Commit | Line | Data |
---|---|---|
60c60c15 MD |
1 | /* |
2 | * test_lfq_baseline.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 <sched.h> | |
38 | #include <errno.h> | |
39 | ||
40 | #include <urcu/arch.h> | |
41 | #include <urcu/list.h> | |
42 | ||
43 | #ifdef __linux__ | |
44 | #include <syscall.h> | |
45 | #endif | |
46 | ||
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 | static volatile int test_go, test_stop; | |
66 | ||
67 | static unsigned long rduration; | |
68 | ||
69 | static unsigned long duration; | |
70 | ||
71 | /* read-side C.S. duration, in loops */ | |
72 | static unsigned long wdelay; | |
73 | ||
74 | static inline void loop_sleep(unsigned long l) | |
75 | { | |
76 | while(l-- != 0) | |
77 | caa_cpu_relax(); | |
78 | } | |
79 | ||
80 | static int verbose_mode; | |
81 | ||
82 | #define printf_verbose(fmt, args...) \ | |
83 | do { \ | |
84 | if (verbose_mode) \ | |
85 | printf(fmt, args); \ | |
86 | } while (0) | |
87 | ||
88 | static unsigned int cpu_affinities[NR_CPUS]; | |
89 | static unsigned int next_aff = 0; | |
90 | static int use_affinity = 0; | |
91 | ||
92 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
93 | ||
94 | #ifndef HAVE_CPU_SET_T | |
95 | typedef unsigned long cpu_set_t; | |
96 | # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0) | |
97 | # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0) | |
98 | #endif | |
99 | ||
100 | static void set_affinity(void) | |
101 | { | |
102 | cpu_set_t mask; | |
103 | int cpu; | |
104 | int ret; | |
105 | ||
106 | if (!use_affinity) | |
107 | return; | |
108 | ||
109 | #if HAVE_SCHED_SETAFFINITY | |
110 | ret = pthread_mutex_lock(&affinity_mutex); | |
111 | if (ret) { | |
112 | perror("Error in pthread mutex lock"); | |
113 | exit(-1); | |
114 | } | |
115 | cpu = cpu_affinities[next_aff++]; | |
116 | ret = pthread_mutex_unlock(&affinity_mutex); | |
117 | if (ret) { | |
118 | perror("Error in pthread mutex unlock"); | |
119 | exit(-1); | |
120 | } | |
121 | ||
122 | CPU_ZERO(&mask); | |
123 | CPU_SET(cpu, &mask); | |
124 | #if SCHED_SETAFFINITY_ARGS == 2 | |
125 | sched_setaffinity(0, &mask); | |
126 | #else | |
127 | sched_setaffinity(0, sizeof(mask), &mask); | |
128 | #endif | |
129 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
130 | } | |
131 | ||
132 | /* | |
133 | * returns 0 if test should end. | |
134 | */ | |
135 | static int test_duration_dequeue(void) | |
136 | { | |
137 | return !test_stop; | |
138 | } | |
139 | ||
140 | static int test_duration_enqueue(void) | |
141 | { | |
142 | return !test_stop; | |
143 | } | |
144 | ||
145 | static unsigned long long __thread nr_dequeues; | |
146 | static unsigned long long __thread nr_enqueues; | |
147 | ||
148 | static unsigned long long __thread nr_successful_dequeues; | |
149 | static unsigned long long __thread nr_successful_enqueues; | |
150 | ||
151 | static unsigned int nr_enqueuers; | |
152 | static unsigned int nr_dequeuers; | |
153 | ||
154 | pthread_mutex_t test_mutex = PTHREAD_MUTEX_INITIALIZER; | |
155 | ||
156 | struct test { | |
157 | struct cds_list_head node; | |
158 | }; | |
159 | ||
160 | static CDS_LIST_HEAD(q); | |
161 | ||
162 | static | |
163 | void test_mutex_lock(void) | |
164 | { | |
165 | int ret; | |
166 | ret = pthread_mutex_lock(&test_mutex); | |
167 | if (ret) { | |
168 | perror("Error in pthread mutex lock"); | |
169 | exit(-1); | |
170 | } | |
171 | } | |
172 | ||
173 | static | |
174 | void test_mutex_unlock(void) | |
175 | { | |
176 | int ret; | |
177 | ||
178 | ret = pthread_mutex_unlock(&test_mutex); | |
179 | if (ret) { | |
180 | perror("Error in pthread mutex unlock"); | |
181 | exit(-1); | |
182 | } | |
183 | } | |
184 | ||
185 | void *thr_enqueuer(void *_count) | |
186 | { | |
187 | unsigned long long *count = _count; | |
188 | ||
189 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
190 | "enqueuer", pthread_self(), (unsigned long)gettid()); | |
191 | ||
192 | set_affinity(); | |
193 | ||
194 | while (!test_go) | |
195 | { | |
196 | } | |
197 | cmm_smp_mb(); | |
198 | ||
199 | for (;;) { | |
200 | struct test *node = malloc(sizeof(*node)); | |
201 | if (!node) | |
202 | goto fail; | |
203 | test_mutex_lock(); | |
204 | cds_list_add_tail(&node->node, &q); | |
205 | test_mutex_unlock(); | |
206 | nr_successful_enqueues++; | |
207 | ||
208 | if (unlikely(wdelay)) | |
209 | loop_sleep(wdelay); | |
210 | fail: | |
211 | nr_enqueues++; | |
212 | if (unlikely(!test_duration_enqueue())) | |
213 | break; | |
214 | } | |
215 | ||
216 | count[0] = nr_enqueues; | |
217 | count[1] = nr_successful_enqueues; | |
218 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " | |
219 | "enqueues %llu successful_enqueues %llu\n", | |
220 | pthread_self(), (unsigned long)gettid(), nr_enqueues, | |
221 | nr_successful_enqueues); | |
222 | return ((void*)1); | |
223 | ||
224 | } | |
225 | ||
226 | void *thr_dequeuer(void *_count) | |
227 | { | |
228 | unsigned long long *count = _count; | |
229 | int ret; | |
230 | ||
231 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
232 | "dequeuer", pthread_self(), (unsigned long)gettid()); | |
233 | ||
234 | set_affinity(); | |
235 | ||
236 | while (!test_go) | |
237 | { | |
238 | } | |
239 | cmm_smp_mb(); | |
240 | ||
241 | for (;;) { | |
242 | struct test *node; | |
243 | ||
244 | test_mutex_lock(); | |
245 | node = cds_list_first_entry(&q, struct test, node); | |
72b972e8 MD |
246 | if (node) |
247 | cds_list_del(&node->node); | |
60c60c15 MD |
248 | test_mutex_unlock(); |
249 | ||
250 | if (node) { | |
251 | free(node); | |
252 | nr_successful_dequeues++; | |
253 | } | |
254 | ||
255 | nr_dequeues++; | |
256 | if (unlikely(!test_duration_dequeue())) | |
257 | break; | |
258 | if (unlikely(rduration)) | |
259 | loop_sleep(rduration); | |
260 | } | |
261 | ||
262 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " | |
263 | "dequeues %llu, successful_dequeues %llu\n", | |
264 | pthread_self(), (unsigned long)gettid(), nr_dequeues, | |
265 | nr_successful_dequeues); | |
266 | count[0] = nr_dequeues; | |
267 | count[1] = nr_successful_dequeues; | |
268 | return ((void*)2); | |
269 | } | |
270 | ||
271 | void test_end(unsigned long long *nr_dequeues) | |
272 | { | |
273 | struct test *node; | |
274 | ||
275 | do { | |
276 | node = cds_list_first_entry(&q, struct test, node); | |
60c60c15 | 277 | if (node) { |
72b972e8 | 278 | cds_list_del(&node->node); |
60c60c15 MD |
279 | free(node); /* no more concurrent access */ |
280 | (*nr_dequeues)++; | |
281 | } | |
282 | } while (node); | |
283 | } | |
284 | ||
285 | void show_usage(int argc, char **argv) | |
286 | { | |
287 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]); | |
288 | printf(" [-d delay] (enqueuer period (in loops))"); | |
289 | printf(" [-c duration] (dequeuer period (in loops))"); | |
290 | printf(" [-v] (verbose output)"); | |
291 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
292 | printf("\n"); | |
293 | } | |
294 | ||
295 | int main(int argc, char **argv) | |
296 | { | |
297 | int err; | |
298 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
299 | void *tret; | |
300 | unsigned long long *count_enqueuer, *count_dequeuer; | |
301 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
302 | unsigned long long tot_successful_enqueues = 0, | |
303 | tot_successful_dequeues = 0; | |
304 | unsigned long long end_dequeues = 0; | |
305 | int i, a; | |
306 | ||
307 | if (argc < 4) { | |
308 | show_usage(argc, argv); | |
309 | return -1; | |
310 | } | |
311 | ||
312 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
313 | if (err != 1) { | |
314 | show_usage(argc, argv); | |
315 | return -1; | |
316 | } | |
317 | ||
318 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
319 | if (err != 1) { | |
320 | show_usage(argc, argv); | |
321 | return -1; | |
322 | } | |
323 | ||
324 | err = sscanf(argv[3], "%lu", &duration); | |
325 | if (err != 1) { | |
326 | show_usage(argc, argv); | |
327 | return -1; | |
328 | } | |
329 | ||
330 | for (i = 4; i < argc; i++) { | |
331 | if (argv[i][0] != '-') | |
332 | continue; | |
333 | switch (argv[i][1]) { | |
334 | case 'a': | |
335 | if (argc < i + 2) { | |
336 | show_usage(argc, argv); | |
337 | return -1; | |
338 | } | |
339 | a = atoi(argv[++i]); | |
340 | cpu_affinities[next_aff++] = a; | |
341 | use_affinity = 1; | |
342 | printf_verbose("Adding CPU %d affinity\n", a); | |
343 | break; | |
344 | case 'c': | |
345 | if (argc < i + 2) { | |
346 | show_usage(argc, argv); | |
347 | return -1; | |
348 | } | |
349 | rduration = atol(argv[++i]); | |
350 | break; | |
351 | case 'd': | |
352 | if (argc < i + 2) { | |
353 | show_usage(argc, argv); | |
354 | return -1; | |
355 | } | |
356 | wdelay = atol(argv[++i]); | |
357 | break; | |
358 | case 'v': | |
359 | verbose_mode = 1; | |
360 | break; | |
361 | } | |
362 | } | |
363 | ||
364 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
365 | "%u dequeuers.\n", | |
366 | duration, nr_enqueuers, nr_dequeuers); | |
367 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
368 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
369 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
370 | "main", pthread_self(), (unsigned long)gettid()); | |
371 | ||
372 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
373 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
374 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
375 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
376 | ||
377 | next_aff = 0; | |
378 | ||
379 | for (i = 0; i < nr_enqueuers; i++) { | |
380 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
381 | &count_enqueuer[2 * i]); | |
382 | if (err != 0) | |
383 | exit(1); | |
384 | } | |
385 | for (i = 0; i < nr_dequeuers; i++) { | |
386 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
387 | &count_dequeuer[2 * i]); | |
388 | if (err != 0) | |
389 | exit(1); | |
390 | } | |
391 | ||
392 | cmm_smp_mb(); | |
393 | ||
394 | test_go = 1; | |
395 | ||
396 | for (i = 0; i < duration; i++) { | |
397 | sleep(1); | |
398 | if (verbose_mode) | |
399 | write (1, ".", 1); | |
400 | } | |
401 | ||
402 | test_stop = 1; | |
403 | ||
404 | for (i = 0; i < nr_enqueuers; i++) { | |
405 | err = pthread_join(tid_enqueuer[i], &tret); | |
406 | if (err != 0) | |
407 | exit(1); | |
408 | tot_enqueues += count_enqueuer[2 * i]; | |
409 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
410 | } | |
411 | for (i = 0; i < nr_dequeuers; i++) { | |
412 | err = pthread_join(tid_dequeuer[i], &tret); | |
413 | if (err != 0) | |
414 | exit(1); | |
415 | tot_dequeues += count_dequeuer[2 * i]; | |
416 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
417 | } | |
418 | ||
419 | test_end(&end_dequeues); | |
60c60c15 MD |
420 | |
421 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
422 | tot_enqueues, tot_dequeues); | |
423 | printf_verbose("total number of successful enqueues : %llu, " | |
424 | "successful dequeues %llu\n", | |
425 | tot_successful_enqueues, tot_successful_dequeues); | |
426 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
427 | "nr_dequeuers %3u " | |
428 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
429 | "successful enqueues %12llu successful dequeues %12llu " | |
430 | "end_dequeues %llu nr_ops %12llu\n", | |
431 | argv[0], duration, nr_enqueuers, wdelay, | |
432 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
433 | tot_successful_enqueues, | |
434 | tot_successful_dequeues, end_dequeues, | |
435 | tot_enqueues + tot_dequeues); | |
436 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
437 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
438 | "succ. dequeues + end dequeues %llu.\n", | |
439 | tot_successful_enqueues, | |
440 | tot_successful_dequeues + end_dequeues); | |
441 | ||
442 | free(count_enqueuer); | |
443 | free(count_dequeuer); | |
444 | free(tid_enqueuer); | |
445 | free(tid_dequeuer); | |
446 | return 0; | |
447 | } |