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 | ||
453629a9 MD |
24 | #include <stdio.h> |
25 | #include <pthread.h> | |
26 | #include <stdlib.h> | |
27 | #include <stdint.h> | |
28 | #include <stdbool.h> | |
29 | #include <string.h> | |
30 | #include <sys/types.h> | |
31 | #include <sys/wait.h> | |
32 | #include <unistd.h> | |
33 | #include <stdio.h> | |
34 | #include <assert.h> | |
453629a9 MD |
35 | #include <errno.h> |
36 | ||
37 | #include <urcu/arch.h> | |
bd252a04 | 38 | #include <urcu/tls-compat.h> |
2953b501 | 39 | #include "cpuset.h" |
94df6318 | 40 | #include "thread-id.h" |
65fcc7e9 | 41 | |
453629a9 MD |
42 | /* hardcoded number of CPUs */ |
43 | #define NR_CPUS 16384 | |
44 | ||
453629a9 MD |
45 | #ifndef DYNAMIC_LINK_TEST |
46 | #define _LGPL_SOURCE | |
47 | #endif | |
48 | #include <urcu.h> | |
e5a7d709 | 49 | #include <urcu/cds.h> |
453629a9 MD |
50 | |
51 | static volatile int test_go, test_stop; | |
52 | ||
53 | static unsigned long rduration; | |
54 | ||
55 | static unsigned long duration; | |
56 | ||
57 | /* read-side C.S. duration, in loops */ | |
58 | static unsigned long wdelay; | |
59 | ||
ab0aacbe | 60 | static inline void loop_sleep(unsigned long loops) |
453629a9 | 61 | { |
ab0aacbe | 62 | while (loops-- != 0) |
06f22bdb | 63 | caa_cpu_relax(); |
453629a9 MD |
64 | } |
65 | ||
66 | static int verbose_mode; | |
67 | ||
68 | #define printf_verbose(fmt, args...) \ | |
69 | do { \ | |
70 | if (verbose_mode) \ | |
71 | printf(fmt, args); \ | |
72 | } while (0) | |
73 | ||
74 | static unsigned int cpu_affinities[NR_CPUS]; | |
75 | static unsigned int next_aff = 0; | |
76 | static int use_affinity = 0; | |
77 | ||
78 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
79 | ||
453629a9 MD |
80 | static void set_affinity(void) |
81 | { | |
95bc7fb9 | 82 | #if HAVE_SCHED_SETAFFINITY |
453629a9 | 83 | cpu_set_t mask; |
95bc7fb9 MD |
84 | int cpu, ret; |
85 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
453629a9 MD |
86 | |
87 | if (!use_affinity) | |
88 | return; | |
89 | ||
90 | #if HAVE_SCHED_SETAFFINITY | |
91 | ret = pthread_mutex_lock(&affinity_mutex); | |
92 | if (ret) { | |
93 | perror("Error in pthread mutex lock"); | |
94 | exit(-1); | |
95 | } | |
96 | cpu = cpu_affinities[next_aff++]; | |
97 | ret = pthread_mutex_unlock(&affinity_mutex); | |
98 | if (ret) { | |
99 | perror("Error in pthread mutex unlock"); | |
100 | exit(-1); | |
101 | } | |
102 | ||
103 | CPU_ZERO(&mask); | |
104 | CPU_SET(cpu, &mask); | |
105 | #if SCHED_SETAFFINITY_ARGS == 2 | |
106 | sched_setaffinity(0, &mask); | |
107 | #else | |
108 | sched_setaffinity(0, sizeof(mask), &mask); | |
109 | #endif | |
110 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
111 | } | |
112 | ||
113 | /* | |
114 | * returns 0 if test should end. | |
115 | */ | |
116 | static int test_duration_dequeue(void) | |
117 | { | |
118 | return !test_stop; | |
119 | } | |
120 | ||
121 | static int test_duration_enqueue(void) | |
122 | { | |
123 | return !test_stop; | |
124 | } | |
125 | ||
bd252a04 MD |
126 | static DEFINE_URCU_TLS(unsigned long long, nr_dequeues); |
127 | static DEFINE_URCU_TLS(unsigned long long, nr_enqueues); | |
453629a9 | 128 | |
bd252a04 MD |
129 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues); |
130 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues); | |
453629a9 MD |
131 | |
132 | static unsigned int nr_enqueuers; | |
133 | static unsigned int nr_dequeuers; | |
134 | ||
94aec122 MD |
135 | struct test { |
136 | struct cds_lfq_node_rcu list; | |
137 | struct rcu_head rcu; | |
138 | }; | |
139 | ||
16aa9ee8 | 140 | static struct cds_lfq_queue_rcu q; |
453629a9 MD |
141 | |
142 | void *thr_enqueuer(void *_count) | |
143 | { | |
144 | unsigned long long *count = _count; | |
145 | ||
94df6318 MD |
146 | printf_verbose("thread_begin %s, tid %lu\n", |
147 | "enqueuer", urcu_get_thread_id()); | |
453629a9 MD |
148 | |
149 | set_affinity(); | |
150 | ||
151 | rcu_register_thread(); | |
152 | ||
153 | while (!test_go) | |
154 | { | |
155 | } | |
5481ddb3 | 156 | cmm_smp_mb(); |
453629a9 MD |
157 | |
158 | for (;;) { | |
94aec122 | 159 | struct test *node = malloc(sizeof(*node)); |
453629a9 MD |
160 | if (!node) |
161 | goto fail; | |
94aec122 | 162 | cds_lfq_node_init_rcu(&node->list); |
6e5f88cf | 163 | rcu_read_lock(); |
94aec122 | 164 | cds_lfq_enqueue_rcu(&q, &node->list); |
6e5f88cf | 165 | rcu_read_unlock(); |
bd252a04 | 166 | URCU_TLS(nr_successful_enqueues)++; |
453629a9 | 167 | |
a0b7f7ea | 168 | if (caa_unlikely(wdelay)) |
453629a9 MD |
169 | loop_sleep(wdelay); |
170 | fail: | |
bd252a04 | 171 | URCU_TLS(nr_enqueues)++; |
a0b7f7ea | 172 | if (caa_unlikely(!test_duration_enqueue())) |
453629a9 MD |
173 | break; |
174 | } | |
175 | ||
176 | rcu_unregister_thread(); | |
177 | ||
bd252a04 MD |
178 | count[0] = URCU_TLS(nr_enqueues); |
179 | count[1] = URCU_TLS(nr_successful_enqueues); | |
94df6318 MD |
180 | printf_verbose("enqueuer thread_end, tid %lu, " |
181 | "enqueues %llu successful_enqueues %llu\n", | |
182 | urcu_get_thread_id(), | |
183 | URCU_TLS(nr_enqueues), | |
184 | URCU_TLS(nr_successful_enqueues)); | |
453629a9 MD |
185 | return ((void*)1); |
186 | ||
187 | } | |
188 | ||
94aec122 MD |
189 | static |
190 | void free_node_cb(struct rcu_head *head) | |
191 | { | |
192 | struct test *node = | |
193 | caa_container_of(head, struct test, rcu); | |
194 | free(node); | |
195 | } | |
196 | ||
453629a9 MD |
197 | void *thr_dequeuer(void *_count) |
198 | { | |
199 | unsigned long long *count = _count; | |
200 | ||
94df6318 MD |
201 | printf_verbose("thread_begin %s, tid %lu\n", |
202 | "dequeuer", urcu_get_thread_id()); | |
453629a9 MD |
203 | |
204 | set_affinity(); | |
205 | ||
453629a9 MD |
206 | rcu_register_thread(); |
207 | ||
208 | while (!test_go) | |
209 | { | |
210 | } | |
5481ddb3 | 211 | cmm_smp_mb(); |
453629a9 MD |
212 | |
213 | for (;;) { | |
94aec122 | 214 | struct cds_lfq_node_rcu *qnode; |
d9b52143 | 215 | |
6e5f88cf | 216 | rcu_read_lock(); |
94aec122 | 217 | qnode = cds_lfq_dequeue_rcu(&q); |
6e5f88cf MD |
218 | rcu_read_unlock(); |
219 | ||
747b5930 LJ |
220 | if (qnode) { |
221 | struct test *node; | |
222 | ||
223 | node = caa_container_of(qnode, struct test, list); | |
94aec122 | 224 | call_rcu(&node->rcu, free_node_cb); |
bd252a04 | 225 | URCU_TLS(nr_successful_dequeues)++; |
453629a9 MD |
226 | } |
227 | ||
bd252a04 | 228 | URCU_TLS(nr_dequeues)++; |
a0b7f7ea | 229 | if (caa_unlikely(!test_duration_dequeue())) |
453629a9 | 230 | break; |
a0b7f7ea | 231 | if (caa_unlikely(rduration)) |
453629a9 MD |
232 | loop_sleep(rduration); |
233 | } | |
234 | ||
235 | rcu_unregister_thread(); | |
94df6318 MD |
236 | printf_verbose("dequeuer thread_end, tid %lu, " |
237 | "dequeues %llu, successful_dequeues %llu\n", | |
238 | urcu_get_thread_id(), | |
239 | URCU_TLS(nr_dequeues), | |
240 | URCU_TLS(nr_successful_dequeues)); | |
bd252a04 MD |
241 | count[0] = URCU_TLS(nr_dequeues); |
242 | count[1] = URCU_TLS(nr_successful_dequeues); | |
453629a9 MD |
243 | return ((void*)2); |
244 | } | |
245 | ||
16aa9ee8 | 246 | void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues) |
453629a9 | 247 | { |
94aec122 | 248 | struct cds_lfq_node_rcu *snode; |
453629a9 MD |
249 | |
250 | do { | |
94aec122 MD |
251 | snode = cds_lfq_dequeue_rcu(q); |
252 | if (snode) { | |
253 | struct test *node; | |
254 | ||
255 | node = caa_container_of(snode, struct test, list); | |
e17d9985 | 256 | free(node); /* no more concurrent access */ |
453629a9 MD |
257 | (*nr_dequeues)++; |
258 | } | |
94aec122 | 259 | } while (snode); |
453629a9 MD |
260 | } |
261 | ||
262 | void show_usage(int argc, char **argv) | |
263 | { | |
06637138 MD |
264 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n", |
265 | argv[0]); | |
266 | printf("OPTIONS:\n"); | |
267 | printf(" [-d delay] (enqueuer period (in loops))\n"); | |
268 | printf(" [-c duration] (dequeuer period (in loops))\n"); | |
269 | printf(" [-v] (verbose output)\n"); | |
270 | printf(" [-a cpu#] [-a cpu#]... (affinity)\n"); | |
453629a9 MD |
271 | printf("\n"); |
272 | } | |
273 | ||
274 | int main(int argc, char **argv) | |
275 | { | |
276 | int err; | |
277 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
278 | void *tret; | |
279 | unsigned long long *count_enqueuer, *count_dequeuer; | |
280 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
281 | unsigned long long tot_successful_enqueues = 0, | |
282 | tot_successful_dequeues = 0; | |
283 | unsigned long long end_dequeues = 0; | |
284 | int i, a; | |
83e334d0 | 285 | unsigned int i_thr; |
453629a9 MD |
286 | |
287 | if (argc < 4) { | |
288 | show_usage(argc, argv); | |
289 | return -1; | |
290 | } | |
291 | ||
292 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
293 | if (err != 1) { | |
294 | show_usage(argc, argv); | |
295 | return -1; | |
296 | } | |
297 | ||
298 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
299 | if (err != 1) { | |
300 | show_usage(argc, argv); | |
301 | return -1; | |
302 | } | |
83e334d0 | 303 | |
453629a9 MD |
304 | err = sscanf(argv[3], "%lu", &duration); |
305 | if (err != 1) { | |
306 | show_usage(argc, argv); | |
307 | return -1; | |
308 | } | |
309 | ||
310 | for (i = 4; i < argc; i++) { | |
311 | if (argv[i][0] != '-') | |
312 | continue; | |
313 | switch (argv[i][1]) { | |
314 | case 'a': | |
315 | if (argc < i + 2) { | |
316 | show_usage(argc, argv); | |
317 | return -1; | |
318 | } | |
319 | a = atoi(argv[++i]); | |
320 | cpu_affinities[next_aff++] = a; | |
321 | use_affinity = 1; | |
322 | printf_verbose("Adding CPU %d affinity\n", a); | |
323 | break; | |
324 | case 'c': | |
325 | if (argc < i + 2) { | |
326 | show_usage(argc, argv); | |
327 | return -1; | |
328 | } | |
329 | rduration = atol(argv[++i]); | |
330 | break; | |
331 | case 'd': | |
332 | if (argc < i + 2) { | |
333 | show_usage(argc, argv); | |
334 | return -1; | |
335 | } | |
336 | wdelay = atol(argv[++i]); | |
337 | break; | |
338 | case 'v': | |
339 | verbose_mode = 1; | |
340 | break; | |
341 | } | |
342 | } | |
343 | ||
344 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
345 | "%u dequeuers.\n", | |
346 | duration, nr_enqueuers, nr_dequeuers); | |
347 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
348 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
94df6318 MD |
349 | printf_verbose("thread %-6s, tid %lu\n", |
350 | "main", urcu_get_thread_id()); | |
453629a9 | 351 | |
9aa14175 MD |
352 | tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer)); |
353 | tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer)); | |
354 | count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer)); | |
355 | count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer)); | |
6e5f88cf | 356 | cds_lfq_init_rcu(&q, call_rcu); |
94aec122 | 357 | err = create_all_cpu_call_rcu_data(0); |
8bcbd94a MD |
358 | if (err) { |
359 | printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n"); | |
360 | } | |
453629a9 MD |
361 | |
362 | next_aff = 0; | |
363 | ||
83e334d0 MJ |
364 | for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) { |
365 | err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer, | |
366 | &count_enqueuer[2 * i_thr]); | |
453629a9 MD |
367 | if (err != 0) |
368 | exit(1); | |
369 | } | |
83e334d0 MJ |
370 | for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) { |
371 | err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer, | |
372 | &count_dequeuer[2 * i_thr]); | |
453629a9 MD |
373 | if (err != 0) |
374 | exit(1); | |
375 | } | |
376 | ||
5481ddb3 | 377 | cmm_smp_mb(); |
453629a9 MD |
378 | |
379 | test_go = 1; | |
380 | ||
83e334d0 | 381 | for (i_thr = 0; i_thr < duration; i_thr++) { |
453629a9 | 382 | sleep(1); |
4b665350 MD |
383 | if (verbose_mode) { |
384 | fwrite(".", sizeof(char), 1, stdout); | |
385 | fflush(stdout); | |
386 | } | |
453629a9 MD |
387 | } |
388 | ||
389 | test_stop = 1; | |
390 | ||
83e334d0 MJ |
391 | for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) { |
392 | err = pthread_join(tid_enqueuer[i_thr], &tret); | |
453629a9 MD |
393 | if (err != 0) |
394 | exit(1); | |
83e334d0 MJ |
395 | tot_enqueues += count_enqueuer[2 * i_thr]; |
396 | tot_successful_enqueues += count_enqueuer[2 * i_thr + 1]; | |
453629a9 | 397 | } |
83e334d0 MJ |
398 | for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) { |
399 | err = pthread_join(tid_dequeuer[i_thr], &tret); | |
453629a9 MD |
400 | if (err != 0) |
401 | exit(1); | |
83e334d0 MJ |
402 | tot_dequeues += count_dequeuer[2 * i_thr]; |
403 | tot_successful_dequeues += count_dequeuer[2 * i_thr + 1]; | |
453629a9 | 404 | } |
83e334d0 | 405 | |
453629a9 | 406 | test_end(&q, &end_dequeues); |
e17d9985 MD |
407 | err = cds_lfq_destroy_rcu(&q); |
408 | assert(!err); | |
453629a9 MD |
409 | |
410 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
411 | tot_enqueues, tot_dequeues); | |
412 | printf_verbose("total number of successful enqueues : %llu, " | |
413 | "successful dequeues %llu\n", | |
414 | tot_successful_enqueues, tot_successful_dequeues); | |
415 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
416 | "nr_dequeuers %3u " | |
417 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
418 | "successful enqueues %12llu successful dequeues %12llu " | |
419 | "end_dequeues %llu nr_ops %12llu\n", | |
420 | argv[0], duration, nr_enqueuers, wdelay, | |
421 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
422 | tot_successful_enqueues, | |
423 | tot_successful_dequeues, end_dequeues, | |
424 | tot_enqueues + tot_dequeues); | |
425 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
426 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
427 | "succ. dequeues + end dequeues %llu.\n", | |
428 | tot_successful_enqueues, | |
429 | tot_successful_dequeues + end_dequeues); | |
430 | ||
0938c541 | 431 | free_all_cpu_call_rcu_data(); |
453629a9 MD |
432 | free(count_enqueuer); |
433 | free(count_dequeuer); | |
434 | free(tid_enqueuer); | |
435 | free(tid_dequeuer); | |
83e334d0 | 436 | |
453629a9 MD |
437 | return 0; |
438 | } |