Commit | Line | Data |
---|---|---|
453629a9 MD |
1 | /* |
2 | * test_urcu_lfs.c | |
3 | * | |
cb8818f0 | 4 | * Userspace RCU library - example lock-free stack |
453629a9 | 5 | * |
cb8818f0 | 6 | * Copyright 2010-2012 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
453629a9 MD |
7 | * Copyright February 2010 - Paolo Bonzini <pbonzini@redhat.com> |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License along | |
20 | * with this program; if not, write to the Free Software Foundation, Inc., | |
21 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
22 | */ | |
23 | ||
24 | #define _GNU_SOURCE | |
25 | #include "../config.h" | |
26 | #include <stdio.h> | |
27 | #include <pthread.h> | |
28 | #include <stdlib.h> | |
29 | #include <stdint.h> | |
30 | #include <stdbool.h> | |
31 | #include <string.h> | |
32 | #include <sys/types.h> | |
33 | #include <sys/wait.h> | |
34 | #include <unistd.h> | |
35 | #include <stdio.h> | |
36 | #include <assert.h> | |
453629a9 MD |
37 | #include <errno.h> |
38 | ||
39 | #include <urcu/arch.h> | |
bd252a04 | 40 | #include <urcu/tls-compat.h> |
2953b501 | 41 | #include "cpuset.h" |
453629a9 | 42 | |
65fcc7e9 MD |
43 | #ifdef __linux__ |
44 | #include <syscall.h> | |
45 | #endif | |
46 | ||
453629a9 MD |
47 | /* hardcoded number of CPUs */ |
48 | #define NR_CPUS 16384 | |
49 | ||
50 | #if defined(_syscall0) | |
51 | _syscall0(pid_t, gettid) | |
52 | #elif defined(__NR_gettid) | |
53 | static inline pid_t gettid(void) | |
54 | { | |
55 | return syscall(__NR_gettid); | |
56 | } | |
57 | #else | |
58 | #warning "use pid as tid" | |
59 | static inline pid_t gettid(void) | |
60 | { | |
61 | return getpid(); | |
62 | } | |
63 | #endif | |
64 | ||
65 | #ifndef DYNAMIC_LINK_TEST | |
66 | #define _LGPL_SOURCE | |
67 | #endif | |
68 | #include <urcu.h> | |
e5a7d709 | 69 | #include <urcu/cds.h> |
453629a9 | 70 | |
111ce0c3 MD |
71 | /* |
72 | * External synchronization used. | |
73 | */ | |
74 | enum test_sync { | |
75 | TEST_SYNC_NONE = 0, | |
76 | TEST_SYNC_RCU, | |
77 | }; | |
78 | ||
79 | static enum test_sync test_sync; | |
80 | ||
453629a9 MD |
81 | static volatile int test_go, test_stop; |
82 | ||
83 | static unsigned long rduration; | |
84 | ||
85 | static unsigned long duration; | |
86 | ||
87 | /* read-side C.S. duration, in loops */ | |
88 | static unsigned long wdelay; | |
89 | ||
ab0aacbe | 90 | static inline void loop_sleep(unsigned long loops) |
453629a9 | 91 | { |
ab0aacbe | 92 | while (loops-- != 0) |
06f22bdb | 93 | caa_cpu_relax(); |
453629a9 MD |
94 | } |
95 | ||
96 | static int verbose_mode; | |
97 | ||
111ce0c3 MD |
98 | static int test_pop, test_pop_all; |
99 | ||
453629a9 MD |
100 | #define printf_verbose(fmt, args...) \ |
101 | do { \ | |
102 | if (verbose_mode) \ | |
111ce0c3 | 103 | printf(fmt, ## args); \ |
453629a9 MD |
104 | } while (0) |
105 | ||
106 | static unsigned int cpu_affinities[NR_CPUS]; | |
107 | static unsigned int next_aff = 0; | |
108 | static int use_affinity = 0; | |
109 | ||
110 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
111 | ||
453629a9 MD |
112 | static void set_affinity(void) |
113 | { | |
95bc7fb9 | 114 | #if HAVE_SCHED_SETAFFINITY |
453629a9 | 115 | cpu_set_t mask; |
95bc7fb9 MD |
116 | int cpu, ret; |
117 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
453629a9 MD |
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_dequeue(void) | |
149 | { | |
150 | return !test_stop; | |
151 | } | |
152 | ||
153 | static int test_duration_enqueue(void) | |
154 | { | |
155 | return !test_stop; | |
156 | } | |
157 | ||
bd252a04 MD |
158 | static DEFINE_URCU_TLS(unsigned long long, nr_dequeues); |
159 | static DEFINE_URCU_TLS(unsigned long long, nr_enqueues); | |
453629a9 | 160 | |
bd252a04 MD |
161 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues); |
162 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues); | |
453629a9 MD |
163 | |
164 | static unsigned int nr_enqueuers; | |
165 | static unsigned int nr_dequeuers; | |
166 | ||
94aec122 | 167 | struct test { |
cb8818f0 | 168 | struct cds_lfs_node list; |
94aec122 MD |
169 | struct rcu_head rcu; |
170 | }; | |
171 | ||
cb8818f0 | 172 | static struct cds_lfs_stack s; |
453629a9 | 173 | |
eb314468 | 174 | static void *thr_enqueuer(void *_count) |
453629a9 MD |
175 | { |
176 | unsigned long long *count = _count; | |
177 | ||
178 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
179 | "enqueuer", (unsigned long) pthread_self(), |
180 | (unsigned long) gettid()); | |
453629a9 MD |
181 | |
182 | set_affinity(); | |
183 | ||
184 | rcu_register_thread(); | |
185 | ||
186 | while (!test_go) | |
187 | { | |
188 | } | |
5481ddb3 | 189 | cmm_smp_mb(); |
453629a9 MD |
190 | |
191 | for (;;) { | |
94aec122 | 192 | struct test *node = malloc(sizeof(*node)); |
453629a9 MD |
193 | if (!node) |
194 | goto fail; | |
cb8818f0 MD |
195 | cds_lfs_node_init(&node->list); |
196 | cds_lfs_push(&s, &node->list); | |
bd252a04 | 197 | URCU_TLS(nr_successful_enqueues)++; |
453629a9 | 198 | |
a0b7f7ea | 199 | if (caa_unlikely(wdelay)) |
453629a9 MD |
200 | loop_sleep(wdelay); |
201 | fail: | |
bd252a04 | 202 | URCU_TLS(nr_enqueues)++; |
a0b7f7ea | 203 | if (caa_unlikely(!test_duration_enqueue())) |
453629a9 MD |
204 | break; |
205 | } | |
206 | ||
207 | rcu_unregister_thread(); | |
208 | ||
bd252a04 MD |
209 | count[0] = URCU_TLS(nr_enqueues); |
210 | count[1] = URCU_TLS(nr_successful_enqueues); | |
453629a9 MD |
211 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " |
212 | "enqueues %llu successful_enqueues %llu\n", | |
6ff854ac MD |
213 | pthread_self(), |
214 | (unsigned long) gettid(), | |
bd252a04 | 215 | URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues)); |
453629a9 MD |
216 | return ((void*)1); |
217 | ||
218 | } | |
219 | ||
94aec122 MD |
220 | static |
221 | void free_node_cb(struct rcu_head *head) | |
222 | { | |
223 | struct test *node = | |
224 | caa_container_of(head, struct test, rcu); | |
225 | free(node); | |
226 | } | |
227 | ||
111ce0c3 MD |
228 | static |
229 | void do_test_pop(enum test_sync sync) | |
230 | { | |
231 | struct cds_lfs_node *snode; | |
232 | ||
233 | if (sync == TEST_SYNC_RCU) | |
234 | rcu_read_lock(); | |
235 | snode = __cds_lfs_pop(&s); | |
236 | if (sync == TEST_SYNC_RCU) | |
237 | rcu_read_unlock(); | |
238 | if (snode) { | |
239 | struct test *node; | |
240 | ||
241 | node = caa_container_of(snode, | |
242 | struct test, list); | |
243 | if (sync == TEST_SYNC_RCU) | |
244 | call_rcu(&node->rcu, free_node_cb); | |
245 | else | |
246 | free(node); | |
247 | URCU_TLS(nr_successful_dequeues)++; | |
248 | } | |
249 | URCU_TLS(nr_dequeues)++; | |
250 | } | |
251 | ||
252 | static | |
253 | void do_test_pop_all(enum test_sync sync) | |
254 | { | |
255 | struct cds_lfs_node *snode; | |
256 | struct cds_lfs_head *head; | |
257 | struct cds_lfs_node *n; | |
258 | ||
259 | head = __cds_lfs_pop_all(&s); | |
260 | cds_lfs_for_each_safe(head, snode, n) { | |
261 | struct test *node; | |
262 | ||
263 | node = caa_container_of(snode, struct test, list); | |
264 | if (sync == TEST_SYNC_RCU) | |
265 | call_rcu(&node->rcu, free_node_cb); | |
266 | else | |
267 | free(node); | |
268 | URCU_TLS(nr_successful_dequeues)++; | |
269 | URCU_TLS(nr_dequeues)++; | |
270 | } | |
271 | ||
272 | } | |
273 | ||
eb314468 | 274 | static void *thr_dequeuer(void *_count) |
453629a9 MD |
275 | { |
276 | unsigned long long *count = _count; | |
277 | ||
278 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
279 | "dequeuer", (unsigned long) pthread_self(), |
280 | (unsigned long) gettid()); | |
453629a9 MD |
281 | |
282 | set_affinity(); | |
283 | ||
453629a9 MD |
284 | rcu_register_thread(); |
285 | ||
286 | while (!test_go) | |
287 | { | |
288 | } | |
5481ddb3 | 289 | cmm_smp_mb(); |
453629a9 | 290 | |
111ce0c3 | 291 | assert(test_pop || test_pop_all); |
f3eb6ef1 | 292 | |
111ce0c3 MD |
293 | for (;;) { |
294 | unsigned int counter = 0; | |
295 | ||
296 | if (test_pop && test_pop_all) { | |
297 | /* both pop and pop all */ | |
298 | if (counter & 1) | |
299 | do_test_pop(test_sync); | |
300 | else | |
301 | do_test_pop_all(test_sync); | |
302 | counter++; | |
303 | } else { | |
304 | if (test_pop) | |
305 | do_test_pop(test_sync); | |
306 | else | |
307 | do_test_pop_all(test_sync); | |
453629a9 | 308 | } |
111ce0c3 | 309 | |
a0b7f7ea | 310 | if (caa_unlikely(!test_duration_dequeue())) |
453629a9 | 311 | break; |
a0b7f7ea | 312 | if (caa_unlikely(rduration)) |
453629a9 MD |
313 | loop_sleep(rduration); |
314 | } | |
315 | ||
316 | rcu_unregister_thread(); | |
453629a9 MD |
317 | |
318 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " | |
319 | "dequeues %llu, successful_dequeues %llu\n", | |
6ff854ac MD |
320 | pthread_self(), |
321 | (unsigned long) gettid(), | |
bd252a04 MD |
322 | URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues)); |
323 | count[0] = URCU_TLS(nr_dequeues); | |
324 | count[1] = URCU_TLS(nr_successful_dequeues); | |
453629a9 MD |
325 | return ((void*)2); |
326 | } | |
327 | ||
eb314468 | 328 | static void test_end(struct cds_lfs_stack *s, unsigned long long *nr_dequeues) |
453629a9 | 329 | { |
cb8818f0 | 330 | struct cds_lfs_node *snode; |
453629a9 MD |
331 | |
332 | do { | |
7294103b | 333 | snode = __cds_lfs_pop(s); |
94aec122 MD |
334 | if (snode) { |
335 | struct test *node; | |
336 | ||
337 | node = caa_container_of(snode, struct test, list); | |
453629a9 MD |
338 | free(node); |
339 | (*nr_dequeues)++; | |
340 | } | |
94aec122 | 341 | } while (snode); |
453629a9 MD |
342 | } |
343 | ||
eb314468 | 344 | static void show_usage(int argc, char **argv) |
453629a9 | 345 | { |
06637138 MD |
346 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n", |
347 | argv[0]); | |
348 | printf("OPTIONS:\n"); | |
349 | printf(" [-d delay] (enqueuer period (in loops))\n"); | |
350 | printf(" [-c duration] (dequeuer period (in loops))\n"); | |
351 | printf(" [-v] (verbose output)\n"); | |
352 | printf(" [-a cpu#] [-a cpu#]... (affinity)\n"); | |
353 | printf(" [-p] (test pop)\n"); | |
354 | printf(" [-P] (test pop_all, enabled by default)\n"); | |
355 | printf(" [-R] (use RCU external synchronization)\n"); | |
356 | printf(" Note: default: no external synchronization used.\n"); | |
453629a9 MD |
357 | printf("\n"); |
358 | } | |
359 | ||
360 | int main(int argc, char **argv) | |
361 | { | |
362 | int err; | |
363 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
364 | void *tret; | |
365 | unsigned long long *count_enqueuer, *count_dequeuer; | |
366 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
367 | unsigned long long tot_successful_enqueues = 0, | |
368 | tot_successful_dequeues = 0; | |
369 | unsigned long long end_dequeues = 0; | |
370 | int i, a; | |
371 | ||
372 | if (argc < 4) { | |
373 | show_usage(argc, argv); | |
374 | return -1; | |
375 | } | |
376 | ||
377 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
378 | if (err != 1) { | |
379 | show_usage(argc, argv); | |
380 | return -1; | |
381 | } | |
382 | ||
383 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
384 | if (err != 1) { | |
385 | show_usage(argc, argv); | |
386 | return -1; | |
387 | } | |
388 | ||
389 | err = sscanf(argv[3], "%lu", &duration); | |
390 | if (err != 1) { | |
391 | show_usage(argc, argv); | |
392 | return -1; | |
393 | } | |
394 | ||
395 | for (i = 4; i < argc; i++) { | |
396 | if (argv[i][0] != '-') | |
397 | continue; | |
398 | switch (argv[i][1]) { | |
399 | case 'a': | |
400 | if (argc < i + 2) { | |
401 | show_usage(argc, argv); | |
402 | return -1; | |
403 | } | |
404 | a = atoi(argv[++i]); | |
405 | cpu_affinities[next_aff++] = a; | |
406 | use_affinity = 1; | |
407 | printf_verbose("Adding CPU %d affinity\n", a); | |
408 | break; | |
409 | case 'c': | |
410 | if (argc < i + 2) { | |
411 | show_usage(argc, argv); | |
412 | return -1; | |
413 | } | |
414 | rduration = atol(argv[++i]); | |
415 | break; | |
416 | case 'd': | |
417 | if (argc < i + 2) { | |
418 | show_usage(argc, argv); | |
419 | return -1; | |
420 | } | |
421 | wdelay = atol(argv[++i]); | |
422 | break; | |
423 | case 'v': | |
424 | verbose_mode = 1; | |
425 | break; | |
111ce0c3 MD |
426 | case 'p': |
427 | test_pop = 1; | |
428 | break; | |
429 | case 'P': | |
430 | test_pop_all = 1; | |
431 | break; | |
432 | case 'R': | |
433 | test_sync = TEST_SYNC_RCU; | |
434 | break; | |
453629a9 MD |
435 | } |
436 | } | |
437 | ||
111ce0c3 MD |
438 | /* activate pop_all test by default */ |
439 | if (!test_pop && !test_pop_all) | |
440 | test_pop_all = 1; | |
441 | ||
453629a9 MD |
442 | printf_verbose("running test for %lu seconds, %u enqueuers, " |
443 | "%u dequeuers.\n", | |
444 | duration, nr_enqueuers, nr_dequeuers); | |
111ce0c3 MD |
445 | if (test_pop) |
446 | printf_verbose("pop test activated.\n"); | |
447 | if (test_pop_all) | |
448 | printf_verbose("pop_all test activated.\n"); | |
eb314468 MD |
449 | if (test_sync == TEST_SYNC_RCU) |
450 | printf_verbose("External sync: RCU.\n"); | |
451 | else | |
452 | printf_verbose("External sync: none.\n"); | |
453629a9 MD |
453 | printf_verbose("Writer delay : %lu loops.\n", rduration); |
454 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
455 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
456 | "main", (unsigned long) pthread_self(), |
457 | (unsigned long) gettid()); | |
453629a9 MD |
458 | |
459 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
460 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
461 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
462 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
cb8818f0 | 463 | cds_lfs_init(&s); |
94aec122 | 464 | err = create_all_cpu_call_rcu_data(0); |
8bcbd94a MD |
465 | if (err) { |
466 | printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n"); | |
467 | } | |
453629a9 MD |
468 | |
469 | next_aff = 0; | |
470 | ||
471 | for (i = 0; i < nr_enqueuers; i++) { | |
472 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
473 | &count_enqueuer[2 * i]); | |
474 | if (err != 0) | |
475 | exit(1); | |
476 | } | |
477 | for (i = 0; i < nr_dequeuers; i++) { | |
478 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
479 | &count_dequeuer[2 * i]); | |
480 | if (err != 0) | |
481 | exit(1); | |
482 | } | |
483 | ||
5481ddb3 | 484 | cmm_smp_mb(); |
453629a9 MD |
485 | |
486 | test_go = 1; | |
487 | ||
488 | for (i = 0; i < duration; i++) { | |
489 | sleep(1); | |
490 | if (verbose_mode) | |
491 | write (1, ".", 1); | |
492 | } | |
493 | ||
494 | test_stop = 1; | |
495 | ||
496 | for (i = 0; i < nr_enqueuers; i++) { | |
497 | err = pthread_join(tid_enqueuer[i], &tret); | |
498 | if (err != 0) | |
499 | exit(1); | |
500 | tot_enqueues += count_enqueuer[2 * i]; | |
501 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
502 | } | |
503 | for (i = 0; i < nr_dequeuers; i++) { | |
504 | err = pthread_join(tid_dequeuer[i], &tret); | |
505 | if (err != 0) | |
506 | exit(1); | |
507 | tot_dequeues += count_dequeuer[2 * i]; | |
508 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
509 | } | |
510 | ||
511 | test_end(&s, &end_dequeues); | |
512 | ||
513 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
514 | tot_enqueues, tot_dequeues); | |
515 | printf_verbose("total number of successful enqueues : %llu, " | |
516 | "successful dequeues %llu\n", | |
517 | tot_successful_enqueues, tot_successful_dequeues); | |
518 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
519 | "nr_dequeuers %3u " | |
520 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
521 | "successful enqueues %12llu successful dequeues %12llu " | |
522 | "end_dequeues %llu nr_ops %12llu\n", | |
523 | argv[0], duration, nr_enqueuers, wdelay, | |
524 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
525 | tot_successful_enqueues, | |
526 | tot_successful_dequeues, end_dequeues, | |
527 | tot_enqueues + tot_dequeues); | |
528 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
529 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
530 | "succ. dequeues + end dequeues %llu.\n", | |
531 | tot_successful_enqueues, | |
532 | tot_successful_dequeues + end_dequeues); | |
533 | ||
0938c541 | 534 | free_all_cpu_call_rcu_data(); |
453629a9 MD |
535 | free(count_enqueuer); |
536 | free(count_dequeuer); | |
537 | free(tid_enqueuer); | |
538 | free(tid_dequeuer); | |
539 | return 0; | |
540 | } |