| 1 | #ifndef _URCU_STATIC_H |
| 2 | #define _URCU_STATIC_H |
| 3 | |
| 4 | /* |
| 5 | * urcu-static.h |
| 6 | * |
| 7 | * Userspace RCU header. |
| 8 | * |
| 9 | * TO BE INCLUDED ONLY IN CODE THAT IS TO BE RECOMPILED ON EACH LIBURCU |
| 10 | * RELEASE. See urcu.h for linking 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 <unistd.h> |
| 35 | #include <stdint.h> |
| 36 | |
| 37 | #include <urcu/config.h> |
| 38 | #include <urcu/compiler.h> |
| 39 | #include <urcu/arch.h> |
| 40 | #include <urcu/system.h> |
| 41 | #include <urcu/uatomic.h> |
| 42 | #include <urcu/list.h> |
| 43 | #include <urcu/futex.h> |
| 44 | #include <urcu/tls-compat.h> |
| 45 | #include <urcu/debug.h> |
| 46 | |
| 47 | #ifdef __cplusplus |
| 48 | extern "C" { |
| 49 | #endif |
| 50 | |
| 51 | /* Default is RCU_MEMBARRIER */ |
| 52 | #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL) |
| 53 | #define RCU_MEMBARRIER |
| 54 | #endif |
| 55 | |
| 56 | /* |
| 57 | * This code section can only be included in LGPL 2.1 compatible source code. |
| 58 | * See below for the function call wrappers which can be used in code meant to |
| 59 | * be only linked with the Userspace RCU library. This comes with a small |
| 60 | * performance degradation on the read-side due to the added function calls. |
| 61 | * This is required to permit relinking with newer versions of the library. |
| 62 | */ |
| 63 | |
| 64 | /* |
| 65 | * The signal number used by the RCU library can be overridden with |
| 66 | * -DSIGRCU= when compiling the library. |
| 67 | * Provide backward compatibility for liburcu 0.3.x SIGURCU. |
| 68 | */ |
| 69 | #ifdef SIGURCU |
| 70 | #define SIGRCU SIGURCU |
| 71 | #endif |
| 72 | |
| 73 | #ifndef SIGRCU |
| 74 | #define SIGRCU SIGUSR1 |
| 75 | #endif |
| 76 | |
| 77 | enum rcu_state { |
| 78 | RCU_READER_ACTIVE_CURRENT, |
| 79 | RCU_READER_ACTIVE_OLD, |
| 80 | RCU_READER_INACTIVE, |
| 81 | }; |
| 82 | |
| 83 | /* |
| 84 | * Slave barriers are only guaranteed to be ordered wrt master barriers. |
| 85 | * |
| 86 | * The pair ordering is detailed as (O: ordered, X: not ordered) : |
| 87 | * slave master |
| 88 | * slave X O |
| 89 | * master O O |
| 90 | */ |
| 91 | |
| 92 | #ifdef RCU_MEMBARRIER |
| 93 | #ifdef CONFIG_RCU_FORCE_SYS_MEMBARRIER |
| 94 | #define rcu_has_sys_membarrier_memb 1 |
| 95 | #else |
| 96 | extern int rcu_has_sys_membarrier_memb; |
| 97 | #endif |
| 98 | |
| 99 | static inline void smp_mb_slave(void) |
| 100 | { |
| 101 | if (caa_likely(rcu_has_sys_membarrier_memb)) |
| 102 | cmm_barrier(); |
| 103 | else |
| 104 | cmm_smp_mb(); |
| 105 | } |
| 106 | #endif |
| 107 | |
| 108 | #ifdef RCU_MB |
| 109 | static inline void smp_mb_slave(void) |
| 110 | { |
| 111 | cmm_smp_mb(); |
| 112 | } |
| 113 | #endif |
| 114 | |
| 115 | #ifdef RCU_SIGNAL |
| 116 | static inline void smp_mb_slave(void) |
| 117 | { |
| 118 | cmm_barrier(); |
| 119 | } |
| 120 | #endif |
| 121 | |
| 122 | /* |
| 123 | * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use |
| 124 | * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits. |
| 125 | */ |
| 126 | #define RCU_GP_COUNT (1UL << 0) |
| 127 | /* Use the amount of bits equal to half of the architecture long size */ |
| 128 | #define RCU_GP_CTR_PHASE (1UL << (sizeof(unsigned long) << 2)) |
| 129 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_PHASE - 1) |
| 130 | |
| 131 | struct rcu_gp { |
| 132 | /* |
| 133 | * Global grace period counter. |
| 134 | * Contains the current RCU_GP_CTR_PHASE. |
| 135 | * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path. |
| 136 | * Written to only by writer with mutex taken. |
| 137 | * Read by both writer and readers. |
| 138 | */ |
| 139 | unsigned long ctr; |
| 140 | |
| 141 | int32_t futex; |
| 142 | } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
| 143 | |
| 144 | extern struct rcu_gp rcu_gp; |
| 145 | |
| 146 | struct rcu_reader { |
| 147 | /* Data used by both reader and synchronize_rcu() */ |
| 148 | unsigned long ctr; |
| 149 | char need_mb; |
| 150 | /* Data used for registry */ |
| 151 | struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
| 152 | pthread_t tid; |
| 153 | /* Reader registered flag, for internal checks. */ |
| 154 | unsigned int registered:1; |
| 155 | }; |
| 156 | |
| 157 | extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader); |
| 158 | |
| 159 | /* |
| 160 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. |
| 161 | */ |
| 162 | static inline void wake_up_gp(void) |
| 163 | { |
| 164 | if (caa_unlikely(uatomic_read(&rcu_gp.futex) == -1)) { |
| 165 | uatomic_set(&rcu_gp.futex, 0); |
| 166 | /* |
| 167 | * Ignoring return value until we can make this function |
| 168 | * return something (because urcu_die() is not publicly |
| 169 | * exposed). |
| 170 | */ |
| 171 | (void) futex_async(&rcu_gp.futex, FUTEX_WAKE, 1, |
| 172 | NULL, NULL, 0); |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | static inline enum rcu_state rcu_reader_state(unsigned long *ctr) |
| 177 | { |
| 178 | unsigned long v; |
| 179 | |
| 180 | /* |
| 181 | * Make sure both tests below are done on the same version of *value |
| 182 | * to insure consistency. |
| 183 | */ |
| 184 | v = CMM_LOAD_SHARED(*ctr); |
| 185 | if (!(v & RCU_GP_CTR_NEST_MASK)) |
| 186 | return RCU_READER_INACTIVE; |
| 187 | if (!((v ^ rcu_gp.ctr) & RCU_GP_CTR_PHASE)) |
| 188 | return RCU_READER_ACTIVE_CURRENT; |
| 189 | return RCU_READER_ACTIVE_OLD; |
| 190 | } |
| 191 | |
| 192 | /* |
| 193 | * Helper for _rcu_read_lock(). The format of rcu_gp.ctr (as well as |
| 194 | * the per-thread rcu_reader.ctr) has the upper bits containing a count of |
| 195 | * _rcu_read_lock() nesting, and a lower-order bit that contains either zero |
| 196 | * or RCU_GP_CTR_PHASE. The smp_mb_slave() ensures that the accesses in |
| 197 | * _rcu_read_lock() happen before the subsequent read-side critical section. |
| 198 | */ |
| 199 | static inline void _rcu_read_lock_update(unsigned long tmp) |
| 200 | { |
| 201 | if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) { |
| 202 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp.ctr)); |
| 203 | smp_mb_slave(); |
| 204 | } else |
| 205 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp + RCU_GP_COUNT); |
| 206 | } |
| 207 | |
| 208 | /* |
| 209 | * Enter an RCU read-side critical section. |
| 210 | * |
| 211 | * The first cmm_barrier() call ensures that the compiler does not reorder |
| 212 | * the body of _rcu_read_lock() with a mutex. |
| 213 | * |
| 214 | * This function and its helper are both less than 10 lines long. The |
| 215 | * intent is that this function meets the 10-line criterion in LGPL, |
| 216 | * allowing this function to be invoked directly from non-LGPL code. |
| 217 | */ |
| 218 | static inline void _rcu_read_lock(void) |
| 219 | { |
| 220 | unsigned long tmp; |
| 221 | |
| 222 | urcu_assert(URCU_TLS(rcu_reader).registered); |
| 223 | cmm_barrier(); |
| 224 | tmp = URCU_TLS(rcu_reader).ctr; |
| 225 | urcu_assert((tmp & RCU_GP_CTR_NEST_MASK) != RCU_GP_CTR_NEST_MASK); |
| 226 | _rcu_read_lock_update(tmp); |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * This is a helper function for _rcu_read_unlock(). |
| 231 | * |
| 232 | * The first smp_mb_slave() call ensures that the critical section is |
| 233 | * seen to precede the store to rcu_reader.ctr. |
| 234 | * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr |
| 235 | * before reading the update-side futex. |
| 236 | */ |
| 237 | static inline void _rcu_read_unlock_update_and_wakeup(unsigned long tmp) |
| 238 | { |
| 239 | if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) { |
| 240 | smp_mb_slave(); |
| 241 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT); |
| 242 | smp_mb_slave(); |
| 243 | wake_up_gp(); |
| 244 | } else |
| 245 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT); |
| 246 | } |
| 247 | |
| 248 | /* |
| 249 | * Exit an RCU read-side crtical section. Both this function and its |
| 250 | * helper are smaller than 10 lines of code, and are intended to be |
| 251 | * usable by non-LGPL code, as called out in LGPL. |
| 252 | */ |
| 253 | static inline void _rcu_read_unlock(void) |
| 254 | { |
| 255 | unsigned long tmp; |
| 256 | |
| 257 | urcu_assert(URCU_TLS(rcu_reader).registered); |
| 258 | tmp = URCU_TLS(rcu_reader).ctr; |
| 259 | urcu_assert(tmp & RCU_GP_CTR_NEST_MASK); |
| 260 | _rcu_read_unlock_update_and_wakeup(tmp); |
| 261 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
| 262 | } |
| 263 | |
| 264 | /* |
| 265 | * Returns whether within a RCU read-side critical section. |
| 266 | * |
| 267 | * This function is less than 10 lines long. The intent is that this |
| 268 | * function meets the 10-line criterion for LGPL, allowing this function |
| 269 | * to be invoked directly from non-LGPL code. |
| 270 | */ |
| 271 | static inline int _rcu_read_ongoing(void) |
| 272 | { |
| 273 | return URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK; |
| 274 | } |
| 275 | |
| 276 | #ifdef __cplusplus |
| 277 | } |
| 278 | #endif |
| 279 | |
| 280 | #endif /* _URCU_STATIC_H */ |