| 1 | #ifndef _URCU_BP_STATIC_H |
| 2 | #define _URCU_BP_STATIC_H |
| 3 | |
| 4 | /* |
| 5 | * urcu-bp-static.h |
| 6 | * |
| 7 | * Userspace RCU header. |
| 8 | * |
| 9 | * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See urcu.h for linking |
| 10 | * dynamically with the userspace rcu library. |
| 11 | * |
| 12 | * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 13 | * Copyright (c) 2009 Paul E. McKenney, IBM Corporation. |
| 14 | * |
| 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 |
| 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> |
| 34 | #include <syscall.h> |
| 35 | #include <unistd.h> |
| 36 | |
| 37 | #include <urcu/compiler.h> |
| 38 | #include <urcu/arch.h> |
| 39 | #include <urcu/system.h> |
| 40 | #include <urcu/uatomic_arch.h> |
| 41 | #include <urcu/list.h> |
| 42 | |
| 43 | /* |
| 44 | * This code section can only be included in LGPL 2.1 compatible source code. |
| 45 | * See below for the function call wrappers which can be used in code meant to |
| 46 | * be only linked with the Userspace RCU library. This comes with a small |
| 47 | * performance degradation on the read-side due to the added function calls. |
| 48 | * This is required to permit relinking with newer versions of the library. |
| 49 | */ |
| 50 | |
| 51 | #ifdef __cplusplus |
| 52 | extern "C" { |
| 53 | #endif |
| 54 | |
| 55 | /* |
| 56 | * Active attempts to check for reader Q.S. before calling sleep(). |
| 57 | */ |
| 58 | #define RCU_QS_ACTIVE_ATTEMPTS 100 |
| 59 | |
| 60 | #ifdef DEBUG_RCU |
| 61 | #define rcu_assert(args...) assert(args) |
| 62 | #else |
| 63 | #define rcu_assert(args...) |
| 64 | #endif |
| 65 | |
| 66 | #ifdef DEBUG_YIELD |
| 67 | #include <sched.h> |
| 68 | #include <time.h> |
| 69 | #include <pthread.h> |
| 70 | #include <unistd.h> |
| 71 | |
| 72 | #define YIELD_READ (1 << 0) |
| 73 | #define YIELD_WRITE (1 << 1) |
| 74 | |
| 75 | /* |
| 76 | * Updates without RCU_MB are much slower. Account this in |
| 77 | * the delay. |
| 78 | */ |
| 79 | /* maximum sleep delay, in us */ |
| 80 | #define MAX_SLEEP 50 |
| 81 | |
| 82 | extern unsigned int yield_active; |
| 83 | extern unsigned int __thread rand_yield; |
| 84 | |
| 85 | static inline void debug_yield_read(void) |
| 86 | { |
| 87 | if (yield_active & YIELD_READ) |
| 88 | if (rand_r(&rand_yield) & 0x1) |
| 89 | usleep(rand_r(&rand_yield) % MAX_SLEEP); |
| 90 | } |
| 91 | |
| 92 | static inline void debug_yield_write(void) |
| 93 | { |
| 94 | if (yield_active & YIELD_WRITE) |
| 95 | if (rand_r(&rand_yield) & 0x1) |
| 96 | usleep(rand_r(&rand_yield) % MAX_SLEEP); |
| 97 | } |
| 98 | |
| 99 | static inline void debug_yield_init(void) |
| 100 | { |
| 101 | rand_yield = time(NULL) ^ pthread_self(); |
| 102 | } |
| 103 | #else |
| 104 | static inline void debug_yield_read(void) |
| 105 | { |
| 106 | } |
| 107 | |
| 108 | static inline void debug_yield_write(void) |
| 109 | { |
| 110 | } |
| 111 | |
| 112 | static inline void debug_yield_init(void) |
| 113 | { |
| 114 | |
| 115 | } |
| 116 | #endif |
| 117 | |
| 118 | /* |
| 119 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use a |
| 120 | * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. |
| 121 | */ |
| 122 | #define RCU_GP_COUNT (1UL << 0) |
| 123 | /* Use the amount of bits equal to half of the architecture long size */ |
| 124 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(long) << 2)) |
| 125 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) |
| 126 | |
| 127 | /* |
| 128 | * Used internally by _rcu_read_lock. |
| 129 | */ |
| 130 | extern void rcu_bp_register(void); |
| 131 | |
| 132 | /* |
| 133 | * Global quiescent period counter with low-order bits unused. |
| 134 | * Using a int rather than a char to eliminate false register dependencies |
| 135 | * causing stalls on some architectures. |
| 136 | */ |
| 137 | extern long rcu_gp_ctr; |
| 138 | |
| 139 | struct rcu_reader { |
| 140 | /* Data used by both reader and synchronize_rcu() */ |
| 141 | long ctr; |
| 142 | /* Data used for registry */ |
| 143 | struct list_head node __attribute__((aligned(CACHE_LINE_SIZE))); |
| 144 | pthread_t tid; |
| 145 | int alloc; /* registry entry allocated */ |
| 146 | }; |
| 147 | |
| 148 | /* |
| 149 | * Bulletproof version keeps a pointer to a registry not part of the TLS. |
| 150 | * Adds a pointer dereference on the read-side, but won't require to unregister |
| 151 | * the reader thread. |
| 152 | */ |
| 153 | extern struct rcu_reader __thread *rcu_reader; |
| 154 | |
| 155 | static inline int rcu_old_gp_ongoing(long *value) |
| 156 | { |
| 157 | long v; |
| 158 | |
| 159 | if (value == NULL) |
| 160 | return 0; |
| 161 | /* |
| 162 | * Make sure both tests below are done on the same version of *value |
| 163 | * to insure consistency. |
| 164 | */ |
| 165 | v = LOAD_SHARED(*value); |
| 166 | return (v & RCU_GP_CTR_NEST_MASK) && |
| 167 | ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE); |
| 168 | } |
| 169 | |
| 170 | static inline void _rcu_read_lock(void) |
| 171 | { |
| 172 | long tmp; |
| 173 | |
| 174 | /* Check if registered */ |
| 175 | if (unlikely(!rcu_reader)) |
| 176 | rcu_bp_register(); |
| 177 | |
| 178 | tmp = rcu_reader->ctr; |
| 179 | /* |
| 180 | * rcu_gp_ctr is |
| 181 | * RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE) |
| 182 | */ |
| 183 | if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
| 184 | _STORE_SHARED(rcu_reader->ctr, _LOAD_SHARED(rcu_gp_ctr)); |
| 185 | /* |
| 186 | * Set active readers count for outermost nesting level before |
| 187 | * accessing the pointer. |
| 188 | */ |
| 189 | smp_mb(); |
| 190 | } else { |
| 191 | _STORE_SHARED(rcu_reader->ctr, tmp + RCU_GP_COUNT); |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | static inline void _rcu_read_unlock(void) |
| 196 | { |
| 197 | /* |
| 198 | * Finish using rcu before decrementing the pointer. |
| 199 | */ |
| 200 | smp_mb(); |
| 201 | _STORE_SHARED(rcu_reader->ctr, rcu_reader->ctr - RCU_GP_COUNT); |
| 202 | } |
| 203 | |
| 204 | #ifdef __cplusplus |
| 205 | } |
| 206 | #endif |
| 207 | |
| 208 | #endif /* _URCU_BP_STATIC_H */ |