wtf/TCSpinLock.h

   1 // Copyright (c) 2005, 2006, Google Inc.
   2 // All rights reserved.
   3 //
   4 // Redistribution and use in source and binary forms, with or without
   5 // modification, are permitted provided that the following conditions are
   6 // met:
   7 //
   8 //     * Redistributions of source code must retain the above copyright
   9 // notice, this list of conditions and the following disclaimer.
  10 //     * Redistributions in binary form must reproduce the above
  11 // copyright notice, this list of conditions and the following disclaimer
  12 // in the documentation and/or other materials provided with the
  13 // distribution.
  14 //     * Neither the name of Google Inc. nor the names of its
  15 // contributors may be used to endorse or promote products derived from
  16 // this software without specific prior written permission.
  17 //
  18 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  21 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  22 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  23 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  24 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  28 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29
  30 // ---
  31 // Author: Sanjay Ghemawat <opensource@google.com>
  32
  33 #ifndef TCMALLOC_INTERNAL_SPINLOCK_H__
  34 #define TCMALLOC_INTERNAL_SPINLOCK_H__
  35
  36 #if (PLATFORM(X86) || PLATFORM(PPC)) && (COMPILER(GCC) || COMPILER(MSVC))
  37
  38 #include <time.h>       /* For nanosleep() */
  39
  40 #include <sched.h>      /* For sched_yield() */
  41
  42 #if HAVE(STDINT_H)
  43 #include <stdint.h>
  44 #elif HAVE(INTTYPES_H)
  45 #include <inttypes.h>
  46 #else
  47 #include <sys/types.h>
  48 #endif
  49 #include <stdlib.h>     /* for abort() */
  50
  51 #if COMPILER(MSVC)
  52 #ifndef WIN32_LEAN_AND_MEAN
  53 #define WIN32_LEAN_AND_MEAN
  54 #endif
  55 #include <windows.h>
  56 #endif
  57
  58 static void TCMalloc_SlowLock(volatile unsigned int* lockword);
  59
  60 // The following is a struct so that it can be initialized at compile time
  61 struct TCMalloc_SpinLock {
  62
  63   inline void Lock() {
  64     int r;
  65 #if COMPILER(GCC)
  66 #if PLATFORM(X86)
  67     __asm__ __volatile__
  68       ("xchgl %0, %1"
  69        : "=r"(r), "=m"(lockword_)
  70        : "0"(1), "m"(lockword_)
  71        : "memory");
  72 #else
  73     volatile unsigned int *lockword_ptr = &lockword_;
  74     __asm__ __volatile__
  75         ("1: lwarx %0, 0, %1\n\t"
  76          "stwcx. %2, 0, %1\n\t"
  77          "bne- 1b\n\t"
  78          "isync"
  79          : "=&r" (r), "=r" (lockword_ptr)
  80          : "r" (1), "1" (lockword_ptr)
  81          : "memory");
  82 #endif
  83 #elif COMPILER(MSVC)
  84     __asm {
  85         mov eax, this    ; store &lockword_ (which is this+0) in eax
  86         mov ebx, 1       ; store 1 in ebx
  87         xchg [eax], ebx  ; exchange lockword_ and 1
  88         mov r, ebx       ; store old value of lockword_ in r
  89     }
  90 #endif
  91     if (r) TCMalloc_SlowLock(&lockword_);
  92   }
  93
  94   inline void Unlock() {
  95 #if COMPILER(GCC)
  96 #if PLATFORM(X86)
  97     __asm__ __volatile__
  98       ("movl $0, %0"
  99        : "=m"(lockword_)
 100        : "m" (lockword_)
 101        : "memory");
 102 #else
 103     __asm__ __volatile__
 104       ("isync\n\t"
 105        "eieio\n\t"
 106        "stw %1, %0"
 107 #if PLATFORM(DARWIN)
 108        : "=o" (lockword_)
 109 #else
 110        : "=m" (lockword_)
 111 #endif
 112        : "r" (0)
 113        : "memory");
 114 #endif
 115 #elif COMPILER(MSVC)
 116       __asm {
 117           mov eax, this  ; store &lockword_ (which is this+0) in eax
 118           mov [eax], 0   ; set lockword_ to 0
 119       }
 120 #endif
 121   }
 122     // Report if we think the lock can be held by this thread.
 123     // When the lock is truly held by the invoking thread
 124     // we will always return true.
 125     // Indended to be used as CHECK(lock.IsHeld());
 126     inline bool IsHeld() const {
 127         return lockword_ != 0;
 128     }
 129
 130     inline void Init() { lockword_ = 0; }
 131
 132     volatile unsigned int lockword_;
 133 };
 134
 135 #define SPINLOCK_INITIALIZER { 0 }
 136
 137 static void TCMalloc_SlowLock(volatile unsigned int* lockword) {
 138   sched_yield();        // Yield immediately since fast path failed
 139   while (true) {
 140     int r;
 141 #if COMPILER(GCC)
 142 #if PLATFORM(X86)
 143     __asm__ __volatile__
 144       ("xchgl %0, %1"
 145        : "=r"(r), "=m"(*lockword)
 146        : "0"(1), "m"(*lockword)
 147        : "memory");
 148
 149 #else
 150     int tmp = 1;
 151     __asm__ __volatile__
 152         ("1: lwarx %0, 0, %1\n\t"
 153          "stwcx. %2, 0, %1\n\t"
 154          "bne- 1b\n\t"
 155          "isync"
 156          : "=&r" (r), "=r" (lockword)
 157          : "r" (tmp), "1" (lockword)
 158          : "memory");
 159 #endif
 160 #elif COMPILER(MSVC)
 161     __asm {
 162         mov eax, lockword     ; assign lockword into eax
 163         mov ebx, 1            ; assign 1 into ebx
 164         xchg [eax], ebx       ; exchange *lockword and 1
 165         mov r, ebx            ; store old value of *lockword in r
 166     }
 167 #endif
 168     if (!r) {
 169       return;
 170     }
 171
 172     // This code was adapted from the ptmalloc2 implementation of
 173     // spinlocks which would sched_yield() upto 50 times before
 174     // sleeping once for a few milliseconds.  Mike Burrows suggested
 175     // just doing one sched_yield() outside the loop and always
 176     // sleeping after that.  This change helped a great deal on the
 177     // performance of spinlocks under high contention.  A test program
 178     // with 10 threads on a dual Xeon (four virtual processors) went
 179     // from taking 30 seconds to 16 seconds.
 180
 181     // Sleep for a few milliseconds
 182 #if COMPILER(MSVC)
 183     Sleep(2);
 184 #else
 185     struct timespec tm;
 186     tm.tv_sec = 0;
 187     tm.tv_nsec = 2000001;
 188     nanosleep(&tm, NULL);
 189 #endif
 190   }
 191 }
 192
 193 #else
 194
 195 #include <pthread.h>
 196
 197 // Portable version
 198 struct TCMalloc_SpinLock {
 199   pthread_mutex_t private_lock_;
 200
 201   inline void Init() {
 202     if (pthread_mutex_init(&private_lock_, NULL) != 0) abort();
 203   }
 204   inline void Finalize() {
 205     if (pthread_mutex_destroy(&private_lock_) != 0) abort();
 206   }
 207   inline void Lock() {
 208     if (pthread_mutex_lock(&private_lock_) != 0) abort();
 209   }
 210   inline void Unlock() {
 211     if (pthread_mutex_unlock(&private_lock_) != 0) abort();
 212   }
 213 };
 214
 215 #define SPINLOCK_INITIALIZER { PTHREAD_MUTEX_INITIALIZER }
 216
 217 #endif
 218
 219 // Corresponding locker object that arranges to acquire a spinlock for
 220 // the duration of a C++ scope.
 221 class TCMalloc_SpinLockHolder {
 222  private:
 223   TCMalloc_SpinLock* lock_;
 224  public:
 225   inline explicit TCMalloc_SpinLockHolder(TCMalloc_SpinLock* l)
 226     : lock_(l) { l->Lock(); }
 227   inline ~TCMalloc_SpinLockHolder() { lock_->Unlock(); }
 228 };
 229
 230 // Short-hands for convenient use by tcmalloc.cc
 231 typedef TCMalloc_SpinLock SpinLock;
 232 typedef TCMalloc_SpinLockHolder SpinLockHolder;
 233
 234 #endif  // TCMALLOC_INTERNAL_SPINLOCK_H__