Add `urcu_posix_assert()` as `assert()` replacement
[urcu.git] / tests / benchmark / test_urcu_lfs_rcu.c
1 /*
2 * test_urcu_lfs_rcu.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 #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 <errno.h>
35
36 #include <urcu/arch.h>
37 #include <urcu/assert.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
49 /* Remove deprecation warnings from test build. */
50 #define CDS_LFS_RCU_DEPRECATED
51
52 #include <urcu/cds.h>
53
54 static volatile int test_go, test_stop;
55
56 static unsigned long rduration;
57
58 static unsigned long duration;
59
60 /* read-side C.S. duration, in loops */
61 static unsigned long wdelay;
62
63 static inline void loop_sleep(unsigned long loops)
64 {
65 while (loops-- != 0)
66 caa_cpu_relax();
67 }
68
69 static int verbose_mode;
70
71 #define printf_verbose(fmt, args...) \
72 do { \
73 if (verbose_mode) \
74 printf(fmt, args); \
75 } while (0)
76
77 static unsigned int cpu_affinities[NR_CPUS];
78 static unsigned int next_aff = 0;
79 static int use_affinity = 0;
80
81 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
82
83 static void set_affinity(void)
84 {
85 #ifdef HAVE_SCHED_SETAFFINITY
86 cpu_set_t mask;
87 int cpu, ret;
88 #endif /* HAVE_SCHED_SETAFFINITY */
89
90 if (!use_affinity)
91 return;
92
93 #ifdef HAVE_SCHED_SETAFFINITY
94 ret = pthread_mutex_lock(&affinity_mutex);
95 if (ret) {
96 perror("Error in pthread mutex lock");
97 exit(-1);
98 }
99 cpu = cpu_affinities[next_aff++];
100 ret = pthread_mutex_unlock(&affinity_mutex);
101 if (ret) {
102 perror("Error in pthread mutex unlock");
103 exit(-1);
104 }
105
106 CPU_ZERO(&mask);
107 CPU_SET(cpu, &mask);
108 sched_setaffinity(0, sizeof(mask), &mask);
109 #endif /* HAVE_SCHED_SETAFFINITY */
110 }
111
112 /*
113 * returns 0 if test should end.
114 */
115 static int test_duration_dequeue(void)
116 {
117 return !test_stop;
118 }
119
120 static int test_duration_enqueue(void)
121 {
122 return !test_stop;
123 }
124
125 static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
126 static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
127
128 static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
129 static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
130
131 static unsigned int nr_enqueuers;
132 static unsigned int nr_dequeuers;
133
134 struct test {
135 struct cds_lfs_node_rcu list;
136 struct rcu_head rcu;
137 };
138
139 static struct cds_lfs_stack_rcu s;
140
141 static
142 void *thr_enqueuer(void *_count)
143 {
144 unsigned long long *count = _count;
145
146 printf_verbose("thread_begin %s, tid %lu\n",
147 "enqueuer", urcu_get_thread_id());
148
149 set_affinity();
150
151 rcu_register_thread();
152
153 while (!test_go)
154 {
155 }
156 cmm_smp_mb();
157
158 for (;;) {
159 struct test *node = malloc(sizeof(*node));
160 if (!node)
161 goto fail;
162 cds_lfs_node_init_rcu(&node->list);
163 /* No rcu read-side is needed for push */
164 cds_lfs_push_rcu(&s, &node->list);
165 URCU_TLS(nr_successful_enqueues)++;
166
167 if (caa_unlikely(wdelay))
168 loop_sleep(wdelay);
169 fail:
170 URCU_TLS(nr_enqueues)++;
171 if (caa_unlikely(!test_duration_enqueue()))
172 break;
173 }
174
175 rcu_unregister_thread();
176
177 count[0] = URCU_TLS(nr_enqueues);
178 count[1] = URCU_TLS(nr_successful_enqueues);
179 printf_verbose("enqueuer thread_end, tid %lu, "
180 "enqueues %llu successful_enqueues %llu\n",
181 urcu_get_thread_id(),
182 URCU_TLS(nr_enqueues),
183 URCU_TLS(nr_successful_enqueues));
184 return ((void*)1);
185
186 }
187
188 static
189 void free_node_cb(struct rcu_head *head)
190 {
191 struct test *node =
192 caa_container_of(head, struct test, rcu);
193 free(node);
194 }
195
196 static
197 void *thr_dequeuer(void *_count)
198 {
199 unsigned long long *count = _count;
200
201 printf_verbose("thread_begin %s, tid %lu\n",
202 "dequeuer", urcu_get_thread_id());
203
204 set_affinity();
205
206 rcu_register_thread();
207
208 while (!test_go)
209 {
210 }
211 cmm_smp_mb();
212
213 for (;;) {
214 struct cds_lfs_node_rcu *snode;
215
216 rcu_read_lock();
217 snode = cds_lfs_pop_rcu(&s);
218 rcu_read_unlock();
219 if (snode) {
220 struct test *node;
221
222 node = caa_container_of(snode, struct test, list);
223 call_rcu(&node->rcu, free_node_cb);
224 URCU_TLS(nr_successful_dequeues)++;
225 }
226 URCU_TLS(nr_dequeues)++;
227 if (caa_unlikely(!test_duration_dequeue()))
228 break;
229 if (caa_unlikely(rduration))
230 loop_sleep(rduration);
231 }
232
233 rcu_unregister_thread();
234
235 printf_verbose("dequeuer thread_end, tid %lu, "
236 "dequeues %llu, successful_dequeues %llu\n",
237 urcu_get_thread_id(),
238 URCU_TLS(nr_dequeues),
239 URCU_TLS(nr_successful_dequeues));
240 count[0] = URCU_TLS(nr_dequeues);
241 count[1] = URCU_TLS(nr_successful_dequeues);
242 return ((void*)2);
243 }
244
245 static
246 void test_end(unsigned long long *nr_dequeues_l)
247 {
248 struct cds_lfs_node_rcu *snode;
249
250 do {
251 snode = cds_lfs_pop_rcu(&s);
252 if (snode) {
253 struct test *node;
254
255 node = caa_container_of(snode, struct test, list);
256 free(node);
257 (*nr_dequeues_l)++;
258 }
259 } while (snode);
260 }
261
262 static
263 void show_usage(char **argv)
264 {
265 printf("Usage : %s nr_dequeuers nr_enqueuers duration (s) <OPTIONS>\n",
266 argv[0]);
267 printf("OPTIONS:\n");
268 printf(" [-d delay] (enqueuer period (in loops))\n");
269 printf(" [-c duration] (dequeuer period (in loops))\n");
270 printf(" [-v] (verbose output)\n");
271 printf(" [-a cpu#] [-a cpu#]... (affinity)\n");
272 printf("\n");
273 }
274
275 int main(int argc, char **argv)
276 {
277 int err;
278 pthread_t *tid_enqueuer, *tid_dequeuer;
279 void *tret;
280 unsigned long long *count_enqueuer, *count_dequeuer;
281 unsigned long long tot_enqueues = 0, tot_dequeues = 0;
282 unsigned long long tot_successful_enqueues = 0,
283 tot_successful_dequeues = 0;
284 unsigned long long end_dequeues = 0;
285 int i, a;
286 unsigned int i_thr;
287
288 if (argc < 4) {
289 show_usage(argv);
290 return -1;
291 }
292
293 err = sscanf(argv[1], "%u", &nr_dequeuers);
294 if (err != 1) {
295 show_usage(argv);
296 return -1;
297 }
298
299 err = sscanf(argv[2], "%u", &nr_enqueuers);
300 if (err != 1) {
301 show_usage(argv);
302 return -1;
303 }
304
305 err = sscanf(argv[3], "%lu", &duration);
306 if (err != 1) {
307 show_usage(argv);
308 return -1;
309 }
310
311 for (i = 4; i < argc; i++) {
312 if (argv[i][0] != '-')
313 continue;
314 switch (argv[i][1]) {
315 case 'a':
316 if (argc < i + 2) {
317 show_usage(argv);
318 return -1;
319 }
320 a = atoi(argv[++i]);
321 cpu_affinities[next_aff++] = a;
322 use_affinity = 1;
323 printf_verbose("Adding CPU %d affinity\n", a);
324 break;
325 case 'c':
326 if (argc < i + 2) {
327 show_usage(argv);
328 return -1;
329 }
330 rduration = atol(argv[++i]);
331 break;
332 case 'd':
333 if (argc < i + 2) {
334 show_usage(argv);
335 return -1;
336 }
337 wdelay = atol(argv[++i]);
338 break;
339 case 'v':
340 verbose_mode = 1;
341 break;
342 }
343 }
344
345 printf_verbose("running test for %lu seconds, %u enqueuers, "
346 "%u dequeuers.\n",
347 duration, nr_enqueuers, nr_dequeuers);
348 printf_verbose("Writer delay : %lu loops.\n", rduration);
349 printf_verbose("Reader duration : %lu loops.\n", wdelay);
350 printf_verbose("thread %-6s, tid %lu\n",
351 "main", urcu_get_thread_id());
352
353 tid_enqueuer = calloc(nr_enqueuers, sizeof(*tid_enqueuer));
354 tid_dequeuer = calloc(nr_dequeuers, sizeof(*tid_dequeuer));
355 count_enqueuer = calloc(nr_enqueuers, 2 * sizeof(*count_enqueuer));
356 count_dequeuer = calloc(nr_dequeuers, 2 * sizeof(*count_dequeuer));
357 cds_lfs_init_rcu(&s);
358 err = create_all_cpu_call_rcu_data(0);
359 if (err) {
360 printf("Per-CPU call_rcu() worker threads unavailable. Using default global worker thread.\n");
361 }
362
363 next_aff = 0;
364
365 for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
366 err = pthread_create(&tid_enqueuer[i_thr], NULL, thr_enqueuer,
367 &count_enqueuer[2 * i_thr]);
368 if (err != 0)
369 exit(1);
370 }
371 for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
372 err = pthread_create(&tid_dequeuer[i_thr], NULL, thr_dequeuer,
373 &count_dequeuer[2 * i_thr]);
374 if (err != 0)
375 exit(1);
376 }
377
378 cmm_smp_mb();
379
380 test_go = 1;
381
382 for (i_thr = 0; i_thr < duration; i_thr++) {
383 sleep(1);
384 if (verbose_mode) {
385 fwrite(".", sizeof(char), 1, stdout);
386 fflush(stdout);
387 }
388 }
389
390 test_stop = 1;
391
392 for (i_thr = 0; i_thr < nr_enqueuers; i_thr++) {
393 err = pthread_join(tid_enqueuer[i_thr], &tret);
394 if (err != 0)
395 exit(1);
396 tot_enqueues += count_enqueuer[2 * i_thr];
397 tot_successful_enqueues += count_enqueuer[2 * i_thr + 1];
398 }
399 for (i_thr = 0; i_thr < nr_dequeuers; i_thr++) {
400 err = pthread_join(tid_dequeuer[i_thr], &tret);
401 if (err != 0)
402 exit(1);
403 tot_dequeues += count_dequeuer[2 * i_thr];
404 tot_successful_dequeues += count_dequeuer[2 * i_thr + 1];
405 }
406
407 test_end(&end_dequeues);
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 return 0;
436 }
This page took 0.038308 seconds and 4 git commands to generate.