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