urcu-static.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 LGPL-COMPATIBLE CODE. See urcu.h for linking
  10  * dynamically with the userspace rcu library.
  11  *
  12  * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
  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 <syscall.h>
  35 #include <unistd.h>
  36
  37 #include <urcu/compiler.h>
  38 #include <urcu/arch.h>
  39 #include <urcu/system.h>
  40 #include <urcu/uatomic_arch.h>
  41 #include <urcu/list.h>
  42 #include <urcu/urcu-futex.h>
  43
  44 #ifdef __cplusplus
  45 extern "C" {
  46 #endif
  47
  48 /*
  49  * This code section can only be included in LGPL 2.1 compatible source code.
  50  * See below for the function call wrappers which can be used in code meant to
  51  * be only linked with the Userspace RCU library. This comes with a small
  52  * performance degradation on the read-side due to the added function calls.
  53  * This is required to permit relinking with newer versions of the library.
  54  */
  55
  56 /*
  57  * The signal number used by the RCU library can be overridden with
  58  * -DSIGURCU= when compiling the library.
  59  */
  60 #ifndef SIGURCU
  61 #define SIGURCU SIGUSR1
  62 #endif
  63
  64 /*
  65  * If a reader is really non-cooperative and refuses to commit its
  66  * urcu_active_readers count to memory (there is no barrier in the reader
  67  * per-se), kick it after a few loops waiting for it.
  68  */
  69 #define KICK_READER_LOOPS 10000
  70
  71 /*
  72  * Active attempts to check for reader Q.S. before calling futex().
  73  */
  74 #define RCU_QS_ACTIVE_ATTEMPTS 100
  75
  76 #ifdef DEBUG_RCU
  77 #define rcu_assert(args...)     assert(args)
  78 #else
  79 #define rcu_assert(args...)
  80 #endif
  81
  82 #ifdef DEBUG_YIELD
  83 #include <sched.h>
  84 #include <time.h>
  85 #include <pthread.h>
  86 #include <unistd.h>
  87
  88 #define YIELD_READ      (1 << 0)
  89 #define YIELD_WRITE     (1 << 1)
  90
  91 /*
  92  * Updates without URCU_MB are much slower. Account this in
  93  * the delay.
  94  */
  95 #ifdef URCU_MB
  96 /* maximum sleep delay, in us */
  97 #define MAX_SLEEP 50
  98 #else
  99 #define MAX_SLEEP 30000
 100 #endif
 101
 102 extern unsigned int yield_active;
 103 extern unsigned int __thread rand_yield;
 104
 105 static inline void debug_yield_read(void)
 106 {
 107         if (yield_active & YIELD_READ)
 108                 if (rand_r(&rand_yield) & 0x1)
 109                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 110 }
 111
 112 static inline void debug_yield_write(void)
 113 {
 114         if (yield_active & YIELD_WRITE)
 115                 if (rand_r(&rand_yield) & 0x1)
 116                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 117 }
 118
 119 static inline void debug_yield_init(void)
 120 {
 121         rand_yield = time(NULL) ^ pthread_self();
 122 }
 123 #else
 124 static inline void debug_yield_read(void)
 125 {
 126 }
 127
 128 static inline void debug_yield_write(void)
 129 {
 130 }
 131
 132 static inline void debug_yield_init(void)
 133 {
 134
 135 }
 136 #endif
 137
 138 #ifdef URCU_MB
 139 static inline void reader_barrier()
 140 {
 141         smp_mb();
 142 }
 143 #else
 144 static inline void reader_barrier()
 145 {
 146         barrier();
 147 }
 148 #endif
 149
 150 /*
 151  * The trick here is that RCU_GP_CTR_BIT must be a multiple of 8 so we can use a
 152  * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 153  */
 154 #define RCU_GP_COUNT            (1UL << 0)
 155 /* Use the amount of bits equal to half of the architecture long size */
 156 #define RCU_GP_CTR_BIT          (1UL << (sizeof(long) << 2))
 157 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_BIT - 1)
 158
 159 /*
 160  * Global quiescent period counter with low-order bits unused.
 161  * Using a int rather than a char to eliminate false register dependencies
 162  * causing stalls on some architectures.
 163  */
 164 extern long urcu_gp_ctr;
 165
 166 struct urcu_reader {
 167         /* Data used by both reader and synchronize_rcu() */
 168         long ctr;
 169         char need_mb;
 170         /* Data used for registry */
 171         struct list_head head __attribute__((aligned(CACHE_LINE_SIZE)));
 172         pthread_t tid;
 173 };
 174
 175 extern struct urcu_reader __thread urcu_reader;
 176
 177 extern int gp_futex;
 178
 179 /*
 180  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 181  */
 182 static inline void wake_up_gp(void)
 183 {
 184         if (unlikely(uatomic_read(&gp_futex) == -1)) {
 185                 uatomic_set(&gp_futex, 0);
 186                 futex_async(&gp_futex, FUTEX_WAKE, 1,
 187                       NULL, NULL, 0);
 188         }
 189 }
 190
 191 static inline int rcu_old_gp_ongoing(long *value)
 192 {
 193         long v;
 194
 195         if (value == NULL)
 196                 return 0;
 197         /*
 198          * Make sure both tests below are done on the same version of *value
 199          * to insure consistency.
 200          */
 201         v = LOAD_SHARED(*value);
 202         return (v & RCU_GP_CTR_NEST_MASK) &&
 203                  ((v ^ urcu_gp_ctr) & RCU_GP_CTR_BIT);
 204 }
 205
 206 static inline void _rcu_read_lock(void)
 207 {
 208         long tmp;
 209
 210         tmp = urcu_reader.ctr;
 211         /* urcu_gp_ctr = RCU_GP_COUNT | (~RCU_GP_CTR_BIT or RCU_GP_CTR_BIT) */
 212         if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 213                 _STORE_SHARED(urcu_reader.ctr, _LOAD_SHARED(urcu_gp_ctr));
 214                 /*
 215                  * Set active readers count for outermost nesting level before
 216                  * accessing the pointer. See force_mb_all_threads().
 217                  */
 218                 reader_barrier();
 219         } else {
 220                 _STORE_SHARED(urcu_reader.ctr, tmp + RCU_GP_COUNT);
 221         }
 222 }
 223
 224 static inline void _rcu_read_unlock(void)
 225 {
 226         long tmp;
 227
 228         tmp = urcu_reader.ctr;
 229         /*
 230          * Finish using rcu before decrementing the pointer.
 231          * See force_mb_all_threads().
 232          */
 233         if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 234                 reader_barrier();
 235                 _STORE_SHARED(urcu_reader.ctr, urcu_reader.ctr - RCU_GP_COUNT);
 236                 /* write urcu_reader.ctr before read futex */
 237                 reader_barrier();
 238                 wake_up_gp();
 239         } else {
 240                 _STORE_SHARED(urcu_reader.ctr, urcu_reader.ctr - RCU_GP_COUNT);
 241         }
 242 }
 243
 244 #ifdef __cplusplus
 245 }
 246 #endif
 247
 248 #endif /* _URCU_STATIC_H */