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/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/debug.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 /*
  83  * Slave barriers are only guaranteed to be ordered wrt master barriers.
  84  *
  85  * The pair ordering is detailed as (O: ordered, X: not ordered) :
  86  *               slave  master
  87  *        slave    X      O
  88  *        master   O      O
  89  */
  90
  91 #ifdef RCU_MEMBARRIER
  92 extern int rcu_has_sys_membarrier;
  93
  94 static inline void smp_mb_slave(void)
  95 {
  96         if (caa_likely(rcu_has_sys_membarrier))
  97                 cmm_barrier();
  98         else
  99                 cmm_smp_mb();
 100 }
 101 #endif
 102
 103 #ifdef RCU_MB
 104 static inline void smp_mb_slave(void)
 105 {
 106         cmm_smp_mb();
 107 }
 108 #endif
 109
 110 #ifdef RCU_SIGNAL
 111 static inline void smp_mb_slave(void)
 112 {
 113         cmm_barrier();
 114 }
 115 #endif
 116
 117 /*
 118  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
 119  * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 120  */
 121 #define RCU_GP_COUNT            (1UL << 0)
 122 /* Use the amount of bits equal to half of the architecture long size */
 123 #define RCU_GP_CTR_PHASE        (1UL << (sizeof(unsigned long) << 2))
 124 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 125
 126 struct rcu_gp {
 127         /*
 128          * Global grace period counter.
 129          * Contains the current RCU_GP_CTR_PHASE.
 130          * Also has a RCU_GP_COUNT of 1, to accelerate the reader fast path.
 131          * Written to only by writer with mutex taken.
 132          * Read by both writer and readers.
 133          */
 134         unsigned long ctr;
 135
 136         int32_t futex;
 137 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 138
 139 extern struct rcu_gp rcu_gp;
 140
 141 struct rcu_reader {
 142         /* Data used by both reader and synchronize_rcu() */
 143         unsigned long ctr;
 144         char need_mb;
 145         /* Data used for registry */
 146         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 147         pthread_t tid;
 148         /* Reader registered flag, for internal checks. */
 149         unsigned int registered:1;
 150 };
 151
 152 extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 153
 154 /*
 155  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 156  */
 157 static inline void wake_up_gp(void)
 158 {
 159         if (caa_unlikely(uatomic_read(&rcu_gp.futex) == -1)) {
 160                 uatomic_set(&rcu_gp.futex, 0);
 161                 /*
 162                  * Ignoring return value until we can make this function
 163                  * return something (because urcu_die() is not publicly
 164                  * exposed).
 165                  */
 166                 (void) futex_async(&rcu_gp.futex, FUTEX_WAKE, 1,
 167                                 NULL, NULL, 0);
 168         }
 169 }
 170
 171 static inline enum rcu_state rcu_reader_state(unsigned long *ctr)
 172 {
 173         unsigned long v;
 174
 175         /*
 176          * Make sure both tests below are done on the same version of *value
 177          * to insure consistency.
 178          */
 179         v = CMM_LOAD_SHARED(*ctr);
 180         if (!(v & RCU_GP_CTR_NEST_MASK))
 181                 return RCU_READER_INACTIVE;
 182         if (!((v ^ rcu_gp.ctr) & RCU_GP_CTR_PHASE))
 183                 return RCU_READER_ACTIVE_CURRENT;
 184         return RCU_READER_ACTIVE_OLD;
 185 }
 186
 187 /*
 188  * Helper for _rcu_read_lock().  The format of rcu_gp.ctr (as well as
 189  * the per-thread rcu_reader.ctr) has the upper bits containing a count of
 190  * _rcu_read_lock() nesting, and a lower-order bit that contains either zero
 191  * or RCU_GP_CTR_PHASE.  The smp_mb_slave() ensures that the accesses in
 192  * _rcu_read_lock() happen before the subsequent read-side critical section.
 193  */
 194 static inline void _rcu_read_lock_update(unsigned long tmp)
 195 {
 196         if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 197                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp.ctr));
 198                 smp_mb_slave();
 199         } else
 200                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp + RCU_GP_COUNT);
 201 }
 202
 203 /*
 204  * Enter an RCU read-side critical section.
 205  *
 206  * The first cmm_barrier() call ensures that the compiler does not reorder
 207  * the body of _rcu_read_lock() with a mutex.
 208  *
 209  * This function and its helper are both less than 10 lines long.  The
 210  * intent is that this function meets the 10-line criterion in LGPL,
 211  * allowing this function to be invoked directly from non-LGPL code.
 212  */
 213 static inline void _rcu_read_lock(void)
 214 {
 215         unsigned long tmp;
 216
 217         urcu_assert(URCU_TLS(rcu_reader).registered);
 218         cmm_barrier();
 219         tmp = URCU_TLS(rcu_reader).ctr;
 220         urcu_assert((tmp & RCU_GP_CTR_NEST_MASK) != RCU_GP_CTR_NEST_MASK);
 221         _rcu_read_lock_update(tmp);
 222 }
 223
 224 /*
 225  * This is a helper function for _rcu_read_unlock().
 226  *
 227  * The first smp_mb_slave() call ensures that the critical section is
 228  * seen to precede the store to rcu_reader.ctr.
 229  * The second smp_mb_slave() call ensures that we write to rcu_reader.ctr
 230  * before reading the update-side futex.
 231  */
 232 static inline void _rcu_read_unlock_update_and_wakeup(unsigned long tmp)
 233 {
 234         if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 235                 smp_mb_slave();
 236                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT);
 237                 smp_mb_slave();
 238                 wake_up_gp();
 239         } else
 240                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp - RCU_GP_COUNT);
 241 }
 242
 243 /*
 244  * Exit an RCU read-side crtical section.  Both this function and its
 245  * helper are smaller than 10 lines of code, and are intended to be
 246  * usable by non-LGPL code, as called out in LGPL.
 247  */
 248 static inline void _rcu_read_unlock(void)
 249 {
 250         unsigned long tmp;
 251
 252         urcu_assert(URCU_TLS(rcu_reader).registered);
 253         tmp = URCU_TLS(rcu_reader).ctr;
 254         urcu_assert(tmp & RCU_GP_CTR_NEST_MASK);
 255         _rcu_read_unlock_update_and_wakeup(tmp);
 256         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 257 }
 258
 259 /*
 260  * Returns whether within a RCU read-side critical section.
 261  *
 262  * This function is less than 10 lines long.  The intent is that this
 263  * function meets the 10-line criterion for LGPL, allowing this function
 264  * to be invoked directly from non-LGPL code.
 265  */
 266 static inline int _rcu_read_ongoing(void)
 267 {
 268         return URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK;
 269 }
 270
 271 #ifdef __cplusplus
 272 }
 273 #endif
 274
 275 #endif /* _URCU_STATIC_H */