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 | |
f0708810 MD |
59 | # define MEMBARRIER_EXPEDITED (1 << 0) |
60 | # define MEMBARRIER_DELAYED (1 << 1) | |
61 | # define membarrier(...) syscall(__NR_membarrier, __VA_ARGS__) | |
fdf01eed | 62 | #else |
f0708810 MD |
63 | # undef RCU_MEMBARRIER |
64 | # define RCU_MB | |
fdf01eed MD |
65 | #endif |
66 | #endif | |
67 | ||
adcfce54 MD |
68 | /* |
69 | * This code section can only be included in LGPL 2.1 compatible source code. | |
70 | * See below for the function call wrappers which can be used in code meant to | |
71 | * be only linked with the Userspace RCU library. This comes with a small | |
72 | * performance degradation on the read-side due to the added function calls. | |
73 | * This is required to permit relinking with newer versions of the library. | |
74 | */ | |
75 | ||
76 | /* | |
77 | * The signal number used by the RCU library can be overridden with | |
02be5561 | 78 | * -DSIGRCU= when compiling the library. |
adcfce54 | 79 | */ |
02be5561 MD |
80 | #ifndef SIGRCU |
81 | #define SIGRCU SIGUSR1 | |
adcfce54 MD |
82 | #endif |
83 | ||
84 | /* | |
85 | * If a reader is really non-cooperative and refuses to commit its | |
02be5561 | 86 | * rcu_active_readers count to memory (there is no barrier in the reader |
adcfce54 MD |
87 | * per-se), kick it after a few loops waiting for it. |
88 | */ | |
89 | #define KICK_READER_LOOPS 10000 | |
90 | ||
bc6c15bb MD |
91 | /* |
92 | * Active attempts to check for reader Q.S. before calling futex(). | |
93 | */ | |
94 | #define RCU_QS_ACTIVE_ATTEMPTS 100 | |
95 | ||
7ac06cef MD |
96 | #ifdef DEBUG_RCU |
97 | #define rcu_assert(args...) assert(args) | |
98 | #else | |
99 | #define rcu_assert(args...) | |
100 | #endif | |
101 | ||
adcfce54 MD |
102 | #ifdef DEBUG_YIELD |
103 | #include <sched.h> | |
104 | #include <time.h> | |
105 | #include <pthread.h> | |
106 | #include <unistd.h> | |
107 | ||
108 | #define YIELD_READ (1 << 0) | |
109 | #define YIELD_WRITE (1 << 1) | |
110 | ||
b4ce1526 | 111 | /* |
fdf01eed | 112 | * Updates with RCU_SIGNAL are much slower. Account this in the delay. |
b4ce1526 | 113 | */ |
fdf01eed | 114 | #ifdef RCU_SIGNAL |
adcfce54 | 115 | /* maximum sleep delay, in us */ |
adcfce54 | 116 | #define MAX_SLEEP 30000 |
fdf01eed MD |
117 | #else |
118 | #define MAX_SLEEP 50 | |
adcfce54 MD |
119 | #endif |
120 | ||
121 | extern unsigned int yield_active; | |
122 | extern unsigned int __thread rand_yield; | |
123 | ||
124 | static inline void debug_yield_read(void) | |
125 | { | |
126 | if (yield_active & YIELD_READ) | |
127 | if (rand_r(&rand_yield) & 0x1) | |
128 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
129 | } | |
130 | ||
131 | static inline void debug_yield_write(void) | |
132 | { | |
133 | if (yield_active & YIELD_WRITE) | |
134 | if (rand_r(&rand_yield) & 0x1) | |
135 | usleep(rand_r(&rand_yield) % MAX_SLEEP); | |
136 | } | |
137 | ||
138 | static inline void debug_yield_init(void) | |
139 | { | |
140 | rand_yield = time(NULL) ^ pthread_self(); | |
141 | } | |
142 | #else | |
143 | static inline void debug_yield_read(void) | |
144 | { | |
145 | } | |
146 | ||
147 | static inline void debug_yield_write(void) | |
148 | { | |
149 | } | |
150 | ||
151 | static inline void debug_yield_init(void) | |
152 | { | |
153 | ||
154 | } | |
155 | #endif | |
156 | ||
fdf01eed MD |
157 | #ifdef RCU_MEMBARRIER |
158 | extern int has_sys_membarrier; | |
159 | ||
160 | static inline void smp_mb_light() | |
161 | { | |
162 | if (likely(has_sys_membarrier)) | |
163 | barrier(); | |
164 | else | |
165 | smp_mb(); | |
166 | } | |
167 | #endif | |
168 | ||
02be5561 | 169 | #ifdef RCU_MB |
9d7e3f89 | 170 | static inline void smp_mb_light() |
adcfce54 MD |
171 | { |
172 | smp_mb(); | |
173 | } | |
fdf01eed MD |
174 | #endif |
175 | ||
176 | #ifdef RCU_SIGNAL | |
9d7e3f89 | 177 | static inline void smp_mb_light() |
adcfce54 MD |
178 | { |
179 | barrier(); | |
180 | } | |
181 | #endif | |
182 | ||
183 | /* | |
02be5561 MD |
184 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use |
185 | * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. | |
adcfce54 MD |
186 | */ |
187 | #define RCU_GP_COUNT (1UL << 0) | |
188 | /* Use the amount of bits equal to half of the architecture long size */ | |
02be5561 MD |
189 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(long) << 2)) |
190 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) | |
adcfce54 MD |
191 | |
192 | /* | |
193 | * Global quiescent period counter with low-order bits unused. | |
194 | * Using a int rather than a char to eliminate false register dependencies | |
195 | * causing stalls on some architectures. | |
196 | */ | |
02be5561 | 197 | extern long rcu_gp_ctr; |
adcfce54 | 198 | |
02be5561 | 199 | struct rcu_reader { |
bd1a5e15 | 200 | /* Data used by both reader and synchronize_rcu() */ |
e3b0cef0 | 201 | long ctr; |
e3b0cef0 | 202 | char need_mb; |
bd1a5e15 MD |
203 | /* Data used for registry */ |
204 | struct list_head head __attribute__((aligned(CACHE_LINE_SIZE))); | |
205 | pthread_t tid; | |
e3b0cef0 MD |
206 | }; |
207 | ||
02be5561 | 208 | extern struct rcu_reader __thread rcu_reader; |
adcfce54 | 209 | |
bc6c15bb MD |
210 | extern int gp_futex; |
211 | ||
212 | /* | |
213 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. | |
214 | */ | |
215 | static inline void wake_up_gp(void) | |
216 | { | |
ec4e58a3 MD |
217 | if (unlikely(uatomic_read(&gp_futex) == -1)) { |
218 | uatomic_set(&gp_futex, 0); | |
0854ccff | 219 | futex_async(&gp_futex, FUTEX_WAKE, 1, |
bc6c15bb MD |
220 | NULL, NULL, 0); |
221 | } | |
222 | } | |
223 | ||
adcfce54 MD |
224 | static inline int rcu_old_gp_ongoing(long *value) |
225 | { | |
226 | long v; | |
227 | ||
228 | if (value == NULL) | |
229 | return 0; | |
230 | /* | |
231 | * Make sure both tests below are done on the same version of *value | |
232 | * to insure consistency. | |
233 | */ | |
234 | v = LOAD_SHARED(*value); | |
235 | return (v & RCU_GP_CTR_NEST_MASK) && | |
02be5561 | 236 | ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE); |
adcfce54 MD |
237 | } |
238 | ||
239 | static inline void _rcu_read_lock(void) | |
240 | { | |
241 | long tmp; | |
242 | ||
02be5561 MD |
243 | tmp = rcu_reader.ctr; |
244 | /* | |
245 | * rcu_gp_ctr is | |
246 | * RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE) | |
247 | */ | |
67ef1a2c | 248 | if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
02be5561 | 249 | _STORE_SHARED(rcu_reader.ctr, _LOAD_SHARED(rcu_gp_ctr)); |
67ef1a2c MD |
250 | /* |
251 | * Set active readers count for outermost nesting level before | |
9d7e3f89 | 252 | * accessing the pointer. See smp_mb_heavy(). |
67ef1a2c | 253 | */ |
9d7e3f89 | 254 | smp_mb_light(); |
67ef1a2c | 255 | } else { |
02be5561 | 256 | _STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT); |
67ef1a2c | 257 | } |
adcfce54 MD |
258 | } |
259 | ||
260 | static inline void _rcu_read_unlock(void) | |
261 | { | |
bc6c15bb MD |
262 | long tmp; |
263 | ||
02be5561 | 264 | tmp = rcu_reader.ctr; |
adcfce54 MD |
265 | /* |
266 | * Finish using rcu before decrementing the pointer. | |
9d7e3f89 | 267 | * See smp_mb_heavy(). |
adcfce54 | 268 | */ |
bc6c15bb | 269 | if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) { |
9d7e3f89 | 270 | smp_mb_light(); |
02be5561 MD |
271 | _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
272 | /* write rcu_reader.ctr before read futex */ | |
9d7e3f89 | 273 | smp_mb_light(); |
bc6c15bb MD |
274 | wake_up_gp(); |
275 | } else { | |
02be5561 | 276 | _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT); |
bc6c15bb | 277 | } |
adcfce54 MD |
278 | } |
279 | ||
36bc70a8 MD |
280 | #ifdef __cplusplus |
281 | } | |
282 | #endif | |
283 | ||
adcfce54 | 284 | #endif /* _URCU_STATIC_H */ |