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> | |
34 | #include <sched.h> | |
35 | #include <errno.h> | |
36 | ||
37 | #include <urcu/arch.h> | |
bd252a04 | 38 | #include <urcu/tls-compat.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 | ||
ab0aacbe | 92 | static inline void loop_sleep(unsigned long loops) |
de10a585 | 93 | { |
ab0aacbe | 94 | while (loops-- != 0) |
de10a585 MD |
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 | ||
112 | #ifndef HAVE_CPU_SET_T | |
113 | typedef unsigned long cpu_set_t; | |
114 | # define CPU_ZERO(cpuset) do { *(cpuset) = 0; } while(0) | |
115 | # define CPU_SET(cpu, cpuset) do { *(cpuset) |= (1UL << (cpu)); } while(0) | |
116 | #endif | |
117 | ||
118 | static void set_affinity(void) | |
119 | { | |
95bc7fb9 | 120 | #if HAVE_SCHED_SETAFFINITY |
de10a585 | 121 | cpu_set_t mask; |
95bc7fb9 MD |
122 | int cpu, ret; |
123 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
de10a585 MD |
124 | |
125 | if (!use_affinity) | |
126 | return; | |
127 | ||
128 | #if HAVE_SCHED_SETAFFINITY | |
129 | ret = pthread_mutex_lock(&affinity_mutex); | |
130 | if (ret) { | |
131 | perror("Error in pthread mutex lock"); | |
132 | exit(-1); | |
133 | } | |
134 | cpu = cpu_affinities[next_aff++]; | |
135 | ret = pthread_mutex_unlock(&affinity_mutex); | |
136 | if (ret) { | |
137 | perror("Error in pthread mutex unlock"); | |
138 | exit(-1); | |
139 | } | |
140 | ||
141 | CPU_ZERO(&mask); | |
142 | CPU_SET(cpu, &mask); | |
143 | #if SCHED_SETAFFINITY_ARGS == 2 | |
144 | sched_setaffinity(0, &mask); | |
145 | #else | |
146 | sched_setaffinity(0, sizeof(mask), &mask); | |
147 | #endif | |
148 | #endif /* HAVE_SCHED_SETAFFINITY */ | |
149 | } | |
150 | ||
151 | /* | |
152 | * returns 0 if test should end. | |
153 | */ | |
154 | static int test_duration_write(void) | |
155 | { | |
156 | return !test_stop; | |
157 | } | |
158 | ||
159 | static int test_duration_read(void) | |
160 | { | |
161 | return !test_stop; | |
162 | } | |
163 | ||
bd252a04 MD |
164 | static DEFINE_URCU_TLS(unsigned long long, nr_writes); |
165 | static DEFINE_URCU_TLS(unsigned long long, nr_reads); | |
de10a585 MD |
166 | |
167 | static unsigned int nr_readers; | |
168 | static unsigned int nr_writers; | |
169 | ||
170 | pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER; | |
171 | static | |
172 | unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes; | |
173 | ||
174 | ||
175 | void rcu_copy_mutex_lock(void) | |
176 | { | |
177 | int ret; | |
178 | ret = pthread_mutex_lock(&rcu_copy_mutex); | |
179 | if (ret) { | |
180 | perror("Error in pthread mutex lock"); | |
181 | exit(-1); | |
182 | } | |
183 | } | |
184 | ||
185 | void rcu_copy_mutex_unlock(void) | |
186 | { | |
187 | int ret; | |
188 | ||
189 | ret = pthread_mutex_unlock(&rcu_copy_mutex); | |
190 | if (ret) { | |
191 | perror("Error in pthread mutex unlock"); | |
192 | exit(-1); | |
193 | } | |
194 | } | |
195 | ||
196 | void *thr_reader(void *_count) | |
197 | { | |
198 | unsigned long long *count = _count; | |
199 | struct test_array *local_ptr; | |
200 | ||
201 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
202 | "reader", pthread_self(), (unsigned long)gettid()); | |
203 | ||
204 | set_affinity(); | |
205 | ||
206 | rcu_register_thread(); | |
207 | ||
208 | while (!test_go) | |
209 | { | |
210 | } | |
211 | cmm_smp_mb(); | |
212 | ||
213 | for (;;) { | |
214 | _rcu_read_lock(); | |
215 | local_ptr = _rcu_dereference(test_rcu_pointer); | |
216 | debug_yield_read(); | |
217 | if (local_ptr) | |
218 | assert(local_ptr->a == 8); | |
a0b7f7ea | 219 | if (caa_unlikely(rduration)) |
de10a585 MD |
220 | loop_sleep(rduration); |
221 | _rcu_read_unlock(); | |
bd252a04 | 222 | URCU_TLS(nr_reads)++; |
de10a585 | 223 | /* QS each 1024 reads */ |
bd252a04 | 224 | if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0)) |
de10a585 | 225 | _rcu_quiescent_state(); |
a0b7f7ea | 226 | if (caa_unlikely(!test_duration_read())) |
de10a585 MD |
227 | break; |
228 | } | |
229 | ||
230 | rcu_unregister_thread(); | |
231 | ||
bd252a04 | 232 | *count = URCU_TLS(nr_reads); |
de10a585 MD |
233 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", |
234 | "reader", pthread_self(), (unsigned long)gettid()); | |
235 | return ((void*)1); | |
236 | ||
237 | } | |
238 | ||
239 | static void rcu_gc_clear_queue(unsigned long wtidx) | |
240 | { | |
241 | void **p; | |
242 | ||
243 | /* Wait for Q.S and empty queue */ | |
244 | synchronize_rcu(); | |
245 | ||
246 | for (p = pending_reclaims[wtidx].queue; | |
247 | p < pending_reclaims[wtidx].head; p++) { | |
248 | /* poison */ | |
249 | if (*p) | |
250 | ((struct test_array *)*p)->a = 0; | |
251 | free(*p); | |
252 | } | |
253 | pending_reclaims[wtidx].head = pending_reclaims[wtidx].queue; | |
254 | } | |
255 | ||
256 | /* Using per-thread queue */ | |
257 | static void rcu_gc_reclaim(unsigned long wtidx, void *old) | |
258 | { | |
259 | /* Queue pointer */ | |
260 | *pending_reclaims[wtidx].head = old; | |
261 | pending_reclaims[wtidx].head++; | |
262 | ||
a0b7f7ea | 263 | if (caa_likely(pending_reclaims[wtidx].head - pending_reclaims[wtidx].queue |
de10a585 MD |
264 | < reclaim_batch)) |
265 | return; | |
266 | ||
267 | rcu_gc_clear_queue(wtidx); | |
268 | } | |
269 | ||
270 | void *thr_writer(void *data) | |
271 | { | |
272 | unsigned long wtidx = (unsigned long)data; | |
273 | #ifdef TEST_LOCAL_GC | |
274 | struct test_array *old = NULL; | |
275 | #else | |
276 | struct test_array *new, *old; | |
277 | #endif | |
278 | ||
279 | printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n", | |
280 | "writer", pthread_self(), (unsigned long)gettid()); | |
281 | ||
282 | set_affinity(); | |
283 | ||
284 | while (!test_go) | |
285 | { | |
286 | } | |
287 | cmm_smp_mb(); | |
288 | ||
289 | for (;;) { | |
290 | #ifndef TEST_LOCAL_GC | |
291 | new = malloc(sizeof(*new)); | |
292 | new->a = 8; | |
293 | old = _rcu_xchg_pointer(&test_rcu_pointer, new); | |
294 | #endif | |
a0b7f7ea | 295 | if (caa_unlikely(wduration)) |
de10a585 MD |
296 | loop_sleep(wduration); |
297 | rcu_gc_reclaim(wtidx, old); | |
bd252a04 | 298 | URCU_TLS(nr_writes)++; |
a0b7f7ea | 299 | if (caa_unlikely(!test_duration_write())) |
de10a585 | 300 | break; |
a0b7f7ea | 301 | if (caa_unlikely(wdelay)) |
de10a585 MD |
302 | loop_sleep(wdelay); |
303 | } | |
304 | ||
305 | printf_verbose("thread_end %s, thread id : %lx, tid %lu\n", | |
306 | "writer", pthread_self(), (unsigned long)gettid()); | |
bd252a04 | 307 | tot_nr_writes[wtidx] = URCU_TLS(nr_writes); |
de10a585 MD |
308 | return ((void*)2); |
309 | } | |
310 | ||
311 | void show_usage(int argc, char **argv) | |
312 | { | |
313 | printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]); | |
314 | #ifdef DEBUG_YIELD | |
315 | printf(" [-r] [-w] (yield reader and/or writer)"); | |
316 | #endif | |
317 | printf(" [-b batch] (batch reclaim)"); | |
318 | printf(" [-d delay] (writer period (us))"); | |
319 | printf(" [-c duration] (reader C.S. duration (in loops))"); | |
320 | printf(" [-e duration] (writer C.S. duration (in loops))"); | |
321 | printf(" [-v] (verbose output)"); | |
322 | printf(" [-a cpu#] [-a cpu#]... (affinity)"); | |
323 | printf("\n"); | |
324 | } | |
325 | ||
326 | int main(int argc, char **argv) | |
327 | { | |
328 | int err; | |
329 | pthread_t *tid_reader, *tid_writer; | |
330 | void *tret; | |
331 | unsigned long long *count_reader; | |
332 | unsigned long long tot_reads = 0, tot_writes = 0; | |
333 | int i, a; | |
334 | ||
335 | if (argc < 4) { | |
336 | show_usage(argc, argv); | |
337 | return -1; | |
338 | } | |
339 | ||
340 | err = sscanf(argv[1], "%u", &nr_readers); | |
341 | if (err != 1) { | |
342 | show_usage(argc, argv); | |
343 | return -1; | |
344 | } | |
345 | ||
346 | err = sscanf(argv[2], "%u", &nr_writers); | |
347 | if (err != 1) { | |
348 | show_usage(argc, argv); | |
349 | return -1; | |
350 | } | |
351 | ||
352 | err = sscanf(argv[3], "%lu", &duration); | |
353 | if (err != 1) { | |
354 | show_usage(argc, argv); | |
355 | return -1; | |
356 | } | |
357 | ||
358 | for (i = 4; i < argc; i++) { | |
359 | if (argv[i][0] != '-') | |
360 | continue; | |
361 | switch (argv[i][1]) { | |
362 | #ifdef DEBUG_YIELD | |
363 | case 'r': | |
364 | yield_active |= YIELD_READ; | |
365 | break; | |
366 | case 'w': | |
367 | yield_active |= YIELD_WRITE; | |
368 | break; | |
369 | #endif | |
370 | case 'a': | |
371 | if (argc < i + 2) { | |
372 | show_usage(argc, argv); | |
373 | return -1; | |
374 | } | |
375 | a = atoi(argv[++i]); | |
376 | cpu_affinities[next_aff++] = a; | |
377 | use_affinity = 1; | |
378 | printf_verbose("Adding CPU %d affinity\n", a); | |
379 | break; | |
380 | case 'b': | |
381 | if (argc < i + 2) { | |
382 | show_usage(argc, argv); | |
383 | return -1; | |
384 | } | |
385 | reclaim_batch = atol(argv[++i]); | |
386 | break; | |
387 | case 'c': | |
388 | if (argc < i + 2) { | |
389 | show_usage(argc, argv); | |
390 | return -1; | |
391 | } | |
392 | rduration = atol(argv[++i]); | |
393 | break; | |
394 | case 'd': | |
395 | if (argc < i + 2) { | |
396 | show_usage(argc, argv); | |
397 | return -1; | |
398 | } | |
399 | wdelay = atol(argv[++i]); | |
400 | break; | |
401 | case 'e': | |
402 | if (argc < i + 2) { | |
403 | show_usage(argc, argv); | |
404 | return -1; | |
405 | } | |
406 | wduration = atol(argv[++i]); | |
407 | break; | |
408 | case 'v': | |
409 | verbose_mode = 1; | |
410 | break; | |
411 | } | |
412 | } | |
413 | ||
414 | printf_verbose("running test for %lu seconds, %u readers, %u writers.\n", | |
415 | duration, nr_readers, nr_writers); | |
416 | printf_verbose("Writer delay : %lu loops.\n", wdelay); | |
417 | printf_verbose("Reader duration : %lu loops.\n", rduration); | |
418 | printf_verbose("thread %-6s, thread id : %lx, tid %lu\n", | |
419 | "main", pthread_self(), (unsigned long)gettid()); | |
420 | ||
421 | tid_reader = malloc(sizeof(*tid_reader) * nr_readers); | |
422 | tid_writer = malloc(sizeof(*tid_writer) * nr_writers); | |
423 | count_reader = malloc(sizeof(*count_reader) * nr_readers); | |
424 | tot_nr_writes = malloc(sizeof(*tot_nr_writes) * nr_writers); | |
425 | pending_reclaims = malloc(sizeof(*pending_reclaims) * nr_writers); | |
426 | if (reclaim_batch * sizeof(*pending_reclaims[i].queue) | |
427 | < CAA_CACHE_LINE_SIZE) | |
428 | for (i = 0; i < nr_writers; i++) | |
429 | pending_reclaims[i].queue = calloc(1, CAA_CACHE_LINE_SIZE); | |
430 | else | |
431 | for (i = 0; i < nr_writers; i++) | |
432 | pending_reclaims[i].queue = calloc(reclaim_batch, | |
433 | sizeof(*pending_reclaims[i].queue)); | |
434 | for (i = 0; i < nr_writers; i++) | |
435 | pending_reclaims[i].head = pending_reclaims[i].queue; | |
436 | ||
437 | next_aff = 0; | |
438 | ||
439 | for (i = 0; i < nr_readers; i++) { | |
440 | err = pthread_create(&tid_reader[i], NULL, thr_reader, | |
441 | &count_reader[i]); | |
442 | if (err != 0) | |
443 | exit(1); | |
444 | } | |
445 | for (i = 0; i < nr_writers; i++) { | |
446 | err = pthread_create(&tid_writer[i], NULL, thr_writer, | |
447 | (void *)(long)i); | |
448 | if (err != 0) | |
449 | exit(1); | |
450 | } | |
451 | ||
452 | cmm_smp_mb(); | |
453 | ||
454 | test_go = 1; | |
455 | ||
456 | sleep(duration); | |
457 | ||
458 | test_stop = 1; | |
459 | ||
460 | for (i = 0; i < nr_readers; i++) { | |
461 | err = pthread_join(tid_reader[i], &tret); | |
462 | if (err != 0) | |
463 | exit(1); | |
464 | tot_reads += count_reader[i]; | |
465 | } | |
466 | for (i = 0; i < nr_writers; i++) { | |
467 | err = pthread_join(tid_writer[i], &tret); | |
468 | if (err != 0) | |
469 | exit(1); | |
470 | tot_writes += tot_nr_writes[i]; | |
471 | rcu_gc_clear_queue(i); | |
472 | } | |
473 | ||
474 | printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads, | |
475 | tot_writes); | |
476 | printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu wdur %6lu " | |
477 | "nr_writers %3u " | |
478 | "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu " | |
479 | "batch %u\n", | |
480 | argv[0], duration, nr_readers, rduration, wduration, | |
481 | nr_writers, wdelay, tot_reads, tot_writes, | |
482 | tot_reads + tot_writes, reclaim_batch); | |
483 | free(tid_reader); | |
484 | free(tid_writer); | |
485 | free(count_reader); | |
486 | free(tot_nr_writes); | |
487 | for (i = 0; i < nr_writers; i++) | |
488 | free(pending_reclaims[i].queue); | |
489 | free(pending_reclaims); | |
490 | ||
491 | return 0; | |
492 | } |