Commit | Line | Data |
---|---|---|
d3d3857f MJ |
1 | // SPDX-FileCopyrightText: 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
2 | // SPDX-FileCopyrightText: 2009 Paul E. McKenney, IBM Corporation. | |
3 | // | |
4 | // SPDX-License-Identifier: LGPL-2.1-or-later | |
5 | ||
4477a870 MD |
6 | #ifndef _URCU_MEMB_STATIC_H |
7 | #define _URCU_MEMB_STATIC_H | |
8 | ||
9 | /* | |
4477a870 MD |
10 | * Userspace RCU header. |
11 | * | |
12 | * TO BE INCLUDED ONLY IN CODE THAT IS TO BE RECOMPILED ON EACH LIBURCU | |
13 | * RELEASE. See urcu.h for linking dynamically with the userspace rcu library. | |
14 | * | |
4477a870 MD |
15 | * IBM's contributions to this file may be relicensed under LGPLv2 or later. |
16 | */ | |
17 | ||
18 | #include <stdlib.h> | |
19 | #include <pthread.h> | |
20 | #include <unistd.h> | |
21 | #include <stdint.h> | |
22 | ||
01477510 | 23 | #include <urcu/debug.h> |
4477a870 MD |
24 | #include <urcu/config.h> |
25 | #include <urcu/compiler.h> | |
26 | #include <urcu/arch.h> | |
27 | #include <urcu/system.h> | |
28 | #include <urcu/uatomic.h> | |
29 | #include <urcu/list.h> | |
30 | #include <urcu/futex.h> | |
31 | #include <urcu/tls-compat.h> | |
4477a870 MD |
32 | #include <urcu/static/urcu-common.h> |
33 | ||
34 | #ifdef __cplusplus | |
35 | extern "C" { | |
36 | #endif | |
37 | ||
38 | /* | |
39 | * This code section can only be included in LGPL 2.1 compatible source code. | |
40 | * See below for the function call wrappers which can be used in code meant to | |
41 | * be only linked with the Userspace RCU library. This comes with a small | |
42 | * performance degradation on the read-side due to the added function calls. | |
43 | * This is required to permit relinking with newer versions of the library. | |
44 | */ | |
45 | ||
46 | /* | |
47 | * Slave barriers are only guaranteed to be ordered wrt master barriers. | |
48 | * | |
49 | * The pair ordering is detailed as (O: ordered, X: not ordered) : | |
50 | * slave master | |
51 | * slave X O | |
52 | * master O O | |
53 | */ | |
54 | ||
55 | #ifdef CONFIG_RCU_FORCE_SYS_MEMBARRIER | |
56 | #define urcu_memb_has_sys_membarrier 1 | |
57 | #else | |
58 | extern int urcu_memb_has_sys_membarrier; | |
59 | #endif | |
60 | ||
61 | static inline void urcu_memb_smp_mb_slave(void) | |
62 | { | |
63 | if (caa_likely(urcu_memb_has_sys_membarrier)) | |
64 | cmm_barrier(); | |
65 | else | |
66 | cmm_smp_mb(); | |
67 | } | |
68 | ||
69 | extern struct urcu_gp urcu_memb_gp; | |
70 | ||
71 | extern DECLARE_URCU_TLS(struct urcu_reader, urcu_memb_reader); | |
72 | ||
73 | /* | |
74 | * Helper for _rcu_read_lock(). The format of urcu_memb_gp.ctr (as well as | |
be152bdb LKO |
75 | * the per-thread rcu_reader.ctr) has the lower-order bits containing a count of |
76 | * _rcu_read_lock() nesting, and a single high-order URCU_BP_GP_CTR_PHASE bit | |
77 | * that contains either zero or one. The smp_mb_slave() ensures that the accesses in | |
4477a870 MD |
78 | * _rcu_read_lock() happen before the subsequent read-side critical section. |
79 | */ | |
80 | static inline void _urcu_memb_read_lock_update(unsigned long tmp) | |
81 | { | |
82 | if (caa_likely(!(tmp & URCU_GP_CTR_NEST_MASK))) { | |
83 | _CMM_STORE_SHARED(URCU_TLS(urcu_memb_reader).ctr, _CMM_LOAD_SHARED(urcu_memb_gp.ctr)); | |
84 | urcu_memb_smp_mb_slave(); | |
85 | } else | |
86 | _CMM_STORE_SHARED(URCU_TLS(urcu_memb_reader).ctr, tmp + URCU_GP_COUNT); | |
87 | } | |
88 | ||
89 | /* | |
90 | * Enter an RCU read-side critical section. | |
91 | * | |
92 | * The first cmm_barrier() call ensures that the compiler does not reorder | |
93 | * the body of _rcu_read_lock() with a mutex. | |
94 | * | |
95 | * This function and its helper are both less than 10 lines long. The | |
96 | * intent is that this function meets the 10-line criterion in LGPL, | |
97 | * allowing this function to be invoked directly from non-LGPL code. | |
98 | */ | |
99 | static inline void _urcu_memb_read_lock(void) | |
100 | { | |
101 | unsigned long tmp; | |
102 | ||
2a27e931 | 103 | urcu_assert_debug(URCU_TLS(urcu_memb_reader).registered); |
4477a870 MD |
104 | cmm_barrier(); |
105 | tmp = URCU_TLS(urcu_memb_reader).ctr; | |
2a27e931 | 106 | urcu_assert_debug((tmp & URCU_GP_CTR_NEST_MASK) != URCU_GP_CTR_NEST_MASK); |
4477a870 MD |
107 | _urcu_memb_read_lock_update(tmp); |
108 | } | |
109 | ||
110 | /* | |
111 | * This is a helper function for _rcu_read_unlock(). | |
112 | * | |
113 | * The first smp_mb_slave() call ensures that the critical section is | |
114 | * seen to precede the store to rcu_reader.ctr. | |
115 | * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr | |
116 | * before reading the update-side futex. | |
117 | */ | |
118 | static inline void _urcu_memb_read_unlock_update_and_wakeup(unsigned long tmp) | |
119 | { | |
120 | if (caa_likely((tmp & URCU_GP_CTR_NEST_MASK) == URCU_GP_COUNT)) { | |
121 | urcu_memb_smp_mb_slave(); | |
122 | _CMM_STORE_SHARED(URCU_TLS(urcu_memb_reader).ctr, tmp - URCU_GP_COUNT); | |
123 | urcu_memb_smp_mb_slave(); | |
124 | urcu_common_wake_up_gp(&urcu_memb_gp); | |
125 | } else | |
126 | _CMM_STORE_SHARED(URCU_TLS(urcu_memb_reader).ctr, tmp - URCU_GP_COUNT); | |
127 | } | |
128 | ||
129 | /* | |
f99c6e92 | 130 | * Exit an RCU read-side critical section. Both this function and its |
4477a870 MD |
131 | * helper are smaller than 10 lines of code, and are intended to be |
132 | * usable by non-LGPL code, as called out in LGPL. | |
133 | */ | |
134 | static inline void _urcu_memb_read_unlock(void) | |
135 | { | |
136 | unsigned long tmp; | |
137 | ||
2a27e931 | 138 | urcu_assert_debug(URCU_TLS(urcu_memb_reader).registered); |
4477a870 | 139 | tmp = URCU_TLS(urcu_memb_reader).ctr; |
2a27e931 | 140 | urcu_assert_debug(tmp & URCU_GP_CTR_NEST_MASK); |
4477a870 MD |
141 | _urcu_memb_read_unlock_update_and_wakeup(tmp); |
142 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ | |
143 | } | |
144 | ||
145 | /* | |
146 | * Returns whether within a RCU read-side critical section. | |
147 | * | |
148 | * This function is less than 10 lines long. The intent is that this | |
149 | * function meets the 10-line criterion for LGPL, allowing this function | |
150 | * to be invoked directly from non-LGPL code. | |
151 | */ | |
152 | static inline int _urcu_memb_read_ongoing(void) | |
153 | { | |
154 | return URCU_TLS(urcu_memb_reader).ctr & URCU_GP_CTR_NEST_MASK; | |
155 | } | |
156 | ||
157 | #ifdef __cplusplus | |
158 | } | |
159 | #endif | |
160 | ||
161 | #endif /* _URCU_MEMB_STATIC_H */ |