aboutsummaryrefslogtreecommitdiff
path: root/port/atomic_pointer.h
blob: a9866b2302fd42dc889e5cf49869727e85658a8b (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.

// AtomicPointer provides storage for a lock-free pointer.
// Platform-dependent implementation of AtomicPointer:
// - If the platform provides a cheap barrier, we use it with raw pointers
// - If cstdatomic is present (on newer versions of gcc, it is), we use
//   a cstdatomic-based AtomicPointer.  However we prefer the memory
//   barrier based version, because at least on a gcc 4.4 32-bit build
//   on linux, we have encountered a buggy <cstdatomic>
//   implementation.  Also, some <cstdatomic> implementations are much
//   slower than a memory-barrier based implementation (~16ns for
//   <cstdatomic> based acquire-load vs. ~1ns for a barrier based
//   acquire-load).
// This code is based on atomicops-internals-* in Google's perftools:
// http://code.google.com/p/google-perftools/source/browse/#svn%2Ftrunk%2Fsrc%2Fbase

#ifndef PORT_ATOMIC_POINTER_H_
#define PORT_ATOMIC_POINTER_H_

#include <stdint.h>
#ifdef LEVELDB_CSTDATOMIC_PRESENT
#include <cstdatomic>
#endif
#ifdef OS_WIN
#include <windows.h>
#endif
#ifdef OS_MACOSX
#include <libkern/OSAtomic.h>
#endif

#if defined(_M_X64) || defined(__x86_64__)
#define ARCH_CPU_X86_FAMILY 1
#elif defined(_M_IX86) || defined(__i386__) || defined(__i386)
#define ARCH_CPU_X86_FAMILY 1
#elif defined(__ARMEL__)
#define ARCH_CPU_ARM_FAMILY 1
#elif defined(__ppc__) || defined(__powerpc__) || defined(__powerpc64__)
#define ARCH_CPU_PPC_FAMILY 1
#endif

namespace leveldb {
namespace port {

// Define MemoryBarrier() if available
// Windows on x86
#if defined(OS_WIN) && defined(COMPILER_MSVC) && defined(ARCH_CPU_X86_FAMILY)
// windows.h already provides a MemoryBarrier(void) macro
// http://msdn.microsoft.com/en-us/library/ms684208(v=vs.85).aspx
#define LEVELDB_HAVE_MEMORY_BARRIER

// Mac OS
#elif defined(OS_MACOSX)
inline void MemoryBarrier() {
  OSMemoryBarrier();
}
#define LEVELDB_HAVE_MEMORY_BARRIER

// Gcc on x86
#elif defined(ARCH_CPU_X86_FAMILY) && defined(__GNUC__)
inline void MemoryBarrier() {
  // See http://gcc.gnu.org/ml/gcc/2003-04/msg01180.html for a discussion on
  // this idiom. Also see http://en.wikipedia.org/wiki/Memory_ordering.
  __asm__ __volatile__("" : : : "memory");
}
#define LEVELDB_HAVE_MEMORY_BARRIER

// Sun Studio
#elif defined(ARCH_CPU_X86_FAMILY) && defined(__SUNPRO_CC)
inline void MemoryBarrier() {
  // See http://gcc.gnu.org/ml/gcc/2003-04/msg01180.html for a discussion on
  // this idiom. Also see http://en.wikipedia.org/wiki/Memory_ordering.
  asm volatile("" : : : "memory");
}
#define LEVELDB_HAVE_MEMORY_BARRIER

// ARM Linux
#elif defined(ARCH_CPU_ARM_FAMILY) && defined(__linux__)
typedef void (*LinuxKernelMemoryBarrierFunc)(void);
// The Linux ARM kernel provides a highly optimized device-specific memory
// barrier function at a fixed memory address that is mapped in every
// user-level process.
//
// This beats using CPU-specific instructions which are, on single-core
// devices, un-necessary and very costly (e.g. ARMv7-A "dmb" takes more
// than 180ns on a Cortex-A8 like the one on a Nexus One). Benchmarking
// shows that the extra function call cost is completely negligible on
// multi-core devices.
//
inline void MemoryBarrier() {
  (*(LinuxKernelMemoryBarrierFunc)0xffff0fa0)();
}
#define LEVELDB_HAVE_MEMORY_BARRIER

// PPC
#elif defined(ARCH_CPU_PPC_FAMILY) && defined(__GNUC__)
inline void MemoryBarrier() {
  // TODO for some powerpc expert: is there a cheaper suitable variant?
  // Perhaps by having separate barriers for acquire and release ops.
  asm volatile("sync" : : : "memory");
}
#define LEVELDB_HAVE_MEMORY_BARRIER

#endif

// AtomicPointer built using platform-specific MemoryBarrier()
#if defined(LEVELDB_HAVE_MEMORY_BARRIER)
class AtomicPointer {
 private:
  void* rep_;
 public:
  AtomicPointer() { }
  explicit AtomicPointer(void* p) : rep_(p) {}
  inline void* NoBarrier_Load() const { return rep_; }
  inline void NoBarrier_Store(void* v) { rep_ = v; }
  inline void* Acquire_Load() const {
    void* result = rep_;
    MemoryBarrier();
    return result;
  }
  inline void Release_Store(void* v) {
    MemoryBarrier();
    rep_ = v;
  }
};

// AtomicPointer based on <cstdatomic>
#elif defined(LEVELDB_CSTDATOMIC_PRESENT)
class AtomicPointer {
 private:
  std::atomic<void*> rep_;
 public:
  AtomicPointer() { }
  explicit AtomicPointer(void* v) : rep_(v) { }
  inline void* Acquire_Load() const {
    return rep_.load(std::memory_order_acquire);
  }
  inline void Release_Store(void* v) {
    rep_.store(v, std::memory_order_release);
  }
  inline void* NoBarrier_Load() const {
    return rep_.load(std::memory_order_relaxed);
  }
  inline void NoBarrier_Store(void* v) {
    rep_.store(v, std::memory_order_relaxed);
  }
};

// Atomic pointer based on sparc memory barriers
#elif defined(__sparcv9) && defined(__GNUC__)
class AtomicPointer {
 private:
  void* rep_;
 public:
  AtomicPointer() { }
  explicit AtomicPointer(void* v) : rep_(v) { }
  inline void* Acquire_Load() const {
    void* val;
    __asm__ __volatile__ (
        "ldx [%[rep_]], %[val] \n\t"
         "membar #LoadLoad|#LoadStore \n\t"
        : [val] "=r" (val)
        : [rep_] "r" (&rep_)
        : "memory");
    return val;
  }
  inline void Release_Store(void* v) {
    __asm__ __volatile__ (
        "membar #LoadStore|#StoreStore \n\t"
        "stx %[v], [%[rep_]] \n\t"
        :
        : [rep_] "r" (&rep_), [v] "r" (v)
        : "memory");
  }
  inline void* NoBarrier_Load() const { return rep_; }
  inline void NoBarrier_Store(void* v) { rep_ = v; }
};

// Atomic pointer based on ia64 acq/rel
#elif defined(__ia64) && defined(__GNUC__)
class AtomicPointer {
 private:
  void* rep_;
 public:
  AtomicPointer() { }
  explicit AtomicPointer(void* v) : rep_(v) { }
  inline void* Acquire_Load() const {
    void* val    ;
    __asm__ __volatile__ (
        "ld8.acq %[val] = [%[rep_]] \n\t"
        : [val] "=r" (val)
        : [rep_] "r" (&rep_)
        : "memory"
        );
    return val;
  }
  inline void Release_Store(void* v) {
    __asm__ __volatile__ (
        "st8.rel [%[rep_]] = %[v]  \n\t"
        :
        : [rep_] "r" (&rep_), [v] "r" (v)
        : "memory"
        );
  }
  inline void* NoBarrier_Load() const { return rep_; }
  inline void NoBarrier_Store(void* v) { rep_ = v; }
};

// We have neither MemoryBarrier(), nor <cstdatomic>
#else
#error Please implement AtomicPointer for this platform.

#endif

#undef LEVELDB_HAVE_MEMORY_BARRIER
#undef ARCH_CPU_X86_FAMILY
#undef ARCH_CPU_ARM_FAMILY
#undef ARCH_CPU_PPC_FAMILY

}  // namespace port
}  // namespace leveldb

#endif  // PORT_ATOMIC_POINTER_H_