library/aesce.c

/*
 *  Armv8-A Cryptographic Extension support functions for Aarch64
 *
 *  Copyright The Mbed TLS Contributors
 *  SPDX-License-Identifier: Apache-2.0
 *
 *  Licensed under the Apache License, Version 2.0 (the "License"); you may
 *  not use this file except in compliance with the License.
 *  You may obtain a copy of the License at
 *
 *  http://www.apache.org/licenses/LICENSE-2.0
 *
 *  Unless required by applicable law or agreed to in writing, software
 *  distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
 *  WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 *  See the License for the specific language governing permissions and
 *  limitations under the License.
 */

#if defined(__clang__) &&  (__clang_major__ >= 4)

/* Ideally, we would simply use MBEDTLS_ARCH_IS_ARMV8 in the following #if,
 * but that is defined by build_info.h, and we need this block to happen first. */
#if defined(__ARM_ARCH)
#if __ARM_ARCH >= 8
#define MBEDTLS_AESCE_ARCH_IS_ARMV8
#endif
#endif

#if defined(MBEDTLS_AESCE_ARCH_IS_ARMV8) && !defined(__ARM_FEATURE_CRYPTO)
/* TODO: Re-consider above after https://reviews.llvm.org/D131064 merged.
 *
 * The intrinsic declaration are guarded by predefined ACLE macros in clang:
 * these are normally only enabled by the -march option on the command line.
 * By defining the macros ourselves we gain access to those declarations without
 * requiring -march on the command line.
 *
 * `arm_neon.h` could be included by any header file, so we put these defines
 * at the top of this file, before any includes.
 */
#define __ARM_FEATURE_CRYPTO 1
/* See: https://arm-software.github.io/acle/main/acle.html#cryptographic-extensions
 *
 * `__ARM_FEATURE_CRYPTO` is deprecated, but we need to continue to specify it
 * for older compilers.
 */
#define __ARM_FEATURE_AES    1
#define MBEDTLS_ENABLE_ARM_CRYPTO_EXTENSIONS_COMPILER_FLAG
#endif

#endif /* defined(__clang__) &&  (__clang_major__ >= 4) */

#include <string.h>
#include "common.h"

#if defined(MBEDTLS_AESCE_C)

#include "aesce.h"

#if defined(MBEDTLS_ARCH_IS_ARMV8)

/* Compiler version checks. */
#if defined(__clang__)
#   if __clang_major__ < 4
#       error "Minimum version of Clang for MBEDTLS_AESCE_C is 4.0."
#   endif
#elif defined(__GNUC__)
#   if __GNUC__ < 6
#       error "Minimum version of GCC for MBEDTLS_AESCE_C is 6.0."
#   endif
#elif defined(_MSC_VER)
/* TODO: We haven't verified MSVC from 1920 to 1928. If someone verified that,
 *       please update this and document of `MBEDTLS_AESCE_C` in
 *       `mbedtls_config.h`. */
#   if _MSC_VER < 1929
#       error "Minimum version of MSVC for MBEDTLS_AESCE_C is 2019 version 16.11.2."
#   endif
#endif

#ifdef __ARM_NEON
#include <arm_neon.h>

#if defined(MBEDTLS_ARCH_IS_ARM32)
#if defined(__clang__)
/* On clang for A32/T32, work around some missing intrinsics and types */

#ifndef vreinterpretq_p64_u8
#define vreinterpretq_p64_u8 (poly64x2_t)
#endif
#ifndef vreinterpretq_u8_p128
#define vreinterpretq_u8_p128 (uint8x16_t)
#endif
#ifndef vreinterpretq_u64_p64
#define vreinterpretq_u64_p64 (uint64x2_t)
#endif

typedef uint8x16_t poly128_t;

static inline poly128_t vmull_p64(poly64_t a, poly64_t b)
{
    poly128_t r;
    asm ("vmull.p64	%[r], %[a], %[b]": [r] "=w" (r) : [a] "w" (a), [b] "w" (b) :);
    return r;
}

static inline poly64x1_t vget_low_p64(poly64x2_t a)
{
    return (poly64x1_t) vget_low_u64(vreinterpretq_u64_p64(a));
}

static inline poly128_t vmull_high_p64(poly64x2_t a, poly64x2_t b)
{
    return vmull_p64((poly64_t) (vget_high_u64((uint64x2_t) a)),
                     (poly64_t) (vget_high_u64((uint64x2_t) b)));
}

#endif /* defined(__clang__) */

static inline uint8x16_t vrbitq_u8(uint8x16_t x)
{
    /* There is no vrbitq_u8 instruction in A32/T32, so provide
     * an equivalent non-Neon implementation. Reverse bit order in each
     * byte with 4x rbit, rev. */
    asm ("ldm  %[p], { r2-r5 } \n\t"
         "rbit r2, r2          \n\t"
         "rev  r2, r2          \n\t"
         "rbit r3, r3          \n\t"
         "rev  r3, r3          \n\t"
         "rbit r4, r4          \n\t"
         "rev  r4, r4          \n\t"
         "rbit r5, r5          \n\t"
         "rev  r5, r5          \n\t"
         "stm  %[p], { r2-r5 } \n\t"
         :
         /* Output: 16 bytes of memory pointed to by &x */
         "+m" (*(uint8_t(*)[16]) &x)
         :
         [p] "r" (&x)
         :
         "r2", "r3", "r4", "r5"
         );
    return x;
}

#endif /* defined(MBEDTLS_ARCH_IS_ARM32) */

#else
#error "Target does not support NEON instructions"
#endif

#if !(defined(__ARM_FEATURE_CRYPTO) || defined(__ARM_FEATURE_AES)) || \
    defined(MBEDTLS_ENABLE_ARM_CRYPTO_EXTENSIONS_COMPILER_FLAG)
#   if defined(__ARMCOMPILER_VERSION)
#       if __ARMCOMPILER_VERSION <= 6090000
#           error "Must use minimum -march=armv8-a+crypto for MBEDTLS_AESCE_C"
#       else
#           pragma clang attribute push (__attribute__((target("aes"))), apply_to=function)
#           define MBEDTLS_POP_TARGET_PRAGMA
#       endif
#   elif defined(__clang__)
#       pragma clang attribute push (__attribute__((target("aes"))), apply_to=function)
#       define MBEDTLS_POP_TARGET_PRAGMA
#   elif defined(__GNUC__)
#       pragma GCC push_options
#       pragma GCC target ("+crypto")
#       define MBEDTLS_POP_TARGET_PRAGMA
#   elif defined(_MSC_VER)
#       error "Required feature(__ARM_FEATURE_AES) is not enabled."
#   endif
#endif /* !(__ARM_FEATURE_CRYPTO || __ARM_FEATURE_AES) ||
          MBEDTLS_ENABLE_ARM_CRYPTO_EXTENSIONS_COMPILER_FLAG */

#if defined(__linux__) && !defined(MBEDTLS_AES_USE_HARDWARE_ONLY)

#include <asm/hwcap.h>
#include <sys/auxv.h>

signed char mbedtls_aesce_has_support_result = -1;

#if !defined(MBEDTLS_AES_USE_HARDWARE_ONLY)
/*
 * AES instruction support detection routine
 */
int mbedtls_aesce_has_support_impl(void)
{
    /* To avoid many calls to getauxval, cache the result. This is
     * thread-safe, because we store the result in a char so cannot
     * be vulnerable to non-atomic updates.
     * It is possible that we could end up setting result more than
     * once, but that is harmless.
     */
    if (mbedtls_aesce_has_support_result == -1) {
        unsigned long auxval = getauxval(AT_HWCAP);
        if ((auxval & (HWCAP_ASIMD | HWCAP_AES)) ==
            (HWCAP_ASIMD | HWCAP_AES)) {
            mbedtls_aesce_has_support_result = 1;
        } else {
            mbedtls_aesce_has_support_result = 0;
        }
    }
    return mbedtls_aesce_has_support_result;
}
#endif

#endif /* defined(__linux__) && !defined(MBEDTLS_AES_USE_HARDWARE_ONLY) */

/* Single round of AESCE encryption */
#define AESCE_ENCRYPT_ROUND                   \
    block = vaeseq_u8(block, vld1q_u8(keys)); \
    block = vaesmcq_u8(block);                \
    keys += 16
/* Two rounds of AESCE encryption */
#define AESCE_ENCRYPT_ROUND_X2        AESCE_ENCRYPT_ROUND; AESCE_ENCRYPT_ROUND

MBEDTLS_OPTIMIZE_FOR_PERFORMANCE
static uint8x16_t aesce_encrypt_block(uint8x16_t block,
                                      unsigned char *keys,
                                      int rounds)
{
    /* 10, 12 or 14 rounds. Unroll loop. */
    if (rounds == 10) {
        goto rounds_10;
    }
    if (rounds == 12) {
        goto rounds_12;
    }
    AESCE_ENCRYPT_ROUND_X2;
rounds_12:
    AESCE_ENCRYPT_ROUND_X2;
rounds_10:
    AESCE_ENCRYPT_ROUND_X2;
    AESCE_ENCRYPT_ROUND_X2;
    AESCE_ENCRYPT_ROUND_X2;
    AESCE_ENCRYPT_ROUND_X2;
    AESCE_ENCRYPT_ROUND;

    /* AES AddRoundKey for the previous round.
     * SubBytes, ShiftRows for the final round.  */
    block = vaeseq_u8(block, vld1q_u8(keys));
    keys += 16;

    /* Final round: no MixColumns */

    /* Final AddRoundKey */
    block = veorq_u8(block, vld1q_u8(keys));

    return block;
}

/* Single round of AESCE decryption
 *
 * AES AddRoundKey, SubBytes, ShiftRows
 *
 *      block = vaesdq_u8(block, vld1q_u8(keys));
 *
 * AES inverse MixColumns for the next round.
 *
 * This means that we switch the order of the inverse AddRoundKey and
 * inverse MixColumns operations. We have to do this as AddRoundKey is
 * done in an atomic instruction together with the inverses of SubBytes
 * and ShiftRows.
 *
 * It works because MixColumns is a linear operation over GF(2^8) and
 * AddRoundKey is an exclusive or, which is equivalent to addition over
 * GF(2^8). (The inverse of MixColumns needs to be applied to the
 * affected round keys separately which has been done when the
 * decryption round keys were calculated.)
 *
 *      block = vaesimcq_u8(block);
 */
#define AESCE_DECRYPT_ROUND                   \
    block = vaesdq_u8(block, vld1q_u8(keys)); \
    block = vaesimcq_u8(block);               \
    keys += 16
/* Two rounds of AESCE decryption */
#define AESCE_DECRYPT_ROUND_X2        AESCE_DECRYPT_ROUND; AESCE_DECRYPT_ROUND

static uint8x16_t aesce_decrypt_block(uint8x16_t block,
                                      unsigned char *keys,
                                      int rounds)
{
    /* 10, 12 or 14 rounds. Unroll loop. */
    if (rounds == 10) {
        goto rounds_10;
    }
    if (rounds == 12) {
        goto rounds_12;
    }
    AESCE_DECRYPT_ROUND_X2;
rounds_12:
    AESCE_DECRYPT_ROUND_X2;
rounds_10:
    AESCE_DECRYPT_ROUND_X2;
    AESCE_DECRYPT_ROUND_X2;
    AESCE_DECRYPT_ROUND_X2;
    AESCE_DECRYPT_ROUND_X2;
    AESCE_DECRYPT_ROUND;

    /* The inverses of AES AddRoundKey, SubBytes, ShiftRows finishing up the
     * last full round. */
    block = vaesdq_u8(block, vld1q_u8(keys));
    keys += 16;

    /* Inverse AddRoundKey for inverting the initial round key addition. */
    block = veorq_u8(block, vld1q_u8(keys));

    return block;
}

/*
 * AES-ECB block en(de)cryption
 */
int mbedtls_aesce_crypt_ecb(mbedtls_aes_context *ctx,
                            int mode,
                            const unsigned char input[16],
                            unsigned char output[16])
{
    uint8x16_t block = vld1q_u8(&input[0]);
    unsigned char *keys = (unsigned char *) (ctx->buf + ctx->rk_offset);

    if (mode == MBEDTLS_AES_ENCRYPT) {
        block = aesce_encrypt_block(block, keys, ctx->nr);
    } else {
        block = aesce_decrypt_block(block, keys, ctx->nr);
    }
    vst1q_u8(&output[0], block);

    return 0;
}

/*
 * Compute decryption round keys from encryption round keys
 */
void mbedtls_aesce_inverse_key(unsigned char *invkey,
                               const unsigned char *fwdkey,
                               int nr)
{
    int i, j;
    j = nr;
    vst1q_u8(invkey, vld1q_u8(fwdkey + j * 16));
    for (i = 1, j--; j > 0; i++, j--) {
        vst1q_u8(invkey + i * 16,
                 vaesimcq_u8(vld1q_u8(fwdkey + j * 16)));
    }
    vst1q_u8(invkey + i * 16, vld1q_u8(fwdkey + j * 16));

}

static inline uint32_t aes_rot_word(uint32_t word)
{
    return (word << (32 - 8)) | (word >> 8);
}

static inline uint32_t aes_sub_word(uint32_t in)
{
    uint8x16_t v = vreinterpretq_u8_u32(vdupq_n_u32(in));
    uint8x16_t zero = vdupq_n_u8(0);

    /* vaeseq_u8 does both SubBytes and ShiftRows. Taking the first row yields
     * the correct result as ShiftRows doesn't change the first row. */
    v = vaeseq_u8(zero, v);
    return vgetq_lane_u32(vreinterpretq_u32_u8(v), 0);
}

/*
 * Key expansion function
 */
static void aesce_setkey_enc(unsigned char *rk,
                             const unsigned char *key,
                             const size_t key_bit_length)
{
    static uint8_t const rcon[] = { 0x01, 0x02, 0x04, 0x08, 0x10,
                                    0x20, 0x40, 0x80, 0x1b, 0x36 };
    /* See https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.197.pdf
     *   - Section 5, Nr = Nk + 6
     *   - Section 5.2, the length of round keys is Nb*(Nr+1)
     */
    const uint32_t key_len_in_words = key_bit_length / 32;  /* Nk */
    const size_t round_key_len_in_words = 4;                /* Nb */
    const size_t rounds_needed = key_len_in_words + 6;      /* Nr */
    const size_t round_keys_len_in_words =
        round_key_len_in_words * (rounds_needed + 1);       /* Nb*(Nr+1) */
    const uint32_t *rko_end = (uint32_t *) rk + round_keys_len_in_words;

    memcpy(rk, key, key_len_in_words * 4);

    for (uint32_t *rki = (uint32_t *) rk;
         rki + key_len_in_words < rko_end;
         rki += key_len_in_words) {

        size_t iteration = (rki - (uint32_t *) rk) / key_len_in_words;
        uint32_t *rko;
        rko = rki + key_len_in_words;
        rko[0] = aes_rot_word(aes_sub_word(rki[key_len_in_words - 1]));
        rko[0] ^= rcon[iteration] ^ rki[0];
        rko[1] = rko[0] ^ rki[1];
        rko[2] = rko[1] ^ rki[2];
        rko[3] = rko[2] ^ rki[3];
        if (rko + key_len_in_words > rko_end) {
            /* Do not write overflow words.*/
            continue;
        }
#if !defined(MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH)
        switch (key_bit_length) {
            case 128:
                break;
            case 192:
                rko[4] = rko[3] ^ rki[4];
                rko[5] = rko[4] ^ rki[5];
                break;
            case 256:
                rko[4] = aes_sub_word(rko[3]) ^ rki[4];
                rko[5] = rko[4] ^ rki[5];
                rko[6] = rko[5] ^ rki[6];
                rko[7] = rko[6] ^ rki[7];
                break;
        }
#endif /* !MBEDTLS_AES_ONLY_128_BIT_KEY_LENGTH */
    }
}

/*
 * Key expansion, wrapper
 */
int mbedtls_aesce_setkey_enc(unsigned char *rk,
                             const unsigned char *key,
                             size_t bits)
{
    switch (bits) {
        case 128:
        case 192:
        case 256:
            aesce_setkey_enc(rk, key, bits);
            break;
        default:
            return MBEDTLS_ERR_AES_INVALID_KEY_LENGTH;
    }

    return 0;
}

#if defined(MBEDTLS_GCM_C)

#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ == 5
/* Some intrinsics are not available for GCC 5.X. */
#define vreinterpretq_p64_u8(a) ((poly64x2_t) a)
#define vreinterpretq_u8_p128(a) ((uint8x16_t) a)
static inline poly64_t vget_low_p64(poly64x2_t __a)
{
    uint64x2_t tmp = (uint64x2_t) (__a);
    uint64x1_t lo = vcreate_u64(vgetq_lane_u64(tmp, 0));
    return (poly64_t) (lo);
}
#endif /* !__clang__ && __GNUC__ && __GNUC__ == 5*/

/* vmull_p64/vmull_high_p64 wrappers.
 *
 * Older compilers miss some intrinsic functions for `poly*_t`. We use
 * uint8x16_t and uint8x16x3_t as input/output parameters.
 */
#if defined(__GNUC__) && !defined(__clang__)
/* GCC reports incompatible type error without cast. GCC think poly64_t and
 * poly64x1_t are different, that is different with MSVC and Clang. */
#define MBEDTLS_VMULL_P64(a, b) vmull_p64((poly64_t) a, (poly64_t) b)
#else
/* MSVC reports `error C2440: 'type cast'` with cast. Clang does not report
 * error with/without cast. And I think poly64_t and poly64x1_t are same, no
 * cast for clang also. */
#define MBEDTLS_VMULL_P64(a, b) vmull_p64(a, b)
#endif
static inline uint8x16_t pmull_low(uint8x16_t a, uint8x16_t b)
{

    return vreinterpretq_u8_p128(
        MBEDTLS_VMULL_P64(
            vget_low_p64(vreinterpretq_p64_u8(a)),
            vget_low_p64(vreinterpretq_p64_u8(b))
            ));
}

static inline uint8x16_t pmull_high(uint8x16_t a, uint8x16_t b)
{
    return vreinterpretq_u8_p128(
        vmull_high_p64(vreinterpretq_p64_u8(a),
                       vreinterpretq_p64_u8(b)));
}

/* GHASH does 128b polynomial multiplication on block in GF(2^128) defined by
 * `x^128 + x^7 + x^2 + x + 1`.
 *
 * Arm64 only has 64b->128b polynomial multipliers, we need to do 4 64b
 * multiplies to generate a 128b.
 *
 * `poly_mult_128` executes polynomial multiplication and outputs 256b that
 * represented by 3 128b due to code size optimization.
 *
 * Output layout:
 * |            |             |             |
 * |------------|-------------|-------------|
 * | ret.val[0] | h3:h2:00:00 | high   128b |
 * | ret.val[1] |   :m2:m1:00 | middle 128b |
 * | ret.val[2] |   :  :l1:l0 | low    128b |
 */
static inline uint8x16x3_t poly_mult_128(uint8x16_t a, uint8x16_t b)
{
    uint8x16x3_t ret;
    uint8x16_t h, m, l; /* retval high/middle/low */
    uint8x16_t c, d, e;

    h = pmull_high(a, b);                       /* h3:h2:00:00 = a1*b1 */
    l = pmull_low(a, b);                        /*   :  :l1:l0 = a0*b0 */
    c = vextq_u8(b, b, 8);                      /*      :c1:c0 = b0:b1 */
    d = pmull_high(a, c);                       /*   :d2:d1:00 = a1*b0 */
    e = pmull_low(a, c);                        /*   :e2:e1:00 = a0*b1 */
    m = veorq_u8(d, e);                         /*   :m2:m1:00 = d + e */

    ret.val[0] = h;
    ret.val[1] = m;
    ret.val[2] = l;
    return ret;
}

/*
 * Modulo reduction.
 *
 * See: https://www.researchgate.net/publication/285612706_Implementing_GCM_on_ARMv8
 *
 * Section 4.3
 *
 * Modular reduction is slightly more complex. Write the GCM modulus as f(z) =
 * z^128 +r(z), where r(z) = z^7+z^2+z+ 1. The well known approach is to
 * consider that z^128 ≡r(z) (mod z^128 +r(z)), allowing us to write the 256-bit
 * operand to be reduced as a(z) = h(z)z^128 +l(z)≡h(z)r(z) + l(z). That is, we
 * simply multiply the higher part of the operand by r(z) and add it to l(z). If
 * the result is still larger than 128 bits, we reduce again.
 */
static inline uint8x16_t poly_mult_reduce(uint8x16x3_t input)
{
    uint8x16_t const ZERO = vdupq_n_u8(0);

    uint64x2_t r = vreinterpretq_u64_u8(vdupq_n_u8(0x87));
#if defined(__GNUC__)
    /* use 'asm' as an optimisation barrier to prevent loading MODULO from
     * memory. It is for GNUC compatible compilers.
     */
    asm ("" : "+w" (r));
#endif
    uint8x16_t const MODULO = vreinterpretq_u8_u64(vshrq_n_u64(r, 64 - 8));
    uint8x16_t h, m, l; /* input high/middle/low 128b */
    uint8x16_t c, d, e, f, g, n, o;
    h = input.val[0];            /* h3:h2:00:00                          */
    m = input.val[1];            /*   :m2:m1:00                          */
    l = input.val[2];            /*   :  :l1:l0                          */
    c = pmull_high(h, MODULO);   /*   :c2:c1:00 = reduction of h3        */
    d = pmull_low(h, MODULO);    /*   :  :d1:d0 = reduction of h2        */
    e = veorq_u8(c, m);          /*   :e2:e1:00 = m2:m1:00 + c2:c1:00    */
    f = pmull_high(e, MODULO);   /*   :  :f1:f0 = reduction of e2        */
    g = vextq_u8(ZERO, e, 8);    /*   :  :g1:00 = e1:00                  */
    n = veorq_u8(d, l);          /*   :  :n1:n0 = d1:d0 + l1:l0          */
    o = veorq_u8(n, f);          /*       o1:o0 = f1:f0 + n1:n0          */
    return veorq_u8(o, g);       /*             = o1:o0 + g1:00          */
}

/*
 * GCM multiplication: c = a times b in GF(2^128)
 */
void mbedtls_aesce_gcm_mult(unsigned char c[16],
                            const unsigned char a[16],
                            const unsigned char b[16])
{
    uint8x16_t va, vb, vc;
    va = vrbitq_u8(vld1q_u8(&a[0]));
    vb = vrbitq_u8(vld1q_u8(&b[0]));
    vc = vrbitq_u8(poly_mult_reduce(poly_mult_128(va, vb)));
    vst1q_u8(&c[0], vc);
}

#endif /* MBEDTLS_GCM_C */

#if defined(MBEDTLS_POP_TARGET_PRAGMA)
#if defined(__clang__)
#pragma clang attribute pop
#elif defined(__GNUC__)
#pragma GCC pop_options
#endif
#undef MBEDTLS_POP_TARGET_PRAGMA
#endif

#endif /* MBEDTLS_ARCH_IS_ARMV8 */

#endif /* MBEDTLS_AESCE_C */