Commit | Line | Data |
---|---|---|
453629a9 MD |
1 | /* |
2 | * test_urcu_lfs.c | |
3 | * | |
4 | * Userspace RCU library - example RCU-based lock-free stack | |
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 <sched.h> |
38 | #include <errno.h> | |
39 | ||
40 | #include <urcu/arch.h> | |
bd252a04 | 41 | #include <urcu/tls-compat.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 | ||
100 | #ifndef HAVE_CPU_SET_T | |
101 | typedef unsigned long cpu_set_t; | |
102 | # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0) | |
103 | # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0) | |
104 | #endif | |
105 | ||
106 | static void set_affinity(void) | |
107 | { | |
108 | cpu_set_t mask; | |
109 | int cpu; | |
110 | int ret; | |
111 | ||
112 | if (!use_affinity) | |
113 | return; | |
114 | ||
115 | #if HAVE_SCHED_SETAFFINITY | |
116 | ret = pthread_mutex_lock(&affinity_mutex); | |
117 | if (ret) { | |
118 | perror("Error in pthread mutex lock"); | |
119 | exit(-1); | |
120 | } | |
121 | cpu = cpu_affinities[next_aff++]; | |
122 | ret = pthread_mutex_unlock(&affinity_mutex); | |
123 | if (ret) { | |
124 | perror("Error in pthread mutex unlock"); | |
125 | exit(-1); | |
126 | } | |
127 | ||
128 | CPU_ZERO(&mask); | |
129 | CPU_SET(cpu, &mask); | |
130 | #if SCHED_SETAFFINITY_ARGS == 2 | |
131 | sched_setaffinity(0, &mask); | |
132 | #else | |
133 | sched_setaffinity(0, sizeof(mask), &mask); | |
134 | #endif | |
135 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
136 | } | |
137 | ||
138 | /* | |
139 | * returns 0 if test should end. | |
140 | */ | |
141 | static int test_duration_dequeue(void) | |
142 | { | |
143 | return !test_stop; | |
144 | } | |
145 | ||
146 | static int test_duration_enqueue(void) | |
147 | { | |
148 | return !test_stop; | |
149 | } | |
150 | ||
bd252a04 MD |
151 | static DEFINE_URCU_TLS(unsigned long long, nr_dequeues); |
152 | static DEFINE_URCU_TLS(unsigned long long, nr_enqueues); | |
453629a9 | 153 | |
bd252a04 MD |
154 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues); |
155 | static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues); | |
453629a9 MD |
156 | |
157 | static unsigned int nr_enqueuers; | |
158 | static unsigned int nr_dequeuers; | |
159 | ||
94aec122 MD |
160 | struct test { |
161 | struct cds_lfs_node_rcu list; | |
162 | struct rcu_head rcu; | |
163 | }; | |
164 | ||
16aa9ee8 | 165 | static struct cds_lfs_stack_rcu s; |
453629a9 MD |
166 | |
167 | void *thr_enqueuer(void *_count) | |
168 | { | |
169 | unsigned long long *count = _count; | |
170 | ||
171 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
172 | "enqueuer", pthread_self(), (unsigned long)gettid()); | |
173 | ||
174 | set_affinity(); | |
175 | ||
176 | rcu_register_thread(); | |
177 | ||
178 | while (!test_go) | |
179 | { | |
180 | } | |
5481ddb3 | 181 | cmm_smp_mb(); |
453629a9 MD |
182 | |
183 | for (;;) { | |
94aec122 | 184 | struct test *node = malloc(sizeof(*node)); |
453629a9 MD |
185 | if (!node) |
186 | goto fail; | |
94aec122 | 187 | cds_lfs_node_init_rcu(&node->list); |
d9b52143 | 188 | /* No rcu read-side is needed for push */ |
94aec122 | 189 | cds_lfs_push_rcu(&s, &node->list); |
bd252a04 | 190 | URCU_TLS(nr_successful_enqueues)++; |
453629a9 | 191 | |
a0b7f7ea | 192 | if (caa_unlikely(wdelay)) |
453629a9 MD |
193 | loop_sleep(wdelay); |
194 | fail: | |
bd252a04 | 195 | URCU_TLS(nr_enqueues)++; |
a0b7f7ea | 196 | if (caa_unlikely(!test_duration_enqueue())) |
453629a9 MD |
197 | break; |
198 | } | |
199 | ||
200 | rcu_unregister_thread(); | |
201 | ||
bd252a04 MD |
202 | count[0] = URCU_TLS(nr_enqueues); |
203 | count[1] = URCU_TLS(nr_successful_enqueues); | |
453629a9 MD |
204 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " |
205 | "enqueues %llu successful_enqueues %llu\n", | |
bd252a04 MD |
206 | pthread_self(), (unsigned long)gettid(), |
207 | URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues)); | |
453629a9 MD |
208 | return ((void*)1); |
209 | ||
210 | } | |
211 | ||
94aec122 MD |
212 | static |
213 | void free_node_cb(struct rcu_head *head) | |
214 | { | |
215 | struct test *node = | |
216 | caa_container_of(head, struct test, rcu); | |
217 | free(node); | |
218 | } | |
219 | ||
453629a9 MD |
220 | void *thr_dequeuer(void *_count) |
221 | { | |
222 | unsigned long long *count = _count; | |
223 | ||
224 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
225 | "dequeuer", pthread_self(), (unsigned long)gettid()); | |
226 | ||
227 | set_affinity(); | |
228 | ||
453629a9 MD |
229 | rcu_register_thread(); |
230 | ||
231 | while (!test_go) | |
232 | { | |
233 | } | |
5481ddb3 | 234 | cmm_smp_mb(); |
453629a9 MD |
235 | |
236 | for (;;) { | |
94aec122 | 237 | struct cds_lfs_node_rcu *snode; |
453629a9 | 238 | |
6e5f88cf | 239 | rcu_read_lock(); |
94aec122 | 240 | snode = cds_lfs_pop_rcu(&s); |
6e5f88cf | 241 | rcu_read_unlock(); |
f3eb6ef1 LJ |
242 | if (snode) { |
243 | struct test *node; | |
244 | ||
245 | node = caa_container_of(snode, struct test, list); | |
94aec122 | 246 | call_rcu(&node->rcu, free_node_cb); |
bd252a04 | 247 | URCU_TLS(nr_successful_dequeues)++; |
453629a9 | 248 | } |
bd252a04 | 249 | URCU_TLS(nr_dequeues)++; |
a0b7f7ea | 250 | if (caa_unlikely(!test_duration_dequeue())) |
453629a9 | 251 | break; |
a0b7f7ea | 252 | if (caa_unlikely(rduration)) |
453629a9 MD |
253 | loop_sleep(rduration); |
254 | } | |
255 | ||
256 | rcu_unregister_thread(); | |
453629a9 MD |
257 | |
258 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " | |
259 | "dequeues %llu, successful_dequeues %llu\n", | |
bd252a04 MD |
260 | pthread_self(), (unsigned long)gettid(), |
261 | URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues)); | |
262 | count[0] = URCU_TLS(nr_dequeues); | |
263 | count[1] = URCU_TLS(nr_successful_dequeues); | |
453629a9 MD |
264 | return ((void*)2); |
265 | } | |
266 | ||
16aa9ee8 | 267 | void test_end(struct cds_lfs_stack_rcu *s, unsigned long long *nr_dequeues) |
453629a9 | 268 | { |
94aec122 | 269 | struct cds_lfs_node_rcu *snode; |
453629a9 MD |
270 | |
271 | do { | |
94aec122 MD |
272 | snode = cds_lfs_pop_rcu(s); |
273 | if (snode) { | |
274 | struct test *node; | |
275 | ||
276 | node = caa_container_of(snode, struct test, list); | |
453629a9 MD |
277 | free(node); |
278 | (*nr_dequeues)++; | |
279 | } | |
94aec122 | 280 | } while (snode); |
453629a9 MD |
281 | } |
282 | ||
283 | void show_usage(int argc, char **argv) | |
284 | { | |
285 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]); | |
286 | printf(" [-d delay] (enqueuer period (in loops))"); | |
287 | printf(" [-c duration] (dequeuer period (in loops))"); | |
288 | printf(" [-v] (verbose output)"); | |
289 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
290 | printf("\n"); | |
291 | } | |
292 | ||
293 | int main(int argc, char **argv) | |
294 | { | |
295 | int err; | |
296 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
297 | void *tret; | |
298 | unsigned long long *count_enqueuer, *count_dequeuer; | |
299 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
300 | unsigned long long tot_successful_enqueues = 0, | |
301 | tot_successful_dequeues = 0; | |
302 | unsigned long long end_dequeues = 0; | |
303 | int i, a; | |
304 | ||
305 | if (argc < 4) { | |
306 | show_usage(argc, argv); | |
307 | return -1; | |
308 | } | |
309 | ||
310 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
311 | if (err != 1) { | |
312 | show_usage(argc, argv); | |
313 | return -1; | |
314 | } | |
315 | ||
316 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
317 | if (err != 1) { | |
318 | show_usage(argc, argv); | |
319 | return -1; | |
320 | } | |
321 | ||
322 | err = sscanf(argv[3], "%lu", &duration); | |
323 | if (err != 1) { | |
324 | show_usage(argc, argv); | |
325 | return -1; | |
326 | } | |
327 | ||
328 | for (i = 4; i < argc; i++) { | |
329 | if (argv[i][0] != '-') | |
330 | continue; | |
331 | switch (argv[i][1]) { | |
332 | case 'a': | |
333 | if (argc < i + 2) { | |
334 | show_usage(argc, argv); | |
335 | return -1; | |
336 | } | |
337 | a = atoi(argv[++i]); | |
338 | cpu_affinities[next_aff++] = a; | |
339 | use_affinity = 1; | |
340 | printf_verbose("Adding CPU %d affinity\n", a); | |
341 | break; | |
342 | case 'c': | |
343 | if (argc < i + 2) { | |
344 | show_usage(argc, argv); | |
345 | return -1; | |
346 | } | |
347 | rduration = atol(argv[++i]); | |
348 | break; | |
349 | case 'd': | |
350 | if (argc < i + 2) { | |
351 | show_usage(argc, argv); | |
352 | return -1; | |
353 | } | |
354 | wdelay = atol(argv[++i]); | |
355 | break; | |
356 | case 'v': | |
357 | verbose_mode = 1; | |
358 | break; | |
359 | } | |
360 | } | |
361 | ||
362 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
363 | "%u dequeuers.\n", | |
364 | duration, nr_enqueuers, nr_dequeuers); | |
365 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
366 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
367 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
368 | "main", pthread_self(), (unsigned long)gettid()); | |
369 | ||
370 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
371 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
372 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
373 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
16aa9ee8 | 374 | cds_lfs_init_rcu(&s); |
94aec122 | 375 | err = create_all_cpu_call_rcu_data(0); |
8bcbd94a MD |
376 | if (err) { |
377 | printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n"); | |
378 | } | |
453629a9 MD |
379 | |
380 | next_aff = 0; | |
381 | ||
382 | for (i = 0; i < nr_enqueuers; i++) { | |
383 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
384 | &count_enqueuer[2 * i]); | |
385 | if (err != 0) | |
386 | exit(1); | |
387 | } | |
388 | for (i = 0; i < nr_dequeuers; i++) { | |
389 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
390 | &count_dequeuer[2 * i]); | |
391 | if (err != 0) | |
392 | exit(1); | |
393 | } | |
394 | ||
5481ddb3 | 395 | cmm_smp_mb(); |
453629a9 MD |
396 | |
397 | test_go = 1; | |
398 | ||
399 | for (i = 0; i < duration; i++) { | |
400 | sleep(1); | |
401 | if (verbose_mode) | |
402 | write (1, ".", 1); | |
403 | } | |
404 | ||
405 | test_stop = 1; | |
406 | ||
407 | for (i = 0; i < nr_enqueuers; i++) { | |
408 | err = pthread_join(tid_enqueuer[i], &tret); | |
409 | if (err != 0) | |
410 | exit(1); | |
411 | tot_enqueues += count_enqueuer[2 * i]; | |
412 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
413 | } | |
414 | for (i = 0; i < nr_dequeuers; i++) { | |
415 | err = pthread_join(tid_dequeuer[i], &tret); | |
416 | if (err != 0) | |
417 | exit(1); | |
418 | tot_dequeues += count_dequeuer[2 * i]; | |
419 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
420 | } | |
421 | ||
422 | test_end(&s, &end_dequeues); | |
423 | ||
424 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
425 | tot_enqueues, tot_dequeues); | |
426 | printf_verbose("total number of successful enqueues : %llu, " | |
427 | "successful dequeues %llu\n", | |
428 | tot_successful_enqueues, tot_successful_dequeues); | |
429 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
430 | "nr_dequeuers %3u " | |
431 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
432 | "successful enqueues %12llu successful dequeues %12llu " | |
433 | "end_dequeues %llu nr_ops %12llu\n", | |
434 | argv[0], duration, nr_enqueuers, wdelay, | |
435 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
436 | tot_successful_enqueues, | |
437 | tot_successful_dequeues, end_dequeues, | |
438 | tot_enqueues + tot_dequeues); | |
439 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
440 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
441 | "succ. dequeues + end dequeues %llu.\n", | |
442 | tot_successful_enqueues, | |
443 | tot_successful_dequeues + end_dequeues); | |
444 | ||
0938c541 | 445 | free_all_cpu_call_rcu_data(); |
453629a9 MD |
446 | free(count_enqueuer); |
447 | free(count_dequeuer); | |
448 | free(tid_enqueuer); | |
449 | free(tid_dequeuer); | |
450 | return 0; | |
451 | } |