urcu/static/urcu-qsbr.h

   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 <assert.h>
  35 #include <limits.h>
  36 #include <unistd.h>
  37 #include <stdint.h>
  38
  39 #include <urcu/compiler.h>
  40 #include <urcu/arch.h>
  41 #include <urcu/system.h>
  42 #include <urcu/uatomic.h>
  43 #include <urcu/list.h>
  44 #include <urcu/futex.h>
  45 #include <urcu/tls-compat.h>
  46 #include <urcu/urcu-checker.h>
  47
  48 #ifdef __cplusplus
  49 extern "C" {
  50 #endif
  51
  52 /*
  53  * This code section can only be included in LGPL 2.1 compatible source code.
  54  * See below for the function call wrappers which can be used in code meant to
  55  * be only linked with the Userspace RCU library. This comes with a small
  56  * performance degradation on the read-side due to the added function calls.
  57  * This is required to permit relinking with newer versions of the library.
  58  */
  59
  60 #ifdef DEBUG_RCU
  61 #define rcu_assert(args...)     assert(args)
  62 #else
  63 #define rcu_assert(args...)
  64 #endif
  65
  66 enum rcu_state {
  67         RCU_READER_ACTIVE_CURRENT,
  68         RCU_READER_ACTIVE_OLD,
  69         RCU_READER_INACTIVE,
  70 };
  71
  72 #define RCU_GP_ONLINE           (1UL << 0)
  73 #define RCU_GP_CTR              (1UL << 1)
  74
  75 struct rcu_gp {
  76         /*
  77          * Global quiescent period counter with low-order bits unused.
  78          * Using a int rather than a char to eliminate false register
  79          * dependencies causing stalls on some architectures.
  80          */
  81         unsigned long ctr;
  82
  83         int32_t futex;
  84 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
  85
  86 extern struct rcu_gp rcu_gp;
  87
  88 struct rcu_reader {
  89         /* Data used by both reader and synchronize_rcu() */
  90         unsigned long ctr;
  91         /* Data used for registry */
  92         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
  93         int waiting;
  94         pthread_t tid;
  95 };
  96
  97 extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
  98
  99 /*
 100  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 101  */
 102 static inline void wake_up_gp(void)
 103 {
 104         if (caa_unlikely(_CMM_LOAD_SHARED(URCU_TLS(rcu_reader).waiting))) {
 105                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).waiting, 0);
 106                 cmm_smp_mb();
 107                 if (uatomic_read(&rcu_gp.futex) != -1)
 108                         return;
 109                 uatomic_set(&rcu_gp.futex, 0);
 110                 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 __attribute__((always_inline))
 135 void _rcu_read_lock(void)
 136 {
 137         rcu_read_lock_debug();
 138         rcu_assert(URCU_TLS(rcu_reader).ctr);
 139 }
 140
 141 /*
 142  * Exit an RCU read-side critical section.
 143  *
 144  * This function is less than 10 lines long.  The intent is that this
 145  * function meets the 10-line criterion for LGPL, allowing this function
 146  * to be invoked directly from non-LGPL code.
 147  */
 148 static inline __attribute__((always_inline))
 149 void _rcu_read_unlock(void)
 150 {
 151         rcu_read_unlock_debug();
 152 }
 153
 154 /*
 155  * Returns whether within a RCU read-side critical section.
 156  *
 157  * This function is less than 10 lines long.  The intent is that this
 158  * function meets the 10-line criterion for LGPL, allowing this function
 159  * to be invoked directly from non-LGPL code.
 160  */
 161 static inline int _rcu_read_ongoing(void)
 162 {
 163         return URCU_TLS(rcu_reader).ctr;
 164 }
 165
 166 /*
 167  * This is a helper function for _rcu_quiescent_state().
 168  * The first cmm_smp_mb() ensures memory accesses in the prior read-side
 169  * critical sections are not reordered with store to
 170  * URCU_TLS(rcu_reader).ctr, and ensures that mutexes held within an
 171  * offline section that would happen to end with this
 172  * rcu_quiescent_state() call are not reordered with
 173  * store to URCU_TLS(rcu_reader).ctr.
 174  */
 175 static inline void _rcu_quiescent_state_update_and_wakeup(unsigned long gp_ctr)
 176 {
 177         cmm_smp_mb();
 178         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, gp_ctr);
 179         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 180         wake_up_gp();
 181         cmm_smp_mb();
 182 }
 183
 184 /*
 185  * Inform RCU of a quiescent state.
 186  *
 187  * This function is less than 10 lines long.  The intent is that this
 188  * function meets the 10-line criterion for LGPL, allowing this function
 189  * to be invoked directly from non-LGPL code.
 190  *
 191  * We skip the memory barriers and gp store if our local ctr already
 192  * matches the global rcu_gp.ctr value: this is OK because a prior
 193  * _rcu_quiescent_state() or _rcu_thread_online() already updated it
 194  * within our thread, so we have no quiescent state to report.
 195  */
 196 static inline void _rcu_quiescent_state(void)
 197 {
 198         unsigned long gp_ctr;
 199
 200         if ((gp_ctr = CMM_LOAD_SHARED(rcu_gp.ctr)) == URCU_TLS(rcu_reader).ctr)
 201                 return;
 202         _rcu_quiescent_state_update_and_wakeup(gp_ctr);
 203 }
 204
 205 /*
 206  * Take a thread offline, prohibiting it from entering further RCU
 207  * read-side critical sections.
 208  *
 209  * This function is less than 10 lines long.  The intent is that this
 210  * function meets the 10-line criterion for LGPL, allowing this function
 211  * to be invoked directly from non-LGPL code.
 212  */
 213 static inline void _rcu_thread_offline(void)
 214 {
 215         cmm_smp_mb();
 216         CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, 0);
 217         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 218         wake_up_gp();
 219         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 220 }
 221
 222 /*
 223  * Bring a thread online, allowing it to once again enter RCU
 224  * read-side critical sections.
 225  *
 226  * This function is less than 10 lines long.  The intent is that this
 227  * function meets the 10-line criterion for LGPL, allowing this function
 228  * to be invoked directly from non-LGPL code.
 229  */
 230 static inline void _rcu_thread_online(void)
 231 {
 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 */