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 | * |
d2d23035 MD |
12 | * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> |
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 MD |
34 | #include <syscall.h> |
35 | #include <unistd.h> | |
adcfce54 | 36 | |
ec4e58a3 MD |
37 | #include <urcu/compiler.h> |
38 | #include <urcu/arch.h> | |
7e30abe3 | 39 | #include <urcu/system.h> |
48d848c7 | 40 | #include <urcu/uatomic_arch.h> |
e3b0cef0 | 41 | #include <urcu/list.h> |
0854ccff | 42 | #include <urcu/urcu-futex.h> |
bc6c15bb | 43 | |
36bc70a8 MD |
44 | #ifdef __cplusplus |
45 | extern "C" { | |
46 | #endif | |
47 | ||
fdf01eed MD |
48 | /* Default is RCU_MEMBARRIER */ |
49 | #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL) | |
50 | #define RCU_MEMBARRIER | |
51 | #endif | |
52 | ||
53 | #ifdef RCU_MEMBARRIER | |
54 | #include <unistd.h> | |
55 | #include <sys/syscall.h> | |
56 | ||
57 | /* If the headers do not support SYS_membarrier, statically use RCU_MB */ | |
58 | #ifdef SYS_membarrier | |
cf5271ee MD |
59 | # define MEMBARRIER_EXPEDITED (1 << 0) |
60 | # define MEMBARRIER_DELAYED (1 << 1) | |
61 | # define MEMBARRIER_QUERY (1 << 16) | |
d4fc680a | 62 | # define membarrier(...) syscall(SYS_membarrier, __VA_ARGS__) |
fdf01eed | 63 | #else |
f0708810 MD |
64 | # undef RCU_MEMBARRIER |
65 | # define RCU_MB | |
fdf01eed MD |
66 | #endif |
67 | #endif | |
68 | ||
adcfce54 MD |
69 | /* |
70 | * This code section can only be included in LGPL 2.1 compatible source code. | |
71 | * See below for the function call wrappers which can be used in code meant to | |
72 | * be only linked with the Userspace RCU library. This comes with a small | |
73 | * performance degradation on the read-side due to the added function calls. | |
74 | * This is required to permit relinking with newer versions of the library. | |
75 | */ | |
76 | ||
77 | /* | |
78 | * The signal number used by the RCU library can be overridden with | |
02be5561 | 79 | * -DSIGRCU= when compiling the library. |
ddf7eefb | 80 | * Provide backward compatibility for liburcu 0.3.x SIGURCU. |
adcfce54 | 81 | */ |
ddf7eefb MD |
82 | #ifdef SIGURCU |
83 | #define SIGRCU SIGURCU | |
84 | #endif | |
85 | ||
02be5561 MD |
86 | #ifndef SIGRCU |
87 | #define SIGRCU SIGUSR1 | |
adcfce54 MD |
88 | #endif |
89 | ||
90 | /* | |
91 | * If a reader is really non-cooperative and refuses to commit its | |
02be5561 | 92 | * rcu_active_readers count to memory (there is no barrier in the reader |
adcfce54 MD |
93 | * per-se), kick it after a few loops waiting for it. |
94 | */ | |
95 | #define KICK_READER_LOOPS 10000 | |
96 | ||
bc6c15bb MD |
97 | /* |
98 | * Active attempts to check for reader Q.S. before calling futex(). | |
99 | */ | |
100 | #define RCU_QS_ACTIVE_ATTEMPTS 100 | |
101 | ||
7ac06cef MD |
102 | #ifdef DEBUG_RCU |
103 | #define rcu_assert(args...) assert(args) | |
104 | #else | |
105 | #define rcu_assert(args...) | |
106 | #endif | |
107 | ||
adcfce54 MD |
108 | #ifdef DEBUG_YIELD |
109 | #include <sched.h> | |
110 | #include <time.h> | |
111 | #include <pthread.h> | |
112 | #include <unistd.h> | |
113 | ||
114 | #define YIELD_READ (1 << 0) | |
115 | #define YIELD_WRITE (1 << 1) | |
116 | ||
b4ce1526 | 117 | /* |
fdf01eed | 118 | * Updates with RCU_SIGNAL are much slower. Account this in the delay. |
b4ce1526 | 119 | */ |
fdf01eed | 120 | #ifdef RCU_SIGNAL |
adcfce54 | 121 | /* maximum sleep delay, in us */ |
adcfce54 | 122 | #define MAX_SLEEP 30000 |
fdf01eed MD |
123 | #else |
124 | #define MAX_SLEEP 50 | |
adcfce54 MD |
125 | #endif |
126 | ||
127 | extern unsigned int yield_active; | |
128 | extern unsigned int __thread rand_yield; | |
129 | ||
130 | static inline void debug_yield_read(void) | |
131 | { | |
132 | if (yield_active & YIELD_READ) | |
133 | if (rand_r(&rand_yield) & 0x1) | |
134 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
135 | } | |
136 | ||
137 | static inline void debug_yield_write(void) | |
138 | { | |
139 | if (yield_active & YIELD_WRITE) | |
140 | if (rand_r(&rand_yield) & 0x1) | |
141 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
142 | } | |
143 | ||
144 | static inline void debug_yield_init(void) | |
145 | { | |
146 | rand_yield = time(NULL) ^ pthread_self(); | |
147 | } | |
148 | #else | |
149 | static inline void debug_yield_read(void) | |
150 | { | |
151 | } | |
152 | ||
153 | static inline void debug_yield_write(void) | |
154 | { | |
155 | } | |
156 | ||
157 | static inline void debug_yield_init(void) | |
158 | { | |
159 | ||
160 | } | |
161 | #endif | |
162 | ||
25cc6d18 MD |
163 | /* |
164 | * RCU memory barrier broadcast group. Currently, only broadcast to all process | |
165 | * threads is supported (group 0). | |
166 | * | |
167 | * Slave barriers are only guaranteed to be ordered wrt master barriers. | |
168 | * | |
169 | * The pair ordering is detailed as (O: ordered, X: not ordered) : | |
170 | * slave master | |
171 | * slave X O | |
172 | * master O O | |
173 | */ | |
174 | ||
175 | #define MB_GROUP_ALL 0 | |
176 | #define RCU_MB_GROUP MB_GROUP_ALL | |
177 | ||
fdf01eed MD |
178 | #ifdef RCU_MEMBARRIER |
179 | extern int has_sys_membarrier; | |
180 | ||
25cc6d18 | 181 | static inline void smp_mb_slave(int group) |
fdf01eed MD |
182 | { |
183 | if (likely(has_sys_membarrier)) | |
184 | barrier(); | |
185 | else | |
186 | smp_mb(); | |
187 | } | |
188 | #endif | |
189 | ||
02be5561 | 190 | #ifdef RCU_MB |
25cc6d18 | 191 | static inline void smp_mb_slave(int group) |
adcfce54 MD |
192 | { |
193 | smp_mb(); | |
194 | } | |
fdf01eed MD |
195 | #endif |
196 | ||
197 | #ifdef RCU_SIGNAL | |
25cc6d18 | 198 | static inline void smp_mb_slave(int group) |
adcfce54 MD |
199 | { |
200 | barrier(); | |
201 | } | |
202 | #endif | |
203 | ||
204 | /* | |
02be5561 MD |
205 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use |
206 | * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. | |
adcfce54 MD |
207 | */ |
208 | #define RCU_GP_COUNT (1UL << 0) | |
209 | /* Use the amount of bits equal to half of the architecture long size */ | |
27d65bc5 | 210 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(unsigned long) << 2)) |
02be5561 | 211 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) |
adcfce54 MD |
212 | |
213 | /* | |
214 | * Global quiescent period counter with low-order bits unused. | |
215 | * Using a int rather than a char to eliminate false register dependencies | |
216 | * causing stalls on some architectures. | |
217 | */ | |
27d65bc5 | 218 | extern unsigned long rcu_gp_ctr; |
adcfce54 | 219 | |
02be5561 | 220 | struct rcu_reader { |
bd1a5e15 | 221 | /* Data used by both reader and synchronize_rcu() */ |
27d65bc5 | 222 | unsigned long ctr; |
e3b0cef0 | 223 | char need_mb; |
bd1a5e15 MD |
224 | /* Data used for registry */ |
225 | struct list_head head __attribute__((aligned(CACHE_LINE_SIZE))); | |
226 | pthread_t tid; | |
e3b0cef0 MD |
227 | }; |
228 | ||
02be5561 | 229 | extern struct rcu_reader __thread rcu_reader; |
adcfce54 | 230 | |
bc6c15bb MD |
231 | extern int gp_futex; |
232 | ||
233 | /* | |
234 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. | |
235 | */ | |
236 | static inline void wake_up_gp(void) | |
237 | { | |
ec4e58a3 MD |
238 | if (unlikely(uatomic_read(&gp_futex) == -1)) { |
239 | uatomic_set(&gp_futex, 0); | |
0854ccff | 240 | futex_async(&gp_futex, FUTEX_WAKE, 1, |
bc6c15bb MD |
241 | NULL, NULL, 0); |
242 | } | |
243 | } | |
244 | ||
e26fa029 | 245 | static inline int rcu_gp_ongoing(unsigned long *ctr) |
adcfce54 | 246 | { |
27d65bc5 | 247 | unsigned long v; |
adcfce54 | 248 | |
adcfce54 MD |
249 | /* |
250 | * Make sure both tests below are done on the same version of *value | |
251 | * to insure consistency. | |
252 | */ | |
e26fa029 | 253 | v = LOAD_SHARED(*ctr); |
adcfce54 | 254 | return (v & RCU_GP_CTR_NEST_MASK) && |
02be5561 | 255 | ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE); |
adcfce54 MD |
256 | } |
257 | ||
258 | static inline void _rcu_read_lock(void) | |
259 | { | |
27d65bc5 | 260 | unsigned long tmp; |
adcfce54 | 261 | |
02be5561 MD |
262 | tmp = rcu_reader.ctr; |
263 | /* | |
264 | * rcu_gp_ctr is | |
265 | * RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE) | |
266 | */ | |
67ef1a2c | 267 | if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
02be5561 | 268 | _STORE_SHARED(rcu_reader.ctr, _LOAD_SHARED(rcu_gp_ctr)); |
67ef1a2c MD |
269 | /* |
270 | * Set active readers count for outermost nesting level before | |
25cc6d18 | 271 | * accessing the pointer. See smp_mb_master(). |
67ef1a2c | 272 | */ |
25cc6d18 | 273 | smp_mb_slave(RCU_MB_GROUP); |
67ef1a2c | 274 | } else { |
02be5561 | 275 | _STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT); |
67ef1a2c | 276 | } |
adcfce54 MD |
277 | } |
278 | ||
279 | static inline void _rcu_read_unlock(void) | |
280 | { | |
27d65bc5 | 281 | unsigned long tmp; |
bc6c15bb | 282 | |
02be5561 | 283 | tmp = rcu_reader.ctr; |
adcfce54 MD |
284 | /* |
285 | * Finish using rcu before decrementing the pointer. | |
25cc6d18 | 286 | * See smp_mb_master(). |
adcfce54 | 287 | */ |
bc6c15bb | 288 | if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) { |
25cc6d18 | 289 | smp_mb_slave(RCU_MB_GROUP); |
02be5561 MD |
290 | _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
291 | /* write rcu_reader.ctr before read futex */ | |
25cc6d18 | 292 | smp_mb_slave(RCU_MB_GROUP); |
bc6c15bb MD |
293 | wake_up_gp(); |
294 | } else { | |
02be5561 | 295 | _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
bc6c15bb | 296 | } |
adcfce54 MD |
297 | } |
298 | ||
36bc70a8 MD |
299 | #ifdef __cplusplus |
300 | } | |
301 | #endif | |
302 | ||
adcfce54 | 303 | #endif /* _URCU_STATIC_H */ |