test_rwlock: Add per-thread count to verbose output
[urcu.git] / tests / benchmark / test_urcu_qsbr_gc.c
CommitLineData
de10a585
MD
1/*
2 * test_urcu_gc.c
3 *
4 * Userspace RCU library - test program (with baatch 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
de10a585
MD
23#include <stdio.h>
24#include <pthread.h>
25#include <stdlib.h>
26#include <string.h>
27#include <sys/types.h>
28#include <sys/wait.h>
29#include <unistd.h>
30#include <stdio.h>
31#include <assert.h>
de10a585
MD
32#include <errno.h>
33
34#include <urcu/arch.h>
bd252a04 35#include <urcu/tls-compat.h>
2953b501 36#include "cpuset.h"
94df6318 37#include "thread-id.h"
2650042a 38#include "../common/debug-yield.h"
de10a585
MD
39
40/* hardcoded number of CPUs */
41#define NR_CPUS 16384
42
de10a585
MD
43#define _LGPL_SOURCE
44#include <urcu-qsbr.h>
45
46struct test_array {
47 int a;
48};
49
50static volatile int test_go, test_stop;
51
52static unsigned long wdelay;
53
54static struct test_array *test_rcu_pointer;
55
56static unsigned long duration;
57
58/* read-side C.S. duration, in loops */
59static unsigned long rduration;
60static unsigned int reclaim_batch = 1;
61
62struct reclaim_queue {
63 void **queue; /* Beginning of queue */
64 void **head; /* Insert position */
65};
66
67static struct reclaim_queue *pending_reclaims;
68
69
70/* write-side C.S. duration, in loops */
71static unsigned long wduration;
72
ab0aacbe 73static inline void loop_sleep(unsigned long loops)
de10a585 74{
ab0aacbe 75 while (loops-- != 0)
de10a585
MD
76 caa_cpu_relax();
77}
78
79static int verbose_mode;
80
81#define printf_verbose(fmt, args...) \
82 do { \
83 if (verbose_mode) \
84 printf(fmt, args); \
85 } while (0)
86
87static unsigned int cpu_affinities[NR_CPUS];
88static unsigned int next_aff = 0;
89static int use_affinity = 0;
90
91pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
92
de10a585
MD
93static void set_affinity(void)
94{
95bc7fb9 95#if HAVE_SCHED_SETAFFINITY
de10a585 96 cpu_set_t mask;
95bc7fb9
MD
97 int cpu, ret;
98#endif /* HAVE_SCHED_SETAFFINITY */
de10a585
MD
99
100 if (!use_affinity)
101 return;
102
103#if HAVE_SCHED_SETAFFINITY
104 ret = pthread_mutex_lock(&affinity_mutex);
105 if (ret) {
106 perror("Error in pthread mutex lock");
107 exit(-1);
108 }
109 cpu = cpu_affinities[next_aff++];
110 ret = pthread_mutex_unlock(&affinity_mutex);
111 if (ret) {
112 perror("Error in pthread mutex unlock");
113 exit(-1);
114 }
115
116 CPU_ZERO(&mask);
117 CPU_SET(cpu, &mask);
118#if SCHED_SETAFFINITY_ARGS == 2
119 sched_setaffinity(0, &mask);
120#else
121 sched_setaffinity(0, sizeof(mask), &mask);
122#endif
123#endif /* HAVE_SCHED_SETAFFINITY */
124}
125
126/*
127 * returns 0 if test should end.
128 */
129static int test_duration_write(void)
130{
131 return !test_stop;
132}
133
134static int test_duration_read(void)
135{
136 return !test_stop;
137}
138
bd252a04
MD
139static DEFINE_URCU_TLS(unsigned long long, nr_writes);
140static DEFINE_URCU_TLS(unsigned long long, nr_reads);
de10a585
MD
141
142static unsigned int nr_readers;
143static unsigned int nr_writers;
144
145pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
146static
147unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
148
149
150void rcu_copy_mutex_lock(void)
151{
152 int ret;
153 ret = pthread_mutex_lock(&rcu_copy_mutex);
154 if (ret) {
155 perror("Error in pthread mutex lock");
156 exit(-1);
157 }
158}
159
160void rcu_copy_mutex_unlock(void)
161{
162 int ret;
163
164 ret = pthread_mutex_unlock(&rcu_copy_mutex);
165 if (ret) {
166 perror("Error in pthread mutex unlock");
167 exit(-1);
168 }
169}
170
171void *thr_reader(void *_count)
172{
173 unsigned long long *count = _count;
174 struct test_array *local_ptr;
175
94df6318
MD
176 printf_verbose("thread_begin %s, tid %lu\n",
177 "reader", urcu_get_thread_id());
de10a585
MD
178
179 set_affinity();
180
181 rcu_register_thread();
182
183 while (!test_go)
184 {
185 }
186 cmm_smp_mb();
187
188 for (;;) {
189 _rcu_read_lock();
190 local_ptr = _rcu_dereference(test_rcu_pointer);
1de4df4b 191 rcu_debug_yield_read();
de10a585
MD
192 if (local_ptr)
193 assert(local_ptr->a == 8);
a0b7f7ea 194 if (caa_unlikely(rduration))
de10a585
MD
195 loop_sleep(rduration);
196 _rcu_read_unlock();
bd252a04 197 URCU_TLS(nr_reads)++;
de10a585 198 /* QS each 1024 reads */
bd252a04 199 if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
de10a585 200 _rcu_quiescent_state();
a0b7f7ea 201 if (caa_unlikely(!test_duration_read()))
de10a585
MD
202 break;
203 }
204
205 rcu_unregister_thread();
206
bd252a04 207 *count = URCU_TLS(nr_reads);
94df6318
MD
208 printf_verbose("thread_end %s, tid %lu\n",
209 "reader", urcu_get_thread_id());
de10a585
MD
210 return ((void*)1);
211
212}
213
214static void rcu_gc_clear_queue(unsigned long wtidx)
215{
216 void **p;
217
218 /* Wait for Q.S and empty queue */
219 synchronize_rcu();
220
221 for (p = pending_reclaims[wtidx].queue;
222 p < pending_reclaims[wtidx].head; p++) {
223 /* poison */
224 if (*p)
225 ((struct test_array *)*p)->a = 0;
226 free(*p);
227 }
228 pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue;
229}
230
231/* Using per-thread queue */
232static void rcu_gc_reclaim(unsigned long wtidx, void *old)
233{
234 /* Queue pointer */
235 *pending_reclaims[wtidx].head = old;
236 pending_reclaims[wtidx].head++;
237
a0b7f7ea 238 if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue
de10a585
MD
239 < reclaim_batch))
240 return;
241
242 rcu_gc_clear_queue(wtidx);
243}
244
245void *thr_writer(void *data)
246{
247 unsigned long wtidx = (unsigned long)data;
248#ifdef TEST_LOCAL_GC
249 struct test_array *old = NULL;
250#else
251 struct test_array *new, *old;
252#endif
253
94df6318
MD
254 printf_verbose("thread_begin %s, tid %lu\n",
255 "writer", urcu_get_thread_id());
de10a585
MD
256
257 set_affinity();
258
259 while (!test_go)
260 {
261 }
262 cmm_smp_mb();
263
264 for (;;) {
265#ifndef TEST_LOCAL_GC
266 new = malloc(sizeof(*new));
267 new->a = 8;
268 old = _rcu_xchg_pointer(&test_rcu_pointer, new);
269#endif
a0b7f7ea 270 if (caa_unlikely(wduration))
de10a585
MD
271 loop_sleep(wduration);
272 rcu_gc_reclaim(wtidx, old);
bd252a04 273 URCU_TLS(nr_writes)++;
a0b7f7ea 274 if (caa_unlikely(!test_duration_write()))
de10a585 275 break;
a0b7f7ea 276 if (caa_unlikely(wdelay))
de10a585
MD
277 loop_sleep(wdelay);
278 }
279
94df6318
MD
280 printf_verbose("thread_end %s, tid %lu\n",
281 "writer", urcu_get_thread_id());
bd252a04 282 tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
de10a585
MD
283 return ((void*)2);
284}
285
286void show_usage(int argc, char **argv)
287{
06637138
MD
288 printf("Usage : %s nr_readers nr_writers duration (s) <OPTIONS>\n",
289 argv[0]);
290 printf("OPTIONS:\n");
06637138 291 printf(" [-r] [-w] (yield reader and/or writer)\n");
06637138
MD
292 printf(" [-b batch] (batch reclaim)\n");
293 printf(" [-d delay] (writer period (us))\n");
294 printf(" [-c duration] (reader C.S. duration (in loops))\n");
295 printf(" [-e duration] (writer C.S. duration (in loops))\n");
296 printf(" [-v] (verbose output)\n");
297 printf(" [-a cpu#] [-a cpu#]... (affinity)\n");
de10a585
MD
298 printf("\n");
299}
300
301int main(int argc, char **argv)
302{
303 int err;
304 pthread_t *tid_reader, *tid_writer;
305 void *tret;
306 unsigned long long *count_reader;
307 unsigned long long tot_reads = 0, tot_writes = 0;
308 int i, a;
309
310 if (argc < 4) {
311 show_usage(argc, argv);
312 return -1;
313 }
314
315 err = sscanf(argv[1], "%u", &nr_readers);
316 if (err != 1) {
317 show_usage(argc, argv);
318 return -1;
319 }
320
321 err = sscanf(argv[2], "%u", &nr_writers);
322 if (err != 1) {
323 show_usage(argc, argv);
324 return -1;
325 }
326
327 err = sscanf(argv[3], "%lu", &duration);
328 if (err != 1) {
329 show_usage(argc, argv);
330 return -1;
331 }
332
333 for (i = 4; i < argc; i++) {
334 if (argv[i][0] != '-')
335 continue;
336 switch (argv[i][1]) {
de10a585 337 case 'r':
2650042a 338 rcu_debug_yield_enable(RCU_YIELD_READ);
de10a585
MD
339 break;
340 case 'w':
2650042a 341 rcu_debug_yield_enable(RCU_YIELD_WRITE);
de10a585 342 break;
de10a585
MD
343 case 'a':
344 if (argc < i + 2) {
345 show_usage(argc, argv);
346 return -1;
347 }
348 a = atoi(argv[++i]);
349 cpu_affinities[next_aff++] = a;
350 use_affinity = 1;
351 printf_verbose("Adding CPU %d affinity\n", a);
352 break;
353 case 'b':
354 if (argc < i + 2) {
355 show_usage(argc, argv);
356 return -1;
357 }
358 reclaim_batch = atol(argv[++i]);
359 break;
360 case 'c':
361 if (argc < i + 2) {
362 show_usage(argc, argv);
363 return -1;
364 }
365 rduration = atol(argv[++i]);
366 break;
367 case 'd':
368 if (argc < i + 2) {
369 show_usage(argc, argv);
370 return -1;
371 }
372 wdelay = atol(argv[++i]);
373 break;
374 case 'e':
375 if (argc < i + 2) {
376 show_usage(argc, argv);
377 return -1;
378 }
379 wduration = atol(argv[++i]);
380 break;
381 case 'v':
382 verbose_mode = 1;
383 break;
384 }
385 }
386
387 printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
388 duration, nr_readers, nr_writers);
389 printf_verbose("Writer delay : %lu loops.\n", wdelay);
390 printf_verbose("Reader duration : %lu loops.\n", rduration);
94df6318
MD
391 printf_verbose("thread %-6s, tid %lu\n",
392 "main", urcu_get_thread_id());
de10a585 393
9aa14175
MD
394 tid_reader = calloc(nr_readers, sizeof(*tid_reader));
395 tid_writer = calloc(nr_writers, sizeof(*tid_writer));
396 count_reader = calloc(nr_readers, sizeof(*count_reader));
397 tot_nr_writes = calloc(nr_writers, sizeof(*tot_nr_writes));
398 pending_reclaims = calloc(nr_writers, sizeof(*pending_reclaims));
de10a585
MD
399 if (reclaim_batch * sizeof(*pending_reclaims[i].queue)
400 < CAA_CACHE_LINE_SIZE)
401 for (i = 0; i < nr_writers; i++)
402 pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE);
403 else
404 for (i = 0; i < nr_writers; i++)
405 pending_reclaims[i].queue = calloc(reclaim_batch,
406 sizeof(*pending_reclaims[i].queue));
407 for (i = 0; i < nr_writers; i++)
408 pending_reclaims[i].head = pending_reclaims[i].queue;
409
410 next_aff = 0;
411
412 for (i = 0; i < nr_readers; i++) {
413 err = pthread_create(&tid_reader[i], NULL, thr_reader,
414 &count_reader[i]);
415 if (err != 0)
416 exit(1);
417 }
418 for (i = 0; i < nr_writers; i++) {
419 err = pthread_create(&tid_writer[i], NULL, thr_writer,
420 (void *)(long)i);
421 if (err != 0)
422 exit(1);
423 }
424
425 cmm_smp_mb();
426
427 test_go = 1;
428
429 sleep(duration);
430
431 test_stop = 1;
432
433 for (i = 0; i < nr_readers; i++) {
434 err = pthread_join(tid_reader[i], &tret);
435 if (err != 0)
436 exit(1);
437 tot_reads += count_reader[i];
438 }
439 for (i = 0; i < nr_writers; i++) {
440 err = pthread_join(tid_writer[i], &tret);
441 if (err != 0)
442 exit(1);
443 tot_writes += tot_nr_writes[i];
444 rcu_gc_clear_queue(i);
445 }
446
447 printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
448 tot_writes);
449 printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu "
450 "nr_writers %3u "
451 "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu "
452 "batch %u\n",
453 argv[0], duration, nr_readers, rduration, wduration,
454 nr_writers, wdelay, tot_reads, tot_writes,
455 tot_reads + tot_writes, reclaim_batch);
456 free(tid_reader);
457 free(tid_writer);
458 free(count_reader);
459 free(tot_nr_writes);
460 for (i = 0; i < nr_writers; i++)
461 free(pending_reclaims[i].queue);
462 free(pending_reclaims);
463
464 return 0;
465}
This page took 0.056232 seconds and 4 git commands to generate.