include/urcu/static/urcu-bp.h

   1 // SPDX-FileCopyrightText: 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   2 // SPDX-FileCopyrightText: 2009 Paul E. McKenney, IBM Corporation.
   3 //
   4 // SPDX-License-Identifier: LGPL-2.1-or-later
   5
   6 #ifndef _URCU_BP_STATIC_H
   7 #define _URCU_BP_STATIC_H
   8
   9 /*
  10  * Userspace RCU header.
  11  *
  12  * TO BE INCLUDED ONLY IN CODE THAT IS TO BE RECOMPILED ON EACH LIBURCU
  13  * RELEASE. See urcu.h for linking dynamically with the userspace rcu library.
  14  *
  15  * IBM's contributions to this file may be relicensed under LGPLv2 or later.
  16  */
  17
  18 #include <stdlib.h>
  19 #include <pthread.h>
  20 #include <unistd.h>
  21
  22 #include <urcu/debug.h>
  23 #include <urcu/config.h>
  24 #include <urcu/compiler.h>
  25 #include <urcu/arch.h>
  26 #include <urcu/system.h>
  27 #include <urcu/uatomic.h>
  28 #include <urcu/list.h>
  29 #include <urcu/tls-compat.h>
  30
  31 /*
  32  * This code section can only be included in LGPL 2.1 compatible source code.
  33  * See below for the function call wrappers which can be used in code meant to
  34  * be only linked with the Userspace RCU library. This comes with a small
  35  * performance degradation on the read-side due to the added function calls.
  36  * This is required to permit relinking with newer versions of the library.
  37  */
  38
  39 #ifdef __cplusplus
  40 extern "C" {
  41 #endif
  42
  43 enum urcu_bp_state {
  44         URCU_BP_READER_ACTIVE_CURRENT,
  45         URCU_BP_READER_ACTIVE_OLD,
  46         URCU_BP_READER_INACTIVE,
  47 };
  48
  49 /*
  50  * The trick here is that URCU_BP_GP_CTR_PHASE must be a multiple of 8 so we can use a
  51  * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
  52  */
  53 #define URCU_BP_GP_COUNT                (1UL << 0)
  54 /* Use the amount of bits equal to half of the architecture long size */
  55 #define URCU_BP_GP_CTR_PHASE            (1UL << (sizeof(long) << 2))
  56 #define URCU_BP_GP_CTR_NEST_MASK        (URCU_BP_GP_CTR_PHASE - 1)
  57
  58 /*
  59  * Used internally by _urcu_bp_read_lock.
  60  */
  61 extern void urcu_bp_register(void);
  62
  63 struct urcu_bp_gp {
  64         /*
  65          * Global grace period counter.
  66          * Contains the current URCU_BP_GP_CTR_PHASE.
  67          * Also has a URCU_BP_GP_COUNT of 1, to accelerate the reader fast path.
  68          * Written to only by writer with mutex taken.
  69          * Read by both writer and readers.
  70          */
  71         unsigned long ctr;
  72 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
  73
  74 extern struct urcu_bp_gp urcu_bp_gp;
  75
  76 struct urcu_bp_reader {
  77         /* Data used by both reader and urcu_bp_synchronize_rcu() */
  78         unsigned long ctr;
  79         /* Data used for registry */
  80         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
  81         pthread_t tid;
  82         int alloc;      /* registry entry allocated */
  83 };
  84
  85 /*
  86  * Bulletproof version keeps a pointer to a registry not part of the TLS.
  87  * Adds a pointer dereference on the read-side, but won't require to unregister
  88  * the reader thread.
  89  */
  90 extern DECLARE_URCU_TLS(struct urcu_bp_reader *, urcu_bp_reader);
  91
  92 #ifdef CONFIG_RCU_FORCE_SYS_MEMBARRIER
  93 #define urcu_bp_has_sys_membarrier      1
  94 #else
  95 extern int urcu_bp_has_sys_membarrier;
  96 #endif
  97
  98 static inline void urcu_bp_smp_mb_slave(void)
  99 {
 100         if (caa_likely(urcu_bp_has_sys_membarrier))
 101                 cmm_barrier();
 102         else
 103                 cmm_smp_mb();
 104 }
 105
 106 static inline enum urcu_bp_state urcu_bp_reader_state(unsigned long *ctr)
 107 {
 108         unsigned long v;
 109
 110         if (ctr == NULL)
 111                 return URCU_BP_READER_INACTIVE;
 112         /*
 113          * Make sure both tests below are done on the same version of *value
 114          * to insure consistency.
 115          */
 116         v = CMM_LOAD_SHARED(*ctr);
 117         if (!(v & URCU_BP_GP_CTR_NEST_MASK))
 118                 return URCU_BP_READER_INACTIVE;
 119         if (!((v ^ urcu_bp_gp.ctr) & URCU_BP_GP_CTR_PHASE))
 120                 return URCU_BP_READER_ACTIVE_CURRENT;
 121         return URCU_BP_READER_ACTIVE_OLD;
 122 }
 123
 124 /*
 125  * Helper for _urcu_bp_read_lock().  The format of urcu_bp_gp.ctr (as well as
 126  * the per-thread rcu_reader.ctr) has the lower-order bits containing a count of
 127  * _urcu_bp_read_lock() nesting, and a single high-order URCU_BP_GP_CTR_PHASE bit
 128  * that contains either zero or one.  The smp_mb_slave() ensures that the accesses in
 129  * _urcu_bp_read_lock() happen before the subsequent read-side critical section.
 130  */
 131 static inline void _urcu_bp_read_lock_update(unsigned long tmp)
 132 {
 133         if (caa_likely(!(tmp & URCU_BP_GP_CTR_NEST_MASK))) {
 134                 _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, _CMM_LOAD_SHARED(urcu_bp_gp.ctr));
 135                 urcu_bp_smp_mb_slave();
 136         } else
 137                 _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, tmp + URCU_BP_GP_COUNT);
 138 }
 139
 140 /*
 141  * Enter an RCU read-side critical section.
 142  *
 143  * The first cmm_barrier() call ensures that the compiler does not reorder
 144  * the body of _urcu_bp_read_lock() with a mutex.
 145  *
 146  * This function and its helper are both less than 10 lines long.  The
 147  * intent is that this function meets the 10-line criterion in LGPL,
 148  * allowing this function to be invoked directly from non-LGPL code.
 149  */
 150 static inline void _urcu_bp_read_lock(void)
 151 {
 152         unsigned long tmp;
 153
 154         if (caa_unlikely(!URCU_TLS(urcu_bp_reader)))
 155                 urcu_bp_register(); /* If not yet registered. */
 156         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 157         tmp = URCU_TLS(urcu_bp_reader)->ctr;
 158         urcu_assert_debug((tmp & URCU_BP_GP_CTR_NEST_MASK) != URCU_BP_GP_CTR_NEST_MASK);
 159         _urcu_bp_read_lock_update(tmp);
 160 }
 161
 162 /*
 163  * Exit an RCU read-side critical section.  This function is less than
 164  * 10 lines of code, and is intended to be usable by non-LGPL code, as
 165  * called out in LGPL.
 166  */
 167 static inline void _urcu_bp_read_unlock(void)
 168 {
 169         unsigned long tmp;
 170
 171         tmp = URCU_TLS(urcu_bp_reader)->ctr;
 172         urcu_assert_debug(tmp & URCU_BP_GP_CTR_NEST_MASK);
 173         /* Finish using rcu before decrementing the pointer. */
 174         urcu_bp_smp_mb_slave();
 175         _CMM_STORE_SHARED(URCU_TLS(urcu_bp_reader)->ctr, tmp - URCU_BP_GP_COUNT);
 176         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 177 }
 178
 179 /*
 180  * Returns whether within a RCU read-side critical section.
 181  *
 182  * This function is less than 10 lines long.  The intent is that this
 183  * function meets the 10-line criterion for LGPL, allowing this function
 184  * to be invoked directly from non-LGPL code.
 185  */
 186 static inline int _urcu_bp_read_ongoing(void)
 187 {
 188         if (caa_unlikely(!URCU_TLS(urcu_bp_reader)))
 189                 urcu_bp_register(); /* If not yet registered. */
 190         return URCU_TLS(urcu_bp_reader)->ctr & URCU_BP_GP_CTR_NEST_MASK;
 191 }
 192
 193 #ifdef __cplusplus
 194 }
 195 #endif
 196
 197 #endif /* _URCU_BP_STATIC_H */