target/arm: Implement an IMPDEF pauth algorithm

Without hardware acceleration, a cryptographically strong
algorithm is too expensive for pauth_computepac.

Even with hardware accel, we are not currently expecting
to link the linux-user binaries to any crypto libraries,
and doing so would generally make the --static build fail.

So choose XXH64 as a reasonably quick and decent hash.

Tested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210111235740.462469-2-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Richard Henderson 2021-01-11 13:57:38 -10:00 committed by Peter Maydell
parent f1fcb6851a
commit 283fc52ade
3 changed files with 131 additions and 9 deletions

View file

@ -119,4 +119,102 @@ static inline uint32_t qemu_xxhash6(uint64_t ab, uint64_t cd, uint32_t e,
return qemu_xxhash7(ab, cd, e, f, 0);
}
/*
* Component parts of the XXH64 algorithm from
* https://github.com/Cyan4973/xxHash/blob/v0.8.0/xxhash.h
*
* The complete algorithm looks like
*
* i = 0;
* if (len >= 32) {
* v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
* v2 = seed + XXH_PRIME64_2;
* v3 = seed + 0;
* v4 = seed - XXH_PRIME64_1;
* do {
* v1 = XXH64_round(v1, get64bits(input + i));
* v2 = XXH64_round(v2, get64bits(input + i + 8));
* v3 = XXH64_round(v3, get64bits(input + i + 16));
* v4 = XXH64_round(v4, get64bits(input + i + 24));
* } while ((i += 32) <= len);
* h64 = XXH64_mergerounds(v1, v2, v3, v4);
* } else {
* h64 = seed + XXH_PRIME64_5;
* }
* h64 += len;
*
* for (; i + 8 <= len; i += 8) {
* h64 ^= XXH64_round(0, get64bits(input + i));
* h64 = rol64(h64, 27) * XXH_PRIME64_1 + XXH_PRIME64_4;
* }
* for (; i + 4 <= len; i += 4) {
* h64 ^= get32bits(input + i) * PRIME64_1;
* h64 = rol64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
* }
* for (; i < len; i += 1) {
* h64 ^= get8bits(input + i) * XXH_PRIME64_5;
* h64 = rol64(h64, 11) * XXH_PRIME64_1;
* }
*
* return XXH64_avalanche(h64)
*
* Exposing the pieces instead allows for simplified usage when
* the length is a known constant and the inputs are in registers.
*/
#define XXH_PRIME64_1 0x9E3779B185EBCA87ULL
#define XXH_PRIME64_2 0xC2B2AE3D27D4EB4FULL
#define XXH_PRIME64_3 0x165667B19E3779F9ULL
#define XXH_PRIME64_4 0x85EBCA77C2B2AE63ULL
#define XXH_PRIME64_5 0x27D4EB2F165667C5ULL
static inline uint64_t XXH64_round(uint64_t acc, uint64_t input)
{
return rol64(acc + input * XXH_PRIME64_2, 31) * XXH_PRIME64_1;
}
static inline uint64_t XXH64_mergeround(uint64_t acc, uint64_t val)
{
return (acc ^ XXH64_round(0, val)) * XXH_PRIME64_1 + XXH_PRIME64_4;
}
static inline uint64_t XXH64_mergerounds(uint64_t v1, uint64_t v2,
uint64_t v3, uint64_t v4)
{
uint64_t h64;
h64 = rol64(v1, 1) + rol64(v2, 7) + rol64(v3, 12) + rol64(v4, 18);
h64 = XXH64_mergeround(h64, v1);
h64 = XXH64_mergeround(h64, v2);
h64 = XXH64_mergeround(h64, v3);
h64 = XXH64_mergeround(h64, v4);
return h64;
}
static inline uint64_t XXH64_avalanche(uint64_t h64)
{
h64 ^= h64 >> 33;
h64 *= XXH_PRIME64_2;
h64 ^= h64 >> 29;
h64 *= XXH_PRIME64_3;
h64 ^= h64 >> 32;
return h64;
}
static inline uint64_t qemu_xxhash64_4(uint64_t a, uint64_t b,
uint64_t c, uint64_t d)
{
uint64_t v1 = QEMU_XXHASH_SEED + XXH_PRIME64_1 + XXH_PRIME64_2;
uint64_t v2 = QEMU_XXHASH_SEED + XXH_PRIME64_2;
uint64_t v3 = QEMU_XXHASH_SEED + 0;
uint64_t v4 = QEMU_XXHASH_SEED - XXH_PRIME64_1;
v1 = XXH64_round(v1, a);
v2 = XXH64_round(v2, b);
v3 = XXH64_round(v3, c);
v4 = XXH64_round(v4, d);
return XXH64_avalanche(XXH64_mergerounds(v1, v2, v3, v4));
}
#endif /* QEMU_XXHASH_H */