Commit | Line | Data |
---|---|---|
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 | ||
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> | |
de10a585 MD |
34 | #include <errno.h> |
35 | ||
36 | #include <urcu/arch.h> | |
bd252a04 | 37 | #include <urcu/tls-compat.h> |
95d8822d | 38 | #include "cpuset.h" |
de10a585 MD |
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 | #define _LGPL_SOURCE | |
63 | #include <urcu-qsbr.h> | |
64 | ||
65 | struct test_array { | |
66 | int a; | |
67 | }; | |
68 | ||
69 | static volatile int test_go, test_stop; | |
70 | ||
71 | static unsigned long wdelay; | |
72 | ||
73 | static struct test_array *test_rcu_pointer; | |
74 | ||
75 | static unsigned long duration; | |
76 | ||
77 | /* read-side C.S. duration, in loops */ | |
78 | static unsigned long rduration; | |
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 | ||
89 | /* write-side C.S. duration, in loops */ | |
90 | static unsigned long wduration; | |
91 | ||
92 | static inline void loop_sleep(unsigned long l) | |
93 | { | |
94 | while(l-- != 0) | |
95 | caa_cpu_relax(); | |
96 | } | |
97 | ||
98 | static int verbose_mode; | |
99 | ||
100 | #define printf_verbose(fmt, args...) \ | |
101 | do { \ | |
102 | if (verbose_mode) \ | |
103 | printf(fmt, args); \ | |
104 | } while (0) | |
105 | ||
106 | static unsigned int cpu_affinities[NR_CPUS]; | |
107 | static unsigned int next_aff = 0; | |
108 | static int use_affinity = 0; | |
109 | ||
110 | pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER; | |
111 | ||
de10a585 MD |
112 | static void set_affinity(void) |
113 | { | |
114 | cpu_set_t mask; | |
115 | int cpu; | |
116 | int ret; | |
117 | ||
118 | if (!use_affinity) | |
119 | return; | |
120 | ||
121 | #if HAVE_SCHED_SETAFFINITY | |
122 | ret = pthread_mutex_lock(&affinity_mutex); | |
123 | if (ret) { | |
124 | perror("Error in pthread mutex lock"); | |
125 | exit(-1); | |
126 | } | |
127 | cpu = cpu_affinities[next_aff++]; | |
128 | ret = pthread_mutex_unlock(&affinity_mutex); | |
129 | if (ret) { | |
130 | perror("Error in pthread mutex unlock"); | |
131 | exit(-1); | |
132 | } | |
133 | ||
134 | CPU_ZERO(&mask); | |
135 | CPU_SET(cpu, &mask); | |
136 | #if SCHED_SETAFFINITY_ARGS == 2 | |
137 | sched_setaffinity(0, &mask); | |
138 | #else | |
139 | sched_setaffinity(0, sizeof(mask), &mask); | |
140 | #endif | |
141 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
142 | } | |
143 | ||
144 | /* | |
145 | * returns 0 if test should end. | |
146 | */ | |
147 | static int test_duration_write(void) | |
148 | { | |
149 | return !test_stop; | |
150 | } | |
151 | ||
152 | static int test_duration_read(void) | |
153 | { | |
154 | return !test_stop; | |
155 | } | |
156 | ||
bd252a04 MD |
157 | static DEFINE_URCU_TLS(unsigned long long, nr_writes); |
158 | static DEFINE_URCU_TLS(unsigned long long, nr_reads); | |
de10a585 MD |
159 | |
160 | static unsigned int nr_readers; | |
161 | static unsigned int nr_writers; | |
162 | ||
163 | pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER; | |
164 | static | |
165 | unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes; | |
166 | ||
167 | ||
168 | void rcu_copy_mutex_lock(void) | |
169 | { | |
170 | int ret; | |
171 | ret = pthread_mutex_lock(&rcu_copy_mutex); | |
172 | if (ret) { | |
173 | perror("Error in pthread mutex lock"); | |
174 | exit(-1); | |
175 | } | |
176 | } | |
177 | ||
178 | void rcu_copy_mutex_unlock(void) | |
179 | { | |
180 | int ret; | |
181 | ||
182 | ret = pthread_mutex_unlock(&rcu_copy_mutex); | |
183 | if (ret) { | |
184 | perror("Error in pthread mutex unlock"); | |
185 | exit(-1); | |
186 | } | |
187 | } | |
188 | ||
189 | void *thr_reader(void *_count) | |
190 | { | |
191 | unsigned long long *count = _count; | |
192 | struct test_array *local_ptr; | |
193 | ||
194 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
2d72fae2 MD |
195 | "reader", (unsigned long) pthread_self(), |
196 | (unsigned long) gettid()); | |
de10a585 MD |
197 | |
198 | set_affinity(); | |
199 | ||
200 | rcu_register_thread(); | |
201 | ||
202 | while (!test_go) | |
203 | { | |
204 | } | |
205 | cmm_smp_mb(); | |
206 | ||
207 | for (;;) { | |
208 | _rcu_read_lock(); | |
209 | local_ptr = _rcu_dereference(test_rcu_pointer); | |
210 | debug_yield_read(); | |
211 | if (local_ptr) | |
212 | assert(local_ptr->a == 8); | |
a0b7f7ea | 213 | if (caa_unlikely(rduration)) |
de10a585 MD |
214 | loop_sleep(rduration); |
215 | _rcu_read_unlock(); | |
bd252a04 | 216 | URCU_TLS(nr_reads)++; |
de10a585 | 217 | /* QS each 1024 reads */ |
bd252a04 | 218 | if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0)) |
de10a585 | 219 | _rcu_quiescent_state(); |
a0b7f7ea | 220 | if (caa_unlikely(!test_duration_read())) |
de10a585 MD |
221 | break; |
222 | } | |
223 | ||
224 | rcu_unregister_thread(); | |
225 | ||
bd252a04 | 226 | *count = URCU_TLS(nr_reads); |
de10a585 | 227 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", |
2d72fae2 MD |
228 | "reader", (unsigned long) pthread_self(), |
229 | (unsigned long) gettid()); | |
de10a585 MD |
230 | return ((void*)1); |
231 | ||
232 | } | |
233 | ||
234 | static void rcu_gc_clear_queue(unsigned long wtidx) | |
235 | { | |
236 | void **p; | |
237 | ||
238 | /* Wait for Q.S and empty queue */ | |
239 | synchronize_rcu(); | |
240 | ||
241 | for (p = pending_reclaims[wtidx].queue; | |
242 | p < pending_reclaims[wtidx].head; p++) { | |
243 | /* poison */ | |
244 | if (*p) | |
245 | ((struct test_array *)*p)->a = 0; | |
246 | free(*p); | |
247 | } | |
248 | pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue; | |
249 | } | |
250 | ||
251 | /* Using per-thread queue */ | |
252 | static void rcu_gc_reclaim(unsigned long wtidx, void *old) | |
253 | { | |
254 | /* Queue pointer */ | |
255 | *pending_reclaims[wtidx].head = old; | |
256 | pending_reclaims[wtidx].head++; | |
257 | ||
a0b7f7ea | 258 | if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue |
de10a585 MD |
259 | < reclaim_batch)) |
260 | return; | |
261 | ||
262 | rcu_gc_clear_queue(wtidx); | |
263 | } | |
264 | ||
265 | void *thr_writer(void *data) | |
266 | { | |
267 | unsigned long wtidx = (unsigned long)data; | |
268 | #ifdef TEST_LOCAL_GC | |
269 | struct test_array *old = NULL; | |
270 | #else | |
271 | struct test_array *new, *old; | |
272 | #endif | |
273 | ||
274 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
2d72fae2 MD |
275 | "writer", (unsigned long) pthread_self(), |
276 | (unsigned long) gettid()); | |
de10a585 MD |
277 | |
278 | set_affinity(); | |
279 | ||
280 | while (!test_go) | |
281 | { | |
282 | } | |
283 | cmm_smp_mb(); | |
284 | ||
285 | for (;;) { | |
286 | #ifndef TEST_LOCAL_GC | |
287 | new = malloc(sizeof(*new)); | |
288 | new->a = 8; | |
289 | old = _rcu_xchg_pointer(&test_rcu_pointer, new); | |
290 | #endif | |
a0b7f7ea | 291 | if (caa_unlikely(wduration)) |
de10a585 MD |
292 | loop_sleep(wduration); |
293 | rcu_gc_reclaim(wtidx, old); | |
bd252a04 | 294 | URCU_TLS(nr_writes)++; |
a0b7f7ea | 295 | if (caa_unlikely(!test_duration_write())) |
de10a585 | 296 | break; |
a0b7f7ea | 297 | if (caa_unlikely(wdelay)) |
de10a585 MD |
298 | loop_sleep(wdelay); |
299 | } | |
300 | ||
301 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", | |
2d72fae2 MD |
302 | "writer", (unsigned long) pthread_self(), |
303 | (unsigned long) gettid()); | |
bd252a04 | 304 | tot_nr_writes[wtidx] = URCU_TLS(nr_writes); |
de10a585 MD |
305 | return ((void*)2); |
306 | } | |
307 | ||
308 | void show_usage(int argc, char **argv) | |
309 | { | |
310 | printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]); | |
311 | #ifdef DEBUG_YIELD | |
312 | printf(" [-r] [-w] (yield reader and/or writer)"); | |
313 | #endif | |
314 | printf(" [-b batch] (batch reclaim)"); | |
315 | printf(" [-d delay] (writer period (us))"); | |
316 | printf(" [-c duration] (reader C.S. duration (in loops))"); | |
317 | printf(" [-e duration] (writer C.S. duration (in loops))"); | |
318 | printf(" [-v] (verbose output)"); | |
319 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
320 | printf("\n"); | |
321 | } | |
322 | ||
323 | int main(int argc, char **argv) | |
324 | { | |
325 | int err; | |
326 | pthread_t *tid_reader, *tid_writer; | |
327 | void *tret; | |
328 | unsigned long long *count_reader; | |
329 | unsigned long long tot_reads = 0, tot_writes = 0; | |
330 | int i, a; | |
331 | ||
332 | if (argc < 4) { | |
333 | show_usage(argc, argv); | |
334 | return -1; | |
335 | } | |
336 | ||
337 | err = sscanf(argv[1], "%u", &nr_readers); | |
338 | if (err != 1) { | |
339 | show_usage(argc, argv); | |
340 | return -1; | |
341 | } | |
342 | ||
343 | err = sscanf(argv[2], "%u", &nr_writers); | |
344 | if (err != 1) { | |
345 | show_usage(argc, argv); | |
346 | return -1; | |
347 | } | |
348 | ||
349 | err = sscanf(argv[3], "%lu", &duration); | |
350 | if (err != 1) { | |
351 | show_usage(argc, argv); | |
352 | return -1; | |
353 | } | |
354 | ||
355 | for (i = 4; i < argc; i++) { | |
356 | if (argv[i][0] != '-') | |
357 | continue; | |
358 | switch (argv[i][1]) { | |
359 | #ifdef DEBUG_YIELD | |
360 | case 'r': | |
361 | yield_active |= YIELD_READ; | |
362 | break; | |
363 | case 'w': | |
364 | yield_active |= YIELD_WRITE; | |
365 | break; | |
366 | #endif | |
367 | case 'a': | |
368 | if (argc < i + 2) { | |
369 | show_usage(argc, argv); | |
370 | return -1; | |
371 | } | |
372 | a = atoi(argv[++i]); | |
373 | cpu_affinities[next_aff++] = a; | |
374 | use_affinity = 1; | |
375 | printf_verbose("Adding CPU %d affinity\n", a); | |
376 | break; | |
377 | case 'b': | |
378 | if (argc < i + 2) { | |
379 | show_usage(argc, argv); | |
380 | return -1; | |
381 | } | |
382 | reclaim_batch = atol(argv[++i]); | |
383 | break; | |
384 | case 'c': | |
385 | if (argc < i + 2) { | |
386 | show_usage(argc, argv); | |
387 | return -1; | |
388 | } | |
389 | rduration = atol(argv[++i]); | |
390 | break; | |
391 | case 'd': | |
392 | if (argc < i + 2) { | |
393 | show_usage(argc, argv); | |
394 | return -1; | |
395 | } | |
396 | wdelay = atol(argv[++i]); | |
397 | break; | |
398 | case 'e': | |
399 | if (argc < i + 2) { | |
400 | show_usage(argc, argv); | |
401 | return -1; | |
402 | } | |
403 | wduration = atol(argv[++i]); | |
404 | break; | |
405 | case 'v': | |
406 | verbose_mode = 1; | |
407 | break; | |
408 | } | |
409 | } | |
410 | ||
411 | printf_verbose("running test for %lu seconds, %u readers, %u writers.\n", | |
412 | duration, nr_readers, nr_writers); | |
413 | printf_verbose("Writer delay : %lu loops.\n", wdelay); | |
414 | printf_verbose("Reader duration : %lu loops.\n", rduration); | |
415 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
2d72fae2 MD |
416 | "main", (unsigned long) pthread_self(), |
417 | (unsigned long) gettid()); | |
de10a585 MD |
418 | |
419 | tid_reader = malloc(sizeof(*tid_reader) * nr_readers); | |
420 | tid_writer = malloc(sizeof(*tid_writer) * nr_writers); | |
421 | count_reader = malloc(sizeof(*count_reader) * nr_readers); | |
422 | tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers); | |
423 | pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers); | |
424 | if (reclaim_batch * sizeof(*pending_reclaims[i].queue) | |
425 | < CAA_CACHE_LINE_SIZE) | |
426 | for (i = 0; i < nr_writers; i++) | |
427 | pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE); | |
428 | else | |
429 | for (i = 0; i < nr_writers; i++) | |
430 | pending_reclaims[i].queue = calloc(reclaim_batch, | |
431 | sizeof(*pending_reclaims[i].queue)); | |
432 | for (i = 0; i < nr_writers; i++) | |
433 | pending_reclaims[i].head = pending_reclaims[i].queue; | |
434 | ||
435 | next_aff = 0; | |
436 | ||
437 | for (i = 0; i < nr_readers; i++) { | |
438 | err = pthread_create(&tid_reader[i], NULL, thr_reader, | |
439 | &count_reader[i]); | |
440 | if (err != 0) | |
441 | exit(1); | |
442 | } | |
443 | for (i = 0; i < nr_writers; i++) { | |
444 | err = pthread_create(&tid_writer[i], NULL, thr_writer, | |
445 | (void *)(long)i); | |
446 | if (err != 0) | |
447 | exit(1); | |
448 | } | |
449 | ||
450 | cmm_smp_mb(); | |
451 | ||
452 | test_go = 1; | |
453 | ||
454 | sleep(duration); | |
455 | ||
456 | test_stop = 1; | |
457 | ||
458 | for (i = 0; i < nr_readers; i++) { | |
459 | err = pthread_join(tid_reader[i], &tret); | |
460 | if (err != 0) | |
461 | exit(1); | |
462 | tot_reads += count_reader[i]; | |
463 | } | |
464 | for (i = 0; i < nr_writers; i++) { | |
465 | err = pthread_join(tid_writer[i], &tret); | |
466 | if (err != 0) | |
467 | exit(1); | |
468 | tot_writes += tot_nr_writes[i]; | |
469 | rcu_gc_clear_queue(i); | |
470 | } | |
471 | ||
472 | printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads, | |
473 | tot_writes); | |
474 | printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu " | |
475 | "nr_writers %3u " | |
476 | "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu " | |
477 | "batch %u\n", | |
478 | argv[0], duration, nr_readers, rduration, wduration, | |
479 | nr_writers, wdelay, tot_reads, tot_writes, | |
480 | tot_reads + tot_writes, reclaim_batch); | |
481 | free(tid_reader); | |
482 | free(tid_writer); | |
483 | free(count_reader); | |
484 | free(tot_nr_writes); | |
485 | for (i = 0; i < nr_writers; i++) | |
486 | free(pending_reclaims[i].queue); | |
487 | free(pending_reclaims); | |
488 | ||
489 | return 0; | |
490 | } |