VeraCrypt
aboutsummaryrefslogtreecommitdiff
path: root/src/Crypto
diff options
context:
space:
mode:
authorMounir IDRASSI <mounir.idrassi@idrix.fr>2020-06-24 22:29:09 +0200
committerMounir IDRASSI <mounir.idrassi@idrix.fr>2020-06-25 01:32:39 +0200
commit7d1724e93b17c5095bb01c9f053600b3a85c3cdc (patch)
treea0859bb1632243345c49b136b5c14bbff0408cf2 /src/Crypto
parentcb1efdc9ae47d13d2c1a10bf7ec3a5308ffbfa32 (diff)
downloadVeraCrypt-7d1724e93b17c5095bb01c9f053600b3a85c3cdc.tar.gz
VeraCrypt-7d1724e93b17c5095bb01c9f053600b3a85c3cdc.zip
Crypto: Fix random crash in Streebog in 32-bit, caused by use of aligned SSE2 instructions _mm_load_si128. Now we use _mm_loadu_si128 instead which is not slower than _mm_load_si128 on recent CPUs
Diffstat (limited to 'src/Crypto')
-rw-r--r--src/Crypto/Streebog.c16
1 files changed, 8 insertions, 8 deletions
diff --git a/src/Crypto/Streebog.c b/src/Crypto/Streebog.c
index e443ecae..9a261a68 100644
--- a/src/Crypto/Streebog.c
+++ b/src/Crypto/Streebog.c
@@ -1995,10 +1995,10 @@ VC_INLINE __m128i _mm_set_epi64x_a(uint64 i0, uint64 i1) {
#define LOAD(P, xmm0, xmm1, xmm2, xmm3) { \
const __m128i *__m128p = (const __m128i *) &P[0]; \
- xmm0 = _mm_load_si128(&__m128p[0]); \
- xmm1 = _mm_load_si128(&__m128p[1]); \
- xmm2 = _mm_load_si128(&__m128p[2]); \
- xmm3 = _mm_load_si128(&__m128p[3]); \
+ xmm0 = _mm_loadu_si128(&__m128p[0]); \
+ xmm1 = _mm_loadu_si128(&__m128p[1]); \
+ xmm2 = _mm_loadu_si128(&__m128p[2]); \
+ xmm3 = _mm_loadu_si128(&__m128p[3]); \
}
#define UNLOAD(P, xmm0, xmm1, xmm2, xmm3) { \
@@ -2018,10 +2018,10 @@ VC_INLINE __m128i _mm_set_epi64x_a(uint64 i0, uint64 i1) {
#define X128M(P, xmm0, xmm1, xmm2, xmm3) { \
const __m128i *__m128p = (const __m128i *) &P[0]; \
- xmm0 = _mm_xor_si128(xmm0, _mm_load_si128(&__m128p[0])); \
- xmm1 = _mm_xor_si128(xmm1, _mm_load_si128(&__m128p[1])); \
- xmm2 = _mm_xor_si128(xmm2, _mm_load_si128(&__m128p[2])); \
- xmm3 = _mm_xor_si128(xmm3, _mm_load_si128(&__m128p[3])); \
+ xmm0 = _mm_xor_si128(xmm0, _mm_loadu_si128(&__m128p[0])); \
+ xmm1 = _mm_xor_si128(xmm1, _mm_loadu_si128(&__m128p[1])); \
+ xmm2 = _mm_xor_si128(xmm2, _mm_loadu_si128(&__m128p[2])); \
+ xmm3 = _mm_xor_si128(xmm3, _mm_loadu_si128(&__m128p[3])); \
}
#define _mm_xor_64(mm0, mm1) _mm_xor_si64(mm0, _mm_cvtsi64_m64(mm1))