urcu/static/urcu.h

   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/compiler.h>
  38 #include <urcu/arch.h>
  39 #include <urcu/system.h>
  40 #include <urcu/uatomic.h>
  41 #include <urcu/list.h>
  42 #include <urcu/futex.h>
  43 #include <urcu/tls-compat.h>
  44 #include <urcu/rand-compat.h>
  45
  46 #ifdef __cplusplus
  47 extern "C" {
  48 #endif
  49
  50 /* Default is RCU_MEMBARRIER */
  51 #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL)
  52 #define RCU_MEMBARRIER
  53 #endif
  54
  55 /*
  56  * This code section can only be included in LGPL 2.1 compatible source code.
  57  * See below for the function call wrappers which can be used in code meant to
  58  * be only linked with the Userspace RCU library. This comes with a small
  59  * performance degradation on the read-side due to the added function calls.
  60  * This is required to permit relinking with newer versions of the library.
  61  */
  62
  63 /*
  64  * The signal number used by the RCU library can be overridden with
  65  * -DSIGRCU= when compiling the library.
  66  * Provide backward compatibility for liburcu 0.3.x SIGURCU.
  67  */
  68 #ifdef SIGURCU
  69 #define SIGRCU SIGURCU
  70 #endif
  71
  72 #ifndef SIGRCU
  73 #define SIGRCU SIGUSR1
  74 #endif
  75
  76 enum rcu_state {
  77         RCU_READER_ACTIVE_CURRENT,
  78         RCU_READER_ACTIVE_OLD,
  79         RCU_READER_INACTIVE,
  80 };
  81
  82 #ifdef DEBUG_RCU
  83 #define rcu_assert(args...)     assert(args)
  84 #else
  85 #define rcu_assert(args...)
  86 #endif
  87
  88 /*
  89  * RCU memory barrier broadcast group. Currently, only broadcast to all process
  90  * threads is supported (group 0).
  91  *
  92  * Slave barriers are only guaranteed to be ordered wrt master barriers.
  93  *
  94  * The pair ordering is detailed as (O: ordered, X: not ordered) :
  95  *               slave  master
  96  *        slave    X      O
  97  *        master   O      O
  98  */
  99
 100 #define MB_GROUP_ALL            0
 101 #define RCU_MB_GROUP            MB_GROUP_ALL
 102
 103 #ifdef RCU_MEMBARRIER
 104 extern int rcu_has_sys_membarrier;
 105
 106 static inline void smp_mb_slave(int group)
 107 {
 108         if (caa_likely(rcu_has_sys_membarrier))
 109                 cmm_barrier();
 110         else
 111                 cmm_smp_mb();
 112 }
 113 #endif
 114
 115 #ifdef RCU_MB
 116 static inline void smp_mb_slave(int group)
 117 {
 118         cmm_smp_mb();
 119 }
 120 #endif
 121
 122 #ifdef RCU_SIGNAL
 123 static inline void smp_mb_slave(int group)
 124 {
 125         cmm_barrier();
 126 }
 127 #endif
 128
 129 /*
 130  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
 131  * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 132  */
 133 #define RCU_GP_COUNT            (1UL << 0)
 134 /* Use the amount of bits equal to half of the architecture long size */
 135 #define RCU_GP_CTR_PHASE        (1UL << (sizeof(unsigned long) << 2))
 136 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 137
 138 struct rcu_gp {
 139         /*
 140          * Global grace period counter.
 141          * Contains the current RCU_GP_CTR_PHASE.
 142          * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
 143          * Written to only by writer with mutex taken.
 144          * Read by both writer and readers.
 145          */
 146         unsigned long ctr;
 147
 148         int32_t futex;
 149 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 150
 151 extern struct rcu_gp rcu_gp;
 152
 153 struct rcu_reader {
 154         /* Data used by both reader and synchronize_rcu() */
 155         unsigned long ctr;
 156         char need_mb;
 157         /* Data used for registry */
 158         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 159         pthread_t tid;
 160 };
 161
 162 extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 163
 164 /*
 165  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 166  */
 167 static inline void wake_up_gp(void)
 168 {
 169         if (caa_unlikely(uatomic_read(&rcu_gp.futex) == -1)) {
 170                 uatomic_set(&rcu_gp.futex, 0);
 171                 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(RCU_MB_GROUP);
 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         cmm_barrier();
 223         tmp = URCU_TLS(rcu_reader).ctr;
 224         _rcu_read_lock_update(tmp);
 225 }
 226
 227 /*
 228  * This is a helper function for _rcu_read_unlock().
 229  *
 230  * The first smp_mb_slave() call ensures that the critical section is
 231  * seen to precede the store to rcu_reader.ctr.
 232  * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr
 233  * before reading the update-side futex.
 234  */
 235 static inline void _rcu_read_unlock_update_and_wakeup(unsigned long tmp)
 236 {
 237         if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 238                 smp_mb_slave(RCU_MB_GROUP);
 239                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, URCU_TLS(rcu_reader).ctr - RCU_GP_COUNT);
 240                 smp_mb_slave(RCU_MB_GROUP);
 241                 wake_up_gp();
 242         } else
 243                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, URCU_TLS(rcu_reader).ctr - RCU_GP_COUNT);
 244 }
 245
 246 /*
 247  * Exit an RCU read-side crtical section.  Both this function and its
 248  * helper are smaller than 10 lines of code, and are intended to be
 249  * usable by non-LGPL code, as called out in LGPL.
 250  */
 251 static inline void _rcu_read_unlock(void)
 252 {
 253         unsigned long tmp;
 254
 255         tmp = URCU_TLS(rcu_reader).ctr;
 256         _rcu_read_unlock_update_and_wakeup(tmp);
 257         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 258 }
 259
 260 /*
 261  * Returns whether within a RCU read-side critical section.
 262  *
 263  * This function is less than 10 lines long.  The intent is that this
 264  * function meets the 10-line criterion for LGPL, allowing this function
 265  * to be invoked directly from non-LGPL code.
 266  */
 267 static inline int _rcu_read_ongoing(void)
 268 {
 269         return URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK;
 270 }
 271
 272 #ifdef __cplusplus
 273 }
 274 #endif
 275
 276 #endif /* _URCU_STATIC_H */