/* sha3.c

 *

 * Copyright (C) 2006-2026 wolfSSL Inc.

 *

 * This file is part of wolfSSL.

 *

 * wolfSSL is free software; you can redistribute it and/or modify

 * it under the terms of the GNU General Public License as published by

 * the Free Software Foundation; either version 3 of the License, or

 * (at your option) any later version.

 *

 * wolfSSL is distributed in the hope that it will be useful,

 * but WITHOUT ANY WARRANTY; without even the implied warranty of

 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

 * GNU General Public License for more details.

 *

 * You should have received a copy of the GNU General Public License

 * along with this program; if not, write to the Free Software

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1335, USA

 */

/*

 * SHA-3 Build Options:

 *

 * Core:

 * WOLFSSL_SHA3:             Enable SHA-3 support                  default: off

 * WOLFSSL_SHA3_SMALL:       Use smaller SHA-3 implementation      default: off

 * WOLFSSL_SHAKE128:         Enable SHAKE128 XOF                   default: off

 * WOLFSSL_SHAKE256:         Enable SHAKE256 XOF                   default: off

 * SHA3_BY_SPEC:             Use specification Keccak-f order      default: off

 * WC_SHA3_NO_ASM:           Disable SHA-3 assembly optimizations  default: off

 * WC_SHA3_FAULT_HARDEN:     Harden SHA-3 against fault attacks    default: off

 * WC_SHA3_SPLIT64:          Run the Keccak permutation on 32-bit halves of each

 *                           64-bit lane so a compiler that lowers 64-bit bitwise

 *                           ops to out-of-line helper calls (e.g. cl2000 on TI

 *                           C28x) emits native 32-bit ops instead.  Auto-enabled

 *                           for little-endian WC_16BIT_CPU; the default

 *                           permutation is otherwise unchanged.    default: off

 *

 * Hardware Acceleration (SHA-3-specific):

 * WC_ASYNC_ENABLE_SHA3:     Enable async SHA-3 operations         default: off

 * WOLFSSL_ARMASM_CRYPTO_SHA3: ARM crypto SHA-3 instructions       default: off

 * STM32_HASH_SHA3:          STM32 hardware SHA-3                  default: off

 * PSOC6_HASH_SHA3:          PSoC6 hardware SHA-3                  default: off

 */

#define WC_FIPS_LL_CRYPTO

#define _WC_BUILDING_SHA3_C

#include <wolfssl/wolfcrypt/libwolfssl_sources.h>

#ifdef WC_SHA3_NO_ASM

    #undef USE_INTEL_SPEEDUP

    #undef WOLFSSL_ARMASM

    #undef WOLFSSL_RISCV_ASM

#endif

#ifdef WOLFSSL_X86_BUILD

    #undef USE_INTEL_SPEEDUP

#endif

#if defined(WOLFSSL_PSOC6_CRYPTO)

    #include <wolfssl/wolfcrypt/port/cypress/psoc6_crypto.h>

#endif

#if defined(WOLFSSL_SHA3) && !defined(WOLFSSL_XILINX_CRYPT) && \

   !defined(WOLFSSL_AFALG_XILINX_SHA3)

#if FIPS_VERSION3_GE(2,0,0)

    #ifdef USE_WINDOWS_API

        #pragma code_seg(".fipsA$n")

        #pragma const_seg(".fipsB$n")

    #endif

#endif

#include <wolfssl/wolfcrypt/sha3.h>

#include <wolfssl/wolfcrypt/hash.h>

#ifdef WOLF_CRYPTO_CB

    #include <wolfssl/wolfcrypt/cryptocb.h>

#endif

#ifdef NO_INLINE

    #include <wolfssl/wolfcrypt/misc.h>

#else

    #define WOLFSSL_MISC_INCLUDED

    #include <wolfcrypt/src/misc.c>

#endif

/* Gates the non-WOLFSSL_SHA3_SMALL software Keccak primitives

 * (hash_keccak_r, BlockSha3, InitSha3, Sha3Update, Sha3Final and the

 * Load64* helpers). Compiled when:

 *  - No HW SHA-3 backend is selected (the original baseline), OR

 *  - STM32 HW SHA-3 is selected and SHAKE is enabled - SHAKE on STM32MP13

 *    runs in software because the HASH peripheral's SHAKE support is

 *    fixed-length and does not match wolfSSL's variable-length / iterative

 *    SqueezeBlocks API. SHA-3 still uses the HASH peripheral.

 *

 * Note: the WOLFSSL_SHA3_SMALL branch earlier in this file defines its

 * own hash_keccak_r and BlockSha3 unconditionally inside its #ifdef

 * block, so this macro only controls the non-SMALL implementation. */

#if (!defined(STM32_HASH_SHA3) && !defined(PSOC6_HASH_SHA3)) || \

    (defined(STM32_HASH_SHA3) && \

     (defined(WOLFSSL_SHAKE128) || defined(WOLFSSL_SHAKE256)))

    #define WC_SHA3_SW_KECCAK

#endif

#if FIPS_VERSION3_GE(6,0,0)

    const unsigned int wolfCrypt_FIPS_sha3_ro_sanity[2] =

                                                     { 0x1a2b3c4d, 0x00000016 };

    int wolfCrypt_FIPS_SHA3_sanity(void)

    {

        return 0;

    }

#endif

#if defined(USE_INTEL_SPEEDUP) || (defined(__aarch64__) && \

        defined(WOLFSSL_ARMASM))

    #include <wolfssl/wolfcrypt/cpuid.h>

    static cpuid_flags_t cpuid_flags = WC_CPUID_INITIALIZER;

#ifdef WC_C_DYNAMIC_FALLBACK

    #define SHA3_BLOCK (sha3->sha3_block)

    #define SHA3_BLOCK_N (sha3->sha3_block_n)

#else

    void (*sha3_block)(word64 *s) = NULL;

    void (*sha3_block_n)(word64 *s, const byte* data, word32 n,

        word64 c) = NULL;

    #define SHA3_BLOCK sha3_block

    #define SHA3_BLOCK_N sha3_block_n

#endif

#endif

#ifdef USE_INTEL_SPEEDUP

    /* Block-function selection when USE_INTEL_SPEEDUP: AVX2 on Intel, else

     * BMI2, else the C block.  Measured single-instance Keccak-f[1600]

     * (Ethereum "Optimizing Keccak"; OpenSSL keccak1600-x86_64.pl): AVX2 is

     * ~13-17% faster than BMI2 on Intel Haswell..Skylake, tied on Ice Lake,

     * but ~2x SLOWER on AMD Zen, so AVX2 is Intel-only.  (Single-stream

     * AVX-512 is vpermt2q-bound and slower than BMI2 everywhere measured, so

     * it is not built - see scripts sha3_avx512.rb.)

     * Overrides: WOLFSSL_SHA3_AVX2 forces AVX2 on any vendor with it;

     *            WOLFSSL_SHA3_NO_AVX2 never uses AVX2. */

    /* SHA3_USE_AVX2() is defined in sha3.h - shared with ML-DSA. */

    /* True when the selected block function uses vector registers and so

     * needs the caller to save/restore them.  BMI2 and the C block use only

     * general registers. */

#ifdef WOLFSSL_SHA3_NO_AVX2

    #define SHA3_BLOCK_VREGS(f) 0

#else

    #define SHA3_BLOCK_VREGS(f) ((f) == sha3_block_avx2)

#endif

#endif

#if !defined(WOLFSSL_ARMASM) && !defined(WOLFSSL_RISCV_ASM) && \

    !defined(WOLFSSL_PPC64_ASM) && !defined(WOLFSSL_PPC32_ASM)

#ifdef WOLFSSL_SHA3_SMALL

/* Rotate a 64-bit value left.

 *

 * a  Number to rotate left.

 * r  Number od bits to rotate left.

 * returns the rotated number.

 */

#define ROTL64(a, n)    (((a)<<(n))|((a)>>(64-(n))))

/* An array of values to XOR for block operation. */

static const word64 hash_keccak_r[24] =

{

    0x0000000000000001UL, 0x0000000000008082UL,

    0x800000000000808aUL, 0x8000000080008000UL,

    0x000000000000808bUL, 0x0000000080000001UL,

    0x8000000080008081UL, 0x8000000000008009UL,

    0x000000000000008aUL, 0x0000000000000088UL,

    0x0000000080008009UL, 0x000000008000000aUL,

    0x000000008000808bUL, 0x800000000000008bUL,

    0x8000000000008089UL, 0x8000000000008003UL,

    0x8000000000008002UL, 0x8000000000000080UL,

    0x000000000000800aUL, 0x800000008000000aUL,

    0x8000000080008081UL, 0x8000000000008080UL,

    0x0000000080000001UL, 0x8000000080008008UL

};

/* Indices used in swap and rotate operation. */

#define K_I_0   10

#define K_I_1    7

#define K_I_2   11

#define K_I_3   17

#define K_I_4   18

#define K_I_5    3

#define K_I_6    5

#define K_I_7   16

#define K_I_8    8

#define K_I_9   21

#define K_I_10  24

#define K_I_11   4

#define K_I_12  15

#define K_I_13  23

#define K_I_14  19

#define K_I_15  13

#define K_I_16  12

#define K_I_17   2

#define K_I_18  20

#define K_I_19  14

#define K_I_20  22

#define K_I_21   9

#define K_I_22   6

#define K_I_23   1

/* Number of bits to rotate in swap and rotate operation. */

#define K_R_0    1

#define K_R_1    3

#define K_R_2    6

#define K_R_3   10

#define K_R_4   15

#define K_R_5   21

#define K_R_6   28

#define K_R_7   36

#define K_R_8   45

#define K_R_9   55

#define K_R_10   2

#define K_R_11  14

#define K_R_12  27

#define K_R_13  41

#define K_R_14  56

#define K_R_15   8

#define K_R_16  25

#define K_R_17  43

#define K_R_18  62

#define K_R_19  18

#define K_R_20  39

#define K_R_21  61

#define K_R_22  20

#define K_R_23  44

/* Swap and rotate left operation.

 *

 * s   The state.

 * t1  Temporary value.

 * t2  Second temporary value.

 * i   The index of the loop.

 */

#define SWAP_ROTL(s, t1, t2, i)                                         \

do {                                                                    \

    t2 = s[K_I_##i]; s[K_I_##i] = ROTL64(t1, K_R_##i);                  \

}                                                                       \

while (0)

/* Mix the XOR of the column's values into each number by column.

 *

 * s  The state.

 * b  Temporary array of XORed column values.

 * x  The index of the column.

 * t  Temporary variable.

 */

#define COL_MIX(s, b, x, t)                                             \

do {                                                                    \

    for (x = 0; x < 5; x++)                                             \

        b[x] = s[x + 0] ^ s[x + 5] ^ s[x + 10] ^ s[x + 15] ^ s[x + 20]; \

    for (x = 0; x < 5; x++) {                                           \

        t = b[(x + 4) % 5] ^ ROTL64(b[(x + 1) % 5], 1);                 \

        s[x +  0] ^= t;                                                 \

        s[x +  5] ^= t;                                                 \

        s[x + 10] ^= t;                                                 \

        s[x + 15] ^= t;                                                 \

        s[x + 20] ^= t;                                                 \

    }                                                                   \

}                                                                       \

while (0)

#ifdef SHA3_BY_SPEC

/* Mix the row values.

 * BMI1 has ANDN instruction ((~a) & b) - Haswell and above.

 *

 * s   The state.

 * b   Temporary array of XORed row values.

 * y   The index of the row to work on.

 * x   The index of the column.

 * t0  Temporary variable.

 * t1  Temporary variable.

 */

#define ROW_MIX(s, b, y, x, t0, t1)                                     \

do {                                                                    \

    for (y = 0; y < 5; y++) {                                           \

        for (x = 0; x < 5; x++)                                         \

            b[x] = s[y * 5 + x];                                        \

        for (x = 0; x < 5; x++)                                         \

            s[y * 5 + x] = b[x] ^ (~b[(x + 1) % 5] & b[(x + 2) % 5]);   \

    }                                                                   \

}                                                                       \

while (0)

#else

/* Mix the row values.

 * a ^ (~b & c) == a ^ (c & (b ^ c)) == (a ^ b) ^ (b | c)

 *

 * s   The state.

 * b   Temporary array of XORed row values.

 * y   The index of the row to work on.

 * x   The index of the column.

 * t0  Temporary variable.

 * t1  Temporary variable.

 */

#define ROW_MIX(s, b, y, x, t12, t34)                                   \

do {                                                                    \

    for (y = 0; y < 5; y++) {                                           \

        for (x = 0; x < 5; x++)                                         \

            b[x] = s[y * 5 + x];                                        \

        t12 = (b[1] ^ b[2]); t34 = (b[3] ^ b[4]);                       \

        s[y * 5 + 0] = b[0] ^ (b[2] &  t12);                            \

        s[y * 5 + 1] =  t12 ^ (b[2] | b[3]);                            \

        s[y * 5 + 2] = b[2] ^ (b[4] &  t34);                            \

        s[y * 5 + 3] =  t34 ^ (b[4] | b[0]);                            \

        s[y * 5 + 4] = b[4] ^ (b[1] & (b[0] ^ b[1]));                   \

    }                                                                   \

}                                                                       \

while (0)

#endif /* SHA3_BY_SPEC */

/* The block operation performed on the state.

 *

 * s  The state.

 */

void BlockSha3(word64* s)

{

    byte i, x, y;

    word64 t0, t1;

    word64 b[5];

    for (i = 0; i < 24; i++)

    {

        COL_MIX(s, b, x, t0);

        t0 = s[1];

        SWAP_ROTL(s, t0, t1,  0);

        SWAP_ROTL(s, t1, t0,  1);

        SWAP_ROTL(s, t0, t1,  2);

        SWAP_ROTL(s, t1, t0,  3);

        SWAP_ROTL(s, t0, t1,  4);

        SWAP_ROTL(s, t1, t0,  5);

        SWAP_ROTL(s, t0, t1,  6);

        SWAP_ROTL(s, t1, t0,  7);

        SWAP_ROTL(s, t0, t1,  8);

        SWAP_ROTL(s, t1, t0,  9);

        SWAP_ROTL(s, t0, t1, 10);

        SWAP_ROTL(s, t1, t0, 11);

        SWAP_ROTL(s, t0, t1, 12);

        SWAP_ROTL(s, t1, t0, 13);

        SWAP_ROTL(s, t0, t1, 14);

        SWAP_ROTL(s, t1, t0, 15);

        SWAP_ROTL(s, t0, t1, 16);

        SWAP_ROTL(s, t1, t0, 17);

        SWAP_ROTL(s, t0, t1, 18);

        SWAP_ROTL(s, t1, t0, 19);

        SWAP_ROTL(s, t0, t1, 20);

        SWAP_ROTL(s, t1, t0, 21);

        SWAP_ROTL(s, t0, t1, 22);

        SWAP_ROTL(s, t1, t0, 23);

        ROW_MIX(s, b, y, x, t0, t1);

        s[0] ^= hash_keccak_r[i];

    }

}

#else

/* Rotate a 64-bit value left.

 *

 * a  Number to rotate left.

 * r  Number od bits to rotate left.

 * returns the rotated number.

 */

#define ROTL64(a, n)    (((a)<<(n))|((a)>>(64-(n))))

#ifdef WC_SHA3_SW_KECCAK

/* An array of values to XOR for block operation. */

static const word64 hash_keccak_r[24] =

{

    W64LIT(0x0000000000000001), W64LIT(0x0000000000008082),

    W64LIT(0x800000000000808a), W64LIT(0x8000000080008000),

    W64LIT(0x000000000000808b), W64LIT(0x0000000080000001),

    W64LIT(0x8000000080008081), W64LIT(0x8000000000008009),

    W64LIT(0x000000000000008a), W64LIT(0x0000000000000088),

    W64LIT(0x0000000080008009), W64LIT(0x000000008000000a),

    W64LIT(0x000000008000808b), W64LIT(0x800000000000008b),

    W64LIT(0x8000000000008089), W64LIT(0x8000000000008003),

    W64LIT(0x8000000000008002), W64LIT(0x8000000000000080),

    W64LIT(0x000000000000800a), W64LIT(0x800000008000000a),

    W64LIT(0x8000000080008081), W64LIT(0x8000000000008080),

    W64LIT(0x0000000080000001), W64LIT(0x8000000080008008)

};

#endif

/* Indices used in swap and rotate operation. */

#define KI_0     6

#define KI_1    12

#define KI_2    18

#define KI_3    24

#define KI_4     3

#define KI_5     9

#define KI_6    10

#define KI_7    16

#define KI_8    22

#define KI_9     1

#define KI_10    7

#define KI_11   13

#define KI_12   19

#define KI_13   20

#define KI_14    4

#define KI_15    5

#define KI_16   11

#define KI_17   17

#define KI_18   23

#define KI_19    2

#define KI_20    8

#define KI_21   14

#define KI_22   15

#define KI_23   21

/* Number of bits to rotate in swap and rotate operation. */

#define KR_0    44

#define KR_1    43

#define KR_2    21

#define KR_3    14

#define KR_4    28

#define KR_5    20

#define KR_6     3

#define KR_7    45

#define KR_8    61

#define KR_9     1

#define KR_10    6

#define KR_11   25

#define KR_12    8

#define KR_13   18

#define KR_14   27

#define KR_15   36

#define KR_16   10

#define KR_17   15

#define KR_18   56

#define KR_19   62

#define KR_20   55

#define KR_21   39

#define KR_22   41

#define KR_23    2

/* Mix the XOR of the column's values into each number by column.

 *

 * s  The state.

 * b  Temporary array of XORed column values.

 * x  The index of the column.

 * t  Temporary variable.

 */

#define COL_MIX(s, b, x, t)                                                         \

do {                                                                                \

    (b)[0] = (s)[0] ^ (s)[5] ^ (s)[10] ^ (s)[15] ^ (s)[20];                         \

    (b)[1] = (s)[1] ^ (s)[6] ^ (s)[11] ^ (s)[16] ^ (s)[21];                         \

    (b)[2] = (s)[2] ^ (s)[7] ^ (s)[12] ^ (s)[17] ^ (s)[22];                         \

    (b)[3] = (s)[3] ^ (s)[8] ^ (s)[13] ^ (s)[18] ^ (s)[23];                         \

    (b)[4] = (s)[4] ^ (s)[9] ^ (s)[14] ^ (s)[19] ^ (s)[24];                         \

    (t) = (b)[(0 + 4) % 5] ^ ROTL64((b)[(0 + 1) % 5], 1);                           \

    (s)[ 0] ^= (t); (s)[ 5] ^= (t); (s)[10] ^= (t); (s)[15] ^= (t); (s)[20] ^= (t); \

    (t) = (b)[(1 + 4) % 5] ^ ROTL64((b)[(1 + 1) % 5], 1);                           \

    (s)[ 1] ^= (t); (s)[ 6] ^= (t); (s)[11] ^= (t); (s)[16] ^= (t); (s)[21] ^= (t); \

    (t) = (b)[(2 + 4) % 5] ^ ROTL64((b)[(2 + 1) % 5], 1);                           \

    (s)[ 2] ^= (t); (s)[ 7] ^= (t); (s)[12] ^= (t); (s)[17] ^= (t); (s)[22] ^= (t); \

    (t) = (b)[(3 + 4) % 5] ^ ROTL64((b)[(3 + 1) % 5], 1);                           \

    (s)[ 3] ^= (t); (s)[ 8] ^= (t); (s)[13] ^= (t); (s)[18] ^= (t); (s)[23] ^= (t); \

    (t) = (b)[(4 + 4) % 5] ^ ROTL64((b)[(4 + 1) % 5], 1);                           \

    (s)[ 4] ^= (t); (s)[ 9] ^= (t); (s)[14] ^= (t); (s)[19] ^= (t); (s)[24] ^= (t); \

}                                                                                   \

while (0)

#define S(s1, i) ROTL64((s1)[KI_##i], KR_##i)

#ifdef SHA3_BY_SPEC

/* Mix the row values.

 * BMI1 has ANDN instruction ((~a) & b) - Haswell and above.

 *

 * s2  The new state.

 * s1  The current state.

 * b   Temporary array of XORed row values.

 * t0  Temporary variable. (Unused)

 * t1  Temporary variable. (Unused)

 */

#define ROW_MIX(s2, s1, b, t0, t1)                    \

do {                                                  \

    (b)[0] = (s1)[0];                                 \

    (b)[1] = S((s1), 0);                              \

    (b)[2] = S((s1), 1);                              \

    (b)[3] = S((s1), 2);                              \

    (b)[4] = S((s1), 3);                              \

    (s2)[0] = (b)[0] ^ (~(b)[1] & (b)[2]);            \

    (s2)[1] = (b)[1] ^ (~(b)[2] & (b)[3]);            \

    (s2)[2] = (b)[2] ^ (~(b)[3] & (b)[4]);            \

    (s2)[3] = (b)[3] ^ (~(b)[4] & (b)[0]);            \

    (s2)[4] = (b)[4] ^ (~(b)[0] & (b)[1]);            \

    (b)[0] = S((s1), 4);                              \

    (b)[1] = S((s1), 5);                              \

    (b)[2] = S((s1), 6);                              \

    (b)[3] = S((s1), 7);                              \

    (b)[4] = S((s1), 8);                              \

    (s2)[5] = (b)[0] ^ (~(b)[1] & (b)[2]);            \

    (s2)[6] = (b)[1] ^ (~(b)[2] & (b)[3]);            \

    (s2)[7] = (b)[2] ^ (~(b)[3] & (b)[4]);            \

    (s2)[8] = (b)[3] ^ (~(b)[4] & (b)[0]);            \

    (s2)[9] = (b)[4] ^ (~(b)[0] & (b)[1]);            \

    (b)[0] = S((s1), 9);                              \

    (b)[1] = S((s1), 10);                             \

    (b)[2] = S((s1), 11);                             \

    (b)[3] = S((s1), 12);                             \

    (b)[4] = S((s1), 13);                             \

    (s2)[10] = (b)[0] ^ (~(b)[1] & (b)[2]);           \

    (s2)[11] = (b)[1] ^ (~(b)[2] & (b)[3]);           \

    (s2)[12] = (b)[2] ^ (~(b)[3] & (b)[4]);           \

    (s2)[13] = (b)[3] ^ (~(b)[4] & (b)[0]);           \

    (s2)[14] = (b)[4] ^ (~(b)[0] & (b)[1]);           \

    (b)[0] = S((s1), 14);                             \

    (b)[1] = S((s1), 15);                             \

    (b)[2] = S((s1), 16);                             \

    (b)[3] = S((s1), 17);                             \

    (b)[4] = S((s1), 18);                             \

    (s2)[15] = (b)[0] ^ (~(b)[1] & (b)[2]);           \

    (s2)[16] = (b)[1] ^ (~(b)[2] & (b)[3]);           \

    (s2)[17] = (b)[2] ^ (~(b)[3] & (b)[4]);           \

    (s2)[18] = (b)[3] ^ (~(b)[4] & (b)[0]);           \

    (s2)[19] = (b)[4] ^ (~(b)[0] & (b)[1]);           \

    (b)[0] = S((s1), 19);                             \

    (b)[1] = S((s1), 20);                             \

    (b)[2] = S((s1), 21);                             \

    (b)[3] = S((s1), 22);                             \

    (b)[4] = S((s1), 23);                             \

    (s2)[20] = (b)[0] ^ (~(b)[1] & (b)[2]);           \

    (s2)[21] = (b)[1] ^ (~(b)[2] & (b)[3]);           \

    (s2)[22] = (b)[2] ^ (~(b)[3] & (b)[4]);           \

    (s2)[23] = (b)[3] ^ (~(b)[4] & (b)[0]);           \

    (s2)[24] = (b)[4] ^ (~(b)[0] & (b)[1]);           \

}                                                     \

while (0)

#else

/* Mix the row values.

 * a ^ (~b & c) == a ^ (c & (b ^ c)) == (a ^ b) ^ (b | c)

 *

 * s2  The new state.

 * s1  The current state.

 * b   Temporary array of XORed row values.

 * t12 Temporary variable.

 * t34 Temporary variable.

 */

#define ROW_MIX(s2, s1, b, t12, t34)                      \

do {                                                      \

    (b)[0] = (s1)[0];                                     \

    (b)[1] = S((s1), 0);                                  \

    (b)[2] = S((s1), 1);                                  \

    (b)[3] = S((s1), 2);                                  \

    (b)[4] = S((s1), 3);                                  \

    (t12) = ((b)[1] ^ (b)[2]); (t34) = ((b)[3] ^ (b)[4]); \

    (s2)[0] = (b)[0] ^ ((b)[2] &  (t12));                 \

    (s2)[1] =  (t12) ^ ((b)[2] | (b)[3]);                 \

    (s2)[2] = (b)[2] ^ ((b)[4] &  (t34));                 \

    (s2)[3] =  (t34) ^ ((b)[4] | (b)[0]);                 \

    (s2)[4] = (b)[4] ^ ((b)[1] & ((b)[0] ^ (b)[1]));      \

    (b)[0] = S((s1), 4);                                  \

    (b)[1] = S((s1), 5);                                  \

    (b)[2] = S((s1), 6);                                  \

    (b)[3] = S((s1), 7);                                  \

    (b)[4] = S((s1), 8);                                  \

    (t12) = ((b)[1] ^ (b)[2]); (t34) = ((b)[3] ^ (b)[4]); \

    (s2)[5] = (b)[0] ^ ((b)[2] &  (t12));                 \

    (s2)[6] =  (t12) ^ ((b)[2] | (b)[3]);                 \

    (s2)[7] = (b)[2] ^ ((b)[4] &  (t34));                 \

    (s2)[8] =  (t34) ^ ((b)[4] | (b)[0]);                 \

    (s2)[9] = (b)[4] ^ ((b)[1] & ((b)[0] ^ (b)[1]));      \

    (b)[0] = S((s1), 9);                                  \

    (b)[1] = S((s1), 10);                                 \

    (b)[2] = S((s1), 11);                                 \

    (b)[3] = S((s1), 12);                                 \

    (b)[4] = S((s1), 13);                                 \

    (t12) = ((b)[1] ^ (b)[2]); (t34) = ((b)[3] ^ (b)[4]); \

    (s2)[10] = (b)[0] ^ ((b)[2] &  (t12));                \

    (s2)[11] =  (t12) ^ ((b)[2] | (b)[3]);                \

    (s2)[12] = (b)[2] ^ ((b)[4] &  (t34));                \

    (s2)[13] =  (t34) ^ ((b)[4] | (b)[0]);                \

    (s2)[14] = (b)[4] ^ ((b)[1] & ((b)[0] ^ (b)[1]));     \

    (b)[0] = S((s1), 14);                                 \

    (b)[1] = S((s1), 15);                                 \

    (b)[2] = S((s1), 16);                                 \

    (b)[3] = S((s1), 17);                                 \

    (b)[4] = S((s1), 18);                                 \

    (t12) = ((b)[1] ^ (b)[2]); (t34) = ((b)[3] ^ (b)[4]); \

    (s2)[15] = (b)[0] ^ ((b)[2] &  (t12));                \

    (s2)[16] =  (t12) ^ ((b)[2] | (b)[3]);                \

    (s2)[17] = (b)[2] ^ ((b)[4] &  (t34));                \

    (s2)[18] =  (t34) ^ ((b)[4] | (b)[0]);                \

    (s2)[19] = (b)[4] ^ ((b)[1] & ((b)[0] ^ (b)[1]));     \

    (b)[0] = S((s1), 19);                                 \

    (b)[1] = S((s1), 20);                                 \

    (b)[2] = S((s1), 21);                                 \

    (b)[3] = S((s1), 22);                                 \

    (b)[4] = S((s1), 23);                                 \

    (t12) = ((b)[1] ^ (b)[2]); (t34) = ((b)[3] ^ (b)[4]); \

    (s2)[20] = (b)[0] ^ ((b)[2] &  (t12));                \

    (s2)[21] =  (t12) ^ ((b)[2] | (b)[3]);                \

    (s2)[22] = (b)[2] ^ ((b)[4] &  (t34));                \

    (s2)[23] =  (t34) ^ ((b)[4] | (b)[0]);                \

    (s2)[24] = (b)[4] ^ ((b)[1] & ((b)[0] ^ (b)[1]));     \

}                                                         \

while (0)

#endif /* SHA3_BY_SPEC */

#ifdef WC_SHA3_SW_KECCAK

/* The block operation performed on the state.

 *

 * s  The state.

 */

/* WC_16BIT_CPU (e.g. TI C28x) lowers every 64-bit ^, | and & to an out-of-line

 * runtime-helper call (cl2000: __c28xabi_xorll / _orll / _andll), which

 * dominates the Keccak permutation.  Auto-select a BlockSha3 that runs on

 * 32-bit halves so the compiler emits native 32-bit ops; external state stays

 * word64 s[25].  Auto-enabled only for WOLFSSL_WIDE_BYTE (the hardware-validated

 * targets); other little-endian 16-bit ports keep the long-tested generic

 * permutation but can opt in by defining WC_SHA3_SPLIT64.  Little-endian word

 * layout assumed (lo half first). */

#if !defined(WC_SHA3_SPLIT64) && defined(WOLFSSL_WIDE_BYTE) && \

    !defined(BIG_ENDIAN_ORDER)

    #define WC_SHA3_SPLIT64

#endif

#ifdef WC_SHA3_SPLIT64

/* Rotate the 64-bit value (sl=low, sh=high) left by compile-time constant r in

 * 1..63, r != 32, into (dl, dh).  r is always a Keccak rho offset (never 0 or

 * 32; r==32 would need a plain half-swap), so that case never occurs.  The & 31

 * keeps the shift count in range in the dead (compile-time-eliminated) branch

 * so there is no undefined shift. */

#define WC_SHA3_RL(dl, dh, sl, sh, r)                                        \

    do {                                                                     \

        word32 _l = (sl), _h = (sh);                                         \

        if ((r) < 32) {                                                      \

            (dl) = (word32)((_l << ((r) & 31)) | (_h >> ((32 - (r)) & 31))); \

            (dh) = (word32)((_h << ((r) & 31)) | (_l >> ((32 - (r)) & 31))); \

        }                                                                    \

        else {                                                               \

            (dl) = (word32)((_h << (((r) - 32) & 31)) |                      \

                            (_l >> ((64 - (r)) & 31)));                      \

            (dh) = (word32)((_l << (((r) - 32) & 31)) |                      \

                            (_h >> ((64 - (r)) & 31)));                      \

        }                                                                    \

    } while (0)

/* Chi over the rotated row held in bl[0..4]/bh[0..4], writing five output lanes

 * at (DL,DH)[k..k+4].  a ^ (~b & c) == (a ^ b) ^ (b | c) per half. */

#define WC_SHA3_CHI(DL, DH, k)                                  \

    do {                                                        \

        word32 al = bl[1] ^ bl[2], ah = bh[1] ^ bh[2];          \

        word32 cl = bl[3] ^ bl[4], ch = bh[3] ^ bh[4];          \

        (DL)[(k)+0] = bl[0] ^ (bl[2] &  al);                    \

        (DH)[(k)+0] = bh[0] ^ (bh[2] &  ah);                    \

        (DL)[(k)+1] =  al   ^ (bl[2] | bl[3]);                  \

        (DH)[(k)+1] =  ah   ^ (bh[2] | bh[3]);                  \

        (DL)[(k)+2] = bl[2] ^ (bl[4] &  cl);                    \

        (DH)[(k)+2] = bh[2] ^ (bh[4] &  ch);                    \

        (DL)[(k)+3] =  cl   ^ (bl[4] | bl[0]);                  \

        (DH)[(k)+3] =  ch   ^ (bh[4] | bh[0]);                  \

        (DL)[(k)+4] = bl[4] ^ (bl[1] & (bl[0] ^ bl[1]));        \

        (DH)[(k)+4] = bh[4] ^ (bh[1] & (bh[0] ^ bh[1]));        \

    } while (0)

/* Theta: mix the column parities into split state L (low) / H (high). */

#define WC_SHA3_THETA(L, H)                                                   \

    do {                                                                      \

        int c;                                                                \

        for (c = 0; c < 5; c++) {                                             \

            bl[c] = (L)[c]^(L)[c+5]^(L)[c+10]^(L)[c+15]^(L)[c+20];            \

            bh[c] = (H)[c]^(H)[c+5]^(H)[c+10]^(H)[c+15]^(H)[c+20];            \

        }                                                                     \

        for (c = 0; c < 5; c++) {                                             \

            int d = (c + 1) % 5, e = (c + 4) % 5;                             \

            word32 xl = bl[e] ^ (word32)((bl[d] << 1) | (bh[d] >> 31));       \

            word32 xh = bh[e] ^ (word32)((bh[d] << 1) | (bl[d] >> 31));       \

            (L)[c]   ^= xl; (H)[c]   ^= xh; (L)[c+5]  ^= xl; (H)[c+5]  ^= xh; \

            (L)[c+10]^= xl; (H)[c+10]^= xh; (L)[c+15] ^= xl; (H)[c+15] ^= xh; \

            (L)[c+20]^= xl; (H)[c+20]^= xh;                                   \

        }                                                                     \

    } while (0)

/* Rho + pi + chi: rotate/permute split state SL/SH into DL/DH. */

#define WC_SHA3_ROWMIX(DL, DH, SL, SH)                            \

    do {                                                          \

        bl[0] = (SL)[0]; bh[0] = (SH)[0];                         \

        WC_SHA3_RL(bl[1],bh[1], (SL)[KI_0], (SH)[KI_0],  KR_0);   \

        WC_SHA3_RL(bl[2],bh[2], (SL)[KI_1], (SH)[KI_1],  KR_1);   \

        WC_SHA3_RL(bl[3],bh[3], (SL)[KI_2], (SH)[KI_2],  KR_2);   \

        WC_SHA3_RL(bl[4],bh[4], (SL)[KI_3], (SH)[KI_3],  KR_3);   \

        WC_SHA3_CHI(DL, DH, 0);                                   \

        WC_SHA3_RL(bl[0],bh[0], (SL)[KI_4], (SH)[KI_4],  KR_4);   \

        WC_SHA3_RL(bl[1],bh[1], (SL)[KI_5], (SH)[KI_5],  KR_5);   \

        WC_SHA3_RL(bl[2],bh[2], (SL)[KI_6], (SH)[KI_6],  KR_6);   \

        WC_SHA3_RL(bl[3],bh[3], (SL)[KI_7], (SH)[KI_7],  KR_7);   \

        WC_SHA3_RL(bl[4],bh[4], (SL)[KI_8], (SH)[KI_8],  KR_8);   \

        WC_SHA3_CHI(DL, DH, 5);                                   \

        WC_SHA3_RL(bl[0],bh[0], (SL)[KI_9], (SH)[KI_9],  KR_9);   \

        WC_SHA3_RL(bl[1],bh[1], (SL)[KI_10],(SH)[KI_10], KR_10);  \

        WC_SHA3_RL(bl[2],bh[2], (SL)[KI_11],(SH)[KI_11], KR_11);  \

        WC_SHA3_RL(bl[3],bh[3], (SL)[KI_12],(SH)[KI_12], KR_12);  \

        WC_SHA3_RL(bl[4],bh[4], (SL)[KI_13],(SH)[KI_13], KR_13);  \

        WC_SHA3_CHI(DL, DH, 10);                                  \

        WC_SHA3_RL(bl[0],bh[0], (SL)[KI_14],(SH)[KI_14], KR_14);  \

        WC_SHA3_RL(bl[1],bh[1], (SL)[KI_15],(SH)[KI_15], KR_15);  \

        WC_SHA3_RL(bl[2],bh[2], (SL)[KI_16],(SH)[KI_16], KR_16);  \

        WC_SHA3_RL(bl[3],bh[3], (SL)[KI_17],(SH)[KI_17], KR_17);  \

        WC_SHA3_RL(bl[4],bh[4], (SL)[KI_18],(SH)[KI_18], KR_18);  \

        WC_SHA3_CHI(DL, DH, 15);                                  \

        WC_SHA3_RL(bl[0],bh[0], (SL)[KI_19],(SH)[KI_19], KR_19);  \

        WC_SHA3_RL(bl[1],bh[1], (SL)[KI_20],(SH)[KI_20], KR_20);  \

        WC_SHA3_RL(bl[2],bh[2], (SL)[KI_21],(SH)[KI_21], KR_21);  \

        WC_SHA3_RL(bl[3],bh[3], (SL)[KI_22],(SH)[KI_22], KR_22);  \

        WC_SHA3_RL(bl[4],bh[4], (SL)[KI_23],(SH)[KI_23], KR_23);  \

        WC_SHA3_CHI(DL, DH, 20);                                  \

    } while (0)

void BlockSha3(word64* s)

{

    /* Process the 25 little-endian lanes as 32-bit halves to avoid 64-bit

     * helper calls.  XMEMCPY in/out (aliasing s through word32* is strict-

     * aliasing UB); st[2k] is lane k's low half, st[2k+1] the high half.

     * Round constants are split with shifts for the same reason. */

    word32 st[50];

    word32 sl[25], sh[25], nl[25], nh[25], bl[5], bh[5];

    word32 i, k;

    word64 rc;

    XMEMCPY(st, s, sizeof(st));

    for (k = 0; k < 25; k++) {

        sl[k] = st[2 * k];

        sh[k] = st[2 * k + 1];

    }

    for (i = 0; i < 24; i += 2) {

        WC_SHA3_THETA(sl, sh);

        WC_SHA3_ROWMIX(nl, nh, sl, sh);

        rc = hash_keccak_r[i];

        nl[0] ^= (word32)rc;          nh[0] ^= (word32)(rc >> 32);

        WC_SHA3_THETA(nl, nh);

        WC_SHA3_ROWMIX(sl, sh, nl, nh);

        rc = hash_keccak_r[i + 1];

        sl[0] ^= (word32)rc;          sh[0] ^= (word32)(rc >> 32);

    }

    for (k = 0; k < 25; k++) {

        st[2 * k]     = sl[k];

        st[2 * k + 1] = sh[k];

    }

    XMEMCPY(s, st, sizeof(st));

}

#undef WC_SHA3_RL

#undef WC_SHA3_CHI

#undef WC_SHA3_THETA

#undef WC_SHA3_ROWMIX

#else /* !WC_SHA3_SPLIT64 */

void BlockSha3(word64* s)

{

    word64 n[25];

    word64 b[5];

    word64 t0;

#ifndef SHA3_BY_SPEC

    word64 t1;

#endif

    word32 i;

    for (i = 0; i < 24; i += 2)

    {

        COL_MIX(s, b, x, t0);

        ROW_MIX(n, s, b, t0, t1);

        n[0] ^= hash_keccak_r[i];

        COL_MIX(n, b, x, t0);

        ROW_MIX(s, n, b, t0, t1);

        s[0] ^= hash_keccak_r[i+1];

    }

}

#endif /* WC_SHA3_SPLIT64 */

#endif /* WC_SHA3_SW_KECCAK */

#endif /* !WOLFSSL_SHA3_SMALL */

#endif /* !WOLFSSL_ARMASM && !WOLFSSL_RISCV_ASM && !WOLFSSL_PPC64_ASM &&

        * !WOLFSSL_PPC32_ASM */

#if defined(WOLFSSL_PPC64_ASM)

#if defined(WOLFSSL_PPC64_ASM_POWER8)

/* PowerPC64 provides two Keccak-f[1600] implementations: the scalar

 * BlockSha3_base and a POWER8 (PowerISA 2.07) VSX BlockSha3_power8 (which uses

 * vrld/mtvsrd).  Select the POWER8 one at run time when the CPU is POWER8 or

 * later.

 *

 * A run-time flag with direct calls is used rather than a function pointer: an

 * indirect call would require an ELFv1 function descriptor, whereas direct

 * calls work under both the ELFv1 and ELFv2 ABIs. */

#include <wolfssl/wolfcrypt/cpuid.h>

/* -1 = not yet determined, 0 = base, 1 = POWER8 */

static int sha3_use_power8 = -1;

void BlockSha3(word64* s)

{

    if (sha3_use_power8 < 0) {

        word32 f = cpuid_get_flags();

        /* The VSX permutation is only worthwhile where the scalar issue width

         * does not already win.  POWER9 (PowerISA 3.0 but not 3.1) has enough

         * scalar throughput that BlockSha3_base is faster, so use the VSX path

         * only on POWER8 and on POWER10 (3.1) or later. */

        sha3_use_power8 = IS_PPC64_ARCH_2_07(f) &&

            (!IS_PPC64_ARCH_3_00(f) || IS_PPC64_ARCH_3_1(f));

    }

    if (sha3_use_power8)

        BlockSha3_power8(s);

    else

        BlockSha3_base(s);

}

#else

/* Only the scalar implementation is built; call it directly (no run-time

 * dispatch, no function pointer). */

void BlockSha3(word64* s)

{

    BlockSha3_base(s);

}

#endif

#endif

/* Scalar PowerPC32 assembly provides BlockSha3 directly (see

 * wolfcrypt/src/port/ppc32/ppc32-sha3-asm.S), so nothing is needed here. */

#ifdef WC_SHA3_SW_KECCAK

#if defined(BIG_ENDIAN_ORDER) || defined(WOLFSSL_WIDE_BYTE)

static WC_INLINE word64 Load64Unaligned(const unsigned char *a)

{

    return ((word64)a[0] <<  0) |

           ((word64)a[1] <<  8) |

           ((word64)a[2] << 16) |

           ((word64)a[3] << 24) |

           ((word64)a[4] << 32) |

           ((word64)a[5] << 40) |

           ((word64)a[6] << 48) |

           ((word64)a[7] << 56);

}

/* Convert the array of bytes, in little-endian order, to a 64-bit integer.

 *

 * a  Array of bytes.

 * returns a 64-bit integer.

 */

static word64 Load64BitLittleEndian(const byte* a)

{

    word64 n = 0;

    int i;

    for (i = 0; i < 8; i++)

        n |= (word64)a[i] << (8 * i);

    return n;

}

#elif defined(WC_SHA3_FAULT_HARDEN)

static WC_INLINE word64 Load64Unaligned(const unsigned char *a) {

    return readUnalignedWord64(a);

}

/* Convert the array of bytes, in little-endian order, to a 64-bit integer.

 *

 * a  Array of bytes.

 * returns a 64-bit integer.

 */

static word64 Load64BitLittleEndian(const byte* a)

{

    return Load64Unaligned(a);

}

#endif

/* Initialize the state for a SHA3-224 hash operation.

 *

 * sha3   wc_Sha3 object holding state.

 * returns 0 on success.

 */

static int InitSha3(wc_Sha3* sha3)

{

    int i;

    for (i = 0; i < 25; i++)

        sha3->s[i] = 0;

    XMEMSET(sha3->t, 0, sizeof(sha3->t));

    sha3->i = 0;

#ifdef WOLFSSL_HASH_FLAGS

    sha3->flags = 0;

#endif

#ifdef WOLF_CRYPTO_CB

    /* Cached hash variant is tied to sponge state; clear it whenever the

     * state is reset so reuse for a different SHA3 variant dispatches

     * correctly through the crypto callback. */

    sha3->hashType = WC_HASH_TYPE_NONE;

#endif

#ifdef USE_INTEL_SPEEDUP

    {

        int cpuid_flags_were_updated = cpuid_get_flags_ex(&cpuid_flags);

#ifdef WC_C_DYNAMIC_FALLBACK

        (void)cpuid_flags_were_updated;

        if (! CAN_SAVE_VECTOR_REGISTERS()) {

            SHA3_BLOCK = BlockSha3;

            SHA3_BLOCK_N = NULL;

        }

        else

#else

        if ((! cpuid_flags_were_updated) && (SHA3_BLOCK != NULL)) {

        }

        else

#endif

        /* See the selection comment above: AVX2 on Intel, otherwise BMI2. */

        if (SHA3_USE_AVX2(cpuid_flags)) {

            SHA3_BLOCK = sha3_block_avx2;

            SHA3_BLOCK_N = sha3_block_n_avx2;

        }

        else if (IS_INTEL_BMI1(cpuid_flags) && IS_INTEL_BMI2(cpuid_flags)) {

            SHA3_BLOCK = sha3_block_bmi2;

            SHA3_BLOCK_N = sha3_block_n_bmi2;

        }

        else {

            SHA3_BLOCK = BlockSha3;

            SHA3_BLOCK_N = NULL;

        }

    }

#define SHA3_FUNC_PTR

#endif /* USE_INTEL_SPEEDUP */

#if defined(__aarch64__) && defined(WOLFSSL_ARMASM)

    {

        int cpuid_flags_were_updated = cpuid_get_flags_ex(&cpuid_flags);

        if ((! cpuid_flags_were_updated) && (SHA3_BLOCK != NULL)) {

        }

        else

    #ifdef WOLFSSL_ARMASM_CRYPTO_SHA3

        if (IS_AARCH64_SHA3(cpuid_flags)) {

            SHA3_BLOCK = BlockSha3_crypto;

            SHA3_BLOCK_N = NULL;

        }

        else

    #endif

        {

            SHA3_BLOCK = BlockSha3_base;

            SHA3_BLOCK_N = NULL;

        }

    }

#define SHA3_FUNC_PTR

#endif

    return 0;

}

#if defined(__aarch64__) && defined(WOLFSSL_ARMASM)

void BlockSha3(word64* s)

{

    (*SHA3_BLOCK)(s);

}

#endif

/* Update the SHA-3 hash state with message data.

 *

 * sha3  wc_Sha3 object holding state.

 * data  Message data to be hashed.

 * len   Length of the message data.

 * p     Number of 64-bit numbers in a block of data to process.

 * returns 0 on success.

 */

static int Sha3Update(wc_Sha3* sha3, const byte* data, word32 len, word32 p)

{

    word32 i;

    word32 blocks;

    int ret = 0;

#ifdef WC_SHA3_FAULT_HARDEN

    word32 check = 0;

    word32 total_check = 0;

#endif

#ifdef USE_INTEL_SPEEDUP

#ifdef WC_C_DYNAMIC_FALLBACK

    void (*sha3_block)(word64 *s) = SHA3_BLOCK;

    void (*sha3_block_n)(word64 *s, const byte* data, word32 n,

        word64 c) = SHA3_BLOCK_N;

#endif

#endif /* USE_INTEL_SPEEDUP */

    if ((p < WC_SHA3_512_COUNT) || (p > WC_SHA3_128_COUNT))

        return BAD_STATE_E;

#ifdef USE_INTEL_SPEEDUP

    if (SHA3_BLOCK_VREGS(sha3_block)) {

        ret = SAVE_VECTOR_REGISTERS2();

        if (ret != 0) {

#ifdef WC_C_DYNAMIC_FALLBACK

            sha3_block = BlockSha3;

            sha3_block_n = NULL;

            ret = 0;

#else

            return ret;

#endif

        }

    }

#endif /* USE_INTEL_SPEEDUP */

    if (sha3->i > 0) {

        byte *t;

        word32 l;

        if (p * 8 < sha3->i) {

            ret = BAD_STATE_E;

            goto out;

        }

        l = (p * 8 - sha3->i);

        if (l > len) {

            l = len;

        }

        t = &sha3->t[sha3->i];

        for (i = 0; i < l; i++) {

            t[i] = data[i];

    #ifdef WC_SHA3_FAULT_HARDEN

            check++;

    #endif

        }

    #ifdef WC_SHA3_FAULT_HARDEN

        if (check != l) {

            ret = BAD_COND_E;

            goto out;

        }

        total_check += l;

    #endif

        data += i;

        len -= i;

        sha3->i += i;

        if (sha3->i == p * 8) {

    #if !defined(BIG_ENDIAN_ORDER) && !defined(WC_SHA3_FAULT_HARDEN) && \

        !defined(WOLFSSL_WIDE_BYTE)

            xorbuf(sha3->s, sha3->t, (word32)(p * 8));

    #else

            for (i = 0; i < p; i++) {

                sha3->s[i] ^= Load64BitLittleEndian(sha3->t + 8 * i);

            #ifdef WC_SHA3_FAULT_HARDEN

                check++;

            #endif

            }

        #ifdef WC_SHA3_FAULT_HARDEN

            if (check != p + l) {

                ret = BAD_COND_E;

                goto out;

            }

            total_check += p;

        #endif

    #endif

        #ifdef SHA3_FUNC_PTR

            (*sha3_block)(sha3->s);

        #else

            BlockSha3(sha3->s);

        #endif

            sha3->i = 0;

        }

    }

    blocks = len / (p * 8U);

    #ifdef SHA3_FUNC_PTR

    if ((sha3_block_n != NULL) && (blocks > 0)) {

        (*sha3_block_n)(sha3->s, data, blocks, p * 8U);

        len -= blocks * (p * 8U);

        data += blocks * (p * 8U);

        blocks = 0;

    }

    #endif

#ifdef WC_SHA3_FAULT_HARDEN

    total_check += blocks * p;

#endif

    for (; blocks > 0; blocks--) {

#if !defined(BIG_ENDIAN_ORDER) && !defined(WC_SHA3_FAULT_HARDEN) && \

    !defined(WOLFSSL_WIDE_BYTE)

        xorbuf(sha3->s, data, (word32)(p * 8));

#else