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