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