urcu-defer-static.h

   1 #ifndef _URCU_DEFER_STATIC_H
   2 #define _URCU_DEFER_STATIC_H
   3
   4 /*
   5  * urcu-defer-static.h
   6  *
   7  * Userspace RCU header - memory reclamation.
   8  *
   9  * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See urcu-defer.h for linking
  10  * dynamically with the userspace rcu reclamation 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
  35 #include <compiler.h>
  36 #include <arch.h>
  37
  38
  39 /*
  40  * Number of entries in the per-thread defer queue. Must be power of 2.
  41  */
  42 #define DEFER_QUEUE_SIZE        (1 << 12)
  43 #define DEFER_QUEUE_MASK        (DEFER_QUEUE_SIZE - 1)
  44
  45 /*
  46  * Typically, data is aligned at least on the architecture size.
  47  * Use lowest bit to indicate that the current callback is changing.
  48  * Assumes that (void *)-2L is not used often. Used to encode non-aligned
  49  * functions and non-aligned data using extra space.
  50  * We encode the (void *)-2L fct as: -2L, fct, data.
  51  * We encode the (void *)-2L data as: -2L, fct, data.
  52  * Here, DQ_FCT_MARK == ~DQ_FCT_BIT. Required for the test order.
  53  */
  54 #define DQ_FCT_BIT              (1 << 0)
  55 #define DQ_IS_FCT_BIT(x)        ((unsigned long)(x) & DQ_FCT_BIT)
  56 #define DQ_SET_FCT_BIT(x)       \
  57         (x = (void *)((unsigned long)(x) | DQ_FCT_BIT))
  58 #define DQ_CLEAR_FCT_BIT(x)     \
  59         (x = (void *)((unsigned long)(x) & ~DQ_FCT_BIT))
  60 #define DQ_FCT_MARK             ((void *)(~DQ_FCT_BIT))
  61
  62 /*
  63  * Identify a shared load. A smp_rmc() or smp_mc() should come before the load.
  64  */
  65 #define _LOAD_SHARED(p)        ACCESS_ONCE(p)
  66
  67 /*
  68  * Load a data from shared memory, doing a cache flush if required.
  69  */
  70 #define LOAD_SHARED(p)                  \
  71         ({                              \
  72                 smp_rmc();              \
  73                 _LOAD_SHARED(p);        \
  74         })
  75
  76 /*
  77  * Identify a shared store. A smp_wmc() or smp_mc() should follow the store.
  78  */
  79 #define _STORE_SHARED(x, v)     ({ ACCESS_ONCE(x) = (v); })
  80
  81 /*
  82  * Store v into x, where x is located in shared memory. Performs the required
  83  * cache flush after writing. Returns v.
  84  */
  85 #define STORE_SHARED(x, v)              \
  86         ({                              \
  87                 _STORE_SHARED(x, v);    \
  88                 smp_wmc();              \
  89                 (v);                    \
  90         })
  91
  92 /*
  93  * This code section can only be included in LGPL 2.1 compatible source code.
  94  * See below for the function call wrappers which can be used in code meant to
  95  * be only linked with the Userspace RCU library. This comes with a small
  96  * performance degradation on the read-side due to the added function calls.
  97  * This is required to permit relinking with newer versions of the library.
  98  */
  99
 100 #ifdef DEBUG_RCU
 101 #define rcu_assert(args...)     assert(args)
 102 #else
 103 #define rcu_assert(args...)
 104 #endif
 105
 106 /*
 107  * defer queue.
 108  * Contains pointers. Encoded to save space when same callback is often used.
 109  * When looking up the next item:
 110  * - if DQ_FCT_BIT is set, set the current callback to DQ_CLEAR_FCT_BIT(ptr)
 111  *   - next element contains pointer to data.
 112  * - else if item == DQ_FCT_MARK
 113  *   - set the current callback to next element ptr
 114  *   - following next element contains pointer to data.
 115  * - else current element contains data
 116  */
 117 struct defer_queue {
 118         unsigned long head;     /* add element at head */
 119         void *last_fct_in;      /* last fct pointer encoded */
 120         unsigned long tail;     /* next element to remove at tail */
 121         void *last_fct_out;     /* last fct pointer encoded */
 122         void **q;
 123 };
 124
 125 extern struct defer_queue __thread defer_queue;
 126
 127 extern void rcu_defer_barrier_thread(void);
 128
 129 /*
 130  * not signal-safe.
 131  */
 132 static inline void _rcu_defer_queue(void (*fct)(void *p), void *p)
 133 {
 134         unsigned long head, tail;
 135
 136         /*
 137          * Head is only modified by ourself. Tail can be modified by reclamation
 138          * thread.
 139          */
 140         head = defer_queue.head;
 141         tail = LOAD_SHARED(defer_queue.tail);
 142
 143         /*
 144          * If queue is full, empty it ourself.
 145          * Worse-case: must allow 2 supplementary entries for fct pointer.
 146          */
 147         if (unlikely(head - tail >= DEFER_QUEUE_SIZE - 2)) {
 148                 assert(head - tail <= DEFER_QUEUE_SIZE);
 149                 rcu_defer_barrier_thread();
 150                 assert(head - LOAD_SHARED(defer_queue.tail) == 0);
 151         }
 152
 153         if (unlikely(defer_queue.last_fct_in != fct)) {
 154                 defer_queue.last_fct_in = fct;
 155                 if (unlikely(DQ_IS_FCT_BIT(fct) || fct == DQ_FCT_MARK)) {
 156                         /*
 157                          * If the function to encode is not aligned or the
 158                          * marker, write DQ_FCT_MARK followed by the function
 159                          * pointer.
 160                          */
 161                         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
 162                                       DQ_FCT_MARK);
 163                         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
 164                                       fct);
 165                 } else {
 166                         DQ_SET_FCT_BIT(fct);
 167                         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
 168                                       fct);
 169                 }
 170         } else {
 171                 if (unlikely(DQ_IS_FCT_BIT(p) || p == DQ_FCT_MARK)) {
 172                         /*
 173                          * If the data to encode is not aligned or the marker,
 174                          * write DQ_FCT_MARK followed by the function pointer.
 175                          */
 176                         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
 177                                       DQ_FCT_MARK);
 178                         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
 179                                       fct);
 180                 }
 181         }
 182         _STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK], p);
 183         smp_wmb();      /* Publish new pointer before head */
 184                         /* Write q[] before head. */
 185         STORE_SHARED(defer_queue.head, head);
 186 }
 187
 188 #endif /* _URCU_DEFER_STATIC_H */