urcu/static/urcu-bp.h

   1 #ifndef _URCU_BP_STATIC_H
   2 #define _URCU_BP_STATIC_H
   3
   4 /*
   5  * urcu-bp-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
  36 #include <urcu/compiler.h>
  37 #include <urcu/arch.h>
  38 #include <urcu/system.h>
  39 #include <urcu/uatomic.h>
  40 #include <urcu/list.h>
  41
  42 /*
  43  * This code section can only be included in LGPL 2.1 compatible source code.
  44  * See below for the function call wrappers which can be used in code meant to
  45  * be only linked with the Userspace RCU library. This comes with a small
  46  * performance degradation on the read-side due to the added function calls.
  47  * This is required to permit relinking with newer versions of the library.
  48  */
  49
  50 #ifdef __cplusplus
  51 extern "C" {
  52 #endif
  53
  54 /*
  55  * Active attempts to check for reader Q.S. before calling sleep().
  56  */
  57 #define RCU_QS_ACTIVE_ATTEMPTS 100
  58
  59 #ifdef DEBUG_RCU
  60 #define rcu_assert(args...)     assert(args)
  61 #else
  62 #define rcu_assert(args...)
  63 #endif
  64
  65 #ifdef DEBUG_YIELD
  66 #include <sched.h>
  67 #include <time.h>
  68 #include <pthread.h>
  69 #include <unistd.h>
  70
  71 #define YIELD_READ      (1 << 0)
  72 #define YIELD_WRITE     (1 << 1)
  73
  74 /*
  75  * Updates without RCU_MB are much slower. Account this in
  76  * the delay.
  77  */
  78 /* maximum sleep delay, in us */
  79 #define MAX_SLEEP 50
  80
  81 extern unsigned int yield_active;
  82 extern unsigned int __thread rand_yield;
  83
  84 static inline void debug_yield_read(void)
  85 {
  86         if (yield_active & YIELD_READ)
  87                 if (rand_r(&rand_yield) & 0x1)
  88                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
  89 }
  90
  91 static inline void debug_yield_write(void)
  92 {
  93         if (yield_active & YIELD_WRITE)
  94                 if (rand_r(&rand_yield) & 0x1)
  95                         usleep(rand_r(&rand_yield) % MAX_SLEEP);
  96 }
  97
  98 static inline void debug_yield_init(void)
  99 {
 100         rand_yield = time(NULL) ^ pthread_self();
 101 }
 102 #else
 103 static inline void debug_yield_read(void)
 104 {
 105 }
 106
 107 static inline void debug_yield_write(void)
 108 {
 109 }
 110
 111 static inline void debug_yield_init(void)
 112 {
 113
 114 }
 115 #endif
 116
 117 /*
 118  * The trick here is that RCU_GP_CTR_PHASE must be a multiple of 8 so we can use a
 119  * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
 120  */
 121 #define RCU_GP_COUNT            (1UL << 0)
 122 /* Use the amount of bits equal to half of the architecture long size */
 123 #define RCU_GP_CTR_PHASE                (1UL << (sizeof(long) << 2))
 124 #define RCU_GP_CTR_NEST_MASK    (RCU_GP_CTR_PHASE - 1)
 125
 126 /*
 127  * Used internally by _rcu_read_lock.
 128  */
 129 extern void rcu_bp_register(void);
 130
 131 /*
 132  * Global quiescent period counter with low-order bits unused.
 133  * Using a int rather than a char to eliminate false register dependencies
 134  * causing stalls on some architectures.
 135  */
 136 extern long rcu_gp_ctr;
 137
 138 struct rcu_reader {
 139         /* Data used by both reader and synchronize_rcu() */
 140         long ctr;
 141         /* Data used for registry */
 142         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 143         pthread_t tid;
 144         int alloc;      /* registry entry allocated */
 145 };
 146
 147 /*
 148  * Bulletproof version keeps a pointer to a registry not part of the TLS.
 149  * Adds a pointer dereference on the read-side, but won't require to unregister
 150  * the reader thread.
 151  */
 152 extern struct rcu_reader __thread *rcu_reader;
 153
 154 static inline int rcu_old_gp_ongoing(long *value)
 155 {
 156         long v;
 157
 158         if (value == NULL)
 159                 return 0;
 160         /*
 161          * Make sure both tests below are done on the same version of *value
 162          * to insure consistency.
 163          */
 164         v = CMM_LOAD_SHARED(*value);
 165         return (v & RCU_GP_CTR_NEST_MASK) &&
 166                  ((v ^ rcu_gp_ctr) & RCU_GP_CTR_PHASE);
 167 }
 168
 169 static inline void _rcu_read_lock(void)
 170 {
 171         long tmp;
 172
 173         /* Check if registered */
 174         if (unlikely(!rcu_reader))
 175                 rcu_bp_register();
 176
 177         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 178         tmp = rcu_reader->ctr;
 179         /*
 180          * rcu_gp_ctr is
 181          *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 182          */
 183         if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
 184                 _CMM_STORE_SHARED(rcu_reader->ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
 185                 /*
 186                  * Set active readers count for outermost nesting level before
 187                  * accessing the pointer.
 188                  */
 189                 cmm_smp_mb();
 190         } else {
 191                 _CMM_STORE_SHARED(rcu_reader->ctr, tmp + RCU_GP_COUNT);
 192         }
 193 }
 194
 195 static inline void _rcu_read_unlock(void)
 196 {
 197         /*
 198          * Finish using rcu before decrementing the pointer.
 199          */
 200         cmm_smp_mb();
 201         _CMM_STORE_SHARED(rcu_reader->ctr, rcu_reader->ctr - RCU_GP_COUNT);
 202         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 203 }
 204
 205 #ifdef __cplusplus
 206 }
 207 #endif
 208
 209 #endif /* _URCU_BP_STATIC_H */