Commit | Line | Data |
---|---|---|
a5eb5ffc MD |
1 | /* |
2 | * test_urcu_wfs.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> | |
a5eb5ffc MD |
37 | #include <sched.h> |
38 | #include <errno.h> | |
39 | ||
40 | #include <urcu/arch.h> | |
bd252a04 | 41 | #include <urcu/tls-compat.h> |
a5eb5ffc | 42 | |
65fcc7e9 MD |
43 | #ifdef __linux__ |
44 | #include <syscall.h> | |
45 | #endif | |
46 | ||
a5eb5ffc 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> | |
cb3f3d6b | 69 | #include <urcu/wfstack.h> |
a5eb5ffc 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) |
a5eb5ffc | 81 | { |
ab0aacbe | 82 | while (loops-- != 0) |
06f22bdb | 83 | caa_cpu_relax(); |
a5eb5ffc 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); | |
a5eb5ffc | 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); | |
a5eb5ffc MD |
156 | |
157 | static unsigned int nr_enqueuers; | |
158 | static unsigned int nr_dequeuers; | |
159 | ||
16aa9ee8 | 160 | static struct cds_wfs_stack s; |
a5eb5ffc MD |
161 | |
162 | void *thr_enqueuer(void *_count) | |
163 | { | |
164 | unsigned long long *count = _count; | |
165 | ||
166 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
167 | "enqueuer", pthread_self(), (unsigned long)gettid()); | |
168 | ||
169 | set_affinity(); | |
170 | ||
a5eb5ffc MD |
171 | while (!test_go) |
172 | { | |
173 | } | |
5481ddb3 | 174 | cmm_smp_mb(); |
a5eb5ffc MD |
175 | |
176 | for (;;) { | |
16aa9ee8 | 177 | struct cds_wfs_node *node = malloc(sizeof(*node)); |
a5eb5ffc MD |
178 | if (!node) |
179 | goto fail; | |
16aa9ee8 DG |
180 | cds_wfs_node_init(node); |
181 | cds_wfs_push(&s, node); | |
bd252a04 | 182 | URCU_TLS(nr_successful_enqueues)++; |
a5eb5ffc | 183 | |
a0b7f7ea | 184 | if (caa_unlikely(wdelay)) |
a5eb5ffc MD |
185 | loop_sleep(wdelay); |
186 | fail: | |
bd252a04 | 187 | URCU_TLS(nr_enqueues)++; |
a0b7f7ea | 188 | if (caa_unlikely(!test_duration_enqueue())) |
a5eb5ffc MD |
189 | break; |
190 | } | |
191 | ||
bd252a04 MD |
192 | count[0] = URCU_TLS(nr_enqueues); |
193 | count[1] = URCU_TLS(nr_successful_enqueues); | |
a5eb5ffc MD |
194 | printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, " |
195 | "enqueues %llu successful_enqueues %llu\n", | |
bd252a04 MD |
196 | pthread_self(), (unsigned long)gettid(), |
197 | URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues)); | |
a5eb5ffc MD |
198 | return ((void*)1); |
199 | ||
200 | } | |
201 | ||
202 | void *thr_dequeuer(void *_count) | |
203 | { | |
204 | unsigned long long *count = _count; | |
205 | ||
206 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
207 | "dequeuer", pthread_self(), (unsigned long)gettid()); | |
208 | ||
209 | set_affinity(); | |
210 | ||
a5eb5ffc MD |
211 | while (!test_go) |
212 | { | |
213 | } | |
5481ddb3 | 214 | cmm_smp_mb(); |
a5eb5ffc MD |
215 | |
216 | for (;;) { | |
16aa9ee8 | 217 | struct cds_wfs_node *node = cds_wfs_pop_blocking(&s); |
a5eb5ffc MD |
218 | |
219 | if (node) { | |
cb3f3d6b | 220 | free(node); |
bd252a04 | 221 | URCU_TLS(nr_successful_dequeues)++; |
a5eb5ffc MD |
222 | } |
223 | ||
bd252a04 | 224 | URCU_TLS(nr_dequeues)++; |
a0b7f7ea | 225 | if (caa_unlikely(!test_duration_dequeue())) |
a5eb5ffc | 226 | break; |
a0b7f7ea | 227 | if (caa_unlikely(rduration)) |
a5eb5ffc MD |
228 | loop_sleep(rduration); |
229 | } | |
230 | ||
a5eb5ffc MD |
231 | printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, " |
232 | "dequeues %llu, successful_dequeues %llu\n", | |
bd252a04 MD |
233 | pthread_self(), (unsigned long)gettid(), |
234 | URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues)); | |
235 | count[0] = URCU_TLS(nr_dequeues); | |
236 | count[1] = URCU_TLS(nr_successful_dequeues); | |
a5eb5ffc MD |
237 | return ((void*)2); |
238 | } | |
239 | ||
16aa9ee8 | 240 | void test_end(struct cds_wfs_stack *s, unsigned long long *nr_dequeues) |
a5eb5ffc | 241 | { |
16aa9ee8 | 242 | struct cds_wfs_node *node; |
a5eb5ffc MD |
243 | |
244 | do { | |
16aa9ee8 | 245 | node = cds_wfs_pop_blocking(s); |
a5eb5ffc MD |
246 | if (node) { |
247 | free(node); | |
248 | (*nr_dequeues)++; | |
249 | } | |
250 | } while (node); | |
251 | } | |
252 | ||
253 | void show_usage(int argc, char **argv) | |
254 | { | |
255 | printf("Usage : %s nr_dequeuers nr_enqueuers duration (s)", argv[0]); | |
256 | printf(" [-d delay] (enqueuer period (in loops))"); | |
257 | printf(" [-c duration] (dequeuer period (in loops))"); | |
258 | printf(" [-v] (verbose output)"); | |
259 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
260 | printf("\n"); | |
261 | } | |
262 | ||
263 | int main(int argc, char **argv) | |
264 | { | |
265 | int err; | |
266 | pthread_t *tid_enqueuer, *tid_dequeuer; | |
267 | void *tret; | |
268 | unsigned long long *count_enqueuer, *count_dequeuer; | |
269 | unsigned long long tot_enqueues = 0, tot_dequeues = 0; | |
270 | unsigned long long tot_successful_enqueues = 0, | |
271 | tot_successful_dequeues = 0; | |
272 | unsigned long long end_dequeues = 0; | |
273 | int i, a; | |
274 | ||
275 | if (argc < 4) { | |
276 | show_usage(argc, argv); | |
277 | return -1; | |
278 | } | |
279 | ||
280 | err = sscanf(argv[1], "%u", &nr_dequeuers); | |
281 | if (err != 1) { | |
282 | show_usage(argc, argv); | |
283 | return -1; | |
284 | } | |
285 | ||
286 | err = sscanf(argv[2], "%u", &nr_enqueuers); | |
287 | if (err != 1) { | |
288 | show_usage(argc, argv); | |
289 | return -1; | |
290 | } | |
291 | ||
292 | err = sscanf(argv[3], "%lu", &duration); | |
293 | if (err != 1) { | |
294 | show_usage(argc, argv); | |
295 | return -1; | |
296 | } | |
297 | ||
298 | for (i = 4; i < argc; i++) { | |
299 | if (argv[i][0] != '-') | |
300 | continue; | |
301 | switch (argv[i][1]) { | |
302 | case 'a': | |
303 | if (argc < i + 2) { | |
304 | show_usage(argc, argv); | |
305 | return -1; | |
306 | } | |
307 | a = atoi(argv[++i]); | |
308 | cpu_affinities[next_aff++] = a; | |
309 | use_affinity = 1; | |
310 | printf_verbose("Adding CPU %d affinity\n", a); | |
311 | break; | |
312 | case 'c': | |
313 | if (argc < i + 2) { | |
314 | show_usage(argc, argv); | |
315 | return -1; | |
316 | } | |
317 | rduration = atol(argv[++i]); | |
318 | break; | |
319 | case 'd': | |
320 | if (argc < i + 2) { | |
321 | show_usage(argc, argv); | |
322 | return -1; | |
323 | } | |
324 | wdelay = atol(argv[++i]); | |
325 | break; | |
326 | case 'v': | |
327 | verbose_mode = 1; | |
328 | break; | |
329 | } | |
330 | } | |
331 | ||
332 | printf_verbose("running test for %lu seconds, %u enqueuers, " | |
333 | "%u dequeuers.\n", | |
334 | duration, nr_enqueuers, nr_dequeuers); | |
335 | printf_verbose("Writer delay : %lu loops.\n", rduration); | |
336 | printf_verbose("Reader duration : %lu loops.\n", wdelay); | |
337 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
338 | "main", pthread_self(), (unsigned long)gettid()); | |
339 | ||
340 | tid_enqueuer = malloc(sizeof(*tid_enqueuer) * nr_enqueuers); | |
341 | tid_dequeuer = malloc(sizeof(*tid_dequeuer) * nr_dequeuers); | |
342 | count_enqueuer = malloc(2 * sizeof(*count_enqueuer) * nr_enqueuers); | |
343 | count_dequeuer = malloc(2 * sizeof(*count_dequeuer) * nr_dequeuers); | |
16aa9ee8 | 344 | cds_wfs_init(&s); |
a5eb5ffc MD |
345 | |
346 | next_aff = 0; | |
347 | ||
348 | for (i = 0; i < nr_enqueuers; i++) { | |
349 | err = pthread_create(&tid_enqueuer[i], NULL, thr_enqueuer, | |
350 | &count_enqueuer[2 * i]); | |
351 | if (err != 0) | |
352 | exit(1); | |
353 | } | |
354 | for (i = 0; i < nr_dequeuers; i++) { | |
355 | err = pthread_create(&tid_dequeuer[i], NULL, thr_dequeuer, | |
356 | &count_dequeuer[2 * i]); | |
357 | if (err != 0) | |
358 | exit(1); | |
359 | } | |
360 | ||
5481ddb3 | 361 | cmm_smp_mb(); |
a5eb5ffc MD |
362 | |
363 | test_go = 1; | |
364 | ||
365 | for (i = 0; i < duration; i++) { | |
366 | sleep(1); | |
367 | if (verbose_mode) | |
368 | write (1, ".", 1); | |
369 | } | |
370 | ||
371 | test_stop = 1; | |
372 | ||
373 | for (i = 0; i < nr_enqueuers; i++) { | |
374 | err = pthread_join(tid_enqueuer[i], &tret); | |
375 | if (err != 0) | |
376 | exit(1); | |
377 | tot_enqueues += count_enqueuer[2 * i]; | |
378 | tot_successful_enqueues += count_enqueuer[2 * i + 1]; | |
379 | } | |
380 | for (i = 0; i < nr_dequeuers; i++) { | |
381 | err = pthread_join(tid_dequeuer[i], &tret); | |
382 | if (err != 0) | |
383 | exit(1); | |
384 | tot_dequeues += count_dequeuer[2 * i]; | |
385 | tot_successful_dequeues += count_dequeuer[2 * i + 1]; | |
386 | } | |
387 | ||
388 | test_end(&s, &end_dequeues); | |
389 | ||
390 | printf_verbose("total number of enqueues : %llu, dequeues %llu\n", | |
391 | tot_enqueues, tot_dequeues); | |
392 | printf_verbose("total number of successful enqueues : %llu, " | |
393 | "successful dequeues %llu\n", | |
394 | tot_successful_enqueues, tot_successful_dequeues); | |
395 | printf("SUMMARY %-25s testdur %4lu nr_enqueuers %3u wdelay %6lu " | |
396 | "nr_dequeuers %3u " | |
397 | "rdur %6lu nr_enqueues %12llu nr_dequeues %12llu " | |
398 | "successful enqueues %12llu successful dequeues %12llu " | |
399 | "end_dequeues %llu nr_ops %12llu\n", | |
400 | argv[0], duration, nr_enqueuers, wdelay, | |
401 | nr_dequeuers, rduration, tot_enqueues, tot_dequeues, | |
402 | tot_successful_enqueues, | |
403 | tot_successful_dequeues, end_dequeues, | |
404 | tot_enqueues + tot_dequeues); | |
405 | if (tot_successful_enqueues != tot_successful_dequeues + end_dequeues) | |
406 | printf("WARNING! Discrepancy between nr succ. enqueues %llu vs " | |
407 | "succ. dequeues + end dequeues %llu.\n", | |
408 | tot_successful_enqueues, | |
409 | tot_successful_dequeues + end_dequeues); | |
410 | ||
411 | free(count_enqueuer); | |
412 | free(count_dequeuer); | |
413 | free(tid_enqueuer); | |
414 | free(tid_dequeuer); | |
415 | return 0; | |
416 | } |