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 LGPL-COMPATIBLE CODE. See urcu.h for linking
  10  * 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
  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 /*
  54  * RCU_MEMBARRIER is only possibly available on Linux. Fallback to RCU_MB
  55  * otherwise.
  56  */
  57 #if !defined(__linux__) && defined(RCU_MEMBARRIER)
  58 #undef RCU_MEMBARRIER
  59 #define RCU_MB
  60 #endif
  61
  62 #ifdef RCU_MEMBARRIER
  63 #include <syscall.h>
  64
  65 /* If the headers do not support SYS_membarrier, statically use RCU_MB */
  66 #ifdef SYS_membarrier
  67 # define MEMBARRIER_EXPEDITED           (1 << 0)
  68 # define MEMBARRIER_DELAYED             (1 << 1)
  69 # define MEMBARRIER_QUERY               (1 << 16)
  70 # define membarrier(...)                syscall(SYS_membarrier, __VA_ARGS__)
  71 #else
  72 # undef RCU_MEMBARRIER
  73 # define RCU_MB
  74 #endif
  75 #endif
  76
  77 /*
  78  * This code section can only be included in LGPL 2.1 compatible source code.
  79  * See below for the function call wrappers which can be used in code meant to
  80  * be only linked with the Userspace RCU library. This comes with a small
  81  * performance degradation on the read-side due to the added function calls.
  82  * This is required to permit relinking with newer versions of the library.
  83  */
  84
  85 /*
  86  * The signal number used by the RCU library can be overridden with
  87  * -DSIGRCU= when compiling the library.
  88  * Provide backward compatibility for liburcu 0.3.x SIGURCU.
  89  */
  90 #ifdef SIGURCU
  91 #define SIGRCU SIGURCU
  92 #endif
  93
  94 #ifndef SIGRCU
  95 #define SIGRCU SIGUSR1
  96 #endif
  97
  98 /*
  99  * If a reader is really non-cooperative and refuses to commit its
 100  * rcu_active_readers count to memory (there is no barrier in the reader
 101  * per-se), kick it after a few loops waiting for it.
 102  */
 103 #define KICK_READER_LOOPS 10000
 104
 105 /*
 106  * Active attempts to check for reader Q.S. before calling futex().
 107  */
 108 #define RCU_QS_ACTIVE_ATTEMPTS 100
 109
 110 #ifdef DEBUG_RCU
 111 #define rcu_assert(args...)     assert(args)
 112 #else
 113 #define rcu_assert(args...)
 114 #endif
 115
 116 #ifdef DEBUG_YIELD
 117 #include <sched.h>
 118 #include <time.h>
 119 #include <pthread.h>
 120 #include <unistd.h>
 121
 122 #define YIELD_READ      (1 << 0)
 123 #define YIELD_WRITE     (1 << 1)
 124
 125 /*
 126  * Updates with RCU_SIGNAL are much slower. Account this in the delay.
 127  */
 128 #ifdef RCU_SIGNAL
 129 /* maximum sleep delay, in us */
 130 #define MAX_SLEEP 30000
 131 #else
 132 #define MAX_SLEEP 50
 133 #endif
 134
 135 extern unsigned int yield_active;
 136 extern unsigned int __thread rand_yield;
 137
 138 static inline void debug_yield_read(void)
 139 {
 140         if (yield_active & YIELD_READ)
 141                 if (rand_r(&rand_yield) & 0x1)
 142                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 143 }
 144
 145 static inline void debug_yield_write(void)
 146 {
 147         if (yield_active & YIELD_WRITE)
 148                 if (rand_r(&rand_yield) & 0x1)
 149                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
 150 }
 151
 152 static inline void debug_yield_init(void)
 153 {
 154         rand_yield = time(NULL) ^ (unsigned long) pthread_self();
 155 }
 156 #else
 157 static inline void debug_yield_read(void)
 158 {
 159 }
 160
 161 static inline void debug_yield_write(void)
 162 {
 163 }
 164
 165 static inline void debug_yield_init(void)
 166 {
 167
 168 }
 169 #endif
 170
 171 /*
 172  * RCU memory barrier broadcast group. Currently, only broadcast to all process
 173  * threads is supported (group 0).
 174  *
 175  * Slave barriers are only guaranteed to be ordered wrt master barriers.
 176  *
 177  * The pair ordering is detailed as (O: ordered, X: not ordered) :
 178  *               slave  master
 179  *        slave    X      O
 180  *        master   O      O
 181  */
 182
 183 #define MB_GROUP_ALL            0
 184 #define RCU_MB_GROUP            MB_GROUP_ALL
 185
 186 #ifdef RCU_MEMBARRIER
 187 extern int has_sys_membarrier;
 188
 189 static inline void smp_mb_slave(int group)
 190 {
 191         if (likely(has_sys_membarrier))
 192                 cmm_barrier();
 193         else
 194                 cmm_smp_mb();
 195 }
 196 #endif
 197
 198 #ifdef RCU_MB
 199 static inline void smp_mb_slave(int group)
 200 {
 201         cmm_smp_mb();
 202 }
 203 #endif
 204
 205 #ifdef RCU_SIGNAL
 206 static inline void smp_mb_slave(int group)
 207 {
 208         cmm_barrier();
 209 }
 210 #endif
 211
 212 /*
 213  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use
 214  * a full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 215  */
 216 #define RCU_GP_COUNT            (1UL << 0)
 217 /* Use the amount of bits equal to half of the architecture long size */
 218 #define RCU_GP_CTR_PHASE        (1UL << (sizeof(unsigned long) << 2))
 219 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 220
 221 /*
 222  * Global quiescent period counter with low-order bits unused.
 223  * Using a int rather than a char to eliminate false register dependencies
 224  * causing stalls on some architectures.
 225  */
 226 extern unsigned long rcu_gp_ctr;
 227
 228 struct rcu_reader {
 229         /* Data used by both reader and synchronize_rcu() */
 230         unsigned long ctr;
 231         char need_mb;
 232         /* Data used for registry */
 233         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 234         pthread_t tid;
 235 };
 236
 237 extern struct rcu_reader __thread rcu_reader;
 238
 239 extern int32_t gp_futex;
 240
 241 /*
 242  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 243  */
 244 static inline void wake_up_gp(void)
 245 {
 246         if (unlikely(uatomic_read(&gp_futex) == -1)) {
 247                 uatomic_set(&gp_futex, 0);
 248                 futex_async(&gp_futex, FUTEX_WAKE, 1,
 249                       NULL, NULL, 0);
 250         }
 251 }
 252
 253 static inline int rcu_gp_ongoing(unsigned long *ctr)
 254 {
 255         unsigned long v;
 256
 257         /*
 258          * Make sure both tests below are done on the same version of *value
 259          * to insure consistency.
 260          */
 261         v = CMM_LOAD_SHARED(*ctr);
 262         return (v & RCU_GP_CTR_NEST_MASK) &&
 263                  ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE);
 264 }
 265
 266 static inline void _rcu_read_lock(void)
 267 {
 268         unsigned long tmp;
 269
 270         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 271         tmp = rcu_reader.ctr;
 272         /*
 273          * rcu_gp_ctr is
 274          *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 275          */
 276         if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 277                 _CMM_STORE_SHARED(rcu_reader.ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
 278                 /*
 279                  * Set active readers count for outermost nesting level before
 280                  * accessing the pointer. See smp_mb_master().
 281                  */
 282                 smp_mb_slave(RCU_MB_GROUP);
 283         } else {
 284                 _CMM_STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT);
 285         }
 286 }
 287
 288 static inline void _rcu_read_unlock(void)
 289 {
 290         unsigned long tmp;
 291
 292         tmp = rcu_reader.ctr;
 293         /*
 294          * Finish using rcu before decrementing the pointer.
 295          * See smp_mb_master().
 296          */
 297         if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 298                 smp_mb_slave(RCU_MB_GROUP);
 299                 _CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 300                 /* write rcu_reader.ctr before read futex */
 301                 smp_mb_slave(RCU_MB_GROUP);
 302                 wake_up_gp();
 303         } else {
 304                 _CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
 305         }
 306         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 307 }
 308
 309 #ifdef __cplusplus
 310 }
 311 #endif
 312
 313 #endif /* _URCU_STATIC_H */