Commit | Line | Data |
---|---|---|
fdee2e6d MD |
1 | /* |
2 | * urcu-bp.c | |
3 | * | |
4 | * Userspace RCU library, "bulletproof" version. | |
5 | * | |
6982d6d7 | 6 | * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
fdee2e6d MD |
7 | * Copyright (c) 2009 Paul E. McKenney, IBM Corporation. |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library 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 GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | * | |
23 | * IBM's contributions to this file may be relicensed under LGPLv2 or later. | |
24 | */ | |
25 | ||
0617bf4c | 26 | #define _GNU_SOURCE |
71c811bf | 27 | #define _LGPL_SOURCE |
fdee2e6d MD |
28 | #include <stdio.h> |
29 | #include <pthread.h> | |
30 | #include <signal.h> | |
31 | #include <assert.h> | |
32 | #include <stdlib.h> | |
33 | #include <string.h> | |
34 | #include <errno.h> | |
35 | #include <poll.h> | |
36 | #include <unistd.h> | |
37 | #include <sys/mman.h> | |
38 | ||
d73fb81f | 39 | #include "urcu/wfcqueue.h" |
57760d44 | 40 | #include "urcu/map/urcu-bp.h" |
af7c2dbe | 41 | #include "urcu/static/urcu-bp.h" |
618b2595 | 42 | #include "urcu-pointer.h" |
bd252a04 | 43 | #include "urcu/tls-compat.h" |
71c811bf | 44 | |
4a6d7378 MD |
45 | #include "urcu-die.h" |
46 | ||
fdee2e6d | 47 | /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */ |
71c811bf | 48 | #undef _LGPL_SOURCE |
fdee2e6d | 49 | #include "urcu-bp.h" |
71c811bf | 50 | #define _LGPL_SOURCE |
fdee2e6d | 51 | |
4c1ae2ea MD |
52 | #ifndef MAP_ANONYMOUS |
53 | #define MAP_ANONYMOUS MAP_ANON | |
54 | #endif | |
55 | ||
c7eaf61c MD |
56 | #ifdef __linux__ |
57 | static | |
58 | void *mremap_wrapper(void *old_address, size_t old_size, | |
59 | size_t new_size, int flags) | |
60 | { | |
61 | return mremap(old_address, old_size, new_size, flags); | |
62 | } | |
63 | #else | |
45a4872f MD |
64 | |
65 | #define MREMAP_MAYMOVE 1 | |
66 | #define MREMAP_FIXED 2 | |
67 | ||
68 | /* | |
69 | * mremap wrapper for non-Linux systems. Maps a RW, anonymous private mapping. | |
70 | * This is not generic. | |
71 | */ | |
c7eaf61c MD |
72 | static |
73 | void *mremap_wrapper(void *old_address, size_t old_size, | |
74 | size_t new_size, int flags) | |
45a4872f | 75 | { |
dc745ef6 | 76 | void *new_address; |
45a4872f MD |
77 | |
78 | assert(flags & MREMAP_MAYMOVE); | |
79 | assert(!(flags & MREMAP_FIXED)); | |
80 | new_address = mmap(old_address, new_size, | |
81 | PROT_READ | PROT_WRITE, | |
82 | MAP_ANONYMOUS | MAP_PRIVATE, | |
83 | -1, 0); | |
84 | if (new_address == MAP_FAILED) | |
85 | return MAP_FAILED; | |
86 | if (old_address) { | |
87 | memcpy(new_address, old_address, old_size); | |
88 | munmap(old_address, old_size); | |
89 | } | |
90 | return new_address; | |
91 | } | |
92 | #endif | |
93 | ||
fdee2e6d MD |
94 | /* Sleep delay in us */ |
95 | #define RCU_SLEEP_DELAY 1000 | |
96 | #define ARENA_INIT_ALLOC 16 | |
97 | ||
b7b6a8f5 PB |
98 | /* |
99 | * Active attempts to check for reader Q.S. before calling sleep(). | |
100 | */ | |
101 | #define RCU_QS_ACTIVE_ATTEMPTS 100 | |
102 | ||
02be5561 | 103 | void __attribute__((destructor)) rcu_bp_exit(void); |
fdee2e6d | 104 | |
6abb4bd5 | 105 | static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER; |
fdee2e6d MD |
106 | |
107 | #ifdef DEBUG_YIELD | |
1de4df4b MD |
108 | unsigned int rcu_yield_active; |
109 | DEFINE_URCU_TLS(unsigned int, rcu_rand_yield); | |
fdee2e6d MD |
110 | #endif |
111 | ||
112 | /* | |
113 | * Global grace period counter. | |
02be5561 | 114 | * Contains the current RCU_GP_CTR_PHASE. |
fdee2e6d MD |
115 | * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path. |
116 | * Written to only by writer with mutex taken. Read by both writer and readers. | |
117 | */ | |
52c75091 | 118 | unsigned long rcu_gp_ctr = RCU_GP_COUNT; |
fdee2e6d MD |
119 | |
120 | /* | |
121 | * Pointer to registry elements. Written to only by each individual reader. Read | |
122 | * by both the reader and the writers. | |
123 | */ | |
bd252a04 | 124 | DEFINE_URCU_TLS(struct rcu_reader *, rcu_reader); |
fdee2e6d | 125 | |
16aa9ee8 | 126 | static CDS_LIST_HEAD(registry); |
fdee2e6d MD |
127 | |
128 | struct registry_arena { | |
129 | void *p; | |
130 | size_t len; | |
131 | size_t used; | |
132 | }; | |
133 | ||
134 | static struct registry_arena registry_arena; | |
135 | ||
4cf1675f MD |
136 | /* Saved fork signal mask, protected by rcu_gp_lock */ |
137 | static sigset_t saved_fork_signal_mask; | |
138 | ||
fdee2e6d MD |
139 | static void rcu_gc_registry(void); |
140 | ||
6abb4bd5 | 141 | static void mutex_lock(pthread_mutex_t *mutex) |
fdee2e6d MD |
142 | { |
143 | int ret; | |
144 | ||
145 | #ifndef DISTRUST_SIGNALS_EXTREME | |
6abb4bd5 | 146 | ret = pthread_mutex_lock(mutex); |
4a6d7378 MD |
147 | if (ret) |
148 | urcu_die(ret); | |
fdee2e6d | 149 | #else /* #ifndef DISTRUST_SIGNALS_EXTREME */ |
6abb4bd5 | 150 | while ((ret = pthread_mutex_trylock(mutex)) != 0) { |
4a6d7378 MD |
151 | if (ret != EBUSY && ret != EINTR) |
152 | urcu_die(ret); | |
fdee2e6d MD |
153 | poll(NULL,0,10); |
154 | } | |
155 | #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */ | |
156 | } | |
157 | ||
6abb4bd5 | 158 | static void mutex_unlock(pthread_mutex_t *mutex) |
fdee2e6d MD |
159 | { |
160 | int ret; | |
161 | ||
6abb4bd5 | 162 | ret = pthread_mutex_unlock(mutex); |
4a6d7378 MD |
163 | if (ret) |
164 | urcu_die(ret); | |
fdee2e6d MD |
165 | } |
166 | ||
52c75091 MD |
167 | static void wait_for_readers(struct cds_list_head *input_readers, |
168 | struct cds_list_head *cur_snap_readers, | |
169 | struct cds_list_head *qsreaders) | |
fdee2e6d | 170 | { |
fdee2e6d | 171 | int wait_loops = 0; |
02be5561 | 172 | struct rcu_reader *index, *tmp; |
fdee2e6d | 173 | |
fdee2e6d | 174 | /* |
dd61d077 MD |
175 | * Wait for each thread URCU_TLS(rcu_reader).ctr to either |
176 | * indicate quiescence (not nested), or observe the current | |
177 | * rcu_gp_ctr value. | |
fdee2e6d MD |
178 | */ |
179 | for (;;) { | |
180 | wait_loops++; | |
52c75091 MD |
181 | cds_list_for_each_entry_safe(index, tmp, input_readers, node) { |
182 | switch (rcu_reader_state(&index->ctr)) { | |
183 | case RCU_READER_ACTIVE_CURRENT: | |
184 | if (cur_snap_readers) { | |
185 | cds_list_move(&index->node, | |
186 | cur_snap_readers); | |
187 | break; | |
188 | } | |
189 | /* Fall-through */ | |
190 | case RCU_READER_INACTIVE: | |
191 | cds_list_move(&index->node, qsreaders); | |
192 | break; | |
193 | case RCU_READER_ACTIVE_OLD: | |
194 | /* | |
195 | * Old snapshot. Leaving node in | |
196 | * input_readers will make us busy-loop | |
197 | * until the snapshot becomes current or | |
198 | * the reader becomes inactive. | |
199 | */ | |
200 | break; | |
201 | } | |
fdee2e6d MD |
202 | } |
203 | ||
52c75091 | 204 | if (cds_list_empty(input_readers)) { |
fdee2e6d MD |
205 | break; |
206 | } else { | |
207 | if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) | |
208 | usleep(RCU_SLEEP_DELAY); | |
209 | else | |
06f22bdb | 210 | caa_cpu_relax(); |
fdee2e6d MD |
211 | } |
212 | } | |
fdee2e6d MD |
213 | } |
214 | ||
215 | void synchronize_rcu(void) | |
216 | { | |
52c75091 MD |
217 | CDS_LIST_HEAD(cur_snap_readers); |
218 | CDS_LIST_HEAD(qsreaders); | |
fdee2e6d MD |
219 | sigset_t newmask, oldmask; |
220 | int ret; | |
221 | ||
222 | ret = sigemptyset(&newmask); | |
223 | assert(!ret); | |
224 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
225 | assert(!ret); | |
226 | ||
6abb4bd5 | 227 | mutex_lock(&rcu_gp_lock); |
fdee2e6d | 228 | |
16aa9ee8 | 229 | if (cds_list_empty(®istry)) |
2dfb8b5e | 230 | goto out; |
fdee2e6d MD |
231 | |
232 | /* All threads should read qparity before accessing data structure | |
2dfb8b5e | 233 | * where new ptr points to. */ |
fdee2e6d | 234 | /* Write new ptr before changing the qparity */ |
5481ddb3 | 235 | cmm_smp_mb(); |
fdee2e6d | 236 | |
2dfb8b5e MD |
237 | /* Remove old registry elements */ |
238 | rcu_gc_registry(); | |
fdee2e6d MD |
239 | |
240 | /* | |
dd61d077 MD |
241 | * Wait for readers to observe original parity or be quiescent. |
242 | */ | |
52c75091 | 243 | wait_for_readers(®istry, &cur_snap_readers, &qsreaders); |
dd61d077 MD |
244 | |
245 | /* | |
246 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the | |
247 | * model easier to understand. It does not have a big performance impact | |
248 | * anyway, given this is the write-side. | |
249 | */ | |
250 | cmm_smp_mb(); | |
251 | ||
252 | /* Switch parity: 0 -> 1, 1 -> 0 */ | |
253 | CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE); | |
254 | ||
255 | /* | |
256 | * Must commit qparity update to memory before waiting for other parity | |
257 | * quiescent state. Failure to do so could result in the writer waiting | |
258 | * forever while new readers are always accessing data (no progress). | |
259 | * Ensured by CMM_STORE_SHARED and CMM_LOAD_SHARED. | |
fdee2e6d | 260 | */ |
fdee2e6d MD |
261 | |
262 | /* | |
5481ddb3 | 263 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the |
fdee2e6d MD |
264 | * model easier to understand. It does not have a big performance impact |
265 | * anyway, given this is the write-side. | |
266 | */ | |
5481ddb3 | 267 | cmm_smp_mb(); |
fdee2e6d | 268 | |
fdee2e6d | 269 | /* |
dd61d077 | 270 | * Wait for readers to observe new parity or be quiescent. |
fdee2e6d | 271 | */ |
52c75091 MD |
272 | wait_for_readers(&cur_snap_readers, NULL, &qsreaders); |
273 | ||
274 | /* | |
275 | * Put quiescent reader list back into registry. | |
276 | */ | |
277 | cds_list_splice(&qsreaders, ®istry); | |
fdee2e6d MD |
278 | |
279 | /* | |
2dfb8b5e MD |
280 | * Finish waiting for reader threads before letting the old ptr being |
281 | * freed. | |
fdee2e6d | 282 | */ |
5481ddb3 | 283 | cmm_smp_mb(); |
2dfb8b5e | 284 | out: |
6abb4bd5 | 285 | mutex_unlock(&rcu_gp_lock); |
fdee2e6d MD |
286 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); |
287 | assert(!ret); | |
288 | } | |
289 | ||
290 | /* | |
291 | * library wrappers to be used by non-LGPL compatible source code. | |
292 | */ | |
293 | ||
294 | void rcu_read_lock(void) | |
295 | { | |
296 | _rcu_read_lock(); | |
297 | } | |
298 | ||
299 | void rcu_read_unlock(void) | |
300 | { | |
301 | _rcu_read_unlock(); | |
302 | } | |
303 | ||
304 | /* | |
305 | * only grow for now. | |
306 | */ | |
307 | static void resize_arena(struct registry_arena *arena, size_t len) | |
308 | { | |
309 | void *new_arena; | |
310 | ||
0617bf4c MD |
311 | if (!arena->p) |
312 | new_arena = mmap(arena->p, len, | |
313 | PROT_READ | PROT_WRITE, | |
314 | MAP_ANONYMOUS | MAP_PRIVATE, | |
315 | -1, 0); | |
316 | else | |
c7eaf61c MD |
317 | new_arena = mremap_wrapper(arena->p, arena->len, |
318 | len, MREMAP_MAYMOVE); | |
0617bf4c MD |
319 | assert(new_arena != MAP_FAILED); |
320 | ||
fdee2e6d MD |
321 | /* |
322 | * re-used the same region ? | |
323 | */ | |
324 | if (new_arena == arena->p) | |
325 | return; | |
326 | ||
fdee2e6d MD |
327 | bzero(new_arena + arena->len, len - arena->len); |
328 | arena->p = new_arena; | |
329 | } | |
330 | ||
331 | /* Called with signals off and mutex locked */ | |
332 | static void add_thread(void) | |
333 | { | |
02be5561 | 334 | struct rcu_reader *rcu_reader_reg; |
fdee2e6d MD |
335 | |
336 | if (registry_arena.len | |
02be5561 | 337 | < registry_arena.used + sizeof(struct rcu_reader)) |
fdee2e6d | 338 | resize_arena(®istry_arena, |
2f8a5ae7 | 339 | caa_max(registry_arena.len << 1, ARENA_INIT_ALLOC)); |
fdee2e6d MD |
340 | /* |
341 | * Find a free spot. | |
342 | */ | |
02be5561 MD |
343 | for (rcu_reader_reg = registry_arena.p; |
344 | (void *)rcu_reader_reg < registry_arena.p + registry_arena.len; | |
345 | rcu_reader_reg++) { | |
346 | if (!rcu_reader_reg->alloc) | |
fdee2e6d MD |
347 | break; |
348 | } | |
02be5561 MD |
349 | rcu_reader_reg->alloc = 1; |
350 | registry_arena.used += sizeof(struct rcu_reader); | |
fdee2e6d MD |
351 | |
352 | /* Add to registry */ | |
02be5561 MD |
353 | rcu_reader_reg->tid = pthread_self(); |
354 | assert(rcu_reader_reg->ctr == 0); | |
16aa9ee8 | 355 | cds_list_add(&rcu_reader_reg->node, ®istry); |
bd252a04 | 356 | URCU_TLS(rcu_reader) = rcu_reader_reg; |
fdee2e6d MD |
357 | } |
358 | ||
359 | /* Called with signals off and mutex locked */ | |
360 | static void rcu_gc_registry(void) | |
361 | { | |
02be5561 | 362 | struct rcu_reader *rcu_reader_reg; |
fdee2e6d MD |
363 | pthread_t tid; |
364 | int ret; | |
365 | ||
02be5561 MD |
366 | for (rcu_reader_reg = registry_arena.p; |
367 | (void *)rcu_reader_reg < registry_arena.p + registry_arena.len; | |
368 | rcu_reader_reg++) { | |
369 | if (!rcu_reader_reg->alloc) | |
fdee2e6d | 370 | continue; |
02be5561 | 371 | tid = rcu_reader_reg->tid; |
fdee2e6d MD |
372 | ret = pthread_kill(tid, 0); |
373 | assert(ret != EINVAL); | |
374 | if (ret == ESRCH) { | |
16aa9ee8 | 375 | cds_list_del(&rcu_reader_reg->node); |
79266051 | 376 | rcu_reader_reg->ctr = 0; |
02be5561 MD |
377 | rcu_reader_reg->alloc = 0; |
378 | registry_arena.used -= sizeof(struct rcu_reader); | |
fdee2e6d MD |
379 | } |
380 | } | |
381 | } | |
382 | ||
383 | /* Disable signals, take mutex, add to registry */ | |
384 | void rcu_bp_register(void) | |
385 | { | |
386 | sigset_t newmask, oldmask; | |
387 | int ret; | |
388 | ||
389 | ret = sigemptyset(&newmask); | |
390 | assert(!ret); | |
391 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
392 | assert(!ret); | |
393 | ||
394 | /* | |
395 | * Check if a signal concurrently registered our thread since | |
396 | * the check in rcu_read_lock(). */ | |
bd252a04 | 397 | if (URCU_TLS(rcu_reader)) |
fdee2e6d MD |
398 | goto end; |
399 | ||
6abb4bd5 | 400 | mutex_lock(&rcu_gp_lock); |
fdee2e6d | 401 | add_thread(); |
6abb4bd5 | 402 | mutex_unlock(&rcu_gp_lock); |
fdee2e6d MD |
403 | end: |
404 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
405 | assert(!ret); | |
406 | } | |
407 | ||
9380711a | 408 | void rcu_bp_exit(void) |
fdee2e6d | 409 | { |
9380711a MD |
410 | if (registry_arena.p) |
411 | munmap(registry_arena.p, registry_arena.len); | |
fdee2e6d | 412 | } |
4cf1675f MD |
413 | |
414 | /* | |
415 | * Holding the rcu_gp_lock across fork will make sure we fork() don't race with | |
416 | * a concurrent thread executing with this same lock held. This ensures that the | |
417 | * registry is in a coherent state in the child. | |
418 | */ | |
419 | void rcu_bp_before_fork(void) | |
420 | { | |
421 | sigset_t newmask, oldmask; | |
422 | int ret; | |
423 | ||
424 | ret = sigemptyset(&newmask); | |
425 | assert(!ret); | |
426 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
427 | assert(!ret); | |
428 | mutex_lock(&rcu_gp_lock); | |
429 | saved_fork_signal_mask = oldmask; | |
430 | } | |
431 | ||
432 | void rcu_bp_after_fork_parent(void) | |
433 | { | |
434 | sigset_t oldmask; | |
435 | int ret; | |
436 | ||
437 | oldmask = saved_fork_signal_mask; | |
438 | mutex_unlock(&rcu_gp_lock); | |
439 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
440 | assert(!ret); | |
441 | } | |
442 | ||
443 | void rcu_bp_after_fork_child(void) | |
444 | { | |
445 | sigset_t oldmask; | |
446 | int ret; | |
447 | ||
448 | rcu_gc_registry(); | |
449 | oldmask = saved_fork_signal_mask; | |
450 | mutex_unlock(&rcu_gp_lock); | |
451 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
452 | assert(!ret); | |
453 | } | |
5e77fc1f | 454 | |
9b7981bb MD |
455 | void *rcu_dereference_sym_bp(void *p) |
456 | { | |
457 | return _rcu_dereference(p); | |
458 | } | |
459 | ||
5efd3cd2 MD |
460 | void *rcu_set_pointer_sym_bp(void **p, void *v) |
461 | { | |
462 | cmm_wmb(); | |
424d4ed5 MD |
463 | uatomic_set(p, v); |
464 | return v; | |
5efd3cd2 MD |
465 | } |
466 | ||
467 | void *rcu_xchg_pointer_sym_bp(void **p, void *v) | |
468 | { | |
469 | cmm_wmb(); | |
470 | return uatomic_xchg(p, v); | |
471 | } | |
472 | ||
473 | void *rcu_cmpxchg_pointer_sym_bp(void **p, void *old, void *_new) | |
474 | { | |
475 | cmm_wmb(); | |
476 | return uatomic_cmpxchg(p, old, _new); | |
477 | } | |
478 | ||
5e6b23a6 | 479 | DEFINE_RCU_FLAVOR(rcu_flavor); |
541d828d | 480 | |
5e77fc1f | 481 | #include "urcu-call-rcu-impl.h" |
0376e7b2 | 482 | #include "urcu-defer-impl.h" |