Commit | Line | Data |
---|---|---|
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> |
a77f7d82 | 44 | #include <urcu/debug.h> |
bc6c15bb | 45 | |
36bc70a8 MD |
46 | #ifdef __cplusplus |
47 | extern "C" { | |
67ecffc0 | 48 | #endif |
36bc70a8 | 49 | |
65f1e634 | 50 | /* Default is RCU_MEMBARRIER */ |
fdf01eed | 51 | #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL) |
65f1e634 MD |
52 | #define RCU_MEMBARRIER |
53 | #endif | |
54 | ||
adcfce54 MD |
55 | /* |
56 | * This code section can only be included in LGPL 2.1 compatible source code. | |
57 | * See below for the function call wrappers which can be used in code meant to | |
58 | * be only linked with the Userspace RCU library. This comes with a small | |
59 | * performance degradation on the read-side due to the added function calls. | |
60 | * This is required to permit relinking with newer versions of the library. | |
61 | */ | |
62 | ||
63 | /* | |
64 | * The signal number used by the RCU library can be overridden with | |
02be5561 | 65 | * -DSIGRCU= when compiling the library. |
ddf7eefb | 66 | * Provide backward compatibility for liburcu 0.3.x SIGURCU. |
adcfce54 | 67 | */ |
ddf7eefb MD |
68 | #ifdef SIGURCU |
69 | #define SIGRCU SIGURCU | |
70 | #endif | |
71 | ||
02be5561 MD |
72 | #ifndef SIGRCU |
73 | #define SIGRCU SIGUSR1 | |
adcfce54 MD |
74 | #endif |
75 | ||
fd189fa5 MD |
76 | enum rcu_state { |
77 | RCU_READER_ACTIVE_CURRENT, | |
78 | RCU_READER_ACTIVE_OLD, | |
79 | RCU_READER_INACTIVE, | |
80 | }; | |
81 | ||
25cc6d18 | 82 | /* |
25cc6d18 MD |
83 | * Slave barriers are only guaranteed to be ordered wrt master barriers. |
84 | * | |
85 | * The pair ordering is detailed as (O: ordered, X: not ordered) : | |
86 | * slave master | |
87 | * slave X O | |
88 | * master O O | |
89 | */ | |
90 | ||
fdf01eed | 91 | #ifdef RCU_MEMBARRIER |
1de4df4b | 92 | extern int rcu_has_sys_membarrier; |
fdf01eed | 93 | |
a4922ed9 | 94 | static inline void smp_mb_slave(void) |
fdf01eed | 95 | { |
1de4df4b | 96 | if (caa_likely(rcu_has_sys_membarrier)) |
5481ddb3 | 97 | cmm_barrier(); |
fdf01eed | 98 | else |
5481ddb3 | 99 | cmm_smp_mb(); |
fdf01eed MD |
100 | } |
101 | #endif | |
102 | ||
02be5561 | 103 | #ifdef RCU_MB |
a4922ed9 | 104 | static inline void smp_mb_slave(void) |
adcfce54 | 105 | { |
5481ddb3 | 106 | cmm_smp_mb(); |
adcfce54 | 107 | } |
fdf01eed MD |
108 | #endif |
109 | ||
110 | #ifdef RCU_SIGNAL | |
a4922ed9 | 111 | static inline void smp_mb_slave(void) |
adcfce54 | 112 | { |
5481ddb3 | 113 | cmm_barrier(); |
adcfce54 MD |
114 | } |
115 | #endif | |
116 | ||
117 | /* | |
02be5561 MD |
118 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use |
119 | * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. | |
adcfce54 MD |
120 | */ |
121 | #define RCU_GP_COUNT (1UL << 0) | |
122 | /* Use the amount of bits equal to half of the architecture long size */ | |
27d65bc5 | 123 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(unsigned long) << 2)) |
02be5561 | 124 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) |
adcfce54 | 125 | |
4de0cd31 | 126 | struct rcu_gp { |
ed1b099e LJ |
127 | /* |
128 | * Global grace period counter. | |
129 | * Contains the current RCU_GP_CTR_PHASE. | |
130 | * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path. | |
131 | * Written to only by writer with mutex taken. | |
132 | * Read by both writer and readers. | |
133 | */ | |
134 | unsigned long ctr; | |
135 | ||
136 | int32_t futex; | |
137 | } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); | |
138 | ||
4de0cd31 | 139 | extern struct rcu_gp rcu_gp; |
adcfce54 | 140 | |
02be5561 | 141 | struct rcu_reader { |
bd1a5e15 | 142 | /* Data used by both reader and synchronize_rcu() */ |
27d65bc5 | 143 | unsigned long ctr; |
e3b0cef0 | 144 | char need_mb; |
bd1a5e15 | 145 | /* Data used for registry */ |
16aa9ee8 | 146 | struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
bd1a5e15 | 147 | pthread_t tid; |
a77f7d82 MD |
148 | /* Reader registered flag, for internal checks. */ |
149 | unsigned int registered:1; | |
e3b0cef0 MD |
150 | }; |
151 | ||
bd252a04 | 152 | extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader); |
adcfce54 | 153 | |
bc6c15bb MD |
154 | /* |
155 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. | |
156 | */ | |
157 | static inline void wake_up_gp(void) | |
158 | { | |
ed1b099e LJ |
159 | if (caa_unlikely(uatomic_read(&rcu_gp.futex) == -1)) { |
160 | uatomic_set(&rcu_gp.futex, 0); | |
b0a841b4 MD |
161 | /* |
162 | * Ignoring return value until we can make this function | |
163 | * return something (because urcu_die() is not publicly | |
164 | * exposed). | |
165 | */ | |
166 | (void) futex_async(&rcu_gp.futex, FUTEX_WAKE, 1, | |
167 | NULL, NULL, 0); | |
bc6c15bb MD |
168 | } |
169 | } | |
170 | ||
fd189fa5 | 171 | static inline enum rcu_state rcu_reader_state(unsigned long *ctr) |
adcfce54 | 172 | { |
27d65bc5 | 173 | unsigned long v; |
adcfce54 | 174 | |
adcfce54 MD |
175 | /* |
176 | * Make sure both tests below are done on the same version of *value | |
177 | * to insure consistency. | |
178 | */ | |
6cf3827c | 179 | v = CMM_LOAD_SHARED(*ctr); |
fd189fa5 MD |
180 | if (!(v & RCU_GP_CTR_NEST_MASK)) |
181 | return RCU_READER_INACTIVE; | |
ed1b099e | 182 | if (!((v ^ rcu_gp.ctr) & RCU_GP_CTR_PHASE)) |
fd189fa5 MD |
183 | return RCU_READER_ACTIVE_CURRENT; |
184 | return RCU_READER_ACTIVE_OLD; | |
adcfce54 MD |
185 | } |
186 | ||
a5a9f428 | 187 | /* |
ed1b099e | 188 | * Helper for _rcu_read_lock(). The format of rcu_gp.ctr (as well as |
a5a9f428 PM |
189 | * the per-thread rcu_reader.ctr) has the upper bits containing a count of |
190 | * _rcu_read_lock() nesting, and a lower-order bit that contains either zero | |
191 | * or RCU_GP_CTR_PHASE. The smp_mb_slave() ensures that the accesses in | |
192 | * _rcu_read_lock() happen before the subsequent read-side critical section. | |
193 | */ | |
194 | static inline void _rcu_read_lock_update(unsigned long tmp) | |
adcfce54 | 195 | { |
a0b7f7ea | 196 | if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
ed1b099e | 197 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp.ctr)); |
a4922ed9 | 198 | smp_mb_slave(); |
a5a9f428 | 199 | } else |
bd252a04 | 200 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp + RCU_GP_COUNT); |
adcfce54 MD |
201 | } |
202 | ||
a5a9f428 PM |
203 | /* |
204 | * Enter an RCU read-side critical section. | |
205 | * | |
206 | * The first cmm_barrier() call ensures that the compiler does not reorder | |
207 | * the body of _rcu_read_lock() with a mutex. | |
208 | * | |
209 | * This function and its helper are both less than 10 lines long. The | |
210 | * intent is that this function meets the 10-line criterion in LGPL, | |
211 | * allowing this function to be invoked directly from non-LGPL code. | |
212 | */ | |
213 | static inline void _rcu_read_lock(void) | |
adcfce54 | 214 | { |
27d65bc5 | 215 | unsigned long tmp; |
bc6c15bb | 216 | |
a77f7d82 | 217 | urcu_assert(URCU_TLS(rcu_reader).registered); |
a5a9f428 | 218 | cmm_barrier(); |
bd252a04 | 219 | tmp = URCU_TLS(rcu_reader).ctr; |
6ef5ddc9 | 220 | urcu_assert((tmp & RCU_GP_CTR_NEST_MASK) != RCU_GP_CTR_NEST_MASK); |
a5a9f428 PM |
221 | _rcu_read_lock_update(tmp); |
222 | } | |
223 | ||
224 | /* | |
225 | * This is a helper function for _rcu_read_unlock(). | |
226 | * | |
227 | * The first smp_mb_slave() call ensures that the critical section is | |
228 | * seen to precede the store to rcu_reader.ctr. | |
229 | * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr | |
230 | * before reading the update-side futex. | |
231 | */ | |
232 | static inline void _rcu_read_unlock_update_and_wakeup(unsigned long tmp) | |
233 | { | |
a0b7f7ea | 234 | if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) { |
a4922ed9 | 235 | smp_mb_slave(); |
343c8b13 | 236 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT); |
a4922ed9 | 237 | smp_mb_slave(); |
bc6c15bb | 238 | wake_up_gp(); |
a5a9f428 | 239 | } else |
343c8b13 | 240 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT); |
a5a9f428 PM |
241 | } |
242 | ||
243 | /* | |
244 | * Exit an RCU read-side crtical section. Both this function and its | |
245 | * helper are smaller than 10 lines of code, and are intended to be | |
246 | * usable by non-LGPL code, as called out in LGPL. | |
247 | */ | |
248 | static inline void _rcu_read_unlock(void) | |
249 | { | |
250 | unsigned long tmp; | |
251 | ||
a77f7d82 | 252 | urcu_assert(URCU_TLS(rcu_reader).registered); |
a5a9f428 | 253 | tmp = URCU_TLS(rcu_reader).ctr; |
343c8b13 | 254 | urcu_assert(tmp & RCU_GP_CTR_NEST_MASK); |
a5a9f428 | 255 | _rcu_read_unlock_update_and_wakeup(tmp); |
5481ddb3 | 256 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
adcfce54 MD |
257 | } |
258 | ||
882f3357 MD |
259 | /* |
260 | * Returns whether within a RCU read-side critical section. | |
261 | * | |
262 | * This function is less than 10 lines long. The intent is that this | |
263 | * function meets the 10-line criterion for LGPL, allowing this function | |
264 | * to be invoked directly from non-LGPL code. | |
265 | */ | |
266 | static inline int _rcu_read_ongoing(void) | |
267 | { | |
268 | return URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK; | |
269 | } | |
270 | ||
ad918eeb | 271 | #ifdef __cplusplus |
36bc70a8 MD |
272 | } |
273 | #endif | |
274 | ||
adcfce54 | 275 | #endif /* _URCU_STATIC_H */ |