[176] | 1 | /* -*- C++ -*- */ |
---|
| 2 | |
---|
| 3 | /* |
---|
| 4 | |
---|
| 5 | Heap Layers: An Extensible Memory Allocation Infrastructure |
---|
| 6 | |
---|
| 7 | Copyright (C) 2000-2006 by Emery Berger |
---|
| 8 | http://www.cs.umass.edu/~emery |
---|
| 9 | emery@cs.umass.edu |
---|
| 10 | |
---|
| 11 | This program is free software; you can redistribute it and/or modify |
---|
| 12 | it under the terms of the GNU General Public License as published by |
---|
| 13 | the Free Software Foundation; either version 2 of the License, or |
---|
| 14 | (at your option) any later version. |
---|
| 15 | |
---|
| 16 | This program is distributed in the hope that it will be useful, |
---|
| 17 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
| 18 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
---|
| 19 | GNU General Public License for more details. |
---|
| 20 | |
---|
| 21 | You should have received a copy of the GNU General Public License |
---|
| 22 | along with this program; if not, write to the Free Software |
---|
| 23 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
---|
| 24 | |
---|
| 25 | */ |
---|
| 26 | |
---|
| 27 | #ifndef _SPINLOCK_H_ |
---|
| 28 | #define _SPINLOCK_H_ |
---|
| 29 | |
---|
| 30 | #if defined(unix) |
---|
| 31 | #include <sched.h> |
---|
| 32 | #endif |
---|
| 33 | |
---|
| 34 | #if defined(__SVR4) |
---|
| 35 | #include <thread.h> |
---|
| 36 | #endif |
---|
| 37 | |
---|
| 38 | #if defined(__sgi) |
---|
| 39 | #include <mutex.h> |
---|
| 40 | #endif |
---|
| 41 | |
---|
| 42 | #if defined(__APPLE__) |
---|
| 43 | #include <libkern/OSAtomic.h> |
---|
| 44 | #endif |
---|
| 45 | |
---|
| 46 | #include "hldefines.h" |
---|
| 47 | #include "cpuinfo.h" |
---|
| 48 | |
---|
| 49 | #if defined(_MSC_VER) |
---|
| 50 | |
---|
| 51 | #if !defined(NO_INLINE) |
---|
| 52 | #pragma inline_depth(255) |
---|
| 53 | #define NO_INLINE __declspec(noinline) |
---|
| 54 | #define INLINE __forceinline |
---|
| 55 | #define inline __forceinline |
---|
| 56 | #endif // !defined(NO_INLINE) |
---|
| 57 | |
---|
| 58 | #else |
---|
| 59 | |
---|
| 60 | #endif // defined(_MSC_VER) |
---|
| 61 | |
---|
| 62 | |
---|
| 63 | #if defined(__SUNPRO_CC) |
---|
| 64 | // x86-interchange.il, x86_64-interchange.il contributed by Markus Bernhardt. |
---|
| 65 | extern "C" unsigned long MyInterlockedExchange (unsigned long * oldval, |
---|
| 66 | unsigned long newval); |
---|
| 67 | #endif |
---|
| 68 | |
---|
| 69 | #if defined(_WIN32) && !defined(_WIN64) |
---|
| 70 | #define _WIN32_WINNT 0x0500 |
---|
| 71 | |
---|
| 72 | // NOTE: Below is the new "pause" instruction, which is inocuous for |
---|
| 73 | // previous architectures, but crucial for Intel chips with |
---|
| 74 | // hyperthreading. See |
---|
| 75 | // http://www.usenix.org/events/wiess02/tech/full_papers/nakajima/nakajima.pdf |
---|
| 76 | // for discussion. |
---|
| 77 | |
---|
| 78 | #define _MM_PAUSE {__asm{_emit 0xf3};__asm {_emit 0x90}} |
---|
| 79 | #include <windows.h> |
---|
| 80 | #else |
---|
| 81 | #define _MM_PAUSE |
---|
| 82 | #endif // defined(_WIN32) && !defined(_WIN64) |
---|
| 83 | |
---|
| 84 | extern volatile int anyThreadCreated; |
---|
| 85 | |
---|
| 86 | namespace HL { |
---|
| 87 | |
---|
| 88 | class SpinLockType { |
---|
| 89 | private: |
---|
| 90 | |
---|
| 91 | enum { UNLOCKED = 0, LOCKED = 1 }; |
---|
| 92 | |
---|
| 93 | public: |
---|
| 94 | |
---|
| 95 | SpinLockType (void) |
---|
| 96 | #if defined(__APPLE__) |
---|
| 97 | : mutex (OS_SPINLOCK_INIT) |
---|
| 98 | #else |
---|
| 99 | : mutex (UNLOCKED) |
---|
| 100 | #endif |
---|
| 101 | {} |
---|
| 102 | |
---|
| 103 | ~SpinLockType (void) |
---|
| 104 | {} |
---|
| 105 | |
---|
| 106 | inline void lock (void) { |
---|
| 107 | if (anyThreadCreated) { |
---|
| 108 | #if defined(__APPLE__) |
---|
| 109 | OSSpinLockLock(&mutex); |
---|
| 110 | #else |
---|
| 111 | // A yielding lock (with an initial spin). |
---|
| 112 | if (MyInterlockedExchange (const_cast<unsigned long *>(&mutex), LOCKED) |
---|
| 113 | != UNLOCKED) { |
---|
| 114 | contendedLock(); |
---|
| 115 | } |
---|
| 116 | #endif |
---|
| 117 | } else { |
---|
| 118 | mutex = LOCKED; |
---|
| 119 | } |
---|
| 120 | } |
---|
| 121 | |
---|
| 122 | |
---|
| 123 | inline void unlock (void) { |
---|
| 124 | #if defined(__APPLE__) |
---|
| 125 | OSSpinLockUnlock(&mutex); |
---|
| 126 | #else |
---|
| 127 | if (anyThreadCreated) { |
---|
| 128 | #if defined(_WIN32) && !defined(_WIN64) |
---|
| 129 | __asm {} |
---|
| 130 | #elif defined(__GNUC__) |
---|
| 131 | asm volatile ("" : : : "memory"); |
---|
| 132 | #endif |
---|
| 133 | } |
---|
| 134 | mutex = UNLOCKED; |
---|
| 135 | #endif |
---|
| 136 | } |
---|
| 137 | |
---|
| 138 | |
---|
| 139 | #if !defined(__SUNPRO_CC) |
---|
| 140 | inline static volatile unsigned long MyInterlockedExchange (unsigned long *,unsigned long); |
---|
| 141 | #endif |
---|
| 142 | |
---|
| 143 | private: |
---|
| 144 | |
---|
| 145 | #if defined(__APPLE__) |
---|
| 146 | OSSpinLock mutex; |
---|
| 147 | |
---|
| 148 | #else |
---|
| 149 | |
---|
| 150 | NO_INLINE |
---|
| 151 | void contendedLock (void) { |
---|
| 152 | while (1) { |
---|
| 153 | if (MyInterlockedExchange (const_cast<unsigned long *>(&mutex), LOCKED) |
---|
| 154 | == UNLOCKED) { |
---|
| 155 | return; |
---|
| 156 | } |
---|
| 157 | while (mutex == LOCKED) { |
---|
| 158 | _MM_PAUSE; |
---|
| 159 | yieldProcessor(); |
---|
| 160 | } |
---|
| 161 | } |
---|
| 162 | } |
---|
| 163 | |
---|
| 164 | // Is this system a multiprocessor? |
---|
| 165 | inline bool onMultiprocessor (void) { |
---|
| 166 | static CPUInfo cpuInfo; |
---|
| 167 | return (cpuInfo.getNumProcessors() > 1); |
---|
| 168 | } |
---|
| 169 | |
---|
| 170 | inline void yieldProcessor (void) { |
---|
| 171 | #if defined(_WIN32) |
---|
| 172 | Sleep(0); |
---|
| 173 | #elif defined(__SVR4) |
---|
| 174 | thr_yield(); |
---|
| 175 | #else |
---|
| 176 | sched_yield(); |
---|
| 177 | #endif |
---|
| 178 | } |
---|
| 179 | |
---|
| 180 | enum { MAX_SPIN_LIMIT = 1024 }; |
---|
| 181 | |
---|
| 182 | union { |
---|
| 183 | volatile unsigned long mutex; |
---|
| 184 | }; |
---|
| 185 | #endif |
---|
| 186 | |
---|
| 187 | }; |
---|
| 188 | |
---|
| 189 | } |
---|
| 190 | |
---|
| 191 | // Atomically: |
---|
| 192 | // retval = *oldval; |
---|
| 193 | // *oldval = newval; |
---|
| 194 | // return retval; |
---|
| 195 | |
---|
| 196 | #if !defined(__SUNPRO_CC) && !defined(__APPLE__) |
---|
| 197 | inline volatile unsigned long |
---|
| 198 | HL::SpinLockType::MyInterlockedExchange (unsigned long * oldval, |
---|
| 199 | unsigned long newval) |
---|
| 200 | { |
---|
| 201 | #if defined(_WIN32) && defined(_MSC_VER) |
---|
| 202 | return InterlockedExchange ((volatile LONG *) oldval, newval); |
---|
| 203 | |
---|
| 204 | #elif defined(__sparc) |
---|
| 205 | asm volatile ("swap [%1],%0" |
---|
| 206 | :"=r" (newval) |
---|
| 207 | :"r" (oldval), "0" (newval) |
---|
| 208 | : "memory"); |
---|
| 209 | |
---|
| 210 | #elif defined(__i386__) |
---|
| 211 | asm volatile ("lock; xchgl %0, %1" |
---|
| 212 | : "=r" (newval) |
---|
| 213 | : "m" (*oldval), "0" (newval) |
---|
| 214 | : "memory"); |
---|
| 215 | |
---|
| 216 | #elif defined(__sgi) |
---|
| 217 | newval = test_and_set (oldval, newval); |
---|
| 218 | |
---|
| 219 | #elif defined(__x86_64__) |
---|
| 220 | // Contributed by Kurt Roeckx. |
---|
| 221 | asm volatile ("lock; xchgq %0, %1" |
---|
| 222 | : "=r" (newval) |
---|
| 223 | : "m" (*oldval), "0" (newval) |
---|
| 224 | : "memory"); |
---|
| 225 | |
---|
| 226 | #elif defined(__ppc) || defined(__powerpc__) || defined(PPC) |
---|
| 227 | // PPC assembly contributed by Maged Michael. |
---|
| 228 | int ret; |
---|
| 229 | asm volatile ( |
---|
| 230 | "La..%=0: lwarx %0,0,%1 ;" |
---|
| 231 | " cmpw %0,%2;" |
---|
| 232 | " beq La..%=1;" |
---|
| 233 | " stwcx. %2,0,%1;" |
---|
| 234 | " bne- La..%=0;" |
---|
| 235 | "La..%=1: isync;" |
---|
| 236 | : "=&r"(ret) |
---|
| 237 | : "r"(oldval), "r"(newval) |
---|
| 238 | : "cr0", "memory"); |
---|
| 239 | return ret; |
---|
| 240 | |
---|
| 241 | #elif defined(__arm__) |
---|
| 242 | // Contributed by Bo Granlund. |
---|
| 243 | long result; |
---|
| 244 | asm volatile ( |
---|
| 245 | "\n\t" |
---|
| 246 | "swp %0,%2,[%1] \n\t" |
---|
| 247 | "" |
---|
| 248 | : "=&r"(result) |
---|
| 249 | : "r"(oldval), "r"(newval) |
---|
| 250 | : "memory"); |
---|
| 251 | return (result); |
---|
| 252 | #else |
---|
| 253 | #error "No spin lock implementation is available for this platform." |
---|
| 254 | #endif |
---|
| 255 | return newval; |
---|
| 256 | } |
---|
| 257 | |
---|
| 258 | |
---|
| 259 | #endif |
---|
| 260 | |
---|
| 261 | #endif // _SPINLOCK_H_ |
---|