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(...)         syscall(__NR_membarrier, __VA_ARGS__)
  60 #else
  61 #undef RCU_MEMBARRIER
  62 #define RCU_MB
  63 #endif
  64 #endif
  65
  66 /*
  67  * This code section can only be included in LGPL 2.1 compatible source code.
  68  * See below for the function call wrappers which can be used in code meant to
  69  * be only linked with the Userspace RCU library. This comes with a small
  70  * performance degradation on the read-side due to the added function calls.
  71  * This is required to permit relinking with newer versions of the library.
  72  */
  73
  74 /*
  75  * The signal number used by the RCU library can be overridden with
  76  * -DSIGRCU= when compiling the library.
  77  */
  78 #ifndef SIGRCU
  79 #define SIGRCU SIGUSR1
  80 #endif
  81
  82 /*
  83  * If a reader is really non-cooperative and refuses to commit its
  84  * rcu_active_readers count to memory (there is no barrier in the reader
  85  * per-se), kick it after a few loops waiting for it.
  86  */
  87 #define KICK_READER_LOOPS 10000
  88
  89 /*
  90  * Active attempts to check for reader Q.S. before calling futex().
  91  */
  92 #define RCU_QS_ACTIVE_ATTEMPTS 100
  93
  94 #ifdef DEBUG_RCU
  95 #define rcu_assert(args...)     assert(args)
  96 #else
  97 #define rcu_assert(args...)
  98 #endif
  99
 100 #ifdef DEBUG_YIELD
 101 #include <sched.h>
 102 #include <time.h>
 103 #include <pthread.h>
 104 #include <unistd.h>
 105
 106 #define YIELD_READ      (1 << 0)
 107 #define YIELD_WRITE     (1 << 1)
 108
 109 /*
 110  * Updates with RCU_SIGNAL are much slower. Account this in the delay.
 111  */
 112 #ifdef RCU_SIGNAL
 113 /* maximum sleep delay, in us */
 114 #define MAX_SLEEP 30000
 115 #else
 116 #define MAX_SLEEP 50
 117 #endif
 118
 119 extern unsigned int yield_active;
 120 extern unsigned int __thread rand_yield;
 121
 122 static inline void debug_yield_read(void)
 123 {
 124         if (yield_active & YIELD_READ)
 125                 if (rand_r(&rand_yield) & 0x1)
 126                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 127 }
 128
 129 static inline void debug_yield_write(void)
 130 {
 131         if (yield_active & YIELD_WRITE)
 132                 if (rand_r(&rand_yield) & 0x1)
 133                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 134 }
 135
 136 static inline void debug_yield_init(void)
 137 {
 138         rand_yield = time(NULL) ^ pthread_self();
 139 }
 140 #else
 141 static inline void debug_yield_read(void)
 142 {
 143 }
 144
 145 static inline void debug_yield_write(void)
 146 {
 147 }
 148
 149 static inline void debug_yield_init(void)
 150 {
 151
 152 }
 153 #endif
 154
 155 #ifdef RCU_MEMBARRIER
 156 extern int has_sys_membarrier;
 157
 158 static inline void smp_mb_light()
 159 {
 160         if (likely(has_sys_membarrier))
 161                 barrier();
 162         else
 163                 smp_mb();
 164 }
 165 #endif
 166
 167 #ifdef RCU_MB
 168 static inline void smp_mb_light()
 169 {
 170         smp_mb();
 171 }
 172 #endif
 173
 174 #ifdef RCU_SIGNAL
 175 static inline void smp_mb_light()
 176 {
 177         barrier();
 178 }
 179 #endif
 180
 181 /*
 182  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
 183  * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 184  */
 185 #define RCU_GP_COUNT            (1UL << 0)
 186 /* Use the amount of bits equal to half of the architecture long size */
 187 #define RCU_GP_CTR_PHASE        (1UL << (sizeof(long) << 2))
 188 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 189
 190 /*
 191  * Global quiescent period counter with low-order bits unused.
 192  * Using a int rather than a char to eliminate false register dependencies
 193  * causing stalls on some architectures.
 194  */
 195 extern long rcu_gp_ctr;
 196
 197 struct rcu_reader {
 198         /* Data used by both reader and synchronize_rcu() */
 199         long ctr;
 200         char need_mb;
 201         /* Data used for registry */
 202         struct list_head head __attribute__((aligned(CACHE_LINE_SIZE)));
 203         pthread_t tid;
 204 };
 205
 206 extern struct rcu_reader __thread rcu_reader;
 207
 208 extern int gp_futex;
 209
 210 /*
 211  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 212  */
 213 static inline void wake_up_gp(void)
 214 {
 215         if (unlikely(uatomic_read(&gp_futex) == -1)) {
 216                 uatomic_set(&gp_futex, 0);
 217                 futex_async(&gp_futex, FUTEX_WAKE, 1,
 218                       NULL, NULL, 0);
 219         }
 220 }
 221
 222 static inline int rcu_old_gp_ongoing(long *value)
 223 {
 224         long v;
 225
 226         if (value == NULL)
 227                 return 0;
 228         /*
 229          * Make sure both tests below are done on the same version of *value
 230          * to insure consistency.
 231          */
 232         v = LOAD_SHARED(*value);
 233         return (v & RCU_GP_CTR_NEST_MASK) &&
 234                  ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE);
 235 }
 236
 237 static inline void _rcu_read_lock(void)
 238 {
 239         long tmp;
 240
 241         tmp = rcu_reader.ctr;
 242         /*
 243          * rcu_gp_ctr is
 244          *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 245          */
 246         if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 247                 _STORE_SHARED(rcu_reader.ctr, _LOAD_SHARED(rcu_gp_ctr));
 248                 /*
 249                  * Set active readers count for outermost nesting level before
 250                  * accessing the pointer. See smp_mb_heavy().
 251                  */
 252                 smp_mb_light();
 253         } else {
 254                 _STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT);
 255         }
 256 }
 257
 258 static inline void _rcu_read_unlock(void)
 259 {
 260         long tmp;
 261
 262         tmp = rcu_reader.ctr;
 263         /*
 264          * Finish using rcu before decrementing the pointer.
 265          * See smp_mb_heavy().
 266          */
 267         if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 268                 smp_mb_light();
 269                 _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 270                 /* write rcu_reader.ctr before read futex */
 271                 smp_mb_light();
 272                 wake_up_gp();
 273         } else {
 274                 _STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 275         }
 276 }
 277
 278 #ifdef __cplusplus
 279 }
 280 #endif
 281
 282 #endif /* _URCU_STATIC_H */