rind 0.1.12

Pure-Rust cryptography for MoQ, forked from graviola
Documentation 
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

// SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT-0

//! Keccak-f[1600] on OpenSSL's `keccak1600-armv4` routines: NEON where the CPU
//! has it, and integer code on bit-interleaved lanes otherwise.
//!
//! The assembly is OpenSSL's, imported by `admin/import-openssl`. It is not
//! formally verified: its correctness rests on its use and testing in
//! OpenSSL, and on rind's tests of SHA-3 and ML-KEM against their vectors and
//! aws-lc-rs.

use super::cpu::HaveNeon;

unsafe extern "C" {
    #[link_name = aws_lc_symbol!("KeccakF1600")]
    fn keccak_f1600_int(a: *mut [[u32; 2]; 25]);

    #[link_name = aws_lc_symbol!("KeccakF1600_neon")]
    fn keccak_f1600_neon_regs();

    #[link_name = aws_lc_symbol!("keccak_f1600_neon")]
    fn keccak_f1600_neon(a: *mut [u64; 25]);
}

// As in `sha_armv4`: ARM mode with NEON, then the target's `.fpu` again.
// OpenSSL's NEON permutation takes the state in d0-d24, so rind's entry loads
// and stores it as OpenSSL's `SHA3_absorb_neon` does around it, using the
// state as the permutation's scratch.
core::arch::global_asm!(
    ".text",
    ".fpu neon",
    ".code 32",
    ".globl {KeccakF1600}",
    ".globl {KeccakF1600_neon}",
    include_str!("openssl/keccak1600-armv4.s"),
    ".globl {keccak_f1600_neon}",
    ".type {keccak_f1600_neon}, %function",
    ".align 5",
    "{keccak_f1600_neon}:",
    "stmdb sp!, {{r4, lr}}",
    "vstmdb sp!, {{d8-d15}}",
    "mov r4, r0",
    "vld1.32 {{d0}}, [r0,:64]!",
    "vld1.32 {{d2}}, [r0,:64]!",
    "vld1.32 {{d4}}, [r0,:64]!",
    "vld1.32 {{d6}}, [r0,:64]!",
    "vld1.32 {{d8}}, [r0,:64]!",
    "vld1.32 {{d1}}, [r0,:64]!",
    "vld1.32 {{d3}}, [r0,:64]!",
    "vld1.32 {{d5}}, [r0,:64]!",
    "vld1.32 {{d7}}, [r0,:64]!",
    "vld1.32 {{d9}}, [r0,:64]!",
    "vld1.32 {{d10}}, [r0,:64]!",
    "vld1.32 {{d12}}, [r0,:64]!",
    "vld1.32 {{d14}}, [r0,:64]!",
    "vld1.32 {{d16}}, [r0,:64]!",
    "vld1.32 {{d18}}, [r0,:64]!",
    "vld1.32 {{d11}}, [r0,:64]!",
    "vld1.32 {{d13}}, [r0,:64]!",
    "vld1.32 {{d15}}, [r0,:64]!",
    "vld1.32 {{d17}}, [r0,:64]!",
    "vld1.32 {{d19}}, [r0,:64]!",
    "vld1.32 {{d20-d23}}, [r0,:64]!",
    "vld1.32 {{d24}}, [r0,:64]",
    "mov r0, r4",
    "bl {KeccakF1600_neon}",
    "mov r0, r4",
    "vst1.32 {{d0}}, [r0,:64]!",
    "vst1.32 {{d2}}, [r0,:64]!",
    "vst1.32 {{d4}}, [r0,:64]!",
    "vst1.32 {{d6}}, [r0,:64]!",
    "vst1.32 {{d8}}, [r0,:64]!",
    "vst1.32 {{d1}}, [r0,:64]!",
    "vst1.32 {{d3}}, [r0,:64]!",
    "vst1.32 {{d5}}, [r0,:64]!",
    "vst1.32 {{d7}}, [r0,:64]!",
    "vst1.32 {{d9}}, [r0,:64]!",
    "vst1.32 {{d10}}, [r0,:64]!",
    "vst1.32 {{d12}}, [r0,:64]!",
    "vst1.32 {{d14}}, [r0,:64]!",
    "vst1.32 {{d16}}, [r0,:64]!",
    "vst1.32 {{d18}}, [r0,:64]!",
    "vst1.32 {{d11}}, [r0,:64]!",
    "vst1.32 {{d13}}, [r0,:64]!",
    "vst1.32 {{d15}}, [r0,:64]!",
    "vst1.32 {{d17}}, [r0,:64]!",
    "vst1.32 {{d19}}, [r0,:64]!",
    "vst1.32 {{d20-d23}}, [r0,:64]!",
    "vst1.32 {{d24}}, [r0,:64]",
    "vldmia sp!, {{d8-d15}}",
    "ldmia sp!, {{r4, pc}}",
    ".size {keccak_f1600_neon}, .-{keccak_f1600_neon}",
    ".fpu vfpv3",
    KeccakF1600 = sym keccak_f1600_int,
    KeccakF1600_neon = sym keccak_f1600_neon_regs,
    keccak_f1600_neon = sym keccak_f1600_neon,
);

/// Keccak-f[1600] on one state.
///
/// `_rc` is Keccak-f's round constants, which the assembly carries itself.
pub(crate) fn sha3_keccak_f1600(a: &mut [u64; 25], _rc: &[u64; 24]) {
    match HaveNeon::check() {
        Some(neon) => permute_neon(neon, a),
        None => permute_int(a),
    }
}

/// Keccak-f[1600] on all four states in `a`, one at a time.
pub(crate) fn sha3_keccak4_f1600(a: &mut [[u64; 25]; 4], rc: &[u64; 24]) {
    sha3_keccak_nof4_f1600::<4>(a, rc)
}

/// Keccak-f[1600] on the first `N` states in `a`, one at a time.
pub(crate) fn sha3_keccak_nof4_f1600<const N: usize>(a: &mut [[u64; 25]; 4], _rc: &[u64; 24]) {
    match HaveNeon::check() {
        Some(neon) => a[..N].iter_mut().for_each(|a| permute_neon(neon, a)),
        None => a[..N].iter_mut().for_each(permute_int),
    }
}

fn permute_neon(_: HaveNeon, a: &mut [u64; 25]) {
    // SAFETY: `a` is 25 writable, 8-byte aligned lanes, and the token proves
    // the CPU has NEON.
    unsafe { keccak_f1600_neon(a) }
}

/// The integer routine keeps each lane as its even bits then its odd bits,
/// whose rotations are 32-bit ones.
fn permute_int(a: &mut [u64; 25]) {
    let mut s = a.map(interleave);
    // SAFETY: `s` is 25 writable interleaved lanes.
    unsafe { keccak_f1600_int(&mut s) };
    *a = s.map(deinterleave);
}

/// Gathers the even bits of `x` into its low half.
fn evens(x: u32) -> u32 {
    let mut x = x & 0x5555_5555;
    x = (x | (x >> 1)) & 0x3333_3333;
    x = (x | (x >> 2)) & 0x0f0f_0f0f;
    x = (x | (x >> 4)) & 0x00ff_00ff;
    (x | (x >> 8)) & 0x0000_ffff
}

/// Spreads the low half of `x` to its even bits.
fn spread(x: u32) -> u32 {
    let mut x = x & 0x0000_ffff;
    x = (x | (x << 8)) & 0x00ff_00ff;
    x = (x | (x << 4)) & 0x0f0f_0f0f;
    x = (x | (x << 2)) & 0x3333_3333;
    (x | (x << 1)) & 0x5555_5555
}

fn interleave(lane: u64) -> [u32; 2] {
    let (lo, hi) = (lane as u32, (lane >> 32) as u32);
    [
        evens(lo) | (evens(hi) << 16),
        evens(lo >> 1) | (evens(hi >> 1) << 16),
    ]
}

fn deinterleave([even, odd]: [u32; 2]) -> u64 {
    let lo = spread(even) | (spread(odd) << 1);
    let hi = spread(even >> 16) | (spread(odd >> 16) << 1);
    lo as u64 | ((hi as u64) << 32)
}

#[cfg(test)]
#[cfg_attr(coverage_nightly, coverage(off))]
mod tests {
    use super::*;

    // Every other test sees only the path the CPU picks, so the two routines
    // are checked against each other here.
    #[test]
    fn neon_matches_int() {
        let Some(neon) = HaveNeon::check() else {
            return;
        };
        let mut want: [u64; 25] =
            core::array::from_fn(|i| (i as u64 + 1).wrapping_mul(0x9e37_79b9_7f4a_7c15));
        let mut got = want;
        for _ in 0..3 {
            permute_int(&mut want);
            permute_neon(neon, &mut got);
            assert_eq!(got, want);
        }
    }

    #[test]
    fn interleave_round_trips() {
        for lane in [0, u64::MAX, 1, 1 << 63, 0x0123_4567_89ab_cdef] {
            assert_eq!(deinterleave(interleave(lane)), lane);
        }
        assert_eq!(interleave(0b10), [0, 1]);
        assert_eq!(interleave(1 << 32), [1 << 16, 0]);
    }
}