| 1 | /* |
| 2 | * test_urcu.c |
| 3 | * |
| 4 | * Userspace RCU library - test program |
| 5 | * |
| 6 | * Copyright February 2009 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 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 "../config.h" |
| 25 | #include <stdio.h> |
| 26 | #include <pthread.h> |
| 27 | #include <stdlib.h> |
| 28 | #include <string.h> |
| 29 | #include <sys/types.h> |
| 30 | #include <sys/wait.h> |
| 31 | #include <unistd.h> |
| 32 | #include <stdio.h> |
| 33 | #include <assert.h> |
| 34 | #include <sched.h> |
| 35 | #include <errno.h> |
| 36 | |
| 37 | #include <urcu/arch.h> |
| 38 | #include <urcu/tls-compat.h> |
| 39 | |
| 40 | #ifdef __linux__ |
| 41 | #include <syscall.h> |
| 42 | #endif |
| 43 | |
| 44 | /* hardcoded number of CPUs */ |
| 45 | #define NR_CPUS 16384 |
| 46 | |
| 47 | #if defined(_syscall0) |
| 48 | _syscall0(pid_t, gettid) |
| 49 | #elif defined(__NR_gettid) |
| 50 | static inline pid_t gettid(void) |
| 51 | { |
| 52 | return syscall(__NR_gettid); |
| 53 | } |
| 54 | #else |
| 55 | #warning "use pid as tid" |
| 56 | static inline pid_t gettid(void) |
| 57 | { |
| 58 | return getpid(); |
| 59 | } |
| 60 | #endif |
| 61 | |
| 62 | #ifndef DYNAMIC_LINK_TEST |
| 63 | #define _LGPL_SOURCE |
| 64 | #else |
| 65 | #define debug_yield_read() |
| 66 | #endif |
| 67 | #include <urcu.h> |
| 68 | |
| 69 | struct test_array { |
| 70 | int a; |
| 71 | }; |
| 72 | |
| 73 | static volatile int test_go, test_stop; |
| 74 | |
| 75 | static unsigned long wdelay; |
| 76 | |
| 77 | static struct test_array *test_rcu_pointer; |
| 78 | |
| 79 | static unsigned long duration; |
| 80 | |
| 81 | /* read-side C.S. duration, in loops */ |
| 82 | static unsigned long rduration; |
| 83 | |
| 84 | /* write-side C.S. duration, in loops */ |
| 85 | static unsigned long wduration; |
| 86 | |
| 87 | static inline void loop_sleep(unsigned long l) |
| 88 | { |
| 89 | while(l-- != 0) |
| 90 | caa_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 | static unsigned int cpu_affinities[NR_CPUS]; |
| 102 | static unsigned int next_aff = 0; |
| 103 | static int use_affinity = 0; |
| 104 | |
| 105 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; |
| 106 | |
| 107 | #ifndef HAVE_CPU_SET_T |
| 108 | typedef unsigned long cpu_set_t; |
| 109 | # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0) |
| 110 | # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0) |
| 111 | #endif |
| 112 | |
| 113 | static void set_affinity(void) |
| 114 | { |
| 115 | cpu_set_t mask; |
| 116 | int cpu; |
| 117 | int ret; |
| 118 | |
| 119 | if (!use_affinity) |
| 120 | return; |
| 121 | |
| 122 | #if HAVE_SCHED_SETAFFINITY |
| 123 | ret = pthread_mutex_lock(&affinity_mutex); |
| 124 | if (ret) { |
| 125 | perror("Error in pthread mutex lock"); |
| 126 | exit(-1); |
| 127 | } |
| 128 | cpu = cpu_affinities[next_aff++]; |
| 129 | ret = pthread_mutex_unlock(&affinity_mutex); |
| 130 | if (ret) { |
| 131 | perror("Error in pthread mutex unlock"); |
| 132 | exit(-1); |
| 133 | } |
| 134 | |
| 135 | CPU_ZERO(&mask); |
| 136 | CPU_SET(cpu, &mask); |
| 137 | #if SCHED_SETAFFINITY_ARGS == 2 |
| 138 | sched_setaffinity(0, &mask); |
| 139 | #else |
| 140 | sched_setaffinity(0, sizeof(mask), &mask); |
| 141 | #endif |
| 142 | #endif /* HAVE_SCHED_SETAFFINITY */ |
| 143 | } |
| 144 | |
| 145 | /* |
| 146 | * returns 0 if test should end. |
| 147 | */ |
| 148 | static int test_duration_write(void) |
| 149 | { |
| 150 | return !test_stop; |
| 151 | } |
| 152 | |
| 153 | static int test_duration_read(void) |
| 154 | { |
| 155 | return !test_stop; |
| 156 | } |
| 157 | |
| 158 | static DEFINE_URCU_TLS(unsigned long long, nr_writes); |
| 159 | static DEFINE_URCU_TLS(unsigned long long, nr_reads); |
| 160 | |
| 161 | static unsigned int nr_readers; |
| 162 | static unsigned int nr_writers; |
| 163 | |
| 164 | pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER; |
| 165 | |
| 166 | void rcu_copy_mutex_lock(void) |
| 167 | { |
| 168 | int ret; |
| 169 | ret = pthread_mutex_lock(&rcu_copy_mutex); |
| 170 | if (ret) { |
| 171 | perror("Error in pthread mutex lock"); |
| 172 | exit(-1); |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | void rcu_copy_mutex_unlock(void) |
| 177 | { |
| 178 | int ret; |
| 179 | |
| 180 | ret = pthread_mutex_unlock(&rcu_copy_mutex); |
| 181 | if (ret) { |
| 182 | perror("Error in pthread mutex unlock"); |
| 183 | exit(-1); |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | /* |
| 188 | * malloc/free are reusing memory areas too quickly, which does not let us |
| 189 | * test races appropriately. Use a large circular array for allocations. |
| 190 | * ARRAY_SIZE is larger than nr_writers, and we keep the mutex across |
| 191 | * both alloc and free, which insures we never run over our tail. |
| 192 | */ |
| 193 | #define ARRAY_SIZE (1048576 * nr_writers) |
| 194 | #define ARRAY_POISON 0xDEADBEEF |
| 195 | static int array_index; |
| 196 | static struct test_array *test_array; |
| 197 | |
| 198 | static struct test_array *test_array_alloc(void) |
| 199 | { |
| 200 | struct test_array *ret; |
| 201 | int index; |
| 202 | |
| 203 | index = array_index % ARRAY_SIZE; |
| 204 | assert(test_array[index].a == ARRAY_POISON || |
| 205 | test_array[index].a == 0); |
| 206 | ret = &test_array[index]; |
| 207 | array_index++; |
| 208 | if (array_index == ARRAY_SIZE) |
| 209 | array_index = 0; |
| 210 | return ret; |
| 211 | } |
| 212 | |
| 213 | static void test_array_free(struct test_array *ptr) |
| 214 | { |
| 215 | if (!ptr) |
| 216 | return; |
| 217 | ptr->a = ARRAY_POISON; |
| 218 | } |
| 219 | |
| 220 | void *thr_reader(void *_count) |
| 221 | { |
| 222 | unsigned long long *count = _count; |
| 223 | struct test_array *local_ptr; |
| 224 | |
| 225 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", |
| 226 | "reader", pthread_self(), (unsigned long)gettid()); |
| 227 | |
| 228 | set_affinity(); |
| 229 | |
| 230 | rcu_register_thread(); |
| 231 | |
| 232 | while (!test_go) |
| 233 | { |
| 234 | } |
| 235 | cmm_smp_mb(); |
| 236 | |
| 237 | for (;;) { |
| 238 | rcu_read_lock(); |
| 239 | local_ptr = rcu_dereference(test_rcu_pointer); |
| 240 | debug_yield_read(); |
| 241 | if (local_ptr) |
| 242 | assert(local_ptr->a == 8); |
| 243 | if (caa_unlikely(rduration)) |
| 244 | loop_sleep(rduration); |
| 245 | rcu_read_unlock(); |
| 246 | URCU_TLS(nr_reads)++; |
| 247 | if (caa_unlikely(!test_duration_read())) |
| 248 | break; |
| 249 | } |
| 250 | |
| 251 | rcu_unregister_thread(); |
| 252 | |
| 253 | /* test extra thread registration */ |
| 254 | rcu_register_thread(); |
| 255 | rcu_unregister_thread(); |
| 256 | |
| 257 | *count = URCU_TLS(nr_reads); |
| 258 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", |
| 259 | "reader", pthread_self(), (unsigned long)gettid()); |
| 260 | return ((void*)1); |
| 261 | |
| 262 | } |
| 263 | |
| 264 | void *thr_writer(void *_count) |
| 265 | { |
| 266 | unsigned long long *count = _count; |
| 267 | struct test_array *new, *old; |
| 268 | |
| 269 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", |
| 270 | "writer", pthread_self(), (unsigned long)gettid()); |
| 271 | |
| 272 | set_affinity(); |
| 273 | |
| 274 | while (!test_go) |
| 275 | { |
| 276 | } |
| 277 | cmm_smp_mb(); |
| 278 | |
| 279 | for (;;) { |
| 280 | rcu_copy_mutex_lock(); |
| 281 | new = test_array_alloc(); |
| 282 | new->a = 8; |
| 283 | old = rcu_xchg_pointer(&test_rcu_pointer, new); |
| 284 | if (caa_unlikely(wduration)) |
| 285 | loop_sleep(wduration); |
| 286 | synchronize_rcu(); |
| 287 | if (old) |
| 288 | old->a = 0; |
| 289 | test_array_free(old); |
| 290 | rcu_copy_mutex_unlock(); |
| 291 | URCU_TLS(nr_writes)++; |
| 292 | if (caa_unlikely(!test_duration_write())) |
| 293 | break; |
| 294 | if (caa_unlikely(wdelay)) |
| 295 | loop_sleep(wdelay); |
| 296 | } |
| 297 | |
| 298 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", |
| 299 | "writer", pthread_self(), (unsigned long)gettid()); |
| 300 | *count = URCU_TLS(nr_writes); |
| 301 | return ((void*)2); |
| 302 | } |
| 303 | |
| 304 | void show_usage(int argc, char **argv) |
| 305 | { |
| 306 | printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]); |
| 307 | #ifdef DEBUG_YIELD |
| 308 | printf(" [-r] [-w] (yield reader and/or writer)"); |
| 309 | #endif |
| 310 | printf(" [-d delay] (writer period (us))"); |
| 311 | printf(" [-c duration] (reader C.S. duration (in loops))"); |
| 312 | printf(" [-e duration] (writer C.S. duration (in loops))"); |
| 313 | printf(" [-v] (verbose output)"); |
| 314 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); |
| 315 | printf("\n"); |
| 316 | } |
| 317 | |
| 318 | int main(int argc, char **argv) |
| 319 | { |
| 320 | int err; |
| 321 | pthread_t *tid_reader, *tid_writer; |
| 322 | void *tret; |
| 323 | unsigned long long *count_reader, *count_writer; |
| 324 | unsigned long long tot_reads = 0, tot_writes = 0; |
| 325 | int i, a; |
| 326 | |
| 327 | if (argc < 4) { |
| 328 | show_usage(argc, argv); |
| 329 | return -1; |
| 330 | } |
| 331 | |
| 332 | err = sscanf(argv[1], "%u", &nr_readers); |
| 333 | if (err != 1) { |
| 334 | show_usage(argc, argv); |
| 335 | return -1; |
| 336 | } |
| 337 | |
| 338 | err = sscanf(argv[2], "%u", &nr_writers); |
| 339 | if (err != 1) { |
| 340 | show_usage(argc, argv); |
| 341 | return -1; |
| 342 | } |
| 343 | |
| 344 | err = sscanf(argv[3], "%lu", &duration); |
| 345 | if (err != 1) { |
| 346 | show_usage(argc, argv); |
| 347 | return -1; |
| 348 | } |
| 349 | |
| 350 | for (i = 4; i < argc; i++) { |
| 351 | if (argv[i][0] != '-') |
| 352 | continue; |
| 353 | switch (argv[i][1]) { |
| 354 | #ifdef DEBUG_YIELD |
| 355 | case 'r': |
| 356 | yield_active |= YIELD_READ; |
| 357 | break; |
| 358 | case 'w': |
| 359 | yield_active |= YIELD_WRITE; |
| 360 | break; |
| 361 | #endif |
| 362 | case 'a': |
| 363 | if (argc < i + 2) { |
| 364 | show_usage(argc, argv); |
| 365 | return -1; |
| 366 | } |
| 367 | a = atoi(argv[++i]); |
| 368 | cpu_affinities[next_aff++] = a; |
| 369 | use_affinity = 1; |
| 370 | printf_verbose("Adding CPU %d affinity\n", a); |
| 371 | break; |
| 372 | case 'c': |
| 373 | if (argc < i + 2) { |
| 374 | show_usage(argc, argv); |
| 375 | return -1; |
| 376 | } |
| 377 | rduration = atol(argv[++i]); |
| 378 | break; |
| 379 | case 'd': |
| 380 | if (argc < i + 2) { |
| 381 | show_usage(argc, argv); |
| 382 | return -1; |
| 383 | } |
| 384 | wdelay = atol(argv[++i]); |
| 385 | break; |
| 386 | case 'e': |
| 387 | if (argc < i + 2) { |
| 388 | show_usage(argc, argv); |
| 389 | return -1; |
| 390 | } |
| 391 | wduration = atol(argv[++i]); |
| 392 | break; |
| 393 | case 'v': |
| 394 | verbose_mode = 1; |
| 395 | break; |
| 396 | } |
| 397 | } |
| 398 | |
| 399 | printf_verbose("running test for %lu seconds, %u readers, %u writers.\n", |
| 400 | duration, nr_readers, nr_writers); |
| 401 | printf_verbose("Writer delay : %lu loops.\n", wdelay); |
| 402 | printf_verbose("Reader duration : %lu loops.\n", rduration); |
| 403 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", |
| 404 | "main", pthread_self(), (unsigned long)gettid()); |
| 405 | |
| 406 | test_array = calloc(1, sizeof(*test_array) * ARRAY_SIZE); |
| 407 | tid_reader = malloc(sizeof(*tid_reader) * nr_readers); |
| 408 | tid_writer = malloc(sizeof(*tid_writer) * nr_writers); |
| 409 | count_reader = malloc(sizeof(*count_reader) * nr_readers); |
| 410 | count_writer = malloc(sizeof(*count_writer) * nr_writers); |
| 411 | |
| 412 | next_aff = 0; |
| 413 | |
| 414 | for (i = 0; i < nr_readers; i++) { |
| 415 | err = pthread_create(&tid_reader[i], NULL, thr_reader, |
| 416 | &count_reader[i]); |
| 417 | if (err != 0) |
| 418 | exit(1); |
| 419 | } |
| 420 | for (i = 0; i < nr_writers; i++) { |
| 421 | err = pthread_create(&tid_writer[i], NULL, thr_writer, |
| 422 | &count_writer[i]); |
| 423 | if (err != 0) |
| 424 | exit(1); |
| 425 | } |
| 426 | |
| 427 | cmm_smp_mb(); |
| 428 | |
| 429 | test_go = 1; |
| 430 | |
| 431 | sleep(duration); |
| 432 | |
| 433 | test_stop = 1; |
| 434 | |
| 435 | for (i = 0; i < nr_readers; i++) { |
| 436 | err = pthread_join(tid_reader[i], &tret); |
| 437 | if (err != 0) |
| 438 | exit(1); |
| 439 | tot_reads += count_reader[i]; |
| 440 | } |
| 441 | for (i = 0; i < nr_writers; i++) { |
| 442 | err = pthread_join(tid_writer[i], &tret); |
| 443 | if (err != 0) |
| 444 | exit(1); |
| 445 | tot_writes += count_writer[i]; |
| 446 | } |
| 447 | |
| 448 | printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads, |
| 449 | tot_writes); |
| 450 | printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu " |
| 451 | "nr_writers %3u " |
| 452 | "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu\n", |
| 453 | argv[0], duration, nr_readers, rduration, wduration, |
| 454 | nr_writers, wdelay, tot_reads, tot_writes, |
| 455 | tot_reads + tot_writes); |
| 456 | test_array_free(test_rcu_pointer); |
| 457 | free(test_array); |
| 458 | free(tid_reader); |
| 459 | free(tid_writer); |
| 460 | free(count_reader); |
| 461 | free(count_writer); |
| 462 | return 0; |
| 463 | } |