Fix tests: finer-grained use of CPU_SET, CPU_ZERO and cpu_set_t
[urcu.git] / tests / test_urcu_gc.c
1 /*
2 * test_urcu_gc.c
3 *
4 * Userspace RCU library - test program (with batch reclamation)
5 *
6 * Copyright February 2009 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
21 */
22
23 #define _GNU_SOURCE
24 #include "../config.h"
25 #include <stdio.h>
26 #include <pthread.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/types.h>
30 #include <sys/wait.h>
31 #include <unistd.h>
32 #include <stdio.h>
33 #include <assert.h>
34 #include <errno.h>
35
36 #include <urcu/arch.h>
37 #include <urcu/tls-compat.h>
38 #include "cpuset.h"
39
40 #ifdef __linux__
41 #include <syscall.h>
42 #endif
43
44 /* hardcoded number of CPUs */
45 #define NR_CPUS 16384
46
47 #if defined(_syscall0)
48 _syscall0(pid_t, gettid)
49 #elif defined(__NR_gettid)
50 static inline pid_t gettid(void)
51 {
52 return syscall(__NR_gettid);
53 }
54 #else
55 #warning "use pid as tid"
56 static inline pid_t gettid(void)
57 {
58 return getpid();
59 }
60 #endif
61
62 #ifndef DYNAMIC_LINK_TEST
63 #define _LGPL_SOURCE
64 #else
65 #define rcu_debug_yield_read()
66 #endif
67 #include <urcu.h>
68
69 struct test_array {
70 int a;
71 };
72
73 static volatile int test_go, test_stop;
74
75 static unsigned long wdelay;
76
77 static struct test_array *test_rcu_pointer;
78
79 static unsigned int reclaim_batch = 1;
80
81 struct reclaim_queue {
82 void **queue; /* Beginning of queue */
83 void **head; /* Insert position */
84 };
85
86 static struct reclaim_queue *pending_reclaims;
87
88 static unsigned long duration;
89
90 /* read-side C.S. duration, in loops */
91 static unsigned long rduration;
92
93 /* write-side C.S. duration, in loops */
94 static unsigned long wduration;
95
96 static inline void loop_sleep(unsigned long loops)
97 {
98 while (loops-- != 0)
99 caa_cpu_relax();
100 }
101
102 static int verbose_mode;
103
104 #define printf_verbose(fmt, args...) \
105 do { \
106 if (verbose_mode) \
107 printf(fmt, args); \
108 } while (0)
109
110 static unsigned int cpu_affinities[NR_CPUS];
111 static unsigned int next_aff = 0;
112 static int use_affinity = 0;
113
114 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
115
116 static void set_affinity(void)
117 {
118 #if HAVE_SCHED_SETAFFINITY
119 cpu_set_t mask;
120 int cpu, ret;
121 #endif /* HAVE_SCHED_SETAFFINITY */
122
123 if (!use_affinity)
124 return;
125
126 #if HAVE_SCHED_SETAFFINITY
127 ret = pthread_mutex_lock(&affinity_mutex);
128 if (ret) {
129 perror("Error in pthread mutex lock");
130 exit(-1);
131 }
132 cpu = cpu_affinities[next_aff++];
133 ret = pthread_mutex_unlock(&affinity_mutex);
134 if (ret) {
135 perror("Error in pthread mutex unlock");
136 exit(-1);
137 }
138
139 CPU_ZERO(&mask);
140 CPU_SET(cpu, &mask);
141 #if SCHED_SETAFFINITY_ARGS == 2
142 sched_setaffinity(0, &mask);
143 #else
144 sched_setaffinity(0, sizeof(mask), &mask);
145 #endif
146 #endif /* HAVE_SCHED_SETAFFINITY */
147 }
148
149 /*
150 * returns 0 if test should end.
151 */
152 static int test_duration_write(void)
153 {
154 return !test_stop;
155 }
156
157 static int test_duration_read(void)
158 {
159 return !test_stop;
160 }
161
162 static DEFINE_URCU_TLS(unsigned long long, nr_writes);
163 static DEFINE_URCU_TLS(unsigned long long, nr_reads);
164
165 static
166 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
167
168 static unsigned int nr_readers;
169 static unsigned int nr_writers;
170
171 pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
172
173 void rcu_copy_mutex_lock(void)
174 {
175 int ret;
176 ret = pthread_mutex_lock(&rcu_copy_mutex);
177 if (ret) {
178 perror("Error in pthread mutex lock");
179 exit(-1);
180 }
181 }
182
183 void rcu_copy_mutex_unlock(void)
184 {
185 int ret;
186
187 ret = pthread_mutex_unlock(&rcu_copy_mutex);
188 if (ret) {
189 perror("Error in pthread mutex unlock");
190 exit(-1);
191 }
192 }
193
194 void *thr_reader(void *_count)
195 {
196 unsigned long long *count = _count;
197 struct test_array *local_ptr;
198
199 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
200 "reader", (unsigned long) pthread_self(),
201 (unsigned long) gettid());
202
203 set_affinity();
204
205 rcu_register_thread();
206
207 while (!test_go)
208 {
209 }
210 cmm_smp_mb();
211
212 for (;;) {
213 rcu_read_lock();
214 local_ptr = rcu_dereference(test_rcu_pointer);
215 rcu_debug_yield_read();
216 if (local_ptr)
217 assert(local_ptr->a == 8);
218 if (caa_unlikely(rduration))
219 loop_sleep(rduration);
220 rcu_read_unlock();
221 URCU_TLS(nr_reads)++;
222 if (caa_unlikely(!test_duration_read()))
223 break;
224 }
225
226 rcu_unregister_thread();
227
228 *count = URCU_TLS(nr_reads);
229 printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
230 "reader", (unsigned long) pthread_self(),
231 (unsigned long) gettid());
232 return ((void*)1);
233
234 }
235
236 static void rcu_gc_clear_queue(unsigned long wtidx)
237 {
238 void **p;
239
240 /* Wait for Q.S and empty queue */
241 synchronize_rcu();
242
243 for (p = pending_reclaims[wtidx].queue;
244 p < pending_reclaims[wtidx].head; p++) {
245 /* poison */
246 if (*p)
247 ((struct test_array *)*p)->a = 0;
248 free(*p);
249 }
250 pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
251 }
252
253 /* Using per-thread queue */
254 static void rcu_gc_reclaim(unsigned long wtidx, void *old)
255 {
256 /* Queue pointer */
257 *pending_reclaims[wtidx].head = old;
258 pending_reclaims[wtidx].head++;
259
260 if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
261 < reclaim_batch))
262 return;
263
264 rcu_gc_clear_queue(wtidx);
265 }
266
267 void *thr_writer(void *data)
268 {
269 unsigned long wtidx = (unsigned long)data;
270 #ifdef TEST_LOCAL_GC
271 struct test_array *old = NULL;
272 #else
273 struct test_array *new, *old;
274 #endif
275
276 printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
277 "writer", (unsigned long) pthread_self(),
278 (unsigned long) gettid());
279
280 set_affinity();
281
282 while (!test_go)
283 {
284 }
285 cmm_smp_mb();
286
287 for (;;) {
288 #ifndef TEST_LOCAL_GC
289 new = malloc(sizeof(*new));
290 new->a = 8;
291 old = rcu_xchg_pointer(&test_rcu_pointer, new);
292 #endif
293 if (caa_unlikely(wduration))
294 loop_sleep(wduration);
295 rcu_gc_reclaim(wtidx, old);
296 URCU_TLS(nr_writes)++;
297 if (caa_unlikely(!test_duration_write()))
298 break;
299 if (caa_unlikely(wdelay))
300 loop_sleep(wdelay);
301 }
302
303 printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
304 "writer", (unsigned long) pthread_self(),
305 (unsigned long) gettid());
306 tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
307 return ((void*)2);
308 }
309
310 void show_usage(int argc, char **argv)
311 {
312 printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
313 #ifdef DEBUG_YIELD
314 printf(" [-r] [-w] (yield reader and/or writer)");
315 #endif
316 printf(" [-d delay] (writer period (us))");
317 printf(" [-c duration] (reader C.S. duration (in loops))");
318 printf(" [-e duration] (writer C.S. duration (in loops))");
319 printf(" [-v] (verbose output)");
320 printf(" [-a cpu#] [-a cpu#]... (affinity)");
321 printf("\n");
322 }
323
324 int main(int argc, char **argv)
325 {
326 int err;
327 pthread_t *tid_reader, *tid_writer;
328 void *tret;
329 unsigned long long *count_reader;
330 unsigned long long tot_reads = 0, tot_writes = 0;
331 int i, a;
332
333 if (argc < 4) {
334 show_usage(argc, argv);
335 return -1;
336 }
337
338 err = sscanf(argv[1], "%u", &nr_readers);
339 if (err != 1) {
340 show_usage(argc, argv);
341 return -1;
342 }
343
344 err = sscanf(argv[2], "%u", &nr_writers);
345 if (err != 1) {
346 show_usage(argc, argv);
347 return -1;
348 }
349
350 err = sscanf(argv[3], "%lu", &duration);
351 if (err != 1) {
352 show_usage(argc, argv);
353 return -1;
354 }
355
356 for (i = 4; i < argc; i++) {
357 if (argv[i][0] != '-')
358 continue;
359 switch (argv[i][1]) {
360 #ifdef DEBUG_YIELD
361 case 'r':
362 rcu_yield_active |= RCU_YIELD_READ;
363 break;
364 case 'w':
365 rcu_yield_active |= RCU_YIELD_WRITE;
366 break;
367 #endif
368 case 'a':
369 if (argc < i + 2) {
370 show_usage(argc, argv);
371 return -1;
372 }
373 a = atoi(argv[++i]);
374 cpu_affinities[next_aff++] = a;
375 use_affinity = 1;
376 printf_verbose("Adding CPU %d affinity\n", a);
377 break;
378 case 'b':
379 if (argc < i + 2) {
380 show_usage(argc, argv);
381 return -1;
382 }
383 reclaim_batch = atol(argv[++i]);
384 break;
385 case 'c':
386 if (argc < i + 2) {
387 show_usage(argc, argv);
388 return -1;
389 }
390 rduration = atol(argv[++i]);
391 break;
392 case 'd':
393 if (argc < i + 2) {
394 show_usage(argc, argv);
395 return -1;
396 }
397 wdelay = atol(argv[++i]);
398 break;
399 case 'e':
400 if (argc < i + 2) {
401 show_usage(argc, argv);
402 return -1;
403 }
404 wduration = atol(argv[++i]);
405 break;
406 case 'v':
407 verbose_mode = 1;
408 break;
409 }
410 }
411
412 printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
413 duration, nr_readers, nr_writers);
414 printf_verbose("Writer delay : %lu loops.\n", wdelay);
415 printf_verbose("Reader duration : %lu loops.\n", rduration);
416 printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
417 "main", (unsigned long) pthread_self(),
418 (unsigned long) gettid());
419
420 tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
421 tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
422 count_reader = malloc(sizeof(*count_reader) * nr_readers);
423 tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers);
424 pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers);
425 if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
426 < CAA_CACHE_LINE_SIZE)
427 for (i = 0; i < nr_writers; i++)
428 pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
429 else
430 for (i = 0; i < nr_writers; i++)
431 pending_reclaims[i].queue = calloc(reclaim_batch,
432 sizeof(*pending_reclaims[i].queue));
433 for (i = 0; i < nr_writers; i++)
434 pending_reclaims[i].head = pending_reclaims[i].queue;
435
436 next_aff = 0;
437
438 for (i = 0; i < nr_readers; i++) {
439 err = pthread_create(&tid_reader[i], NULL, thr_reader,
440 &count_reader[i]);
441 if (err != 0)
442 exit(1);
443 }
444 for (i = 0; i < nr_writers; i++) {
445 err = pthread_create(&tid_writer[i], NULL, thr_writer,
446 (void *)(long)i);
447 if (err != 0)
448 exit(1);
449 }
450
451 cmm_smp_mb();
452
453 test_go = 1;
454
455 sleep(duration);
456
457 test_stop = 1;
458
459 for (i = 0; i < nr_readers; i++) {
460 err = pthread_join(tid_reader[i], &tret);
461 if (err != 0)
462 exit(1);
463 tot_reads += count_reader[i];
464 }
465 for (i = 0; i < nr_writers; i++) {
466 err = pthread_join(tid_writer[i], &tret);
467 if (err != 0)
468 exit(1);
469 tot_writes += tot_nr_writes[i];
470 rcu_gc_clear_queue(i);
471 }
472
473 printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
474 tot_writes);
475 printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
476 "nr_writers %3u "
477 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
478 "batch %u\n",
479 argv[0], duration, nr_readers, rduration, wduration,
480 nr_writers, wdelay, tot_reads, tot_writes,
481 tot_reads + tot_writes, reclaim_batch);
482 free(tid_reader);
483 free(tid_writer);
484 free(count_reader);
485 free(tot_nr_writes);
486 for (i = 0; i < nr_writers; i++)
487 free(pending_reclaims[i].queue);
488 free(pending_reclaims);
489
490 return 0;
491 }
This page took 0.038296 seconds and 4 git commands to generate.