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 |
fdee2e6d MD |
27 | #include <stdio.h> |
28 | #include <pthread.h> | |
29 | #include <signal.h> | |
30 | #include <assert.h> | |
31 | #include <stdlib.h> | |
32 | #include <string.h> | |
33 | #include <errno.h> | |
34 | #include <poll.h> | |
35 | #include <unistd.h> | |
36 | #include <sys/mman.h> | |
37 | ||
57760d44 | 38 | #include "urcu/map/urcu-bp.h" |
5e77fc1f | 39 | |
af7c2dbe | 40 | #include "urcu/static/urcu-bp.h" |
fdee2e6d MD |
41 | /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */ |
42 | #include "urcu-bp.h" | |
43 | ||
4c1ae2ea MD |
44 | #ifndef MAP_ANONYMOUS |
45 | #define MAP_ANONYMOUS MAP_ANON | |
46 | #endif | |
47 | ||
45a4872f MD |
48 | #ifndef __linux__ |
49 | ||
50 | #define MREMAP_MAYMOVE 1 | |
51 | #define MREMAP_FIXED 2 | |
52 | ||
53 | /* | |
54 | * mremap wrapper for non-Linux systems. Maps a RW, anonymous private mapping. | |
55 | * This is not generic. | |
56 | */ | |
57 | void *mremap(void *old_address, size_t old_size, size_t new_size, int flags) | |
58 | { | |
dc745ef6 | 59 | void *new_address; |
45a4872f MD |
60 | |
61 | assert(flags & MREMAP_MAYMOVE); | |
62 | assert(!(flags & MREMAP_FIXED)); | |
63 | new_address = mmap(old_address, new_size, | |
64 | PROT_READ | PROT_WRITE, | |
65 | MAP_ANONYMOUS | MAP_PRIVATE, | |
66 | -1, 0); | |
67 | if (new_address == MAP_FAILED) | |
68 | return MAP_FAILED; | |
69 | if (old_address) { | |
70 | memcpy(new_address, old_address, old_size); | |
71 | munmap(old_address, old_size); | |
72 | } | |
73 | return new_address; | |
74 | } | |
75 | #endif | |
76 | ||
fdee2e6d MD |
77 | /* Sleep delay in us */ |
78 | #define RCU_SLEEP_DELAY 1000 | |
79 | #define ARENA_INIT_ALLOC 16 | |
80 | ||
02be5561 | 81 | void __attribute__((destructor)) rcu_bp_exit(void); |
fdee2e6d | 82 | |
6abb4bd5 | 83 | static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER; |
fdee2e6d MD |
84 | |
85 | #ifdef DEBUG_YIELD | |
86 | unsigned int yield_active; | |
87 | unsigned int __thread rand_yield; | |
88 | #endif | |
89 | ||
90 | /* | |
91 | * Global grace period counter. | |
02be5561 | 92 | * Contains the current RCU_GP_CTR_PHASE. |
fdee2e6d MD |
93 | * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path. |
94 | * Written to only by writer with mutex taken. Read by both writer and readers. | |
95 | */ | |
02be5561 | 96 | long rcu_gp_ctr = RCU_GP_COUNT; |
fdee2e6d MD |
97 | |
98 | /* | |
99 | * Pointer to registry elements. Written to only by each individual reader. Read | |
100 | * by both the reader and the writers. | |
101 | */ | |
02be5561 | 102 | struct rcu_reader __thread *rcu_reader; |
fdee2e6d | 103 | |
16aa9ee8 | 104 | static CDS_LIST_HEAD(registry); |
fdee2e6d MD |
105 | |
106 | struct registry_arena { | |
107 | void *p; | |
108 | size_t len; | |
109 | size_t used; | |
110 | }; | |
111 | ||
112 | static struct registry_arena registry_arena; | |
113 | ||
4cf1675f MD |
114 | /* Saved fork signal mask, protected by rcu_gp_lock */ |
115 | static sigset_t saved_fork_signal_mask; | |
116 | ||
fdee2e6d MD |
117 | static void rcu_gc_registry(void); |
118 | ||
6abb4bd5 | 119 | static void mutex_lock(pthread_mutex_t *mutex) |
fdee2e6d MD |
120 | { |
121 | int ret; | |
122 | ||
123 | #ifndef DISTRUST_SIGNALS_EXTREME | |
6abb4bd5 | 124 | ret = pthread_mutex_lock(mutex); |
fdee2e6d MD |
125 | if (ret) { |
126 | perror("Error in pthread mutex lock"); | |
127 | exit(-1); | |
128 | } | |
129 | #else /* #ifndef DISTRUST_SIGNALS_EXTREME */ | |
6abb4bd5 | 130 | while ((ret = pthread_mutex_trylock(mutex)) != 0) { |
fdee2e6d MD |
131 | if (ret != EBUSY && ret != EINTR) { |
132 | printf("ret = %d, errno = %d\n", ret, errno); | |
133 | perror("Error in pthread mutex lock"); | |
134 | exit(-1); | |
135 | } | |
fdee2e6d MD |
136 | poll(NULL,0,10); |
137 | } | |
138 | #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */ | |
139 | } | |
140 | ||
6abb4bd5 | 141 | static void mutex_unlock(pthread_mutex_t *mutex) |
fdee2e6d MD |
142 | { |
143 | int ret; | |
144 | ||
6abb4bd5 | 145 | ret = pthread_mutex_unlock(mutex); |
fdee2e6d MD |
146 | if (ret) { |
147 | perror("Error in pthread mutex unlock"); | |
148 | exit(-1); | |
149 | } | |
150 | } | |
151 | ||
2dfb8b5e | 152 | void update_counter_and_wait(void) |
fdee2e6d | 153 | { |
16aa9ee8 | 154 | CDS_LIST_HEAD(qsreaders); |
fdee2e6d | 155 | int wait_loops = 0; |
02be5561 | 156 | struct rcu_reader *index, *tmp; |
fdee2e6d | 157 | |
32c15e4e | 158 | /* Switch parity: 0 -> 1, 1 -> 0 */ |
6cf3827c | 159 | CMM_STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE); |
2dfb8b5e MD |
160 | |
161 | /* | |
162 | * Must commit qparity update to memory before waiting for other parity | |
163 | * quiescent state. Failure to do so could result in the writer waiting | |
164 | * forever while new readers are always accessing data (no progress). | |
6cf3827c | 165 | * Ensured by CMM_STORE_SHARED and CMM_LOAD_SHARED. |
2dfb8b5e MD |
166 | */ |
167 | ||
168 | /* | |
5481ddb3 | 169 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the |
2dfb8b5e MD |
170 | * model easier to understand. It does not have a big performance impact |
171 | * anyway, given this is the write-side. | |
172 | */ | |
5481ddb3 | 173 | cmm_smp_mb(); |
2dfb8b5e | 174 | |
fdee2e6d | 175 | /* |
02be5561 | 176 | * Wait for each thread rcu_reader.ctr count to become 0. |
fdee2e6d MD |
177 | */ |
178 | for (;;) { | |
179 | wait_loops++; | |
16aa9ee8 | 180 | cds_list_for_each_entry_safe(index, tmp, ®istry, node) { |
fdee2e6d | 181 | if (!rcu_old_gp_ongoing(&index->ctr)) |
16aa9ee8 | 182 | cds_list_move(&index->node, &qsreaders); |
fdee2e6d MD |
183 | } |
184 | ||
16aa9ee8 | 185 | if (cds_list_empty(®istry)) { |
fdee2e6d MD |
186 | break; |
187 | } else { | |
188 | if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) | |
189 | usleep(RCU_SLEEP_DELAY); | |
190 | else | |
06f22bdb | 191 | caa_cpu_relax(); |
fdee2e6d MD |
192 | } |
193 | } | |
194 | /* put back the reader list in the registry */ | |
16aa9ee8 | 195 | cds_list_splice(&qsreaders, ®istry); |
fdee2e6d MD |
196 | } |
197 | ||
198 | void synchronize_rcu(void) | |
199 | { | |
200 | sigset_t newmask, oldmask; | |
201 | int ret; | |
202 | ||
203 | ret = sigemptyset(&newmask); | |
204 | assert(!ret); | |
205 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
206 | assert(!ret); | |
207 | ||
6abb4bd5 | 208 | mutex_lock(&rcu_gp_lock); |
fdee2e6d | 209 | |
16aa9ee8 | 210 | if (cds_list_empty(®istry)) |
2dfb8b5e | 211 | goto out; |
fdee2e6d MD |
212 | |
213 | /* All threads should read qparity before accessing data structure | |
2dfb8b5e | 214 | * where new ptr points to. */ |
fdee2e6d | 215 | /* Write new ptr before changing the qparity */ |
5481ddb3 | 216 | cmm_smp_mb(); |
fdee2e6d | 217 | |
2dfb8b5e MD |
218 | /* Remove old registry elements */ |
219 | rcu_gc_registry(); | |
fdee2e6d MD |
220 | |
221 | /* | |
222 | * Wait for previous parity to be empty of readers. | |
223 | */ | |
2dfb8b5e | 224 | update_counter_and_wait(); /* 0 -> 1, wait readers in parity 0 */ |
fdee2e6d MD |
225 | |
226 | /* | |
5481ddb3 | 227 | * Adding a cmm_smp_mb() which is _not_ formally required, but makes the |
fdee2e6d MD |
228 | * model easier to understand. It does not have a big performance impact |
229 | * anyway, given this is the write-side. | |
230 | */ | |
5481ddb3 | 231 | cmm_smp_mb(); |
fdee2e6d | 232 | |
fdee2e6d | 233 | /* |
2dfb8b5e | 234 | * Wait for previous parity to be empty of readers. |
fdee2e6d | 235 | */ |
2dfb8b5e | 236 | update_counter_and_wait(); /* 1 -> 0, wait readers in parity 1 */ |
fdee2e6d MD |
237 | |
238 | /* | |
2dfb8b5e MD |
239 | * Finish waiting for reader threads before letting the old ptr being |
240 | * freed. | |
fdee2e6d | 241 | */ |
5481ddb3 | 242 | cmm_smp_mb(); |
2dfb8b5e | 243 | out: |
6abb4bd5 | 244 | mutex_unlock(&rcu_gp_lock); |
fdee2e6d MD |
245 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); |
246 | assert(!ret); | |
247 | } | |
248 | ||
249 | /* | |
250 | * library wrappers to be used by non-LGPL compatible source code. | |
251 | */ | |
252 | ||
253 | void rcu_read_lock(void) | |
254 | { | |
255 | _rcu_read_lock(); | |
256 | } | |
257 | ||
258 | void rcu_read_unlock(void) | |
259 | { | |
260 | _rcu_read_unlock(); | |
261 | } | |
262 | ||
263 | /* | |
264 | * only grow for now. | |
265 | */ | |
266 | static void resize_arena(struct registry_arena *arena, size_t len) | |
267 | { | |
268 | void *new_arena; | |
269 | ||
0617bf4c MD |
270 | if (!arena->p) |
271 | new_arena = mmap(arena->p, len, | |
272 | PROT_READ | PROT_WRITE, | |
273 | MAP_ANONYMOUS | MAP_PRIVATE, | |
274 | -1, 0); | |
275 | else | |
276 | new_arena = mremap(arena->p, arena->len, | |
277 | len, MREMAP_MAYMOVE); | |
278 | assert(new_arena != MAP_FAILED); | |
279 | ||
fdee2e6d MD |
280 | /* |
281 | * re-used the same region ? | |
282 | */ | |
283 | if (new_arena == arena->p) | |
284 | return; | |
285 | ||
fdee2e6d MD |
286 | bzero(new_arena + arena->len, len - arena->len); |
287 | arena->p = new_arena; | |
288 | } | |
289 | ||
290 | /* Called with signals off and mutex locked */ | |
291 | static void add_thread(void) | |
292 | { | |
02be5561 | 293 | struct rcu_reader *rcu_reader_reg; |
fdee2e6d MD |
294 | |
295 | if (registry_arena.len | |
02be5561 | 296 | < registry_arena.used + sizeof(struct rcu_reader)) |
fdee2e6d MD |
297 | resize_arena(®istry_arena, |
298 | max(registry_arena.len << 1, ARENA_INIT_ALLOC)); | |
299 | /* | |
300 | * Find a free spot. | |
301 | */ | |
02be5561 MD |
302 | for (rcu_reader_reg = registry_arena.p; |
303 | (void *)rcu_reader_reg < registry_arena.p + registry_arena.len; | |
304 | rcu_reader_reg++) { | |
305 | if (!rcu_reader_reg->alloc) | |
fdee2e6d MD |
306 | break; |
307 | } | |
02be5561 MD |
308 | rcu_reader_reg->alloc = 1; |
309 | registry_arena.used += sizeof(struct rcu_reader); | |
fdee2e6d MD |
310 | |
311 | /* Add to registry */ | |
02be5561 MD |
312 | rcu_reader_reg->tid = pthread_self(); |
313 | assert(rcu_reader_reg->ctr == 0); | |
16aa9ee8 | 314 | cds_list_add(&rcu_reader_reg->node, ®istry); |
02be5561 | 315 | rcu_reader = rcu_reader_reg; |
fdee2e6d MD |
316 | } |
317 | ||
318 | /* Called with signals off and mutex locked */ | |
319 | static void rcu_gc_registry(void) | |
320 | { | |
02be5561 | 321 | struct rcu_reader *rcu_reader_reg; |
fdee2e6d MD |
322 | pthread_t tid; |
323 | int ret; | |
324 | ||
02be5561 MD |
325 | for (rcu_reader_reg = registry_arena.p; |
326 | (void *)rcu_reader_reg < registry_arena.p + registry_arena.len; | |
327 | rcu_reader_reg++) { | |
328 | if (!rcu_reader_reg->alloc) | |
fdee2e6d | 329 | continue; |
02be5561 | 330 | tid = rcu_reader_reg->tid; |
fdee2e6d MD |
331 | ret = pthread_kill(tid, 0); |
332 | assert(ret != EINVAL); | |
333 | if (ret == ESRCH) { | |
16aa9ee8 | 334 | cds_list_del(&rcu_reader_reg->node); |
79266051 | 335 | rcu_reader_reg->ctr = 0; |
02be5561 MD |
336 | rcu_reader_reg->alloc = 0; |
337 | registry_arena.used -= sizeof(struct rcu_reader); | |
fdee2e6d MD |
338 | } |
339 | } | |
340 | } | |
341 | ||
342 | /* Disable signals, take mutex, add to registry */ | |
343 | void rcu_bp_register(void) | |
344 | { | |
345 | sigset_t newmask, oldmask; | |
346 | int ret; | |
347 | ||
348 | ret = sigemptyset(&newmask); | |
349 | assert(!ret); | |
350 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
351 | assert(!ret); | |
352 | ||
353 | /* | |
354 | * Check if a signal concurrently registered our thread since | |
355 | * the check in rcu_read_lock(). */ | |
02be5561 | 356 | if (rcu_reader) |
fdee2e6d MD |
357 | goto end; |
358 | ||
6abb4bd5 | 359 | mutex_lock(&rcu_gp_lock); |
fdee2e6d | 360 | add_thread(); |
6abb4bd5 | 361 | mutex_unlock(&rcu_gp_lock); |
fdee2e6d MD |
362 | end: |
363 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
364 | assert(!ret); | |
365 | } | |
366 | ||
02be5561 | 367 | void rcu_bp_exit() |
fdee2e6d MD |
368 | { |
369 | munmap(registry_arena.p, registry_arena.len); | |
370 | } | |
4cf1675f MD |
371 | |
372 | /* | |
373 | * Holding the rcu_gp_lock across fork will make sure we fork() don't race with | |
374 | * a concurrent thread executing with this same lock held. This ensures that the | |
375 | * registry is in a coherent state in the child. | |
376 | */ | |
377 | void rcu_bp_before_fork(void) | |
378 | { | |
379 | sigset_t newmask, oldmask; | |
380 | int ret; | |
381 | ||
382 | ret = sigemptyset(&newmask); | |
383 | assert(!ret); | |
384 | ret = pthread_sigmask(SIG_SETMASK, &newmask, &oldmask); | |
385 | assert(!ret); | |
386 | mutex_lock(&rcu_gp_lock); | |
387 | saved_fork_signal_mask = oldmask; | |
388 | } | |
389 | ||
390 | void rcu_bp_after_fork_parent(void) | |
391 | { | |
392 | sigset_t oldmask; | |
393 | int ret; | |
394 | ||
395 | oldmask = saved_fork_signal_mask; | |
396 | mutex_unlock(&rcu_gp_lock); | |
397 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
398 | assert(!ret); | |
399 | } | |
400 | ||
401 | void rcu_bp_after_fork_child(void) | |
402 | { | |
403 | sigset_t oldmask; | |
404 | int ret; | |
405 | ||
406 | rcu_gc_registry(); | |
407 | oldmask = saved_fork_signal_mask; | |
408 | mutex_unlock(&rcu_gp_lock); | |
409 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
410 | assert(!ret); | |
411 | } | |
5e77fc1f PM |
412 | |
413 | #include "urcu-call-rcu-impl.h" | |
0376e7b2 | 414 | #include "urcu-defer-impl.h" |