libkern/gen/OSAtomicOperations.c

   1 /*
   2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #include <libkern/OSAtomic.h>
  30
  31 enum {
  32         false   = 0,
  33         true    = 1
  34 };
  35
  36 #ifndef NULL
  37 #define NULL ((void *)0)
  38 #endif
  39
  40 /*
  41  * atomic operations
  42  *      these are _the_ atomic operations, currently cast atop CompareAndSwap,
  43  *      which is implemented in assembler.      if we are worried about the cost of
  44  *      this layering (we shouldn't be), then all this stuff could be
  45  *      implemented in assembler, as it is in MacOS8/9
  46  *      (derived from SuperMario/NativeLibs/IO/DriverServices/Synchronization.s,
  47  *      which I wrote for NuKernel in a previous life with a different last name...)
  48  *
  49  * native Boolean       CompareAndSwap(UInt32 oldValue, UInt32 newValue, UInt32 * oldValuePtr);
  50  *
  51  * We've since implemented a few more of these -- OSAddAtomic, OSDequeueAtomic,
  52  * OSEnqueueAtomic etc -- in assembler, either for speed or correctness.  See also the
  53  * commpage atomic operations, and the platform specific versions.
  54  * Like standards, there are a lot of atomic ops to choose from!
  55  */
  56
  57 #if defined(__i386__) || defined(__x86_64__)
  58 /* Implemented in assembly for i386 and x86_64 */
  59 #else
  60 #error Unsupported arch
  61 #endif
  62
  63 #undef OSIncrementAtomic
  64 SInt32  OSIncrementAtomic(volatile SInt32 * value)
  65 {
  66         return OSAddAtomic(1, value);
  67 }
  68
  69 #undef OSDecrementAtomic
  70 SInt32  OSDecrementAtomic(volatile SInt32 * value)
  71 {
  72         return OSAddAtomic(-1, value);
  73 }
  74
  75 static UInt32   OSBitwiseAtomic(UInt32 and_mask, UInt32 or_mask, UInt32 xor_mask, volatile UInt32 * value)
  76 {
  77         UInt32  oldValue;
  78         UInt32  newValue;
  79
  80         do {
  81                 oldValue = *value;
  82                 newValue = ((oldValue & and_mask) | or_mask) ^ xor_mask;
  83         } while (! OSCompareAndSwap(oldValue, newValue, value));
  84
  85         return oldValue;
  86 }
  87
  88 #undef OSBitAndAtomic
  89 UInt32  OSBitAndAtomic(UInt32 mask, volatile UInt32 * value)
  90 {
  91         return OSBitwiseAtomic(mask, 0, 0, value);
  92 }
  93
  94 #undef OSBitOrAtomic
  95 UInt32  OSBitOrAtomic(UInt32 mask, volatile UInt32 * value)
  96 {
  97         return OSBitwiseAtomic((UInt32) -1, mask, 0, value);
  98 }
  99
 100 #undef OSBitXorAtomic
 101 UInt32  OSBitXorAtomic(UInt32 mask, volatile UInt32 * value)
 102 {
 103         return OSBitwiseAtomic((UInt32) -1, 0, mask, value);
 104 }
 105
 106 static Boolean OSCompareAndSwap8(UInt8 oldValue8, UInt8 newValue8, volatile UInt8 * value8)
 107 {
 108         UInt32                          mask            = 0x000000ff;
 109         UInt32                          alignment       = (UInt32)((unsigned long) value8) & (sizeof(UInt32) - 1);
 110         UInt32                          shiftValues = (24 << 24) | (16 << 16) | (8 << 8);
 111         int                                     shift           = (UInt32) *(((UInt8 *) &shiftValues) + alignment);
 112         volatile UInt32 *       value32         = (volatile UInt32 *) ((uintptr_t)value8 - alignment);
 113         UInt32                          oldValue;
 114         UInt32                          newValue;
 115
 116         mask <<= shift;
 117
 118         oldValue = *value32;
 119         oldValue = (oldValue & ~mask) | (oldValue8 << shift);
 120         newValue = (oldValue & ~mask) | (newValue8 << shift);
 121
 122         return OSCompareAndSwap(oldValue, newValue, value32);
 123 }
 124
 125 static Boolean  OSTestAndSetClear(UInt32 bit, Boolean wantSet, volatile UInt8 * startAddress)
 126 {
 127         UInt8           mask = 1;
 128         UInt8           oldValue;
 129         UInt8           wantValue;
 130
 131         startAddress += (bit / 8);
 132         mask <<= (7 - (bit % 8));
 133         wantValue = wantSet ? mask : 0;
 134
 135         do {
 136                 oldValue = *startAddress;
 137                 if ((oldValue & mask) == wantValue) {
 138                         break;
 139                 }
 140         } while (! OSCompareAndSwap8(oldValue, (oldValue & ~mask) | wantValue, startAddress));
 141
 142         return (oldValue & mask) == wantValue;
 143 }
 144
 145 Boolean OSTestAndSet(UInt32 bit, volatile UInt8 * startAddress)
 146 {
 147         return OSTestAndSetClear(bit, true, startAddress);
 148 }
 149
 150 Boolean OSTestAndClear(UInt32 bit, volatile UInt8 * startAddress)
 151 {
 152         return OSTestAndSetClear(bit, false, startAddress);
 153 }
 154
 155 /*
 156  * silly unaligned versions
 157  */
 158
 159 SInt8   OSIncrementAtomic8(volatile SInt8 * value)
 160 {
 161         return OSAddAtomic8(1, value);
 162 }
 163
 164 SInt8   OSDecrementAtomic8(volatile SInt8 * value)
 165 {
 166         return OSAddAtomic8(-1, value);
 167 }
 168
 169 SInt8   OSAddAtomic8(SInt32 amount, volatile SInt8 * value)
 170 {
 171         SInt8   oldValue;
 172         SInt8   newValue;
 173
 174         do {
 175                 oldValue = *value;
 176                 newValue = oldValue + amount;
 177         } while (! OSCompareAndSwap8((UInt8) oldValue, (UInt8) newValue, (volatile UInt8 *) value));
 178
 179         return oldValue;
 180 }
 181
 182 static UInt8    OSBitwiseAtomic8(UInt32 and_mask, UInt32 or_mask, UInt32 xor_mask, volatile UInt8 * value)
 183 {
 184         UInt8   oldValue;
 185         UInt8   newValue;
 186
 187         do {
 188                 oldValue = *value;
 189                 newValue = ((oldValue & and_mask) | or_mask) ^ xor_mask;
 190         } while (! OSCompareAndSwap8(oldValue, newValue, value));
 191
 192         return oldValue;
 193 }
 194
 195 UInt8   OSBitAndAtomic8(UInt32 mask, volatile UInt8 * value)
 196 {
 197         return OSBitwiseAtomic8(mask, 0, 0, value);
 198 }
 199
 200 UInt8   OSBitOrAtomic8(UInt32 mask, volatile UInt8 * value)
 201 {
 202         return OSBitwiseAtomic8((UInt32) -1, mask, 0, value);
 203 }
 204
 205 UInt8   OSBitXorAtomic8(UInt32 mask, volatile UInt8 * value)
 206 {
 207         return OSBitwiseAtomic8((UInt32) -1, 0, mask, value);
 208 }
 209
 210 static Boolean OSCompareAndSwap16(UInt16 oldValue16, UInt16 newValue16, volatile UInt16 * value16)
 211 {
 212         UInt32                          mask            = 0x0000ffff;
 213         UInt32                          alignment       = (UInt32)((unsigned long) value16) & (sizeof(UInt32) - 1);
 214         UInt32                          shiftValues = (16 << 24) | (16 << 16);
 215         UInt32                          shift           = (UInt32) *(((UInt8 *) &shiftValues) + alignment);
 216         volatile UInt32 *       value32         = (volatile UInt32 *) (((unsigned long) value16) - alignment);
 217         UInt32                          oldValue;
 218         UInt32                          newValue;
 219
 220         mask <<= shift;
 221
 222         oldValue = *value32;
 223         oldValue = (oldValue & ~mask) | (oldValue16 << shift);
 224         newValue = (oldValue & ~mask) | (newValue16 << shift);
 225
 226         return OSCompareAndSwap(oldValue, newValue, value32);
 227 }
 228
 229 SInt16  OSIncrementAtomic16(volatile SInt16 * value)
 230 {
 231         return OSAddAtomic16(1, value);
 232 }
 233
 234 SInt16  OSDecrementAtomic16(volatile SInt16 * value)
 235 {
 236         return OSAddAtomic16(-1, value);
 237 }
 238
 239 SInt16  OSAddAtomic16(SInt32 amount, volatile SInt16 * value)
 240 {
 241         SInt16  oldValue;
 242         SInt16  newValue;
 243
 244         do {
 245                 oldValue = *value;
 246                 newValue = oldValue + amount;
 247         } while (! OSCompareAndSwap16((UInt16) oldValue, (UInt16) newValue, (volatile UInt16 *) value));
 248
 249         return oldValue;
 250 }
 251
 252 static UInt16   OSBitwiseAtomic16(UInt32 and_mask, UInt32 or_mask, UInt32 xor_mask, volatile UInt16 * value)
 253 {
 254         UInt16  oldValue;
 255         UInt16  newValue;
 256
 257         do {
 258                 oldValue = *value;
 259                 newValue = ((oldValue & and_mask) | or_mask) ^ xor_mask;
 260         } while (! OSCompareAndSwap16(oldValue, newValue, value));
 261
 262         return oldValue;
 263 }
 264
 265 UInt16  OSBitAndAtomic16(UInt32 mask, volatile UInt16 * value)
 266 {
 267         return OSBitwiseAtomic16(mask, 0, 0, value);
 268 }
 269
 270 UInt16  OSBitOrAtomic16(UInt32 mask, volatile UInt16 * value)
 271 {
 272         return OSBitwiseAtomic16((UInt32) -1, mask, 0, value);
 273 }
 274
 275 UInt16  OSBitXorAtomic16(UInt32 mask, volatile UInt16 * value)
 276 {
 277         return OSBitwiseAtomic16((UInt32) -1, 0, mask, value);
 278 }
 279