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 | * |
d2d23035 MD |
9 | * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See urcu.h for linking |
10 | * 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> |
bc6c15bb | 43 | |
36bc70a8 MD |
44 | #ifdef __cplusplus |
45 | extern "C" { | |
46 | #endif | |
47 | ||
4906d398 | 48 | /* Default is RCU_MEMBARRIER on linux */ |
fdf01eed | 49 | #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL) |
4906d398 MD |
50 | # ifdef __linux__ |
51 | # define RCU_MEMBARRIER | |
52 | # else | |
53 | # define RCU_MB | |
54 | # endif | |
fdf01eed MD |
55 | #endif |
56 | ||
57 | #ifdef RCU_MEMBARRIER | |
4906d398 | 58 | #include <syscall.h> |
fdf01eed MD |
59 | |
60 | /* If the headers do not support SYS_membarrier, statically use RCU_MB */ | |
61 | #ifdef SYS_membarrier | |
cf5271ee MD |
62 | # define MEMBARRIER_EXPEDITED (1 << 0) |
63 | # define MEMBARRIER_DELAYED (1 << 1) | |
64 | # define MEMBARRIER_QUERY (1 << 16) | |
d4fc680a | 65 | # define membarrier(...) syscall(SYS_membarrier, __VA_ARGS__) |
fdf01eed | 66 | #else |
f0708810 MD |
67 | # undef RCU_MEMBARRIER |
68 | # define RCU_MB | |
fdf01eed MD |
69 | #endif |
70 | #endif | |
71 | ||
adcfce54 MD |
72 | /* |
73 | * This code section can only be included in LGPL 2.1 compatible source code. | |
74 | * See below for the function call wrappers which can be used in code meant to | |
75 | * be only linked with the Userspace RCU library. This comes with a small | |
76 | * performance degradation on the read-side due to the added function calls. | |
77 | * This is required to permit relinking with newer versions of the library. | |
78 | */ | |
79 | ||
80 | /* | |
81 | * The signal number used by the RCU library can be overridden with | |
02be5561 | 82 | * -DSIGRCU= when compiling the library. |
ddf7eefb | 83 | * Provide backward compatibility for liburcu 0.3.x SIGURCU. |
adcfce54 | 84 | */ |
ddf7eefb MD |
85 | #ifdef SIGURCU |
86 | #define SIGRCU SIGURCU | |
87 | #endif | |
88 | ||
02be5561 MD |
89 | #ifndef SIGRCU |
90 | #define SIGRCU SIGUSR1 | |
adcfce54 MD |
91 | #endif |
92 | ||
93 | /* | |
94 | * If a reader is really non-cooperative and refuses to commit its | |
02be5561 | 95 | * rcu_active_readers count to memory (there is no barrier in the reader |
adcfce54 MD |
96 | * per-se), kick it after a few loops waiting for it. |
97 | */ | |
98 | #define KICK_READER_LOOPS 10000 | |
99 | ||
bc6c15bb MD |
100 | /* |
101 | * Active attempts to check for reader Q.S. before calling futex(). | |
102 | */ | |
103 | #define RCU_QS_ACTIVE_ATTEMPTS 100 | |
104 | ||
7ac06cef MD |
105 | #ifdef DEBUG_RCU |
106 | #define rcu_assert(args...) assert(args) | |
107 | #else | |
108 | #define rcu_assert(args...) | |
109 | #endif | |
110 | ||
adcfce54 MD |
111 | #ifdef DEBUG_YIELD |
112 | #include <sched.h> | |
113 | #include <time.h> | |
114 | #include <pthread.h> | |
115 | #include <unistd.h> | |
116 | ||
117 | #define YIELD_READ (1 << 0) | |
118 | #define YIELD_WRITE (1 << 1) | |
119 | ||
b4ce1526 | 120 | /* |
fdf01eed | 121 | * Updates with RCU_SIGNAL are much slower. Account this in the delay. |
b4ce1526 | 122 | */ |
fdf01eed | 123 | #ifdef RCU_SIGNAL |
adcfce54 | 124 | /* maximum sleep delay, in us */ |
adcfce54 | 125 | #define MAX_SLEEP 30000 |
fdf01eed MD |
126 | #else |
127 | #define MAX_SLEEP 50 | |
adcfce54 MD |
128 | #endif |
129 | ||
130 | extern unsigned int yield_active; | |
131 | extern unsigned int __thread rand_yield; | |
132 | ||
133 | static inline void debug_yield_read(void) | |
134 | { | |
135 | if (yield_active & YIELD_READ) | |
136 | if (rand_r(&rand_yield) & 0x1) | |
137 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
138 | } | |
139 | ||
140 | static inline void debug_yield_write(void) | |
141 | { | |
142 | if (yield_active & YIELD_WRITE) | |
143 | if (rand_r(&rand_yield) & 0x1) | |
144 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
145 | } | |
146 | ||
147 | static inline void debug_yield_init(void) | |
148 | { | |
149 | rand_yield = time(NULL) ^ pthread_self(); | |
150 | } | |
151 | #else | |
152 | static inline void debug_yield_read(void) | |
153 | { | |
154 | } | |
155 | ||
156 | static inline void debug_yield_write(void) | |
157 | { | |
158 | } | |
159 | ||
160 | static inline void debug_yield_init(void) | |
161 | { | |
162 | ||
163 | } | |
164 | #endif | |
165 | ||
25cc6d18 MD |
166 | /* |
167 | * RCU memory barrier broadcast group. Currently, only broadcast to all process | |
168 | * threads is supported (group 0). | |
169 | * | |
170 | * Slave barriers are only guaranteed to be ordered wrt master barriers. | |
171 | * | |
172 | * The pair ordering is detailed as (O: ordered, X: not ordered) : | |
173 | * slave master | |
174 | * slave X O | |
175 | * master O O | |
176 | */ | |
177 | ||
178 | #define MB_GROUP_ALL 0 | |
179 | #define RCU_MB_GROUP MB_GROUP_ALL | |
180 | ||
fdf01eed MD |
181 | #ifdef RCU_MEMBARRIER |
182 | extern int has_sys_membarrier; | |
183 | ||
25cc6d18 | 184 | static inline void smp_mb_slave(int group) |
fdf01eed MD |
185 | { |
186 | if (likely(has_sys_membarrier)) | |
5481ddb3 | 187 | cmm_barrier(); |
fdf01eed | 188 | else |
5481ddb3 | 189 | cmm_smp_mb(); |
fdf01eed MD |
190 | } |
191 | #endif | |
192 | ||
02be5561 | 193 | #ifdef RCU_MB |
25cc6d18 | 194 | static inline void smp_mb_slave(int group) |
adcfce54 | 195 | { |
5481ddb3 | 196 | cmm_smp_mb(); |
adcfce54 | 197 | } |
fdf01eed MD |
198 | #endif |
199 | ||
200 | #ifdef RCU_SIGNAL | |
25cc6d18 | 201 | static inline void smp_mb_slave(int group) |
adcfce54 | 202 | { |
5481ddb3 | 203 | cmm_barrier(); |
adcfce54 MD |
204 | } |
205 | #endif | |
206 | ||
207 | /* | |
02be5561 MD |
208 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use |
209 | * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. | |
adcfce54 MD |
210 | */ |
211 | #define RCU_GP_COUNT (1UL << 0) | |
212 | /* Use the amount of bits equal to half of the architecture long size */ | |
27d65bc5 | 213 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(unsigned long) << 2)) |
02be5561 | 214 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) |
adcfce54 MD |
215 | |
216 | /* | |
217 | * Global quiescent period counter with low-order bits unused. | |
218 | * Using a int rather than a char to eliminate false register dependencies | |
219 | * causing stalls on some architectures. | |
220 | */ | |
27d65bc5 | 221 | extern unsigned long rcu_gp_ctr; |
adcfce54 | 222 | |
02be5561 | 223 | struct rcu_reader { |
bd1a5e15 | 224 | /* Data used by both reader and synchronize_rcu() */ |
27d65bc5 | 225 | unsigned long ctr; |
e3b0cef0 | 226 | char need_mb; |
bd1a5e15 | 227 | /* Data used for registry */ |
16aa9ee8 | 228 | struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
bd1a5e15 | 229 | pthread_t tid; |
e3b0cef0 MD |
230 | }; |
231 | ||
02be5561 | 232 | extern struct rcu_reader __thread rcu_reader; |
adcfce54 | 233 | |
6d841bc2 | 234 | extern int32_t gp_futex; |
bc6c15bb MD |
235 | |
236 | /* | |
237 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. | |
238 | */ | |
239 | static inline void wake_up_gp(void) | |
240 | { | |
ec4e58a3 MD |
241 | if (unlikely(uatomic_read(&gp_futex) == -1)) { |
242 | uatomic_set(&gp_futex, 0); | |
0854ccff | 243 | futex_async(&gp_futex, FUTEX_WAKE, 1, |
bc6c15bb MD |
244 | NULL, NULL, 0); |
245 | } | |
246 | } | |
247 | ||
e26fa029 | 248 | static inline int rcu_gp_ongoing(unsigned long *ctr) |
adcfce54 | 249 | { |
27d65bc5 | 250 | unsigned long v; |
adcfce54 | 251 | |
adcfce54 MD |
252 | /* |
253 | * Make sure both tests below are done on the same version of *value | |
254 | * to insure consistency. | |
255 | */ | |
6cf3827c | 256 | v = CMM_LOAD_SHARED(*ctr); |
adcfce54 | 257 | return (v & RCU_GP_CTR_NEST_MASK) && |
02be5561 | 258 | ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE); |
adcfce54 MD |
259 | } |
260 | ||
261 | static inline void _rcu_read_lock(void) | |
262 | { | |
27d65bc5 | 263 | unsigned long tmp; |
adcfce54 | 264 | |
5481ddb3 | 265 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
02be5561 MD |
266 | tmp = rcu_reader.ctr; |
267 | /* | |
268 | * rcu_gp_ctr is | |
269 | * RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE) | |
270 | */ | |
67ef1a2c | 271 | if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
6cf3827c | 272 | _CMM_STORE_SHARED(rcu_reader.ctr, _CMM_LOAD_SHARED(rcu_gp_ctr)); |
67ef1a2c MD |
273 | /* |
274 | * Set active readers count for outermost nesting level before | |
25cc6d18 | 275 | * accessing the pointer. See smp_mb_master(). |
67ef1a2c | 276 | */ |
25cc6d18 | 277 | smp_mb_slave(RCU_MB_GROUP); |
67ef1a2c | 278 | } else { |
6cf3827c | 279 | _CMM_STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT); |
67ef1a2c | 280 | } |
adcfce54 MD |
281 | } |
282 | ||
283 | static inline void _rcu_read_unlock(void) | |
284 | { | |
27d65bc5 | 285 | unsigned long tmp; |
bc6c15bb | 286 | |
02be5561 | 287 | tmp = rcu_reader.ctr; |
adcfce54 MD |
288 | /* |
289 | * Finish using rcu before decrementing the pointer. | |
25cc6d18 | 290 | * See smp_mb_master(). |
adcfce54 | 291 | */ |
bc6c15bb | 292 | if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) { |
25cc6d18 | 293 | smp_mb_slave(RCU_MB_GROUP); |
6cf3827c | 294 | _CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
02be5561 | 295 | /* write rcu_reader.ctr before read futex */ |
25cc6d18 | 296 | smp_mb_slave(RCU_MB_GROUP); |
bc6c15bb MD |
297 | wake_up_gp(); |
298 | } else { | |
6cf3827c | 299 | _CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
bc6c15bb | 300 | } |
5481ddb3 | 301 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
adcfce54 MD |
302 | } |
303 | ||
ad918eeb | 304 | #ifdef __cplusplus |
36bc70a8 MD |
305 | } |
306 | #endif | |
307 | ||
adcfce54 | 308 | #endif /* _URCU_STATIC_H */ |