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> | |
37 | #include <sys/syscall.h> | |
38 | #include <sched.h> | |
39 | #include <errno.h> | |
40 | ||
41 | #include <urcu/arch.h> | |
42 | ||
43 | /* hardcoded number of CPUs */ | |
44 | #define NR_CPUS 16384 | |
45 | ||
46 | #if defined(_syscall0) | |
47 | _syscall0(pid_t, gettid) | |
48 | #elif defined(__NR_gettid) | |
49 | static inline pid_t gettid(void) | |
50 | { | |
51 | return syscall(__NR_gettid); | |
52 | } | |
53 | #else | |
54 | #warning "use pid as tid" | |
55 | static inline pid_t gettid(void) | |
56 | { | |
57 | return getpid(); | |
58 | } | |
59 | #endif | |
60 | ||
61 | #ifndef DYNAMIC_LINK_TEST | |
62 | #define _LGPL_SOURCE | |
63 | #endif | |
64 | #include <urcu.h> | |
65 | #include <urcu/rculfqueue.h> | |
66 | #include <urcu-defer.h> | |
67 | ||
68 | static volatile int test_go, test_stop; | |
69 | ||
70 | static unsigned long rduration; | |
71 | ||
72 | static unsigned long duration; | |
73 | ||
74 | /* read-side C.S. duration, in loops */ | |
75 | static unsigned long wdelay; | |
76 | ||
77 | static inline void loop_sleep(unsigned long l) | |
78 | { | |
79 | while(l-- != 0) | |
06f22bdb | 80 | caa_cpu_relax(); |
453629a9 MD |
81 | } |
82 | ||
83 | static int verbose_mode; | |
84 | ||
85 | #define printf_verbose(fmt, args...) \ | |
86 | do { \ | |
87 | if (verbose_mode) \ | |
88 | printf(fmt, args); \ | |
89 | } while (0) | |
90 | ||
91 | static unsigned int cpu_affinities[NR_CPUS]; | |
92 | static unsigned int next_aff = 0; | |
93 | static int use_affinity = 0; | |
94 | ||
95 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
96 | ||
97 | #ifndef HAVE_CPU_SET_T | |
98 | typedef unsigned long cpu_set_t; | |
99 | # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0) | |
100 | # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0) | |
101 | #endif | |
102 | ||
103 | static void set_affinity(void) | |
104 | { | |
105 | cpu_set_t mask; | |
106 | int cpu; | |
107 | int ret; | |
108 | ||
109 | if (!use_affinity) | |
110 | return; | |
111 | ||
112 | #if HAVE_SCHED_SETAFFINITY | |
113 | ret = pthread_mutex_lock(&affinity_mutex); | |
114 | if (ret) { | |
115 | perror("Error in pthread mutex lock"); | |
116 | exit(-1); | |
117 | } | |
118 | cpu = cpu_affinities[next_aff++]; | |
119 | ret = pthread_mutex_unlock(&affinity_mutex); | |
120 | if (ret) { | |
121 | perror("Error in pthread mutex unlock"); | |
122 | exit(-1); | |
123 | } | |
124 | ||
125 | CPU_ZERO(&mask); | |
126 | CPU_SET(cpu, &mask); | |
127 | #if SCHED_SETAFFINITY_ARGS == 2 | |
128 | sched_setaffinity(0, &mask); | |
129 | #else | |
130 | sched_setaffinity(0, sizeof(mask), &mask); | |
131 | #endif | |
132 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
133 | } | |
134 | ||
135 | /* | |
136 | * returns 0 if test should end. | |
137 | */ | |
138 | static int test_duration_dequeue(void) | |
139 | { | |
140 | return !test_stop; | |
141 | } | |
142 | ||
143 | static int test_duration_enqueue(void) | |
144 | { | |
145 | return !test_stop; | |
146 | } | |
147 | ||
148 | static unsigned long long __thread nr_dequeues; | |
149 | static unsigned long long __thread nr_enqueues; | |
150 | ||
151 | static unsigned long long __thread nr_successful_dequeues; | |
152 | static unsigned long long __thread nr_successful_enqueues; | |
153 | ||
154 | static unsigned int nr_enqueuers; | |
155 | static unsigned int nr_dequeuers; | |
156 | ||
16aa9ee8 | 157 | static struct cds_lfq_queue_rcu q; |
453629a9 MD |
158 | |
159 | void *thr_enqueuer(void *_count) | |
160 | { | |
161 | unsigned long long *count = _count; | |
162 | ||
163 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
164 | "enqueuer", pthread_self(), (unsigned long)gettid()); | |
165 | ||
166 | set_affinity(); | |
167 | ||
168 | rcu_register_thread(); | |
169 | ||
170 | while (!test_go) | |
171 | { | |
172 | } | |
5481ddb3 | 173 | cmm_smp_mb(); |
453629a9 MD |
174 | |
175 | for (;;) { | |
16aa9ee8 | 176 | struct cds_lfq_node_rcu *node = malloc(sizeof(*node)); |
453629a9 MD |
177 | if (!node) |
178 | goto fail; | |
16aa9ee8 DG |
179 | cds_lfq_node_init_rcu(node); |
180 | cds_lfq_enqueue_rcu(&q, node); | |
453629a9 MD |
181 | nr_successful_enqueues++; |
182 | ||
183 | if (unlikely(wdelay)) | |
184 | loop_sleep(wdelay); | |
185 | fail: | |
186 | nr_enqueues++; | |
187 | if (unlikely(!test_duration_enqueue())) | |
188 | break; | |
189 | } | |
190 | ||
191 | rcu_unregister_thread(); | |
192 | ||
193 | count[0] = nr_enqueues; | |
194 | count[1] = nr_successful_enqueues; | |
195 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " | |
196 | "enqueues %llu successful_enqueues %llu\n", | |
197 | pthread_self(), (unsigned long)gettid(), nr_enqueues, | |
198 | nr_successful_enqueues); | |
199 | return ((void*)1); | |
200 | ||
201 | } | |
202 | ||
203 | static void rcu_release_node(struct urcu_ref *ref) | |
204 | { | |
16aa9ee8 | 205 | struct cds_lfq_node_rcu *node = caa_container_of(ref, struct cds_lfq_node_rcu, ref); |
453629a9 MD |
206 | defer_rcu(free, node); |
207 | //synchronize_rcu(); | |
208 | //free(node); | |
209 | } | |
210 | ||
211 | void *thr_dequeuer(void *_count) | |
212 | { | |
213 | unsigned long long *count = _count; | |
9bfaf0ca | 214 | int ret; |
453629a9 MD |
215 | |
216 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
217 | "dequeuer", pthread_self(), (unsigned long)gettid()); | |
218 | ||
219 | set_affinity(); | |
220 | ||
6ade67d5 MD |
221 | ret = rcu_defer_register_thread(); |
222 | if (ret) { | |
223 | printf("Error in rcu_defer_register_thread\n"); | |
224 | exit(-1); | |
225 | } | |
453629a9 MD |
226 | rcu_register_thread(); |
227 | ||
228 | while (!test_go) | |
229 | { | |
230 | } | |
5481ddb3 | 231 | cmm_smp_mb(); |
453629a9 MD |
232 | |
233 | for (;;) { | |
16aa9ee8 | 234 | struct cds_lfq_node_rcu *node = cds_lfq_dequeue_rcu(&q, |
453629a9 MD |
235 | rcu_release_node); |
236 | ||
237 | if (node) { | |
238 | urcu_ref_put(&node->ref, rcu_release_node); | |
239 | nr_successful_dequeues++; | |
240 | } | |
241 | ||
242 | nr_dequeues++; | |
243 | if (unlikely(!test_duration_dequeue())) | |
244 | break; | |
245 | if (unlikely(rduration)) | |
246 | loop_sleep(rduration); | |
247 | } | |
248 | ||
249 | rcu_unregister_thread(); | |
250 | rcu_defer_unregister_thread(); | |
251 | ||
252 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " | |
253 | "dequeues %llu, successful_dequeues %llu\n", | |
254 | pthread_self(), (unsigned long)gettid(), nr_dequeues, | |
255 | nr_successful_dequeues); | |
256 | count[0] = nr_dequeues; | |
257 | count[1] = nr_successful_dequeues; | |
258 | return ((void*)2); | |
259 | } | |
260 | ||
261 | static void release_node(struct urcu_ref *ref) | |
262 | { | |
16aa9ee8 | 263 | struct cds_lfq_node_rcu *node = caa_container_of(ref, struct cds_lfq_node_rcu, ref); |
453629a9 MD |
264 | free(node); |
265 | } | |
266 | ||
16aa9ee8 | 267 | void test_end(struct cds_lfq_queue_rcu *q, unsigned long long *nr_dequeues) |
453629a9 | 268 | { |
16aa9ee8 | 269 | struct cds_lfq_node_rcu *node; |
453629a9 MD |
270 | |
271 | do { | |
16aa9ee8 | 272 | node = cds_lfq_dequeue_rcu(q, release_node); |
453629a9 MD |
273 | if (node) { |
274 | urcu_ref_put(&node->ref, release_node); | |
275 | (*nr_dequeues)++; | |
276 | } | |
277 | } while (node); | |
278 | } | |
279 | ||
280 | void show_usage(int argc, char **argv) | |
281 | { | |
282 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]); | |
283 | printf(" [-d delay] (enqueuer period (in loops))"); | |
284 | printf(" [-c duration] (dequeuer period (in loops))"); | |
285 | printf(" [-v] (verbose output)"); | |
286 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
287 | printf("\n"); | |
288 | } | |
289 | ||
290 | int main(int argc, char **argv) | |
291 | { | |
292 | int err; | |
293 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
294 | void *tret; | |
295 | unsigned long long *count_enqueuer, *count_dequeuer; | |
296 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
297 | unsigned long long tot_successful_enqueues = 0, | |
298 | tot_successful_dequeues = 0; | |
299 | unsigned long long end_dequeues = 0; | |
300 | int i, a; | |
301 | ||
302 | if (argc < 4) { | |
303 | show_usage(argc, argv); | |
304 | return -1; | |
305 | } | |
306 | ||
307 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
308 | if (err != 1) { | |
309 | show_usage(argc, argv); | |
310 | return -1; | |
311 | } | |
312 | ||
313 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
314 | if (err != 1) { | |
315 | show_usage(argc, argv); | |
316 | return -1; | |
317 | } | |
318 | ||
319 | err = sscanf(argv[3], "%lu", &duration); | |
320 | if (err != 1) { | |
321 | show_usage(argc, argv); | |
322 | return -1; | |
323 | } | |
324 | ||
325 | for (i = 4; i < argc; i++) { | |
326 | if (argv[i][0] != '-') | |
327 | continue; | |
328 | switch (argv[i][1]) { | |
329 | case 'a': | |
330 | if (argc < i + 2) { | |
331 | show_usage(argc, argv); | |
332 | return -1; | |
333 | } | |
334 | a = atoi(argv[++i]); | |
335 | cpu_affinities[next_aff++] = a; | |
336 | use_affinity = 1; | |
337 | printf_verbose("Adding CPU %d affinity\n", a); | |
338 | break; | |
339 | case 'c': | |
340 | if (argc < i + 2) { | |
341 | show_usage(argc, argv); | |
342 | return -1; | |
343 | } | |
344 | rduration = atol(argv[++i]); | |
345 | break; | |
346 | case 'd': | |
347 | if (argc < i + 2) { | |
348 | show_usage(argc, argv); | |
349 | return -1; | |
350 | } | |
351 | wdelay = atol(argv[++i]); | |
352 | break; | |
353 | case 'v': | |
354 | verbose_mode = 1; | |
355 | break; | |
356 | } | |
357 | } | |
358 | ||
359 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
360 | "%u dequeuers.\n", | |
361 | duration, nr_enqueuers, nr_dequeuers); | |
362 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
363 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
364 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
365 | "main", pthread_self(), (unsigned long)gettid()); | |
366 | ||
367 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
368 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
369 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
370 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
16aa9ee8 | 371 | cds_lfq_init_rcu(&q); |
453629a9 MD |
372 | |
373 | next_aff = 0; | |
374 | ||
375 | for (i = 0; i < nr_enqueuers; i++) { | |
376 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
377 | &count_enqueuer[2 * i]); | |
378 | if (err != 0) | |
379 | exit(1); | |
380 | } | |
381 | for (i = 0; i < nr_dequeuers; i++) { | |
382 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
383 | &count_dequeuer[2 * i]); | |
384 | if (err != 0) | |
385 | exit(1); | |
386 | } | |
387 | ||
5481ddb3 | 388 | cmm_smp_mb(); |
453629a9 MD |
389 | |
390 | test_go = 1; | |
391 | ||
392 | for (i = 0; i < duration; i++) { | |
393 | sleep(1); | |
394 | if (verbose_mode) | |
395 | write (1, ".", 1); | |
396 | } | |
397 | ||
398 | test_stop = 1; | |
399 | ||
400 | for (i = 0; i < nr_enqueuers; i++) { | |
401 | err = pthread_join(tid_enqueuer[i], &tret); | |
402 | if (err != 0) | |
403 | exit(1); | |
404 | tot_enqueues += count_enqueuer[2 * i]; | |
405 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
406 | } | |
407 | for (i = 0; i < nr_dequeuers; i++) { | |
408 | err = pthread_join(tid_dequeuer[i], &tret); | |
409 | if (err != 0) | |
410 | exit(1); | |
411 | tot_dequeues += count_dequeuer[2 * i]; | |
412 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
413 | } | |
414 | ||
415 | test_end(&q, &end_dequeues); | |
416 | ||
417 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
418 | tot_enqueues, tot_dequeues); | |
419 | printf_verbose("total number of successful enqueues : %llu, " | |
420 | "successful dequeues %llu\n", | |
421 | tot_successful_enqueues, tot_successful_dequeues); | |
422 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
423 | "nr_dequeuers %3u " | |
424 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
425 | "successful enqueues %12llu successful dequeues %12llu " | |
426 | "end_dequeues %llu nr_ops %12llu\n", | |
427 | argv[0], duration, nr_enqueuers, wdelay, | |
428 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
429 | tot_successful_enqueues, | |
430 | tot_successful_dequeues, end_dequeues, | |
431 | tot_enqueues + tot_dequeues); | |
432 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
433 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
434 | "succ. dequeues + end dequeues %llu.\n", | |
435 | tot_successful_enqueues, | |
436 | tot_successful_dequeues + end_dequeues); | |
437 | ||
438 | free(count_enqueuer); | |
439 | free(count_dequeuer); | |
440 | free(tid_enqueuer); | |
441 | free(tid_dequeuer); | |
442 | return 0; | |
443 | } |