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 /* Default is RCU_MEMBARRIER */
  49 #if !defined(RCU_MEMBARRIER) && !defined(RCU_MB) && !defined(RCU_SIGNAL)
  50 #define RCU_MEMBARRIER
  51 #endif
  52
  53 #ifdef RCU_MEMBARRIER
  54 #include <unistd.h>
  55 #include <sys/syscall.h>
  56
  57 /* If the headers do not support SYS_membarrier, statically use RCU_MB */
  58 #ifdef SYS_membarrier
  59 # define MEMBARRIER_EXPEDITED           (1 << 0)
  60 # define MEMBARRIER_DELAYED             (1 << 1)
  61 # define MEMBARRIER_QUERY               (1 << 16)
  62 # define membarrier(...)                syscall(__NR_membarrier, __VA_ARGS__)
  63 #else
  64 # undef RCU_MEMBARRIER
  65 # define RCU_MB
  66 #endif
  67 #endif
  68
  69 /*
  70  * This code section can only be included in LGPL 2.1 compatible source code.
  71  * See below for the function call wrappers which can be used in code meant to
  72  * be only linked with the Userspace RCU library. This comes with a small
  73  * performance degradation on the read-side due to the added function calls.
  74  * This is required to permit relinking with newer versions of the library.
  75  */
  76
  77 /*
  78  * The signal number used by the RCU library can be overridden with
  79  * -DSIGRCU= when compiling the library.
  80  */
  81 #ifndef SIGRCU
  82 #define SIGRCU SIGUSR1
  83 #endif
  84
  85 /*
  86  * If a reader is really non-cooperative and refuses to commit its
  87  * rcu_active_readers count to memory (there is no barrier in the reader
  88  * per-se), kick it after a few loops waiting for it.
  89  */
  90 #define KICK_READER_LOOPS 10000
  91
  92 /*
  93  * Active attempts to check for reader Q.S. before calling futex().
  94  */
  95 #define RCU_QS_ACTIVE_ATTEMPTS 100
  96
  97 #ifdef DEBUG_RCU
  98 #define rcu_assert(args...)     assert(args)
  99 #else
 100 #define rcu_assert(args...)
 101 #endif
 102
 103 #ifdef DEBUG_YIELD
 104 #include <sched.h>
 105 #include <time.h>
 106 #include <pthread.h>
 107 #include <unistd.h>
 108
 109 #define YIELD_READ      (1 << 0)
 110 #define YIELD_WRITE     (1 << 1)
 111
 112 /*
 113  * Updates with RCU_SIGNAL are much slower. Account this in the delay.
 114  */
 115 #ifdef RCU_SIGNAL
 116 /* maximum sleep delay, in us */
 117 #define MAX_SLEEP 30000
 118 #else
 119 #define MAX_SLEEP 50
 120 #endif
 121
 122 extern unsigned int yield_active;
 123 extern unsigned int __thread rand_yield;
 124
 125 static inline void debug_yield_read(void)
 126 {
 127         if (yield_active & YIELD_READ)
 128                 if (rand_r(&rand_yield) & 0x1)
 129                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 130 }
 131
 132 static inline void debug_yield_write(void)
 133 {
 134         if (yield_active & YIELD_WRITE)
 135                 if (rand_r(&rand_yield) & 0x1)
 136                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 137 }
 138
 139 static inline void debug_yield_init(void)
 140 {
 141         rand_yield = time(NULL) ^ pthread_self();
 142 }
 143 #else
 144 static inline void debug_yield_read(void)
 145 {
 146 }
 147
 148 static inline void debug_yield_write(void)
 149 {
 150 }
 151
 152 static inline void debug_yield_init(void)
 153 {
 154
 155 }
 156 #endif
 157
 158 #ifdef RCU_MEMBARRIER
 159 extern int has_sys_membarrier;
 160
 161 static inline void smp_mb_light()
 162 {
 163         if (likely(has_sys_membarrier))
 164                 barrier();
 165         else
 166                 smp_mb();
 167 }
 168 #endif
 169
 170 #ifdef RCU_MB
 171 static inline void smp_mb_light()
 172 {
 173         smp_mb();
 174 }
 175 #endif
 176
 177 #ifdef RCU_SIGNAL
 178 static inline void smp_mb_light()
 179 {
 180         barrier();
 181 }
 182 #endif
 183
 184 /*
 185  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
 186  * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 187  */
 188 #define RCU_GP_COUNT            (1UL << 0)
 189 /* Use the amount of bits equal to half of the architecture long size */
 190 #define RCU_GP_CTR_PHASE        (1UL << (sizeof(long) << 2))
 191 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 192
 193 /*
 194  * Global quiescent period counter with low-order bits unused.
 195  * Using a int rather than a char to eliminate false register dependencies
 196  * causing stalls on some architectures.
 197  */
 198 extern long rcu_gp_ctr;
 199
 200 struct rcu_reader {
 201         /* Data used by both reader and synchronize_rcu() */
 202         long ctr;
 203         char need_mb;
 204         /* Data used for registry */
 205         struct list_head head __attribute__((aligned(CACHE_LINE_SIZE)));
 206         pthread_t tid;
 207 };
 208
 209 extern struct rcu_reader __thread rcu_reader;
 210
 211 extern int gp_futex;
 212
 213 /*
 214  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 215  */
 216 static inline void wake_up_gp(void)
 217 {
 218         if (unlikely(uatomic_read(&gp_futex) == -1)) {
 219                 uatomic_set(&gp_futex, 0);
 220                 futex_async(&gp_futex, FUTEX_WAKE, 1,
 221                       NULL, NULL, 0);
 222         }
 223 }
 224
 225 static inline int rcu_old_gp_ongoing(long *value)
 226 {
 227         long v;
 228
 229         if (value == NULL)
 230                 return 0;
 231         /*
 232          * Make sure both tests below are done on the same version of *value
 233          * to insure consistency.
 234          */
 235         v = LOAD_SHARED(*value);
 236         return (v & RCU_GP_CTR_NEST_MASK) &&
 237                  ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE);
 238 }
 239
 240 static inline void _rcu_read_lock(void)
 241 {
 242         long tmp;
 243
 244         tmp = rcu_reader.ctr;
 245         /*
 246          * rcu_gp_ctr is
 247          *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 248          */
 249         if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 250                 _STORE_SHARED(rcu_reader.ctr, _LOAD_SHARED(rcu_gp_ctr));
 251                 /*
 252                  * Set active readers count for outermost nesting level before
 253                  * accessing the pointer. See smp_mb_heavy().
 254                  */
 255                 smp_mb_light();
 256         } else {
 257                 _STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT);
 258         }
 259 }
 260
 261 static inline void _rcu_read_unlock(void)
 262 {
 263         long tmp;
 264
 265         tmp = rcu_reader.ctr;
 266         /*
 267          * Finish using rcu before decrementing the pointer.
 268          * See smp_mb_heavy().
 269          */
 270         if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 271                 smp_mb_light();
 272                 _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 273                 /* write rcu_reader.ctr before read futex */
 274                 smp_mb_light();
 275                 wake_up_gp();
 276         } else {
 277                 _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 278         }
 279 }
 280
 281 #ifdef __cplusplus
 282 }
 283 #endif
 284
 285 #endif /* _URCU_STATIC_H */