diff options
Diffstat (limited to '')
| -rw-r--r-- | security/nss/lib/freebl/gcm-arm32-neon.c | 202 |
1 files changed, 202 insertions, 0 deletions
diff --git a/security/nss/lib/freebl/gcm-arm32-neon.c b/security/nss/lib/freebl/gcm-arm32-neon.c new file mode 100644 index 0000000000..be04247701 --- /dev/null +++ b/security/nss/lib/freebl/gcm-arm32-neon.c @@ -0,0 +1,202 @@ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifdef FREEBL_NO_DEPEND +#include "stubs.h" +#endif +#include "blapii.h" +#include "blapit.h" +#include "gcm.h" +#include "secerr.h" +#include "prtypes.h" + +#if defined(IS_LITTLE_ENDIAN) + +#include <arm_neon.h> + +SECStatus +gcm_HashWrite_hw(gcmHashContext *ghash, unsigned char *outbuf) +{ + vst1_u8(outbuf, vrev64_u8(vcreate_u8(ghash->x_high))); + vst1_u8(outbuf + 8, vrev64_u8(vcreate_u8(ghash->x_low))); + return SECSuccess; +} + +/* Carry-less multiplication. a * b = ret. */ +static inline uint8x16_t +clmul(const uint8x8_t a, const uint8x8_t b) +{ + uint8x16_t d, e, f, g, h, i, j, k, l, m, n; + uint8x8_t t_high, t_low; + uint8x16_t t0, t1, t2, t3; + const uint8x8_t k16 = vcreate_u8(0xffff); + const uint8x8_t k32 = vcreate_u8(0xffffffff); + const uint8x8_t k48 = vcreate_u8(0xffffffffffff); + + // D = A * B + d = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(a), + vreinterpret_p8_u8(b))); + // E = A * B1 + e = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(a), + vreinterpret_p8_u8(vext_u8(b, b, 1)))); + // F = A1 * B + f = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(vext_u8(a, a, 1)), + vreinterpret_p8_u8(b))); + // G = A * B2 + g = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(a), + vreinterpret_p8_u8(vext_u8(b, b, 2)))); + // H = A2 * B + h = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(vext_u8(a, a, 2)), + vreinterpret_p8_u8(b))); + // I = A * B3 + i = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(a), + vreinterpret_p8_u8(vext_u8(b, b, 3)))); + // J = A3 * B + j = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(vext_u8(a, a, 3)), + vreinterpret_p8_u8(b))); + // K = A * B4 + k = vreinterpretq_u8_p16(vmull_p8(vreinterpret_p8_u8(a), + vreinterpret_p8_u8(vext_u8(b, b, 4)))); + // L = E + F + l = veorq_u8(e, f); + // M = G + H + m = veorq_u8(g, h); + // N = I + J + n = veorq_u8(i, j); + + // t0 = (L) (P0 + P1) << 8 + t_high = vget_high_u8(l); + t_low = vget_low_u8(l); + t_low = veor_u8(t_low, t_high); + t_high = vand_u8(t_high, k48); + t_low = veor_u8(t_low, t_high); + t0 = vcombine_u8(t_low, t_high); + t0 = vextq_u8(t0, t0, 15); + + // t1 = (M) (P2 + P3) << 16 + t_high = vget_high_u8(m); + t_low = vget_low_u8(m); + t_low = veor_u8(t_low, t_high); + t_high = vand_u8(t_high, k32); + t_low = veor_u8(t_low, t_high); + t1 = vcombine_u8(t_low, t_high); + t1 = vextq_u8(t1, t1, 14); + + // t2 = (N) (P4 + P5) << 24 + t_high = vget_high_u8(n); + t_low = vget_low_u8(n); + t_low = veor_u8(t_low, t_high); + t_high = vand_u8(t_high, k16); + t_low = veor_u8(t_low, t_high); + t2 = vcombine_u8(t_low, t_high); + t2 = vextq_u8(t2, t2, 13); + + // t3 = (K) (P6 + P7) << 32 + t_high = vget_high_u8(k); + t_low = vget_low_u8(k); + t_low = veor_u8(t_low, t_high); + t_high = vdup_n_u8(0); + t3 = vcombine_u8(t_low, t_high); + t3 = vextq_u8(t3, t3, 12); + + t0 = veorq_u8(t0, t1); + t2 = veorq_u8(t2, t3); + return veorq_u8(veorq_u8(d, t0), t2); +} + +SECStatus +gcm_HashMult_hw(gcmHashContext *ghash, const unsigned char *buf, + unsigned int count) +{ + const uint8x8_t h_low = vcreate_u8(ghash->h_low); + const uint8x8_t h_high = vcreate_u8(ghash->h_high); + uint8x16_t ci; + uint8x8_t ci_low; + uint8x8_t ci_high; + uint8x16_t z0, z2, z1a; + uint8x16_t z_high, z_low; + uint8x16_t t; + int64x2_t t1, t2, t3; + uint64x2_t z_low_l, z_low_r, z_high_l, z_high_r; + size_t i; + + ci = vcombine_u8(vcreate_u8(ghash->x_low), vcreate_u8(ghash->x_high)); + + for (i = 0; i < count; i++, buf += 16) { + ci = veorq_u8(ci, vcombine_u8(vrev64_u8(vld1_u8(buf + 8)), + vrev64_u8(vld1_u8(buf)))); + ci_high = vget_high_u8(ci); + ci_low = vget_low_u8(ci); + + /* Do binary mult ghash->X = C * ghash->H (Karatsuba). */ + z0 = clmul(ci_low, h_low); + z2 = clmul(ci_high, h_high); + z1a = clmul(veor_u8(ci_high, ci_low), veor_u8(h_high, h_low)); + z1a = veorq_u8(z0, z1a); + z1a = veorq_u8(z2, z1a); + z_high = vcombine_u8(veor_u8(vget_low_u8(z2), vget_high_u8(z1a)), + vget_high_u8(z2)); + z_low = vcombine_u8(vget_low_u8(z0), + veor_u8(vget_high_u8(z0), vget_low_u8(z1a))); + + /* Shift one (multiply by x) as gcm spec is stupid. */ + z_low_l = vshlq_n_u64(vreinterpretq_u64_u8(z_low), 1); + z_low_r = vshrq_n_u64(vreinterpretq_u64_u8(z_low), 63); + z_high_l = vshlq_n_u64(vreinterpretq_u64_u8(z_high), 1); + z_high_r = vshrq_n_u64(vreinterpretq_u64_u8(z_high), 63); + z_low = vreinterpretq_u8_u64( + vcombine_u64(vget_low_u64(z_low_l), + vorr_u64(vget_high_u64(z_low_l), + vget_low_u64(z_low_r)))); + z_high = vreinterpretq_u8_u64( + vcombine_u64(vorr_u64(vget_low_u64(z_high_l), + vget_high_u64(z_low_r)), + vorr_u64(vget_high_u64(z_high_l), + vget_low_u64(z_high_r)))); + + /* Reduce */ + t1 = vshlq_n_s64(vreinterpretq_s64_u8(z_low), 57); + t2 = vshlq_n_s64(vreinterpretq_s64_u8(z_low), 62); + t3 = vshlq_n_s64(vreinterpretq_s64_u8(z_low), 63); + t = vreinterpretq_u8_s64(veorq_s64(t1, veorq_s64(t2, t3))); + + z_low = vcombine_u8(vget_low_u8(z_low), + veor_u8(vget_high_u8(z_low), vget_low_u8(t))); + z_high = vcombine_u8(veor_u8(vget_low_u8(z_high), vget_high_u8(t)), + vget_high_u8(z_high)); + + t = vreinterpretq_u8_u64(vshrq_n_u64(vreinterpretq_u64_u8(z_low), 1)); + z_high = veorq_u8(z_high, z_low); + z_low = veorq_u8(z_low, t); + t = vreinterpretq_u8_u64(vshrq_n_u64(vreinterpretq_u64_u8(t), 6)); + z_low = vreinterpretq_u8_u64( + vshrq_n_u64(vreinterpretq_u64_u8(z_low), 1)); + z_low = veorq_u8(z_low, z_high); + ci = veorq_u8(z_low, t); + } + + vst1_u8((uint8_t *)&ghash->x_high, vget_high_u8(ci)); + vst1_u8((uint8_t *)&ghash->x_low, vget_low_u8(ci)); + return SECSuccess; +} + +SECStatus +gcm_HashInit_hw(gcmHashContext *ghash) +{ + ghash->ghash_mul = gcm_HashMult_hw; + ghash->x_low = 0; + ghash->x_high = 0; + ghash->hw = PR_TRUE; + return SECSuccess; +} + +SECStatus +gcm_HashZeroX_hw(gcmHashContext *ghash) +{ + ghash->x_low = 0; + ghash->x_high = 0; + return SECSuccess; +} + +#endif /* IS_LITTLE_ENDIAN */ |