Detect RCU read-side underflows
[urcu.git] / urcu / static / urcu.h
CommitLineData
adcfce54
MD
1#ifndef _URCU_STATIC_H
2#define _URCU_STATIC_H
3
4/*
5 * urcu-static.h
6 *
d2d23035 7 * Userspace RCU header.
adcfce54 8 *
a5a9f428
PM
9 * TO BE INCLUDED ONLY IN CODE THAT IS TO BE RECOMPILED ON EACH LIBURCU
10 * RELEASE. See urcu.h for linking dynamically with the userspace rcu library.
adcfce54 11 *
6982d6d7 12 * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
d2d23035 13 * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
adcfce54 14 *
d2d23035
MD
15 * This library is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU Lesser General Public
17 * License as published by the Free Software Foundation; either
18 * version 2.1 of the License, or (at your option) any later version.
19 *
20 * This library is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * Lesser General Public License for more details.
24 *
25 * You should have received a copy of the GNU Lesser General Public
26 * License along with this library; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
adcfce54
MD
28 *
29 * IBM's contributions to this file may be relicensed under LGPLv2 or later.
30 */
31
32#include <stdlib.h>
33#include <pthread.h>
bc6c15bb 34#include <unistd.h>
6d841bc2 35#include <stdint.h>
adcfce54 36
ec4e58a3
MD
37#include <urcu/compiler.h>
38#include <urcu/arch.h>
7e30abe3 39#include <urcu/system.h>
a2e7bf9c 40#include <urcu/uatomic.h>
e3b0cef0 41#include <urcu/list.h>
41849996 42#include <urcu/futex.h>
bd252a04 43#include <urcu/tls-compat.h>
e1c0b55c 44#include <urcu/rand-compat.h>
a77f7d82 45#include <urcu/debug.h>
bc6c15bb 46
36bc70a8
MD
47#ifdef __cplusplus
48extern "C" {
49#endif
50
65f1e634 51/* Default is RCU_MEMBARRIER */
fdf01eed 52#if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL)
65f1e634
MD
53#define RCU_MEMBARRIER
54#endif
55
adcfce54
MD
56/*
57 * This code section can only be included in LGPL 2.1 compatible source code.
58 * See below for the function call wrappers which can be used in code meant to
59 * be only linked with the Userspace RCU library. This comes with a small
60 * performance degradation on the read-side due to the added function calls.
61 * This is required to permit relinking with newer versions of the library.
62 */
63
64/*
65 * The signal number used by the RCU library can be overridden with
02be5561 66 * -DSIGRCU= when compiling the library.
ddf7eefb 67 * Provide backward compatibility for liburcu 0.3.x SIGURCU.
adcfce54 68 */
ddf7eefb
MD
69#ifdef SIGURCU
70#define SIGRCU SIGURCU
71#endif
72
02be5561
MD
73#ifndef SIGRCU
74#define SIGRCU SIGUSR1
adcfce54
MD
75#endif
76
fd189fa5
MD
77enum rcu_state {
78 RCU_READER_ACTIVE_CURRENT,
79 RCU_READER_ACTIVE_OLD,
80 RCU_READER_INACTIVE,
81};
82
25cc6d18
MD
83/*
84 * RCU memory barrier broadcast group. Currently, only broadcast to all process
85 * threads is supported (group 0).
86 *
87 * Slave barriers are only guaranteed to be ordered wrt master barriers.
88 *
89 * The pair ordering is detailed as (O: ordered, X: not ordered) :
90 * slave master
91 * slave X O
92 * master O O
93 */
94
95#define MB_GROUP_ALL 0
96#define RCU_MB_GROUP MB_GROUP_ALL
97
fdf01eed 98#ifdef RCU_MEMBARRIER
1de4df4b 99extern int rcu_has_sys_membarrier;
fdf01eed 100
25cc6d18 101static inline void smp_mb_slave(int group)
fdf01eed 102{
1de4df4b 103 if (caa_likely(rcu_has_sys_membarrier))
5481ddb3 104 cmm_barrier();
fdf01eed 105 else
5481ddb3 106 cmm_smp_mb();
fdf01eed
MD
107}
108#endif
109
02be5561 110#ifdef RCU_MB
25cc6d18 111static inline void smp_mb_slave(int group)
adcfce54 112{
5481ddb3 113 cmm_smp_mb();
adcfce54 114}
fdf01eed
MD
115#endif
116
117#ifdef RCU_SIGNAL
25cc6d18 118static inline void smp_mb_slave(int group)
adcfce54 119{
5481ddb3 120 cmm_barrier();
adcfce54
MD
121}
122#endif
123
124/*
02be5561
MD
125 * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
126 * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
adcfce54
MD
127 */
128#define RCU_GP_COUNT (1UL << 0)
129/* Use the amount of bits equal to half of the architecture long size */
27d65bc5 130#define RCU_GP_CTR_PHASE (1UL << (sizeof(unsigned long) << 2))
02be5561 131#define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1)
adcfce54 132
4de0cd31 133struct rcu_gp {
ed1b099e
LJ
134 /*
135 * Global grace period counter.
136 * Contains the current RCU_GP_CTR_PHASE.
137 * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
138 * Written to only by writer with mutex taken.
139 * Read by both writer and readers.
140 */
141 unsigned long ctr;
142
143 int32_t futex;
144} __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
145
4de0cd31 146extern struct rcu_gp rcu_gp;
adcfce54 147
02be5561 148struct rcu_reader {
bd1a5e15 149 /* Data used by both reader and synchronize_rcu() */
27d65bc5 150 unsigned long ctr;
e3b0cef0 151 char need_mb;
bd1a5e15 152 /* Data used for registry */
16aa9ee8 153 struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
bd1a5e15 154 pthread_t tid;
a77f7d82
MD
155 /* Reader registered flag, for internal checks. */
156 unsigned int registered:1;
e3b0cef0
MD
157};
158
bd252a04 159extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
adcfce54 160
bc6c15bb
MD
161/*
162 * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
163 */
164static inline void wake_up_gp(void)
165{
ed1b099e
LJ
166 if (caa_unlikely(uatomic_read(&rcu_gp.futex) == -1)) {
167 uatomic_set(&rcu_gp.futex, 0);
b0a841b4
MD
168 /*
169 * Ignoring return value until we can make this function
170 * return something (because urcu_die() is not publicly
171 * exposed).
172 */
173 (void) futex_async(&rcu_gp.futex, FUTEX_WAKE, 1,
174 NULL, NULL, 0);
bc6c15bb
MD
175 }
176}
177
fd189fa5 178static inline enum rcu_state rcu_reader_state(unsigned long *ctr)
adcfce54 179{
27d65bc5 180 unsigned long v;
adcfce54 181
adcfce54
MD
182 /*
183 * Make sure both tests below are done on the same version of *value
184 * to insure consistency.
185 */
6cf3827c 186 v = CMM_LOAD_SHARED(*ctr);
fd189fa5
MD
187 if (!(v & RCU_GP_CTR_NEST_MASK))
188 return RCU_READER_INACTIVE;
ed1b099e 189 if (!((v ^ rcu_gp.ctr) & RCU_GP_CTR_PHASE))
fd189fa5
MD
190 return RCU_READER_ACTIVE_CURRENT;
191 return RCU_READER_ACTIVE_OLD;
adcfce54
MD
192}
193
a5a9f428 194/*
ed1b099e 195 * Helper for _rcu_read_lock(). The format of rcu_gp.ctr (as well as
a5a9f428
PM
196 * the per-thread rcu_reader.ctr) has the upper bits containing a count of
197 * _rcu_read_lock() nesting, and a lower-order bit that contains either zero
198 * or RCU_GP_CTR_PHASE. The smp_mb_slave() ensures that the accesses in
199 * _rcu_read_lock() happen before the subsequent read-side critical section.
200 */
201static inline void _rcu_read_lock_update(unsigned long tmp)
adcfce54 202{
a0b7f7ea 203 if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
ed1b099e 204 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp.ctr));
25cc6d18 205 smp_mb_slave(RCU_MB_GROUP);
a5a9f428 206 } else
bd252a04 207 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp + RCU_GP_COUNT);
adcfce54
MD
208}
209
a5a9f428
PM
210/*
211 * Enter an RCU read-side critical section.
212 *
213 * The first cmm_barrier() call ensures that the compiler does not reorder
214 * the body of _rcu_read_lock() with a mutex.
215 *
216 * This function and its helper are both less than 10 lines long. The
217 * intent is that this function meets the 10-line criterion in LGPL,
218 * allowing this function to be invoked directly from non-LGPL code.
219 */
220static inline void _rcu_read_lock(void)
adcfce54 221{
27d65bc5 222 unsigned long tmp;
bc6c15bb 223
a77f7d82 224 urcu_assert(URCU_TLS(rcu_reader).registered);
a5a9f428 225 cmm_barrier();
bd252a04 226 tmp = URCU_TLS(rcu_reader).ctr;
a5a9f428
PM
227 _rcu_read_lock_update(tmp);
228}
229
230/*
231 * This is a helper function for _rcu_read_unlock().
232 *
233 * The first smp_mb_slave() call ensures that the critical section is
234 * seen to precede the store to rcu_reader.ctr.
235 * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr
236 * before reading the update-side futex.
237 */
238static inline void _rcu_read_unlock_update_and_wakeup(unsigned long tmp)
239{
a0b7f7ea 240 if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
25cc6d18 241 smp_mb_slave(RCU_MB_GROUP);
343c8b13 242 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT);
25cc6d18 243 smp_mb_slave(RCU_MB_GROUP);
bc6c15bb 244 wake_up_gp();
a5a9f428 245 } else
343c8b13 246 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT);
a5a9f428
PM
247}
248
249/*
250 * Exit an RCU read-side crtical section. Both this function and its
251 * helper are smaller than 10 lines of code, and are intended to be
252 * usable by non-LGPL code, as called out in LGPL.
253 */
254static inline void _rcu_read_unlock(void)
255{
256 unsigned long tmp;
257
a77f7d82 258 urcu_assert(URCU_TLS(rcu_reader).registered);
a5a9f428 259 tmp = URCU_TLS(rcu_reader).ctr;
343c8b13 260 urcu_assert(tmp & RCU_GP_CTR_NEST_MASK);
a5a9f428 261 _rcu_read_unlock_update_and_wakeup(tmp);
5481ddb3 262 cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */
adcfce54
MD
263}
264
882f3357
MD
265/*
266 * Returns whether within a RCU read-side critical section.
267 *
268 * This function is less than 10 lines long. The intent is that this
269 * function meets the 10-line criterion for LGPL, allowing this function
270 * to be invoked directly from non-LGPL code.
271 */
272static inline int _rcu_read_ongoing(void)
273{
274 return URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK;
275}
276
ad918eeb 277#ifdef __cplusplus
36bc70a8
MD
278}
279#endif
280
adcfce54 281#endif /* _URCU_STATIC_H */
This page took 0.075562 seconds and 4 git commands to generate.