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
|
//! `__sync_*` atomic builtins for Armv6-M.
//!
//! Some Rust ThreadX targets, like `thumbv6m-threadx-eabi`, declares `atomic_cas: true`
//! and emulate them with OS support. Right now, we handle this by disabling interrupts.
//! This should be OK for single core support.
//!
//! <https://llvm.org/docs/Atomics.html>
#![no_std]
#![feature(rustc_private)]
extern crate fusible;
use fusible::interrupt_control::with_disabled;
/// Atomically apply `op` to `*ptr`, returning the previous value.
///
/// # Safety
///
/// `ptr` must be valid for reads and writes and aligned to `size_of::<T>()`.
unsafe fn rmw<T, O>(ptr: *mut T, op: O) -> T
where
T: Copy,
O: FnOnce(T) -> T,
{
// Safety: interrupt control sufficient for single-core access. Caller
// (LLVM) assumed to uphold safety contract.
with_disabled(|| unsafe {
let old = ptr.read();
ptr.write(op(old));
old
})
}
/// Atomically store `new` into `*ptr` if `*ptr` equals `old`, returning the
/// previous value either way.
///
/// # Safety
///
/// `ptr` must be valid for reads and writes and aligned to `size_of::<T>()`.
unsafe fn cas<T>(ptr: *mut T, old: T, new: T) -> T
where
T: Copy + PartialEq,
{
// Safety: Same as rmw.
with_disabled(|| unsafe {
let current = ptr.read();
if current == old {
ptr.write(new);
}
current
})
}
macro_rules! fetch_and_op {
($name:ident, $ty:ty, $op:expr) => {
#[unsafe(no_mangle)]
pub unsafe extern "C" fn $name(ptr: *mut $ty, val: $ty) -> $ty {
unsafe { rmw(ptr, |cur| $op(cur, val)) }
}
};
}
macro_rules! val_compare_and_swap {
($name:ident, $ty:ty) => {
#[unsafe(no_mangle)]
pub unsafe extern "C" fn $name(ptr: *mut $ty, oldval: $ty, newval: $ty) -> $ty {
unsafe { cas(ptr, oldval, newval) }
}
};
}
fetch_and_op!(__sync_fetch_and_add_1, u8, |a: u8, b: u8| a.wrapping_add(b));
fetch_and_op!(__sync_fetch_and_add_2, u16, |a: u16, b: u16| a
.wrapping_add(b));
fetch_and_op!(__sync_fetch_and_add_4, u32, |a: u32, b: u32| a
.wrapping_add(b));
fetch_and_op!(__sync_fetch_and_sub_1, u8, |a: u8, b: u8| a.wrapping_sub(b));
fetch_and_op!(__sync_fetch_and_sub_2, u16, |a: u16, b: u16| a
.wrapping_sub(b));
fetch_and_op!(__sync_fetch_and_sub_4, u32, |a: u32, b: u32| a
.wrapping_sub(b));
fetch_and_op!(__sync_fetch_and_and_1, u8, |a: u8, b: u8| a & b);
fetch_and_op!(__sync_fetch_and_and_2, u16, |a: u16, b: u16| a & b);
fetch_and_op!(__sync_fetch_and_and_4, u32, |a: u32, b: u32| a & b);
fetch_and_op!(__sync_fetch_and_or_1, u8, |a: u8, b: u8| a | b);
fetch_and_op!(__sync_fetch_and_or_2, u16, |a: u16, b: u16| a | b);
fetch_and_op!(__sync_fetch_and_or_4, u32, |a: u32, b: u32| a | b);
fetch_and_op!(__sync_fetch_and_xor_1, u8, |a: u8, b: u8| a ^ b);
fetch_and_op!(__sync_fetch_and_xor_2, u16, |a: u16, b: u16| a ^ b);
fetch_and_op!(__sync_fetch_and_xor_4, u32, |a: u32, b: u32| a ^ b);
fetch_and_op!(__sync_fetch_and_nand_1, u8, |a: u8, b: u8| !(a & b));
fetch_and_op!(__sync_fetch_and_nand_2, u16, |a: u16, b: u16| !(a & b));
fetch_and_op!(__sync_fetch_and_nand_4, u32, |a: u32, b: u32| !(a & b));
fetch_and_op!(__sync_fetch_and_max_1, i8, |a: i8, b: i8| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_max_2, i16, |a: i16, b: i16| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_max_4, i32, |a: i32, b: i32| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umax_1, u8, |a: u8, b: u8| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umax_2, u16, |a: u16, b: u16| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umax_4, u32, |a: u32, b: u32| if a > b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_min_1, i8, |a: i8, b: i8| if a < b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_min_2, i16, |a: i16, b: i16| if a < b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_min_4, i32, |a: i32, b: i32| if a < b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umin_1, u8, |a: u8, b: u8| if a < b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umin_2, u16, |a: u16, b: u16| if a < b {
a
} else {
b
});
fetch_and_op!(__sync_fetch_and_umin_4, u32, |a: u32, b: u32| if a < b {
a
} else {
b
});
//
// From the GCC docs, linked from the LLVM docs:
//
// > This built-in function, as described by Intel, is not a traditional
// > test-and-set operation, but rather an atomic exchange operation. It
// > writes value into *ptr, and returns the previous contents of *ptr.
//
fetch_and_op!(__sync_lock_test_and_set_1, u8, |_: u8, new: u8| new);
fetch_and_op!(__sync_lock_test_and_set_2, u16, |_: u16, new: u16| new);
fetch_and_op!(__sync_lock_test_and_set_4, u32, |_: u32, new: u32| new);
val_compare_and_swap!(__sync_val_compare_and_swap_1, u8);
val_compare_and_swap!(__sync_val_compare_and_swap_2, u16);
val_compare_and_swap!(__sync_val_compare_and_swap_4, u32);
|
