eb97b1ae1062548c8092bf289e10984694e45420
[urcu.git] / tests / benchmark / test_urcu_lfq.c
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 #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>
35 #include <errno.h>
36
37 #include <urcu/arch.h>
38 #include <urcu/tls-compat.h>
39 #include "thread-id.h"
40
41 /* hardcoded number of CPUs */
42 #define NR_CPUS 16384
43
44 #ifndef DYNAMIC_LINK_TEST
45 #define _LGPL_SOURCE
46 #endif
47 #include <urcu.h>
48 #include <urcu/cds.h>
49
50 static volatile int test_go, test_stop;
51
52 static unsigned long rduration;
53
54 static unsigned long duration;
55
56 /* read-side C.S. duration, in loops */
57 static unsigned long wdelay;
58
59 static inline void loop_sleep(unsigned long loops)
60 {
61 while (loops-- != 0)
62 caa_cpu_relax();
63 }
64
65 static int verbose_mode;
66
67 #define printf_verbose(fmt, args...) \
68 do { \
69 if (verbose_mode) \
70 printf(fmt, args); \
71 } while (0)
72
73 static unsigned int cpu_affinities[NR_CPUS];
74 static unsigned int next_aff = 0;
75 static int use_affinity = 0;
76
77 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
78
79 static void set_affinity(void)
80 {
81 #if HAVE_SCHED_SETAFFINITY
82 cpu_set_t mask;
83 int cpu, ret;
84 #endif /* HAVE_SCHED_SETAFFINITY */
85
86 if (!use_affinity)
87 return;
88
89 #if HAVE_SCHED_SETAFFINITY
90 ret = pthread_mutex_lock(&affinity_mutex);
91 if (ret) {
92 perror("Error in pthread mutex lock");
93 exit(-1);
94 }
95 cpu = cpu_affinities[next_aff++];
96 ret = pthread_mutex_unlock(&affinity_mutex);
97 if (ret) {
98 perror("Error in pthread mutex unlock");
99 exit(-1);
100 }
101
102 CPU_ZERO(&mask);
103 CPU_SET(cpu, &mask);
104 sched_setaffinity(0, sizeof(mask), &mask);
105 #endif /* HAVE_SCHED_SETAFFINITY */
106 }
107
108 /*
109 * returns 0 if test should end.
110 */
111 static int test_duration_dequeue(void)
112 {
113 return !test_stop;
114 }
115
116 static int test_duration_enqueue(void)
117 {
118 return !test_stop;
119 }
120
121 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
122 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
123
124 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
125 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
126
127 static unsigned int nr_enqueuers;
128 static unsigned int nr_dequeuers;
129
130 struct test {
131 struct cds_lfq_node_rcu list;
132 struct rcu_head rcu;
133 };
134
135 static struct cds_lfq_queue_rcu q;
136
137 static
138 void *thr_enqueuer(void *_count)
139 {
140 unsigned long long *count = _count;
141
142 printf_verbose("thread_begin %s, tid %lu\n",
143 "enqueuer", urcu_get_thread_id());
144
145 set_affinity();
146
147 rcu_register_thread();
148
149 while (!test_go)
150 {
151 }
152 cmm_smp_mb();
153
154 for (;;) {
155 struct test *node = malloc(sizeof(*node));
156 if (!node)
157 goto fail;
158 cds_lfq_node_init_rcu(&node->list);
159 rcu_read_lock();
160 cds_lfq_enqueue_rcu(&q, &node->list);
161 rcu_read_unlock();
162 URCU_TLS(nr_successful_enqueues)++;
163
164 if (caa_unlikely(wdelay))
165 loop_sleep(wdelay);
166 fail:
167 URCU_TLS(nr_enqueues)++;
168 if (caa_unlikely(!test_duration_enqueue()))
169 break;
170 }
171
172 rcu_unregister_thread();
173
174 count[0] = URCU_TLS(nr_enqueues);
175 count[1] = URCU_TLS(nr_successful_enqueues);
176 printf_verbose("enqueuer thread_end, tid %lu, "
177 "enqueues %llu successful_enqueues %llu\n",
178 urcu_get_thread_id(),
179 URCU_TLS(nr_enqueues),
180 URCU_TLS(nr_successful_enqueues));
181 return ((void*)1);
182
183 }
184
185 static
186 void free_node_cb(struct rcu_head *head)
187 {
188 struct test *node =
189 caa_container_of(head, struct test, rcu);
190 free(node);
191 }
192
193 static
194 void *thr_dequeuer(void *_count)
195 {
196 unsigned long long *count = _count;
197
198 printf_verbose("thread_begin %s, tid %lu\n",
199 "dequeuer", urcu_get_thread_id());
200
201 set_affinity();
202
203 rcu_register_thread();
204
205 while (!test_go)
206 {
207 }
208 cmm_smp_mb();
209
210 for (;;) {
211 struct cds_lfq_node_rcu *qnode;
212
213 rcu_read_lock();
214 qnode = cds_lfq_dequeue_rcu(&q);
215 rcu_read_unlock();
216
217 if (qnode) {
218 struct test *node;
219
220 node = caa_container_of(qnode, struct test, list);
221 call_rcu(&node->rcu, free_node_cb);
222 URCU_TLS(nr_successful_dequeues)++;
223 }
224
225 URCU_TLS(nr_dequeues)++;
226 if (caa_unlikely(!test_duration_dequeue()))
227 break;
228 if (caa_unlikely(rduration))
229 loop_sleep(rduration);
230 }
231
232 rcu_unregister_thread();
233 printf_verbose("dequeuer thread_end, tid %lu, "
234 "dequeues %llu, successful_dequeues %llu\n",
235 urcu_get_thread_id(),
236 URCU_TLS(nr_dequeues),
237 URCU_TLS(nr_successful_dequeues));
238 count[0] = URCU_TLS(nr_dequeues);
239 count[1] = URCU_TLS(nr_successful_dequeues);
240 return ((void*)2);
241 }
242
243 static
244 void test_end(unsigned long long *nr_dequeues_l)
245 {
246 struct cds_lfq_node_rcu *snode;
247
248 do {
249 snode = cds_lfq_dequeue_rcu(&q);
250 if (snode) {
251 struct test *node;
252
253 node = caa_container_of(snode, struct test, list);
254 free(node); /* no more concurrent access */
255 (*nr_dequeues_l)++;
256 }
257 } while (snode);
258 }
259
260 static
261 void show_usage(char **argv)
262 {
263 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
264 argv[0]);
265 printf("OPTIONS:\n");
266 printf(" [-d delay] (enqueuer period (in loops))\n");
267 printf(" [-c duration] (dequeuer period (in loops))\n");
268 printf(" [-v] (verbose output)\n");
269 printf(" [-a cpu#] [-a cpu#]... (affinity)\n");
270 printf("\n");
271 }
272
273 int main(int argc, char **argv)
274 {
275 int err;
276 pthread_t *tid_enqueuer, *tid_dequeuer;
277 void *tret;
278 unsigned long long *count_enqueuer, *count_dequeuer;
279 unsigned long long tot_enqueues = 0, tot_dequeues = 0;
280 unsigned long long tot_successful_enqueues = 0,
281 tot_successful_dequeues = 0;
282 unsigned long long end_dequeues = 0;
283 int i, a;
284 unsigned int i_thr;
285
286 if (argc < 4) {
287 show_usage(argv);
288 return -1;
289 }
290
291 err = sscanf(argv[1], "%u", &nr_dequeuers);
292 if (err != 1) {
293 show_usage(argv);
294 return -1;
295 }
296
297 err = sscanf(argv[2], "%u", &nr_enqueuers);
298 if (err != 1) {
299 show_usage(argv);
300 return -1;
301 }
302
303 err = sscanf(argv[3], "%lu", &duration);
304 if (err != 1) {
305 show_usage(argv);
306 return -1;
307 }
308
309 for (i = 4; i < argc; i++) {
310 if (argv[i][0] != '-')
311 continue;
312 switch (argv[i][1]) {
313 case 'a':
314 if (argc < i + 2) {
315 show_usage(argv);
316 return -1;
317 }
318 a = atoi(argv[++i]);
319 cpu_affinities[next_aff++] = a;
320 use_affinity = 1;
321 printf_verbose("Adding CPU %d affinity\n", a);
322 break;
323 case 'c':
324 if (argc < i + 2) {
325 show_usage(argv);
326 return -1;
327 }
328 rduration = atol(argv[++i]);
329 break;
330 case 'd':
331 if (argc < i + 2) {
332 show_usage(argv);
333 return -1;
334 }
335 wdelay = atol(argv[++i]);
336 break;
337 case 'v':
338 verbose_mode = 1;
339 break;
340 }
341 }
342
343 printf_verbose("running test for %lu seconds, %u enqueuers, "
344 "%u dequeuers.\n",
345 duration, nr_enqueuers, nr_dequeuers);
346 printf_verbose("Writer delay : %lu loops.\n", rduration);
347 printf_verbose("Reader duration : %lu loops.\n", wdelay);
348 printf_verbose("thread %-6s, tid %lu\n",
349 "main", urcu_get_thread_id());
350
351 tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
352 tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
353 count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
354 count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
355 cds_lfq_init_rcu(&q, call_rcu);
356 err = create_all_cpu_call_rcu_data(0);
357 if (err) {
358 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
359 }
360
361 next_aff = 0;
362
363 for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
364 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
365 &count_enqueuer[2 * i_thr]);
366 if (err != 0)
367 exit(1);
368 }
369 for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
370 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
371 &count_dequeuer[2 * i_thr]);
372 if (err != 0)
373 exit(1);
374 }
375
376 cmm_smp_mb();
377
378 test_go = 1;
379
380 for (i_thr = 0; i_thr < duration; i_thr++) {
381 sleep(1);
382 if (verbose_mode) {
383 fwrite(".", sizeof(char), 1, stdout);
384 fflush(stdout);
385 }
386 }
387
388 test_stop = 1;
389
390 for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
391 err = pthread_join(tid_enqueuer[i_thr], &tret);
392 if (err != 0)
393 exit(1);
394 tot_enqueues += count_enqueuer[2 * i_thr];
395 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
396 }
397 for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
398 err = pthread_join(tid_dequeuer[i_thr], &tret);
399 if (err != 0)
400 exit(1);
401 tot_dequeues += count_dequeuer[2 * i_thr];
402 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
403 }
404
405 test_end(&end_dequeues);
406 err = cds_lfq_destroy_rcu(&q);
407 assert(!err);
408
409 printf_verbose("total number of enqueues : %llu, dequeues %llu\n",
410 tot_enqueues, tot_dequeues);
411 printf_verbose("total number of successful enqueues : %llu, "
412 "successful dequeues %llu\n",
413 tot_successful_enqueues, tot_successful_dequeues);
414 printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu "
415 "nr_dequeuers %3u "
416 "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu "
417 "successful enqueues %12llu successful dequeues %12llu "
418 "end_dequeues %llu nr_ops %12llu\n",
419 argv[0], duration, nr_enqueuers, wdelay,
420 nr_dequeuers, rduration, tot_enqueues, tot_dequeues,
421 tot_successful_enqueues,
422 tot_successful_dequeues, end_dequeues,
423 tot_enqueues + tot_dequeues);
424 if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues)
425 printf("WARNING! Discrepancy between nr succ. enqueues %llu vs "
426 "succ. dequeues + end dequeues %llu.\n",
427 tot_successful_enqueues,
428 tot_successful_dequeues + end_dequeues);
429
430 free_all_cpu_call_rcu_data();
431 free(count_enqueuer);
432 free(count_dequeuer);
433 free(tid_enqueuer);
434 free(tid_dequeuer);
435
436 return 0;
437 }
This page took 0.037929 seconds and 4 git commands to generate.