| 1 | #ifndef _URCU_QSBR_STATIC_H |
| 2 | #define _URCU_QSBR_STATIC_H |
| 3 | |
| 4 | /* |
| 5 | * urcu-qsbr-static.h |
| 6 | * |
| 7 | * Userspace RCU QSBR 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 <limits.h> |
| 35 | #include <unistd.h> |
| 36 | #include <stdint.h> |
| 37 | |
| 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 | /* |
| 52 | * This code section can only be included in LGPL 2.1 compatible source code. |
| 53 | * See below for the function call wrappers which can be used in code meant to |
| 54 | * be only linked with the Userspace RCU library. This comes with a small |
| 55 | * performance degradation on the read-side due to the added function calls. |
| 56 | * This is required to permit relinking with newer versions of the library. |
| 57 | */ |
| 58 | |
| 59 | enum rcu_state { |
| 60 | RCU_READER_ACTIVE_CURRENT, |
| 61 | RCU_READER_ACTIVE_OLD, |
| 62 | RCU_READER_INACTIVE, |
| 63 | }; |
| 64 | |
| 65 | #define RCU_GP_ONLINE (1UL << 0) |
| 66 | #define RCU_GP_CTR (1UL << 1) |
| 67 | |
| 68 | struct rcu_gp { |
| 69 | /* |
| 70 | * Global quiescent period counter with low-order bits unused. |
| 71 | * Using a int rather than a char to eliminate false register |
| 72 | * dependencies causing stalls on some architectures. |
| 73 | */ |
| 74 | unsigned long ctr; |
| 75 | |
| 76 | int32_t futex; |
| 77 | } __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
| 78 | |
| 79 | extern struct rcu_gp rcu_gp; |
| 80 | |
| 81 | struct rcu_reader { |
| 82 | /* Data used by both reader and synchronize_rcu() */ |
| 83 | unsigned long ctr; |
| 84 | /* Data used for registry */ |
| 85 | struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE))); |
| 86 | int waiting; |
| 87 | pthread_t tid; |
| 88 | /* Reader registered flag, for internal checks. */ |
| 89 | unsigned int registered:1; |
| 90 | }; |
| 91 | |
| 92 | extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader); |
| 93 | |
| 94 | /* |
| 95 | * Wake-up waiting synchronize_rcu(). Called from many concurrent threads. |
| 96 | */ |
| 97 | static inline void wake_up_gp(void) |
| 98 | { |
| 99 | if (caa_unlikely(_CMM_LOAD_SHARED(URCU_TLS(rcu_reader).waiting))) { |
| 100 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).waiting, 0); |
| 101 | cmm_smp_mb(); |
| 102 | if (uatomic_read(&rcu_gp.futex) != -1) |
| 103 | return; |
| 104 | uatomic_set(&rcu_gp.futex, 0); |
| 105 | /* |
| 106 | * Ignoring return value until we can make this function |
| 107 | * return something (because urcu_die() is not publicly |
| 108 | * exposed). |
| 109 | */ |
| 110 | (void) futex_noasync(&rcu_gp.futex, FUTEX_WAKE, 1, |
| 111 | NULL, NULL, 0); |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | static inline enum rcu_state rcu_reader_state(unsigned long *ctr) |
| 116 | { |
| 117 | unsigned long v; |
| 118 | |
| 119 | v = CMM_LOAD_SHARED(*ctr); |
| 120 | if (!v) |
| 121 | return RCU_READER_INACTIVE; |
| 122 | if (v == rcu_gp.ctr) |
| 123 | return RCU_READER_ACTIVE_CURRENT; |
| 124 | return RCU_READER_ACTIVE_OLD; |
| 125 | } |
| 126 | |
| 127 | /* |
| 128 | * Enter an RCU read-side critical section. |
| 129 | * |
| 130 | * This function is less than 10 lines long. The intent is that this |
| 131 | * function meets the 10-line criterion for LGPL, allowing this function |
| 132 | * to be invoked directly from non-LGPL code. |
| 133 | */ |
| 134 | static inline void _rcu_read_lock(void) |
| 135 | { |
| 136 | urcu_assert(URCU_TLS(rcu_reader).ctr); |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * Exit an RCU read-side critical section. |
| 141 | * |
| 142 | * This function is less than 10 lines long. The intent is that this |
| 143 | * function meets the 10-line criterion for LGPL, allowing this function |
| 144 | * to be invoked directly from non-LGPL code. |
| 145 | */ |
| 146 | static inline void _rcu_read_unlock(void) |
| 147 | { |
| 148 | urcu_assert(URCU_TLS(rcu_reader).ctr); |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * Returns whether within a RCU read-side critical section. |
| 153 | * |
| 154 | * This function is less than 10 lines long. The intent is that this |
| 155 | * function meets the 10-line criterion for LGPL, allowing this function |
| 156 | * to be invoked directly from non-LGPL code. |
| 157 | */ |
| 158 | static inline int _rcu_read_ongoing(void) |
| 159 | { |
| 160 | return URCU_TLS(rcu_reader).ctr; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * This is a helper function for _rcu_quiescent_state(). |
| 165 | * The first cmm_smp_mb() ensures memory accesses in the prior read-side |
| 166 | * critical sections are not reordered with store to |
| 167 | * URCU_TLS(rcu_reader).ctr, and ensures that mutexes held within an |
| 168 | * offline section that would happen to end with this |
| 169 | * rcu_quiescent_state() call are not reordered with |
| 170 | * store to URCU_TLS(rcu_reader).ctr. |
| 171 | */ |
| 172 | static inline void _rcu_quiescent_state_update_and_wakeup(unsigned long gp_ctr) |
| 173 | { |
| 174 | cmm_smp_mb(); |
| 175 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, gp_ctr); |
| 176 | cmm_smp_mb(); /* write URCU_TLS(rcu_reader).ctr before read futex */ |
| 177 | wake_up_gp(); |
| 178 | cmm_smp_mb(); |
| 179 | } |
| 180 | |
| 181 | /* |
| 182 | * Inform RCU of a quiescent state. |
| 183 | * |
| 184 | * This function is less than 10 lines long. The intent is that this |
| 185 | * function meets the 10-line criterion for LGPL, allowing this function |
| 186 | * to be invoked directly from non-LGPL code. |
| 187 | * |
| 188 | * We skip the memory barriers and gp store if our local ctr already |
| 189 | * matches the global rcu_gp.ctr value: this is OK because a prior |
| 190 | * _rcu_quiescent_state() or _rcu_thread_online() already updated it |
| 191 | * within our thread, so we have no quiescent state to report. |
| 192 | */ |
| 193 | static inline void _rcu_quiescent_state(void) |
| 194 | { |
| 195 | unsigned long gp_ctr; |
| 196 | |
| 197 | urcu_assert(URCU_TLS(rcu_reader).registered); |
| 198 | if ((gp_ctr = CMM_LOAD_SHARED(rcu_gp.ctr)) == URCU_TLS(rcu_reader).ctr) |
| 199 | return; |
| 200 | _rcu_quiescent_state_update_and_wakeup(gp_ctr); |
| 201 | } |
| 202 | |
| 203 | /* |
| 204 | * Take a thread offline, prohibiting it from entering further RCU |
| 205 | * read-side critical sections. |
| 206 | * |
| 207 | * This function is less than 10 lines long. The intent is that this |
| 208 | * function meets the 10-line criterion for LGPL, allowing this function |
| 209 | * to be invoked directly from non-LGPL code. |
| 210 | */ |
| 211 | static inline void _rcu_thread_offline(void) |
| 212 | { |
| 213 | urcu_assert(URCU_TLS(rcu_reader).registered); |
| 214 | cmm_smp_mb(); |
| 215 | CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, 0); |
| 216 | cmm_smp_mb(); /* write URCU_TLS(rcu_reader).ctr before read futex */ |
| 217 | wake_up_gp(); |
| 218 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
| 219 | } |
| 220 | |
| 221 | /* |
| 222 | * Bring a thread online, allowing it to once again enter RCU |
| 223 | * read-side critical sections. |
| 224 | * |
| 225 | * This function is less than 10 lines long. The intent is that this |
| 226 | * function meets the 10-line criterion for LGPL, allowing this function |
| 227 | * to be invoked directly from non-LGPL code. |
| 228 | */ |
| 229 | static inline void _rcu_thread_online(void) |
| 230 | { |
| 231 | urcu_assert(URCU_TLS(rcu_reader).registered); |
| 232 | cmm_barrier(); /* Ensure the compiler does not reorder us with mutex */ |
| 233 | _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, CMM_LOAD_SHARED(rcu_gp.ctr)); |
| 234 | cmm_smp_mb(); |
| 235 | } |
| 236 | |
| 237 | #ifdef __cplusplus |
| 238 | } |
| 239 | #endif |
| 240 | |
| 241 | #endif /* _URCU_QSBR_STATIC_H */ |