Commit | Line | Data |
---|---|---|
453629a9 MD |
1 | /* |
2 | * test_urcu_lfq.c | |
3 | * | |
4 | * Userspace RCU library - example RCU-based lock-free queue | |
5 | * | |
6 | * Copyright February 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
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 MD |
70 | |
71 | static volatile int test_go, test_stop; | |
72 | ||
73 | static unsigned long rduration; | |
74 | ||
75 | static unsigned long duration; | |
76 | ||
77 | /* read-side C.S. duration, in loops */ | |
78 | static unsigned long wdelay; | |
79 | ||
ab0aacbe | 80 | static inline void loop_sleep(unsigned long loops) |
453629a9 | 81 | { |
ab0aacbe | 82 | while (loops-- != 0) |
06f22bdb | 83 | caa_cpu_relax(); |
453629a9 MD |
84 | } |
85 | ||
86 | static int verbose_mode; | |
87 | ||
88 | #define printf_verbose(fmt, args...) \ | |
89 | do { \ | |
90 | if (verbose_mode) \ | |
91 | printf(fmt, args); \ | |
92 | } while (0) | |
93 | ||
94 | static unsigned int cpu_affinities[NR_CPUS]; | |
95 | static unsigned int next_aff = 0; | |
96 | static int use_affinity = 0; | |
97 | ||
98 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
99 | ||
453629a9 MD |
100 | static void set_affinity(void) |
101 | { | |
95bc7fb9 | 102 | #if HAVE_SCHED_SETAFFINITY |
453629a9 | 103 | cpu_set_t mask; |
95bc7fb9 MD |
104 | int cpu, ret; |
105 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
453629a9 MD |
106 | |
107 | if (!use_affinity) | |
108 | return; | |
109 | ||
110 | #if HAVE_SCHED_SETAFFINITY | |
111 | ret = pthread_mutex_lock(&affinity_mutex); | |
112 | if (ret) { | |
113 | perror("Error in pthread mutex lock"); | |
114 | exit(-1); | |
115 | } | |
116 | cpu = cpu_affinities[next_aff++]; | |
117 | ret = pthread_mutex_unlock(&affinity_mutex); | |
118 | if (ret) { | |
119 | perror("Error in pthread mutex unlock"); | |
120 | exit(-1); | |
121 | } | |
122 | ||
123 | CPU_ZERO(&mask); | |
124 | CPU_SET(cpu, &mask); | |
125 | #if SCHED_SETAFFINITY_ARGS == 2 | |
126 | sched_setaffinity(0, &mask); | |
127 | #else | |
128 | sched_setaffinity(0, sizeof(mask), &mask); | |
129 | #endif | |
130 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
131 | } | |
132 | ||
133 | /* | |
134 | * returns 0 if test should end. | |
135 | */ | |
136 | static int test_duration_dequeue(void) | |
137 | { | |
138 | return !test_stop; | |
139 | } | |
140 | ||
141 | static int test_duration_enqueue(void) | |
142 | { | |
143 | return !test_stop; | |
144 | } | |
145 | ||
bd252a04 MD |
146 | static DEFINE_URCU_TLS(unsigned long long, nr_dequeues); |
147 | static DEFINE_URCU_TLS(unsigned long long, nr_enqueues); | |
453629a9 | 148 | |
bd252a04 MD |
149 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues); |
150 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues); | |
453629a9 MD |
151 | |
152 | static unsigned int nr_enqueuers; | |
153 | static unsigned int nr_dequeuers; | |
154 | ||
94aec122 MD |
155 | struct test { |
156 | struct cds_lfq_node_rcu list; | |
157 | struct rcu_head rcu; | |
158 | }; | |
159 | ||
16aa9ee8 | 160 | static struct cds_lfq_queue_rcu q; |
453629a9 MD |
161 | |
162 | void *thr_enqueuer(void *_count) | |
163 | { | |
164 | unsigned long long *count = _count; | |
165 | ||
166 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
167 | "enqueuer", (unsigned long) pthread_self(), |
168 | (unsigned long) gettid()); | |
453629a9 MD |
169 | |
170 | set_affinity(); | |
171 | ||
172 | rcu_register_thread(); | |
173 | ||
174 | while (!test_go) | |
175 | { | |
176 | } | |
5481ddb3 | 177 | cmm_smp_mb(); |
453629a9 MD |
178 | |
179 | for (;;) { | |
94aec122 | 180 | struct test *node = malloc(sizeof(*node)); |
453629a9 MD |
181 | if (!node) |
182 | goto fail; | |
94aec122 | 183 | cds_lfq_node_init_rcu(&node->list); |
6e5f88cf | 184 | rcu_read_lock(); |
94aec122 | 185 | cds_lfq_enqueue_rcu(&q, &node->list); |
6e5f88cf | 186 | rcu_read_unlock(); |
bd252a04 | 187 | URCU_TLS(nr_successful_enqueues)++; |
453629a9 | 188 | |
a0b7f7ea | 189 | if (caa_unlikely(wdelay)) |
453629a9 MD |
190 | loop_sleep(wdelay); |
191 | fail: | |
bd252a04 | 192 | URCU_TLS(nr_enqueues)++; |
a0b7f7ea | 193 | if (caa_unlikely(!test_duration_enqueue())) |
453629a9 MD |
194 | break; |
195 | } | |
196 | ||
197 | rcu_unregister_thread(); | |
198 | ||
bd252a04 MD |
199 | count[0] = URCU_TLS(nr_enqueues); |
200 | count[1] = URCU_TLS(nr_successful_enqueues); | |
453629a9 MD |
201 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " |
202 | "enqueues %llu successful_enqueues %llu\n", | |
6ff854ac MD |
203 | pthread_self(), |
204 | (unsigned long) gettid(), | |
bd252a04 | 205 | URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues)); |
453629a9 MD |
206 | return ((void*)1); |
207 | ||
208 | } | |
209 | ||
94aec122 MD |
210 | static |
211 | void free_node_cb(struct rcu_head *head) | |
212 | { | |
213 | struct test *node = | |
214 | caa_container_of(head, struct test, rcu); | |
215 | free(node); | |
216 | } | |
217 | ||
453629a9 MD |
218 | void *thr_dequeuer(void *_count) |
219 | { | |
220 | unsigned long long *count = _count; | |
221 | ||
222 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
223 | "dequeuer", (unsigned long) pthread_self(), |
224 | (unsigned long) gettid()); | |
453629a9 MD |
225 | |
226 | set_affinity(); | |
227 | ||
453629a9 MD |
228 | rcu_register_thread(); |
229 | ||
230 | while (!test_go) | |
231 | { | |
232 | } | |
5481ddb3 | 233 | cmm_smp_mb(); |
453629a9 MD |
234 | |
235 | for (;;) { | |
94aec122 | 236 | struct cds_lfq_node_rcu *qnode; |
d9b52143 | 237 | |
6e5f88cf | 238 | rcu_read_lock(); |
94aec122 | 239 | qnode = cds_lfq_dequeue_rcu(&q); |
6e5f88cf MD |
240 | rcu_read_unlock(); |
241 | ||
747b5930 LJ |
242 | if (qnode) { |
243 | struct test *node; | |
244 | ||
245 | node = caa_container_of(qnode, struct test, list); | |
94aec122 | 246 | call_rcu(&node->rcu, free_node_cb); |
bd252a04 | 247 | URCU_TLS(nr_successful_dequeues)++; |
453629a9 MD |
248 | } |
249 | ||
bd252a04 | 250 | URCU_TLS(nr_dequeues)++; |
a0b7f7ea | 251 | if (caa_unlikely(!test_duration_dequeue())) |
453629a9 | 252 | break; |
a0b7f7ea | 253 | if (caa_unlikely(rduration)) |
453629a9 MD |
254 | loop_sleep(rduration); |
255 | } | |
256 | ||
257 | rcu_unregister_thread(); | |
453629a9 MD |
258 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " |
259 | "dequeues %llu, successful_dequeues %llu\n", | |
6ff854ac MD |
260 | pthread_self(), |
261 | (unsigned long) gettid(), | |
bd252a04 MD |
262 | URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues)); |
263 | count[0] = URCU_TLS(nr_dequeues); | |
264 | count[1] = URCU_TLS(nr_successful_dequeues); | |
453629a9 MD |
265 | return ((void*)2); |
266 | } | |
267 | ||
16aa9ee8 | 268 | void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues) |
453629a9 | 269 | { |
94aec122 | 270 | struct cds_lfq_node_rcu *snode; |
453629a9 MD |
271 | |
272 | do { | |
94aec122 MD |
273 | snode = cds_lfq_dequeue_rcu(q); |
274 | if (snode) { | |
275 | struct test *node; | |
276 | ||
277 | node = caa_container_of(snode, struct test, list); | |
e17d9985 | 278 | free(node); /* no more concurrent access */ |
453629a9 MD |
279 | (*nr_dequeues)++; |
280 | } | |
94aec122 | 281 | } while (snode); |
453629a9 MD |
282 | } |
283 | ||
284 | void show_usage(int argc, char **argv) | |
285 | { | |
286 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]); | |
287 | printf(" [-d delay] (enqueuer period (in loops))"); | |
288 | printf(" [-c duration] (dequeuer period (in loops))"); | |
289 | printf(" [-v] (verbose output)"); | |
290 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
291 | printf("\n"); | |
292 | } | |
293 | ||
294 | int main(int argc, char **argv) | |
295 | { | |
296 | int err; | |
297 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
298 | void *tret; | |
299 | unsigned long long *count_enqueuer, *count_dequeuer; | |
300 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
301 | unsigned long long tot_successful_enqueues = 0, | |
302 | tot_successful_dequeues = 0; | |
303 | unsigned long long end_dequeues = 0; | |
304 | int i, a; | |
305 | ||
306 | if (argc < 4) { | |
307 | show_usage(argc, argv); | |
308 | return -1; | |
309 | } | |
310 | ||
311 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
312 | if (err != 1) { | |
313 | show_usage(argc, argv); | |
314 | return -1; | |
315 | } | |
316 | ||
317 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
318 | if (err != 1) { | |
319 | show_usage(argc, argv); | |
320 | return -1; | |
321 | } | |
322 | ||
323 | err = sscanf(argv[3], "%lu", &duration); | |
324 | if (err != 1) { | |
325 | show_usage(argc, argv); | |
326 | return -1; | |
327 | } | |
328 | ||
329 | for (i = 4; i < argc; i++) { | |
330 | if (argv[i][0] != '-') | |
331 | continue; | |
332 | switch (argv[i][1]) { | |
333 | case 'a': | |
334 | if (argc < i + 2) { | |
335 | show_usage(argc, argv); | |
336 | return -1; | |
337 | } | |
338 | a = atoi(argv[++i]); | |
339 | cpu_affinities[next_aff++] = a; | |
340 | use_affinity = 1; | |
341 | printf_verbose("Adding CPU %d affinity\n", a); | |
342 | break; | |
343 | case 'c': | |
344 | if (argc < i + 2) { | |
345 | show_usage(argc, argv); | |
346 | return -1; | |
347 | } | |
348 | rduration = atol(argv[++i]); | |
349 | break; | |
350 | case 'd': | |
351 | if (argc < i + 2) { | |
352 | show_usage(argc, argv); | |
353 | return -1; | |
354 | } | |
355 | wdelay = atol(argv[++i]); | |
356 | break; | |
357 | case 'v': | |
358 | verbose_mode = 1; | |
359 | break; | |
360 | } | |
361 | } | |
362 | ||
363 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
364 | "%u dequeuers.\n", | |
365 | duration, nr_enqueuers, nr_dequeuers); | |
366 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
367 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
368 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
6ff854ac MD |
369 | "main", (unsigned long) pthread_self(), |
370 | (unsigned long) gettid()); | |
453629a9 MD |
371 | |
372 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
373 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
374 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
375 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
6e5f88cf | 376 | cds_lfq_init_rcu(&q, call_rcu); |
94aec122 | 377 | err = create_all_cpu_call_rcu_data(0); |
8bcbd94a MD |
378 | if (err) { |
379 | printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n"); | |
380 | } | |
453629a9 MD |
381 | |
382 | next_aff = 0; | |
383 | ||
384 | for (i = 0; i < nr_enqueuers; i++) { | |
385 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
386 | &count_enqueuer[2 * i]); | |
387 | if (err != 0) | |
388 | exit(1); | |
389 | } | |
390 | for (i = 0; i < nr_dequeuers; i++) { | |
391 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
392 | &count_dequeuer[2 * i]); | |
393 | if (err != 0) | |
394 | exit(1); | |
395 | } | |
396 | ||
5481ddb3 | 397 | cmm_smp_mb(); |
453629a9 MD |
398 | |
399 | test_go = 1; | |
400 | ||
401 | for (i = 0; i < duration; i++) { | |
402 | sleep(1); | |
403 | if (verbose_mode) | |
404 | write (1, ".", 1); | |
405 | } | |
406 | ||
407 | test_stop = 1; | |
408 | ||
409 | for (i = 0; i < nr_enqueuers; i++) { | |
410 | err = pthread_join(tid_enqueuer[i], &tret); | |
411 | if (err != 0) | |
412 | exit(1); | |
413 | tot_enqueues += count_enqueuer[2 * i]; | |
414 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
415 | } | |
416 | for (i = 0; i < nr_dequeuers; i++) { | |
417 | err = pthread_join(tid_dequeuer[i], &tret); | |
418 | if (err != 0) | |
419 | exit(1); | |
420 | tot_dequeues += count_dequeuer[2 * i]; | |
421 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
422 | } | |
423 | ||
424 | test_end(&q, &end_dequeues); | |
e17d9985 MD |
425 | err = cds_lfq_destroy_rcu(&q); |
426 | assert(!err); | |
453629a9 MD |
427 | |
428 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
429 | tot_enqueues, tot_dequeues); | |
430 | printf_verbose("total number of successful enqueues : %llu, " | |
431 | "successful dequeues %llu\n", | |
432 | tot_successful_enqueues, tot_successful_dequeues); | |
433 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
434 | "nr_dequeuers %3u " | |
435 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
436 | "successful enqueues %12llu successful dequeues %12llu " | |
437 | "end_dequeues %llu nr_ops %12llu\n", | |
438 | argv[0], duration, nr_enqueuers, wdelay, | |
439 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
440 | tot_successful_enqueues, | |
441 | tot_successful_dequeues, end_dequeues, | |
442 | tot_enqueues + tot_dequeues); | |
443 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
444 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
445 | "succ. dequeues + end dequeues %llu.\n", | |
446 | tot_successful_enqueues, | |
447 | tot_successful_dequeues + end_dequeues); | |
448 | ||
0938c541 | 449 | free_all_cpu_call_rcu_data(); |
453629a9 MD |
450 | free(count_enqueuer); |
451 | free(count_dequeuer); | |
452 | free(tid_enqueuer); | |
453 | free(tid_dequeuer); | |
454 | return 0; | |
455 | } |