning;
  74 char argsbuf[64];
  75
  76 #define GOFLAG_INIT 0
  77 #define GOFLAG_RUN  1
  78 #define GOFLAG_STOP 2
  79
  80 int goflag __attribute__((__aligned__(CACHE_LINE_SIZE))) = GOFLAG_INIT;
  81
  82 #define RCU_READ_RUN 1000
  83
  84 //MD
  85 #define RCU_READ_NESTABLE
  86
  87 #ifdef RCU_READ_NESTABLE
  88 #define rcu_read_lock_nest() rcu_read_lock()
  89 #define rcu_read_unlock_nest() rcu_read_unlock()
  90 #else /* #ifdef RCU_READ_NESTABLE */
  91 #define rcu_read_lock_nest()
  92 #define rcu_read_unlock_nest()
  93 #endif /* #else #ifdef RCU_READ_NESTABLE */
  94
  95 #ifdef TORTURE_QSBR
  96 #define mark_rcu_quiescent_state        rcu_quiescent_state
  97 #define put_thread_offline              rcu_thread_offline
  98 #define put_thread_online               rcu_thread_online
  99 #endif
 100
 101 #ifndef mark_rcu_quiescent_state
 102 #define mark_rcu_quiescent_state() do ; while (0)
 103 #endif /* #ifdef mark_rcu_quiescent_state */
 104
 105 #ifndef put_thread_offline
 106 #define put_thread_offline()            do ; while (0)
 107 #define put_thread_online()             do ; while (0)
 108 #define put_thread_online_delay()       do ; while (0)
 109 #else /* #ifndef put_thread_offline */
 110 #define put_thread_online_delay()       synchronize_rcu()
 111 #endif /* #else #ifndef put_thread_offline */
 112
 113 /*
 114  * Performance test.
 115  */
 116
 117 void *rcu_read_perf_test(void *arg)
 118 {
 119         int i;
 120         int me = (long)arg;
 121         long long n_reads_local = 0;
 122
 123         rcu_register_thread();
 124         run_on(me);
 125         uatomic_inc(&nthreadsrunning);
 126         while (goflag == GOFLAG_INIT)
 127                 poll(NULL, 0, 1);
 128         mark_rcu_quiescent_state();
 129         while (goflag == GOFLAG_RUN) {
 130                 for (i = 0; i < RCU_READ_RUN; i++) {
 131                         rcu_read_lock();
 132                         /* rcu_read_lock_nest(); */
 133                         /* rcu_read_unlock_nest(); */
 134                         rcu_read_unlock();
 135                 }
 136                 n_reads_local += RCU_READ_RUN;
 137                 mark_rcu_quiescent_state();
 138         }
 139         __get_thread_var(n_reads_pt) += n_reads_local;
 140         put_thread_offline();
 141         rcu_unregister_thread();
 142
 143         return (NULL);
 144 }
 145
 146 void *rcu_update_perf_test(void *arg)
 147 {
 148         long long n_updates_local = 0;
 149
 150         uatomic_inc(&nthreadsrunning);
 151         while (goflag == GOFLAG_INIT)
 152                 poll(NULL, 0, 1);
 153         while (goflag == GOFLAG_RUN) {
 154                 synchronize_rcu();
 155                 n_updates_local++;
 156         }
 157         __get_thread_var(n_updates_pt) += n_updates_local;
 158         return NULL;
 159 }
 160
 161 void perftestinit(void)
 162 {
 163         init_per_thread(n_reads_pt, 0LL);
 164         init_per_thread(n_updates_pt, 0LL);
 165         uatomic_set(&nthreadsrunning, 0);
 166 }
 167
 168 void perftestrun(int nthreads, int nreaders, int nupdaters)
 169 {
 170         int t;
 171         int duration = 1;
 172
 173         cmm_smp_mb();
 174         while (uatomic_read(&nthreadsrunning) < nthreads)
 175                 poll(NULL, 0, 1);
 176         goflag = GOFLAG_RUN;
 177         cmm_smp_mb();
 178         sleep(duration);
 179         cmm_smp_mb();
 180         goflag = GOFLAG_STOP;
 181         cmm_smp_mb();
 182         wait_all_threads();
 183         for_each_thread(t) {
 184                 n_reads += per_thread(n_reads_pt, t);
 185                 n_updates += per_thread(n_updates_pt, t);
 186         }
 187         printf("n_reads: %lld  n_updates: %ld  nreaders: %d  nupdaters: %d duration: %d\n",
 188                n_reads, n_updates, nreaders, nupdaters, duration);
 189         printf("ns/read: %g  ns/update: %g\n",
 190                ((duration * 1000*1000*1000.*(double)nreaders) /
 191                 (double)n_reads),
 192                ((duration * 1000*1000*1000.*(double)nupdaters) /
 193                 (double)n_updates));
 194         exit(0);
 195 }
 196
 197 void perftest(int nreaders, int cpustride)
 198 {
 199         int i;
 200         long arg;
 201
 202         perftestinit();
 203         for (i = 0; i < nreaders; i++) {
 204                 arg = (long)(i * cpustride);
 205                 create_thread(rcu_read_perf_test, (void *)arg);
 206         }
 207         arg = (long)(i * cpustride);
 208         create_thread(rcu_update_perf_test, (void *)arg);
 209         perftestrun(i + 1, nreaders, 1);
 210 }
 211
 212 void rperftest(int nreaders, int cpustride)
 213 {
 214         int i;
 215         long arg;
 216
 217         perftestinit();
 218         init_per_thread(n_reads_pt, 0LL);
 219         for (i = 0; i < nreaders; i++) {
 220                 arg = (long)(i * cpustride);
 221                 create_thread(rcu_read_perf_test, (void *)arg);
 222         }
 223         perftestrun(i, nreaders, 0);
 224 }
 225
 226 void uperftest(int nupdaters, int cpustride)
 227 {
 228         int i;
 229         long arg;
 230
 231         perftestinit();
 232         init_per_thread(n_reads_pt, 0LL);
 233         for (i = 0; i < nupdaters; i++) {
 234                 arg = (long)(i * cpustride);
 235                 create_thread(rcu_update_perf_test, (void *)arg);
 236         }
 237         perftestrun(i, 0, nupdaters);
 238 }
 239
 240 /*
 241  * Stress test.
 242  */
 243
 244 #define RCU_STRESS_PIPE_LEN 10
 245
 246 struct rcu_stress {
 247         int pipe_count;
 248         int mbtest;
 249 };
 250
 251 struct rcu_stress rcu_stress_array[RCU_STRESS_PIPE_LEN] = { { 0 } };
 252 struct rcu_stress *rcu_stress_current;
 253 int rcu_stress_idx = 0;
 254
 255 int n_mberror = 0;
 256 DEFINE_PER_THREAD(long long [RCU_STRESS_PIPE_LEN + 1], rcu_stress_count);
 257
 258 int garbage = 0;
 259
 260 void *rcu_read_stress_test(void *arg)
 261 {
 262         int i;
 263         int itercnt = 0;
 264         struct rcu_stress *p;
 265         int pc;
 266
 267         rcu_register_thread();
 268         while (goflag == GOFLAG_INIT)
 269                 poll(NULL, 0, 1);
 270         mark_rcu_quiescent_state();
 271         while (goflag == GOFLAG_RUN) {
 272                 rcu_read_lock();
 273                 p = rcu_dereference(rcu_stress_current);
 274                 if (p->mbtest == 0)
 275                         n_mberror++;
 276                 rcu_read_lock_nest();
 277                 for (i = 0; i < 100; i++)
 278                         garbage++;
 279                 rcu_read_unlock_nest();
 280                 pc = p->pipe_count;
 281                 rcu_read_unlock();
 282                 if ((pc > RCU_STRESS_PIPE_LEN) || (pc < 0))
 283                         pc = RCU_STRESS_PIPE_LEN;
 284                 __get_thread_var(rcu_stress_count)[pc]++;
 285                 __get_thread_var(n_reads_pt)++;
 286                 mark_rcu_quiescent_state();
 287                 if ((++itercnt % 0x1000) == 0) {
 288                         put_thread_offline();
 289                         put_thread_online_delay();
 290                         put_thread_online();
 291                 }
 292         }
 293         put_thread_offline();
 294         rcu_unregister_thread();
 295
 296         return (NULL);
 297 }
 298
 299 void *rcu_update_stress_test(void *arg)
 300 {
 301         int i;
 302         struct rcu_stress *p;
 303
 304         while (goflag == GOFLAG_INIT)
 305                 poll(NULL, 0, 1);
 306         while (goflag == GOFLAG_RUN) {
 307                 i = rcu_stress_idx + 1;
 308                 if (i >= RCU_STRESS_PIPE_LEN)
 309                         i = 0;
 310                 p = &rcu_stress_array[i];
 311                 p->mbtest = 0;
 312                 cmm_smp_mb();
 313                 p->pipe_count = 0;
 314                 p->mbtest = 1;
 315                 rcu_assign_pointer(rcu_stress_current, p);
 316                 rcu_stress_idx = i;
 317                 for (i = 0; i < RCU_STRESS_PIPE_LEN; i++)
 318                         if (i != rcu_stress_idx)
 319                                 rcu_stress_array[i].pipe_count++;
 320                 synchronize_rcu();
 321                 n_updates++;
 322         }
 323         return NULL;
 324 }
 325
 326 void *rcu_fake_update_stress_test(void *arg)
 327 {
 328         while (goflag == GOFLAG_INIT)
 329                 poll(NULL, 0, 1);
 330         while (goflag == GOFLAG_RUN) {
 331                 synchronize_rcu();
 332                 poll(NULL, 0, 1);
 333         }
 334         return NULL;
 335 }
 336
 337 void stresstest(int nreaders)
 338 {
 339         int i;
 340         int t;
 341         long long *p;
 342         long long sum;
 343
 344         init_per_thread(n_reads_pt, 0LL);
 345         for_each_thread(t) {
 346                 p = &per_thread(rcu_stress_count,t)[0];
 347                 for (i = 0; i <= RCU_STRESS_PIPE_LEN; i++)
 348                         p[i] = 0LL;
 349         }
 350         rcu_stress_current = &rcu_stress_array[0];
 351         rcu_stress_current->pipe_count = 0;
 352         rcu_stress_current->mbtest = 1;
 353         for (i = 0; i < nreaders; i++)
 354                 create_thread(rcu_read_stress_test, NULL);
 355         create_thread(rcu_update_stress_test, NULL);
 356         for (i = 0; i < 5; i++)
 357                 create_thread(rcu_fake_update_stress_test, NULL);
 358         cmm_smp_mb();
 359         goflag = GOFLAG_RUN;
 360         cmm_smp_mb();
 361         sleep(10);
 362         cmm_smp_mb();
 363         goflag = GOFLAG_STOP;
 364         cmm_smp_mb();
 365         wait_all_threads();
 366         for_each_thread(t)
 367                 n_reads += per_thread(n_reads_pt, t);
 368         printf("n_reads: %lld  n_updates: %ld  n_mberror: %d\n",
 369                n_reads, n_updates, n_mberror);
 370         printf("rcu_stress_count:");
 371         for (i = 0; i <= RCU_STRESS_PIPE_LEN; i++) {
 372                 sum = 0LL;
 373                 for_each_thread(t) {
 374                         sum += per_thread(rcu_stress_count, t)[i];
 375                 }
 376                 printf(" %lld", sum);
 377         }
 378         printf("\n");
 379         exit(0);
 380 }
 381
 382 /*
 383  * Mainprogram.
 384  */
 385
 386 void usage(int argc, char *argv[])
 387 {
 388         fprintf(stderr, "Usage: %s [nreaders [ perf | stress ] ]\n", argv[0]);
 389         exit(-1);
 390 }
 391
 392 int main(int argc, char *argv[])
 393 {
 394         int nreaders = 1;
 395         int cpustride = 1;
 396
 397         smp_init();
 398         //rcu_init();
 399
 400 #ifdef DEBUG_YIELD
 401         yield_active |= YIELD_READ;
 402         yield_active |= YIELD_WRITE;
 403 #endif
 404
 405         if (argc > 1) {
 406                 nreaders = strtoul(argv[1], NULL, 0);
 407                 if (argc == 2)
 408                         perftest(nreaders, cpustride);
 409                 if (argc > 3)
 410                         cpustride = strtoul(argv[3], NULL, 0);
 411                 if (strcmp(argv[2], "perf") == 0)
 412                         perftest(nreaders, cpustride);
 413                 else if (strcmp(argv[2], "rperf") == 0)
 414                         rperftest(nreaders, cpustride);
 415                 else if (strcmp(argv[2], "uperf") == 0)
 416                         uperftest(nreaders, cpustride);
 417                 else if (strcmp(argv[2], "stress") == 0)
 418                         stresstest(nreaders);
 419                 usage(argc, argv);
 420         }
 421         perftest(nreaders, cpustride);
 422         return 0;
 423 }