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
|
#ifndef Py_ATOMIC_H
#define Py_ATOMIC_H
/* XXX: When compilers start offering a stdatomic.h with lock-free
atomic_int and atomic_address types, include that here and rewrite
the atomic operations in terms of it. */
#include "dynamic_annotations.h"
#ifdef __cplusplus
extern "C" {
#endif
/* This is modeled after the atomics interface from C1x, according to
* the draft at
* http://www.open-std.org/JTC1/SC22/wg14/www/docs/n1425.pdf.
* Operations and types are named the same except with a _Py_ prefix
* and have the same semantics.
*
* Beware, the implementations here are deep magic.
*/
typedef enum _Py_memory_order {
_Py_memory_order_relaxed,
_Py_memory_order_acquire,
_Py_memory_order_release,
_Py_memory_order_acq_rel,
_Py_memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
void *_value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
int _value;
} _Py_atomic_int;
/* Only support GCC (for expression statements) and x86 (for simple
* atomic semantics) for now */
#if defined(__GNUC__) && (defined(__i386__) || defined(__amd64))
static __inline__ void
_Py_atomic_signal_fence(_Py_memory_order order)
{
if (order != _Py_memory_order_relaxed)
__asm__ volatile("":::"memory");
}
static __inline__ void
_Py_atomic_thread_fence(_Py_memory_order order)
{
if (order != _Py_memory_order_relaxed)
__asm__ volatile("mfence":::"memory");
}
/* Tell the race checker about this operation's effects. */
static __inline__ void
_Py_ANNOTATE_MEMORY_ORDER(const volatile void *address, _Py_memory_order order)
{
switch(order) {
case _Py_memory_order_release:
case _Py_memory_order_acq_rel:
case _Py_memory_order_seq_cst:
_Py_ANNOTATE_HAPPENS_BEFORE(address);
break;
default:
break;
}
switch(order) {
case _Py_memory_order_acquire:
case _Py_memory_order_acq_rel:
case _Py_memory_order_seq_cst:
_Py_ANNOTATE_HAPPENS_AFTER(address);
break;
default:
break;
}
}
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
__extension__ ({ \
__typeof__(ATOMIC_VAL) atomic_val = ATOMIC_VAL; \
__typeof__(atomic_val->_value) new_val = NEW_VAL;\
volatile __typeof__(new_val) *volatile_data = &atomic_val->_value; \
_Py_memory_order order = ORDER; \
_Py_ANNOTATE_MEMORY_ORDER(atomic_val, order); \
\
/* Perform the operation. */ \
_Py_ANNOTATE_IGNORE_WRITES_BEGIN(); \
switch(order) { \
case _Py_memory_order_release: \
_Py_atomic_signal_fence(_Py_memory_order_release); \
/* fallthrough */ \
case _Py_memory_order_relaxed: \
*volatile_data = new_val; \
break; \
\
case _Py_memory_order_acquire: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
__asm__ volatile("xchg %0, %1" \
: "+r"(new_val) \
: "m"(atomic_val->_value) \
: "memory"); \
break; \
} \
_Py_ANNOTATE_IGNORE_WRITES_END(); \
})
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
__extension__ ({ \
__typeof__(ATOMIC_VAL) atomic_val = ATOMIC_VAL; \
__typeof__(atomic_val->_value) result; \
volatile __typeof__(result) *volatile_data = &atomic_val->_value; \
_Py_memory_order order = ORDER; \
_Py_ANNOTATE_MEMORY_ORDER(atomic_val, order); \
\
/* Perform the operation. */ \
_Py_ANNOTATE_IGNORE_READS_BEGIN(); \
switch(order) { \
case _Py_memory_order_release: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
/* Loads on x86 are not releases by default, so need a */ \
/* thread fence. */ \
_Py_atomic_thread_fence(_Py_memory_order_release); \
break; \
default: \
/* No fence */ \
break; \
} \
result = *volatile_data; \
switch(order) { \
case _Py_memory_order_acquire: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
/* Loads on x86 are automatically acquire operations so */ \
/* can get by with just a compiler fence. */ \
_Py_atomic_signal_fence(_Py_memory_order_acquire); \
break; \
default: \
/* No fence */ \
break; \
} \
_Py_ANNOTATE_IGNORE_READS_END(); \
result; \
})
#else /* !gcc x86 */
/* Fall back to other compilers and processors by assuming that simple
volatile accesses are atomic. This is false, so people should port
this. */
#define _Py_atomic_signal_fence(/*memory_order*/ ORDER) ((void)0)
#define _Py_atomic_thread_fence(/*memory_order*/ ORDER) ((void)0)
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
((ATOMIC_VAL)->_value = NEW_VAL)
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
((ATOMIC_VAL)->_value)
#endif /* !gcc x86 */
/* Standardized shortcuts. */
#define _Py_atomic_store(ATOMIC_VAL, NEW_VAL) \
_Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, _Py_memory_order_seq_cst)
#define _Py_atomic_load(ATOMIC_VAL) \
_Py_atomic_load_explicit(ATOMIC_VAL, _Py_memory_order_seq_cst)
/* Python-local extensions */
#define _Py_atomic_store_relaxed(ATOMIC_VAL, NEW_VAL) \
_Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, _Py_memory_order_relaxed)
#define _Py_atomic_load_relaxed(ATOMIC_VAL) \
_Py_atomic_load_explicit(ATOMIC_VAL, _Py_memory_order_relaxed)
#ifdef __cplusplus
}
#endif
#endif /* Py_ATOMIC_H */
|
