4 * Userspace RCU library
6 * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
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.
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.
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
23 * IBM's contributions to this file may be relicensed under LGPLv2 or later.
39 #include "urcu/wfcqueue.h"
40 #include "urcu/map/urcu.h"
41 #include "urcu/static/urcu.h"
42 #include "urcu-pointer.h"
43 #include "urcu/tls-compat.h"
47 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
53 * If a reader is really non-cooperative and refuses to commit its
54 * rcu_active_readers count to memory (there is no barrier in the reader
55 * per-se), kick it after a few loops waiting for it.
57 #define KICK_READER_LOOPS 10000
60 * Active attempts to check for reader Q.S. before calling futex().
62 #define RCU_QS_ACTIVE_ATTEMPTS 100
66 int rcu_has_sys_membarrier
;
68 void __attribute__((constructor
)) rcu_init(void);
80 void __attribute__((constructor
)) rcu_init(void);
81 void __attribute__((destructor
)) rcu_exit(void);
84 static pthread_mutex_t rcu_gp_lock
= PTHREAD_MUTEX_INITIALIZER
;
89 * Global grace period counter.
90 * Contains the current RCU_GP_CTR_PHASE.
91 * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
92 * Written to only by writer with mutex taken. Read by both writer and readers.
94 unsigned long rcu_gp_ctr
= RCU_GP_COUNT
;
97 * Written to only by each individual reader. Read by both the reader and the
100 DEFINE_URCU_TLS(struct rcu_reader
, rcu_reader
);
103 unsigned int rcu_yield_active
;
104 DEFINE_URCU_TLS(unsigned int, rcu_rand_yield
);
107 static CDS_LIST_HEAD(registry
);
109 static void mutex_lock(pthread_mutex_t
*mutex
)
113 #ifndef DISTRUST_SIGNALS_EXTREME
114 ret
= pthread_mutex_lock(mutex
);
117 #else /* #ifndef DISTRUST_SIGNALS_EXTREME */
118 while ((ret
= pthread_mutex_trylock(mutex
)) != 0) {
119 if (ret
!= EBUSY
&& ret
!= EINTR
)
121 if (CMM_LOAD_SHARED(URCU_TLS(rcu_reader
).need_mb
)) {
123 _CMM_STORE_SHARED(URCU_TLS(rcu_reader
).need_mb
, 0);
128 #endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
131 static void mutex_unlock(pthread_mutex_t
*mutex
)
135 ret
= pthread_mutex_unlock(mutex
);
140 #ifdef RCU_MEMBARRIER
141 static void smp_mb_master(int group
)
143 if (caa_likely(rcu_has_sys_membarrier
))
144 membarrier(MEMBARRIER_EXPEDITED
);
151 static void smp_mb_master(int group
)
158 static void force_mb_all_readers(void)
160 struct rcu_reader
*index
;
163 * Ask for each threads to execute a cmm_smp_mb() so we can consider the
164 * compiler barriers around rcu read lock as real memory barriers.
166 if (cds_list_empty(®istry
))
169 * pthread_kill has a cmm_smp_mb(). But beware, we assume it performs
170 * a cache flush on architectures with non-coherent cache. Let's play
171 * safe and don't assume anything : we use cmm_smp_mc() to make sure the
172 * cache flush is enforced.
174 cds_list_for_each_entry(index
, ®istry
, node
) {
175 CMM_STORE_SHARED(index
->need_mb
, 1);
176 pthread_kill(index
->tid
, SIGRCU
);
179 * Wait for sighandler (and thus mb()) to execute on every thread.
181 * Note that the pthread_kill() will never be executed on systems
182 * that correctly deliver signals in a timely manner. However, it
183 * is not uncommon for kernels to have bugs that can result in
184 * lost or unduly delayed signals.
186 * If you are seeing the below pthread_kill() executing much at
187 * all, we suggest testing the underlying kernel and filing the
188 * relevant bug report. For Linux kernels, we recommend getting
189 * the Linux Test Project (LTP).
191 cds_list_for_each_entry(index
, ®istry
, node
) {
192 while (CMM_LOAD_SHARED(index
->need_mb
)) {
193 pthread_kill(index
->tid
, SIGRCU
);
197 cmm_smp_mb(); /* read ->need_mb before ending the barrier */
200 static void smp_mb_master(int group
)
202 force_mb_all_readers();
204 #endif /* #ifdef RCU_SIGNAL */
207 * synchronize_rcu() waiting. Single thread.
209 static void wait_gp(void)
211 /* Read reader_gp before read futex */
212 smp_mb_master(RCU_MB_GROUP
);
213 if (uatomic_read(&rcu_gp_futex
) == -1)
214 futex_async(&rcu_gp_futex
, FUTEX_WAIT
, -1,
218 static void update_counter_and_wait(void)
220 CDS_LIST_HEAD(qsreaders
);
222 struct rcu_reader
*index
, *tmp
;
224 /* Switch parity: 0 -> 1, 1 -> 0 */
225 CMM_STORE_SHARED(rcu_gp_ctr
, rcu_gp_ctr
^ RCU_GP_CTR_PHASE
);
228 * Must commit rcu_gp_ctr update to memory before waiting for quiescent
229 * state. Failure to do so could result in the writer waiting forever
230 * while new readers are always accessing data (no progress). Enforce
231 * compiler-order of store to rcu_gp_ctr before load rcu_reader ctr.
237 * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
238 * model easier to understand. It does not have a big performance impact
239 * anyway, given this is the write-side.
244 * Wait for each thread URCU_TLS(rcu_reader).ctr count to become 0.
248 if (wait_loops
== RCU_QS_ACTIVE_ATTEMPTS
) {
249 uatomic_dec(&rcu_gp_futex
);
250 /* Write futex before read reader_gp */
251 smp_mb_master(RCU_MB_GROUP
);
254 cds_list_for_each_entry_safe(index
, tmp
, ®istry
, node
) {
255 if (!rcu_gp_ongoing(&index
->ctr
))
256 cds_list_move(&index
->node
, &qsreaders
);
259 #ifndef HAS_INCOHERENT_CACHES
260 if (cds_list_empty(®istry
)) {
261 if (wait_loops
== RCU_QS_ACTIVE_ATTEMPTS
) {
262 /* Read reader_gp before write futex */
263 smp_mb_master(RCU_MB_GROUP
);
264 uatomic_set(&rcu_gp_futex
, 0);
268 if (wait_loops
== RCU_QS_ACTIVE_ATTEMPTS
)
273 #else /* #ifndef HAS_INCOHERENT_CACHES */
275 * BUSY-LOOP. Force the reader thread to commit its
276 * URCU_TLS(rcu_reader).ctr update to memory if we wait
279 if (cds_list_empty(®istry
)) {
280 if (wait_loops
== RCU_QS_ACTIVE_ATTEMPTS
) {
281 /* Read reader_gp before write futex */
282 smp_mb_master(RCU_MB_GROUP
);
283 uatomic_set(&rcu_gp_futex
, 0);
287 switch (wait_loops
) {
288 case RCU_QS_ACTIVE_ATTEMPTS
:
290 break; /* only escape switch */
291 case KICK_READER_LOOPS
:
292 smp_mb_master(RCU_MB_GROUP
);
294 break; /* only escape switch */
299 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
301 /* put back the reader list in the registry */
302 cds_list_splice(&qsreaders
, ®istry
);
305 void synchronize_rcu(void)
307 mutex_lock(&rcu_gp_lock
);
309 if (cds_list_empty(®istry
))
312 /* All threads should read qparity before accessing data structure
313 * where new ptr points to. Must be done within rcu_gp_lock because it
314 * iterates on reader threads.*/
315 /* Write new ptr before changing the qparity */
316 smp_mb_master(RCU_MB_GROUP
);
319 * Wait for previous parity to be empty of readers.
321 update_counter_and_wait(); /* 0 -> 1, wait readers in parity 0 */
324 * Must finish waiting for quiescent state for parity 0 before
325 * committing next rcu_gp_ctr update to memory. Failure to do so could
326 * result in the writer waiting forever while new readers are always
327 * accessing data (no progress). Enforce compiler-order of load
328 * URCU_TLS(rcu_reader).ctr before store to rcu_gp_ctr.
333 * Adding a cmm_smp_mb() which is _not_ formally required, but makes the
334 * model easier to understand. It does not have a big performance impact
335 * anyway, given this is the write-side.
340 * Wait for previous parity to be empty of readers.
342 update_counter_and_wait(); /* 1 -> 0, wait readers in parity 1 */
344 /* Finish waiting for reader threads before letting the old ptr being
345 * freed. Must be done within rcu_gp_lock because it iterates on reader
347 smp_mb_master(RCU_MB_GROUP
);
349 mutex_unlock(&rcu_gp_lock
);
353 * library wrappers to be used by non-LGPL compatible source code.
356 void rcu_read_lock(void)
361 void rcu_read_unlock(void)
366 void rcu_register_thread(void)
368 URCU_TLS(rcu_reader
).tid
= pthread_self();
369 assert(URCU_TLS(rcu_reader
).need_mb
== 0);
370 assert(!(URCU_TLS(rcu_reader
).ctr
& RCU_GP_CTR_NEST_MASK
));
372 mutex_lock(&rcu_gp_lock
);
373 rcu_init(); /* In case gcc does not support constructor attribute */
374 cds_list_add(&URCU_TLS(rcu_reader
).node
, ®istry
);
375 mutex_unlock(&rcu_gp_lock
);
378 void rcu_unregister_thread(void)
380 mutex_lock(&rcu_gp_lock
);
381 cds_list_del(&URCU_TLS(rcu_reader
).node
);
382 mutex_unlock(&rcu_gp_lock
);
385 #ifdef RCU_MEMBARRIER
391 if (!membarrier(MEMBARRIER_EXPEDITED
| MEMBARRIER_QUERY
))
392 rcu_has_sys_membarrier
= 1;
397 static void sigrcu_handler(int signo
, siginfo_t
*siginfo
, void *context
)
400 * Executing this cmm_smp_mb() is the only purpose of this signal handler.
401 * It punctually promotes cmm_barrier() into cmm_smp_mb() on every thread it is
405 _CMM_STORE_SHARED(URCU_TLS(rcu_reader
).need_mb
, 0);
410 * rcu_init constructor. Called when the library is linked, but also when
411 * reader threads are calling rcu_register_thread().
412 * Should only be called by a single thread at a given time. This is ensured by
413 * holing the rcu_gp_lock from rcu_register_thread() or by running at library
414 * load time, which should not be executed by multiple threads nor concurrently
415 * with rcu_register_thread() anyway.
419 struct sigaction act
;
426 act
.sa_sigaction
= sigrcu_handler
;
427 act
.sa_flags
= SA_SIGINFO
| SA_RESTART
;
428 sigemptyset(&act
.sa_mask
);
429 ret
= sigaction(SIGRCU
, &act
, NULL
);
436 struct sigaction act
;
439 ret
= sigaction(SIGRCU
, NULL
, &act
);
442 assert(act
.sa_sigaction
== sigrcu_handler
);
443 assert(cds_list_empty(®istry
));
446 #endif /* #ifdef RCU_SIGNAL */
448 DEFINE_RCU_FLAVOR(rcu_flavor
);
450 #include "urcu-call-rcu-impl.h"
451 #include "urcu-defer-impl.h"