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 | ||
fdee2e6d MD |
6 | #ifndef _URCU_BP_STATIC_H |
7 | #define _URCU_BP_STATIC_H | |
8 | ||
9 | /* | |
fdee2e6d MD |
10 | * Userspace RCU header. |
11 | * | |
a5a9f428 PM |
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. | |
fdee2e6d | 14 | * |
fdee2e6d 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> | |
fdee2e6d MD |
20 | #include <unistd.h> |
21 | ||
01477510 | 22 | #include <urcu/debug.h> |
d8d9a340 | 23 | #include <urcu/config.h> |
fdee2e6d MD |
24 | #include <urcu/compiler.h> |
25 | #include <urcu/arch.h> | |
26 | #include <urcu/system.h> | |
a2e7bf9c | 27 | #include <urcu/uatomic.h> |
fdee2e6d | 28 | #include <urcu/list.h> |
bd252a04 | 29 | #include <urcu/tls-compat.h> |
fdee2e6d MD |
30 | |
31 | /* | |
32 | * This code section can only be included in LGPL 2.1 compatible source code. | |
33 | * See below for the function call wrappers which can be used in code meant to | |
34 | * be only linked with the Userspace RCU library. This comes with a small | |
35 | * performance degradation on the read-side due to the added function calls. | |
36 | * This is required to permit relinking with newer versions of the library. | |
37 | */ | |
38 | ||
36bc70a8 MD |
39 | #ifdef __cplusplus |
40 | extern "C" { | |
41 | #endif | |
42 | ||
4477a870 MD |
43 | enum urcu_bp_state { |
44 | URCU_BP_READER_ACTIVE_CURRENT, | |
45 | URCU_BP_READER_ACTIVE_OLD, | |
46 | URCU_BP_READER_INACTIVE, | |
52c75091 MD |
47 | }; |
48 | ||
fdee2e6d | 49 | /* |
4477a870 | 50 | * The trick here is that URCU_BP_GP_CTR_PHASE must be a multiple of 8 so we can use a |
fdee2e6d MD |
51 | * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. |
52 | */ | |
4477a870 | 53 | #define URCU_BP_GP_COUNT (1UL << 0) |
fdee2e6d | 54 | /* Use the amount of bits equal to half of the architecture long size */ |
4477a870 MD |
55 | #define URCU_BP_GP_CTR_PHASE (1UL << (sizeof(long) << 2)) |
56 | #define URCU_BP_GP_CTR_NEST_MASK (URCU_BP_GP_CTR_PHASE - 1) | |
fdee2e6d MD |
57 | |
58 | /* | |
4477a870 | 59 | * Used internally by _urcu_bp_read_lock. |
fdee2e6d | 60 | */ |
4477a870 | 61 | extern void urcu_bp_register(void); |
fdee2e6d | 62 | |
4477a870 | 63 | struct urcu_bp_gp { |
c13c2e55 MD |
64 | /* |
65 | * Global grace period counter. | |
4477a870 MD |
66 | * Contains the current URCU_BP_GP_CTR_PHASE. |
67 | * Also has a URCU_BP_GP_COUNT of 1, to accelerate the reader fast path. | |
c13c2e55 MD |
68 | * Written to only by writer with mutex taken. |
69 | * Read by both writer and readers. | |
70 | */ | |
71 | unsigned long ctr; | |
72 | } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); | |
73 | ||
4477a870 | 74 | extern struct urcu_bp_gp urcu_bp_gp; |
fdee2e6d | 75 | |
4477a870 MD |
76 | struct urcu_bp_reader { |
77 | /* Data used by both reader and urcu_bp_synchronize_rcu() */ | |
52c75091 | 78 | unsigned long ctr; |
fdee2e6d | 79 | /* Data used for registry */ |
16aa9ee8 | 80 | struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
fdee2e6d MD |
81 | pthread_t tid; |
82 | int alloc; /* registry entry allocated */ | |
83 | }; | |
84 | ||
85 | /* | |
86 | * Bulletproof version keeps a pointer to a registry not part of the TLS. | |
87 | * Adds a pointer dereference on the read-side, but won't require to unregister | |
88 | * the reader thread. | |
89 | */ | |
4477a870 | 90 | extern DECLARE_URCU_TLS(struct urcu_bp_reader *, urcu_bp_reader); |
fdee2e6d | 91 | |
d8d9a340 MD |
92 | #ifdef CONFIG_RCU_FORCE_SYS_MEMBARRIER |
93 | #define urcu_bp_has_sys_membarrier 1 | |
94 | #else | |
f541831e | 95 | extern int urcu_bp_has_sys_membarrier; |
d8d9a340 | 96 | #endif |
f541831e MD |
97 | |
98 | static inline void urcu_bp_smp_mb_slave(void) | |
99 | { | |
100 | if (caa_likely(urcu_bp_has_sys_membarrier)) | |
101 | cmm_barrier(); | |
102 | else | |
103 | cmm_smp_mb(); | |
104 | } | |
105 | ||
4477a870 | 106 | static inline enum urcu_bp_state urcu_bp_reader_state(unsigned long *ctr) |
fdee2e6d | 107 | { |
52c75091 | 108 | unsigned long v; |
fdee2e6d | 109 | |
52c75091 | 110 | if (ctr == NULL) |
4477a870 | 111 | return URCU_BP_READER_INACTIVE; |
fdee2e6d MD |
112 | /* |
113 | * Make sure both tests below are done on the same version of *value | |
114 | * to insure consistency. | |
115 | */ | |
52c75091 | 116 | v = CMM_LOAD_SHARED(*ctr); |
4477a870 MD |
117 | if (!(v & URCU_BP_GP_CTR_NEST_MASK)) |
118 | return URCU_BP_READER_INACTIVE; | |
119 | if (!((v ^ urcu_bp_gp.ctr) & URCU_BP_GP_CTR_PHASE)) | |
120 | return URCU_BP_READER_ACTIVE_CURRENT; | |
121 | return URCU_BP_READER_ACTIVE_OLD; | |
fdee2e6d MD |
122 | } |
123 | ||
a5a9f428 | 124 | /* |
4477a870 | 125 | * Helper for _urcu_bp_read_lock(). The format of urcu_bp_gp.ctr (as well as |
be152bdb LKO |
126 | * the per-thread rcu_reader.ctr) has the lower-order bits containing a count of |
127 | * _urcu_bp_read_lock() nesting, and a single high-order URCU_BP_GP_CTR_PHASE bit | |
128 | * that contains either zero or one. The smp_mb_slave() ensures that the accesses in | |
4477a870 | 129 | * _urcu_bp_read_lock() happen before the subsequent read-side critical section. |
a5a9f428 | 130 | */ |
4477a870 | 131 | static inline void _urcu_bp_read_lock_update(unsigned long tmp) |
a5a9f428 | 132 | { |
4477a870 MD |
133 | if (caa_likely(!(tmp & URCU_BP_GP_CTR_NEST_MASK))) { |
134 | _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, _CMM_LOAD_SHARED(urcu_bp_gp.ctr)); | |
f541831e | 135 | urcu_bp_smp_mb_slave(); |
a5a9f428 | 136 | } else |
4477a870 | 137 | _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, tmp + URCU_BP_GP_COUNT); |
a5a9f428 PM |
138 | } |
139 | ||
140 | /* | |
141 | * Enter an RCU read-side critical section. | |
142 | * | |
143 | * The first cmm_barrier() call ensures that the compiler does not reorder | |
4477a870 | 144 | * the body of _urcu_bp_read_lock() with a mutex. |
a5a9f428 PM |
145 | * |
146 | * This function and its helper are both less than 10 lines long. The | |
147 | * intent is that this function meets the 10-line criterion in LGPL, | |
148 | * allowing this function to be invoked directly from non-LGPL code. | |
149 | */ | |
4477a870 | 150 | static inline void _urcu_bp_read_lock(void) |
fdee2e6d | 151 | { |
52c75091 | 152 | unsigned long tmp; |
fdee2e6d | 153 | |
4477a870 MD |
154 | if (caa_unlikely(!URCU_TLS(urcu_bp_reader))) |
155 | urcu_bp_register(); /* If not yet registered. */ | |
5481ddb3 | 156 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
4477a870 | 157 | tmp = URCU_TLS(urcu_bp_reader)->ctr; |
2a27e931 | 158 | urcu_assert_debug((tmp & URCU_BP_GP_CTR_NEST_MASK) != URCU_BP_GP_CTR_NEST_MASK); |
4477a870 | 159 | _urcu_bp_read_lock_update(tmp); |
fdee2e6d MD |
160 | } |
161 | ||
a5a9f428 PM |
162 | /* |
163 | * Exit an RCU read-side critical section. This function is less than | |
164 | * 10 lines of code, and is intended to be usable by non-LGPL code, as | |
165 | * called out in LGPL. | |
166 | */ | |
4477a870 | 167 | static inline void _urcu_bp_read_unlock(void) |
fdee2e6d | 168 | { |
343c8b13 MD |
169 | unsigned long tmp; |
170 | ||
4477a870 | 171 | tmp = URCU_TLS(urcu_bp_reader)->ctr; |
2a27e931 | 172 | urcu_assert_debug(tmp & URCU_BP_GP_CTR_NEST_MASK); |
343c8b13 | 173 | /* Finish using rcu before decrementing the pointer. */ |
f541831e | 174 | urcu_bp_smp_mb_slave(); |
4477a870 | 175 | _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, tmp - URCU_BP_GP_COUNT); |
5481ddb3 | 176 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
fdee2e6d MD |
177 | } |
178 | ||
882f3357 MD |
179 | /* |
180 | * Returns whether within a RCU read-side critical section. | |
181 | * | |
182 | * This function is less than 10 lines long. The intent is that this | |
183 | * function meets the 10-line criterion for LGPL, allowing this function | |
184 | * to be invoked directly from non-LGPL code. | |
185 | */ | |
4477a870 | 186 | static inline int _urcu_bp_read_ongoing(void) |
882f3357 | 187 | { |
4477a870 MD |
188 | if (caa_unlikely(!URCU_TLS(urcu_bp_reader))) |
189 | urcu_bp_register(); /* If not yet registered. */ | |
190 | return URCU_TLS(urcu_bp_reader)->ctr & URCU_BP_GP_CTR_NEST_MASK; | |
882f3357 MD |
191 | } |
192 | ||
67ecffc0 | 193 | #ifdef __cplusplus |
36bc70a8 MD |
194 | } |
195 | #endif | |
196 | ||
fdee2e6d | 197 | #endif /* _URCU_BP_STATIC_H */ |