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