X-Git-Url: https://www.bearssl.org/gitweb//home/git/?p=BearSSL;a=blobdiff_plain;f=src%2Fec%2Fec_p256_m31.c;h=0462c15b1b64274472ef067a0240ec91bbb4d3d4;hp=0631a135f7d1a0a3b0e1d39e34c967fd884341d2;hb=2b738493bd16d57fdb12d38d03631981370259be;hpb=bd3036844bd20b2b8d7bce7fee5ad010ce401915;ds=sidebyside diff --git a/src/ec/ec_p256_m31.c b/src/ec/ec_p256_m31.c index 0631a13..0462c15 100644 --- a/src/ec/ec_p256_m31.c +++ b/src/ec/ec_p256_m31.c @@ -394,7 +394,7 @@ mul_f256(uint32_t *d, const uint32_t *a, const uint32_t *b) uint32_t t[18]; uint64_t s[18]; uint64_t cc, x; - uint32_t z; + uint32_t z, c; int i; mul9(t, a, b); @@ -465,7 +465,15 @@ mul_f256(uint32_t *d, const uint32_t *a, const uint32_t *b) d[8] &= 0xFFFF; /* - * Subtract cc*p. + * One extra round of reduction, for cc*2^256, which means + * adding cc*(2^224-2^192-2^96+1) to a 256-bit (nonnegative) + * value. If cc is negative, then it may happen (rarely, but + * not neglectibly so) that the result would be negative. In + * order to avoid that, if cc is negative, then we add the + * modulus once. Note that if cc is negative, then propagating + * that carry must yield a value lower than the modulus, so + * adding the modulus once will keep the final result under + * twice the modulus. */ z = (uint32_t)cc; d[3] -= z << 6; @@ -473,6 +481,12 @@ mul_f256(uint32_t *d, const uint32_t *a, const uint32_t *b) d[7] -= ARSH(z, 18); d[7] += (z << 14) & 0x3FFFFFFF; d[8] += ARSH(z, 16); + c = z >> 31; + d[0] -= c; + d[3] += c << 6; + d[6] += c << 12; + d[7] -= c << 14; + d[8] += c << 16; for (i = 0; i < 9; i ++) { uint32_t w; @@ -492,7 +506,7 @@ square_f256(uint32_t *d, const uint32_t *a) uint32_t t[18]; uint64_t s[18]; uint64_t cc, x; - uint32_t z; + uint32_t z, c; int i; square9(t, a); @@ -563,7 +577,15 @@ square_f256(uint32_t *d, const uint32_t *a) d[8] &= 0xFFFF; /* - * Subtract cc*p. + * One extra round of reduction, for cc*2^256, which means + * adding cc*(2^224-2^192-2^96+1) to a 256-bit (nonnegative) + * value. If cc is negative, then it may happen (rarely, but + * not neglectibly so) that the result would be negative. In + * order to avoid that, if cc is negative, then we add the + * modulus once. Note that if cc is negative, then propagating + * that carry must yield a value lower than the modulus, so + * adding the modulus once will keep the final result under + * twice the modulus. */ z = (uint32_t)cc; d[3] -= z << 6; @@ -571,6 +593,12 @@ square_f256(uint32_t *d, const uint32_t *a) d[7] -= ARSH(z, 18); d[7] += (z << 14) & 0x3FFFFFFF; d[8] += ARSH(z, 16); + c = z >> 31; + d[0] -= c; + d[3] += c << 6; + d[6] += c << 12; + d[7] -= c << 14; + d[8] += c << 16; for (i = 0; i < 9; i ++) { uint32_t w;