| 1 | /* |
| 2 | * SPDX-License-Identifier: GPL-2.0-or-later |
| 3 | * |
| 4 | * MIPS Octeon crypto emulation helpers. |
| 5 | * |
| 6 | * Copyright (c) 2026 James Hilliard |
| 7 | */ |
| 8 | |
| 9 | #include "qemu/osdep.h" |
| 10 | #include "cpu.h" |
| 11 | #include "internal.h" |
| 12 | #include "exec/helper-proto.h" |
| 13 | #include "crypto/aes.h" |
| 14 | #include "crypto/clmul.h" |
| 15 | #include "crypto/sm4.h" |
| 16 | #include "qemu/bitops.h" |
| 17 | #include "qemu/host-utils.h" |
| 18 | |
| 19 | #define OCTEON_LLM_NARROW_MASK ((1ULL << 36) - 1) |
| 20 | |
| 21 | static uint64_t octeon_llm_pack_narrow(uint64_t value) |
| 22 | { |
| 23 | value &= OCTEON_LLM_NARROW_MASK; |
| 24 | return value | ((uint64_t)(ctpop64(value) & 1) << 36); |
| 25 | } |
| 26 | |
| 27 | static void octeon_llm_read(MIPSOcteonCryptoState *crypto, unsigned int set, |
| 28 | uint64_t addr, bool wide) |
| 29 | { |
| 30 | uint64_t value; |
| 31 | |
| 32 | if (wide) { |
| 33 | value = mips_octeon_llm_load(crypto->llm64, addr); |
| 34 | } else { |
| 35 | value = octeon_llm_pack_narrow( |
| 36 | mips_octeon_llm_load(crypto->llm36, addr)); |
| 37 | } |
| 38 | |
| 39 | crypto->llm_data[set] = value; |
| 40 | } |
| 41 | |
| 42 | static void octeon_llm_write(MIPSOcteonCryptoState *crypto, unsigned int set, |
| 43 | uint64_t addr, bool wide) |
| 44 | { |
| 45 | uint64_t value = crypto->llm_data[set]; |
| 46 | |
| 47 | if (wide) { |
| 48 | mips_octeon_llm_store(&crypto->llm64, addr, value); |
| 49 | } else { |
| 50 | mips_octeon_llm_store(&crypto->llm36, addr, |
| 51 | value & OCTEON_LLM_NARROW_MASK); |
| 52 | } |
| 53 | } |
| 54 | |
| 55 | static uint32_t octeon_crc_reflect32_by_byte(uint32_t v) |
| 56 | { |
| 57 | return bswap32(revbit32(v)); |
| 58 | } |
| 59 | |
| 60 | static uint32_t octeon_crc_state_reflect(const MIPSOcteonCryptoState *crypto) |
| 61 | { |
| 62 | return octeon_crc_reflect32_by_byte(crypto->crc_iv); |
| 63 | } |
| 64 | |
| 65 | static void octeon_crc_set_state_reflect(MIPSOcteonCryptoState *crypto, |
| 66 | uint32_t state) |
| 67 | { |
| 68 | crypto->crc_iv = octeon_crc_reflect32_by_byte(state); |
| 69 | } |
| 70 | |
| 71 | static void octeon_crc_update_normal(MIPSOcteonCryptoState *crypto, |
| 72 | uint64_t value, unsigned int bytes) |
| 73 | { |
| 74 | uint32_t crc = crypto->crc_iv; |
| 75 | uint32_t poly = crypto->crc_poly; |
| 76 | |
| 77 | for (unsigned int i = 0; i < bytes; i++) { |
| 78 | uint8_t byte = value >> ((bytes - 1 - i) * 8); |
| 79 | |
| 80 | crc ^= (uint32_t)byte << 24; |
| 81 | for (int bit = 0; bit < 8; bit++) { |
| 82 | if (crc & 0x80000000U) { |
| 83 | crc = (crc << 1) ^ poly; |
| 84 | } else { |
| 85 | crc <<= 1; |
| 86 | } |
| 87 | } |
| 88 | } |
| 89 | |
| 90 | crypto->crc_iv = crc; |
| 91 | } |
| 92 | |
| 93 | static void octeon_crc_update_reflect(MIPSOcteonCryptoState *crypto, |
| 94 | uint64_t value, unsigned int bytes) |
| 95 | { |
| 96 | uint32_t crc = octeon_crc_state_reflect(crypto); |
| 97 | uint32_t poly = bswap32(crypto->crc_poly); |
| 98 | |
| 99 | for (unsigned int i = 0; i < bytes; i++) { |
| 100 | uint8_t byte = value >> ((bytes - 1 - i) * 8); |
| 101 | |
| 102 | crc ^= byte; |
| 103 | for (int bit = 0; bit < 8; bit++) { |
| 104 | if (crc & 1U) { |
| 105 | crc = (crc >> 1) ^ poly; |
| 106 | } else { |
| 107 | crc >>= 1; |
| 108 | } |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | octeon_crc_set_state_reflect(crypto, crc); |
| 113 | } |
| 114 | |
| 115 | static void octeon_gfm_mul(const uint64_t x[2], const uint64_t y[2], |
| 116 | uint16_t poly, uint64_t out[2]) |
| 117 | { |
| 118 | uint64_t zh = 0, zl = 0; |
| 119 | uint64_t vh = y[0], vl = y[1]; |
| 120 | uint64_t rh = (uint64_t)poly << 48; |
| 121 | int i; |
| 122 | |
| 123 | /* |
| 124 | * Keep the reflected-shift formulation used by Octeon software: the |
| 125 | * selector polynomial is already in reflected bit order, and the software |
| 126 | * view folds its 16 reduction bits from the top of the high word. |
| 127 | */ |
| 128 | for (i = 0; i < 128; i++) { |
| 129 | bool bit; |
| 130 | bool lsb; |
| 131 | |
| 132 | if (i < 64) { |
| 133 | bit = (x[0] >> (63 - i)) & 1; |
| 134 | } else { |
| 135 | bit = (x[1] >> (127 - i)) & 1; |
| 136 | } |
| 137 | if (bit) { |
| 138 | zh ^= vh; |
| 139 | zl ^= vl; |
| 140 | } |
| 141 | |
| 142 | lsb = vl & 1; |
| 143 | vl = (vh << 63) | (vl >> 1); |
| 144 | vh >>= 1; |
| 145 | if (lsb) { |
| 146 | vh ^= rh; |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | out[0] = zh; |
| 151 | out[1] = zl; |
| 152 | } |
| 153 | |
| 154 | static uint64_t octeon_gfm_reduce64(Int128 product, uint8_t poly) |
| 155 | { |
| 156 | uint64_t lo = int128_getlo(product); |
| 157 | uint64_t hi = int128_gethi(product); |
| 158 | |
| 159 | while (hi) { |
| 160 | int bit = 63 - clz64(hi); |
| 161 | |
| 162 | hi ^= 1ULL << bit; |
| 163 | lo ^= (uint64_t)poly << bit; |
| 164 | if (bit > 56) { |
| 165 | hi ^= (uint64_t)poly >> (64 - bit); |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | return lo; |
| 170 | } |
| 171 | |
| 172 | static void octeon_gfm_mul64_uia2(const uint64_t x[2], const uint64_t y[2], |
| 173 | uint8_t poly, uint64_t out[2]) |
| 174 | { |
| 175 | /* |
| 176 | * SNOW3G UIA2 uses the GFM datapath as a reflected 64-bit multiply in |
| 177 | * the low half of the 128-bit register pair. When RESINP[0], MUL[1], |
| 178 | * and the high polynomial byte are all zero, octeon_gfm_mul() observes |
| 179 | * only x[1], y[0], and the low 8-bit polynomial. Reflect those operands |
| 180 | * into normal carryless-multiply order and reflect the reduced result |
| 181 | * back into RESINP[1]. |
| 182 | */ |
| 183 | uint64_t vx = revbit64(x[1]); |
| 184 | uint64_t vy = revbit64(y[0]); |
| 185 | Int128 product = clmul_64(vx, vy); |
| 186 | uint64_t res = octeon_gfm_reduce64(product, revbit32(poly) >> 24); |
| 187 | |
| 188 | out[0] = 0; |
| 189 | out[1] = revbit64(res); |
| 190 | } |
| 191 | |
| 192 | static uint32_t octeon_hsh_get32(const uint64_t *regs, unsigned int index) |
| 193 | { |
| 194 | return regs[index]; |
| 195 | } |
| 196 | |
| 197 | static void octeon_hsh_set32(uint64_t *regs, unsigned int index, uint32_t value) |
| 198 | { |
| 199 | regs[index] = (regs[index] & ~(uint64_t)UINT32_MAX) | value; |
| 200 | } |
| 201 | |
| 202 | static void octeon_hsh_set_pair(uint64_t *regs, unsigned int index, |
| 203 | uint64_t value) |
| 204 | { |
| 205 | octeon_hsh_set32(regs, index * 2, value >> 32); |
| 206 | octeon_hsh_set32(regs, index * 2 + 1, value); |
| 207 | } |
| 208 | |
| 209 | static void octeon_md5_transform(MIPSOcteonCryptoState *crypto) |
| 210 | { |
| 211 | static const uint32_t k[64] = { |
| 212 | 0xd76aa478U, 0xe8c7b756U, 0x242070dbU, 0xc1bdceeeU, |
| 213 | 0xf57c0fafU, 0x4787c62aU, 0xa8304613U, 0xfd469501U, |
| 214 | 0x698098d8U, 0x8b44f7afU, 0xffff5bb1U, 0x895cd7beU, |
| 215 | 0x6b901122U, 0xfd987193U, 0xa679438eU, 0x49b40821U, |
| 216 | 0xf61e2562U, 0xc040b340U, 0x265e5a51U, 0xe9b6c7aaU, |
| 217 | 0xd62f105dU, 0x02441453U, 0xd8a1e681U, 0xe7d3fbc8U, |
| 218 | 0x21e1cde6U, 0xc33707d6U, 0xf4d50d87U, 0x455a14edU, |
| 219 | 0xa9e3e905U, 0xfcefa3f8U, 0x676f02d9U, 0x8d2a4c8aU, |
| 220 | 0xfffa3942U, 0x8771f681U, 0x6d9d6122U, 0xfde5380cU, |
| 221 | 0xa4beea44U, 0x4bdecfa9U, 0xf6bb4b60U, 0xbebfbc70U, |
| 222 | 0x289b7ec6U, 0xeaa127faU, 0xd4ef3085U, 0x04881d05U, |
| 223 | 0xd9d4d039U, 0xe6db99e5U, 0x1fa27cf8U, 0xc4ac5665U, |
| 224 | 0xf4292244U, 0x432aff97U, 0xab9423a7U, 0xfc93a039U, |
| 225 | 0x655b59c3U, 0x8f0ccc92U, 0xffeff47dU, 0x85845dd1U, |
| 226 | 0x6fa87e4fU, 0xfe2ce6e0U, 0xa3014314U, 0x4e0811a1U, |
| 227 | 0xf7537e82U, 0xbd3af235U, 0x2ad7d2bbU, 0xeb86d391U, |
| 228 | }; |
| 229 | static const uint8_t s[64] = { |
| 230 | 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, |
| 231 | 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, |
| 232 | 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, |
| 233 | 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21, |
| 234 | }; |
| 235 | uint32_t m[16]; |
| 236 | uint32_t a, b, c, d; |
| 237 | uint32_t aa, bb, cc, dd; |
| 238 | int i; |
| 239 | |
| 240 | for (i = 0; i < 16; i++) { |
| 241 | m[i] = bswap32(octeon_hsh_get32(crypto->hsh_dat, i)); |
| 242 | } |
| 243 | |
| 244 | a = bswap32(octeon_hsh_get32(crypto->hsh_iv, 0)); |
| 245 | b = bswap32(octeon_hsh_get32(crypto->hsh_iv, 1)); |
| 246 | c = bswap32(octeon_hsh_get32(crypto->hsh_iv, 2)); |
| 247 | d = bswap32(octeon_hsh_get32(crypto->hsh_iv, 3)); |
| 248 | aa = a; |
| 249 | bb = b; |
| 250 | cc = c; |
| 251 | dd = d; |
| 252 | |
| 253 | for (i = 0; i < 64; i++) { |
| 254 | uint32_t f, g, tmp; |
| 255 | |
| 256 | if (i < 16) { |
| 257 | f = (b & c) | ((~b) & d); |
| 258 | g = i; |
| 259 | } else if (i < 32) { |
| 260 | f = (d & b) | ((~d) & c); |
| 261 | g = (5 * i + 1) & 0xf; |
| 262 | } else if (i < 48) { |
| 263 | f = b ^ c ^ d; |
| 264 | g = (3 * i + 5) & 0xf; |
| 265 | } else { |
| 266 | f = c ^ (b | (~d)); |
| 267 | g = (7 * i) & 0xf; |
| 268 | } |
| 269 | |
| 270 | tmp = d; |
| 271 | d = c; |
| 272 | c = b; |
| 273 | b = b + rol32(a + f + k[i] + m[g], s[i]); |
| 274 | a = tmp; |
| 275 | } |
| 276 | |
| 277 | a += aa; |
| 278 | b += bb; |
| 279 | c += cc; |
| 280 | d += dd; |
| 281 | octeon_hsh_set32(crypto->hsh_iv, 0, bswap32(a)); |
| 282 | octeon_hsh_set32(crypto->hsh_iv, 1, bswap32(b)); |
| 283 | octeon_hsh_set32(crypto->hsh_iv, 2, bswap32(c)); |
| 284 | octeon_hsh_set32(crypto->hsh_iv, 3, bswap32(d)); |
| 285 | } |
| 286 | |
| 287 | static void octeon_sha1_transform(MIPSOcteonCryptoState *crypto) |
| 288 | { |
| 289 | uint32_t w[80]; |
| 290 | uint32_t a, b, c, d, e; |
| 291 | uint32_t orig[5]; |
| 292 | int i; |
| 293 | |
| 294 | for (i = 0; i < 16; i++) { |
| 295 | w[i] = octeon_hsh_get32(crypto->hsh_dat, i); |
| 296 | } |
| 297 | for (i = 16; i < 80; i++) { |
| 298 | w[i] = rol32(w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16], 1); |
| 299 | } |
| 300 | |
| 301 | for (i = 0; i < 5; i++) { |
| 302 | orig[i] = octeon_hsh_get32(crypto->hsh_iv, i); |
| 303 | } |
| 304 | a = orig[0]; |
| 305 | b = orig[1]; |
| 306 | c = orig[2]; |
| 307 | d = orig[3]; |
| 308 | e = orig[4]; |
| 309 | |
| 310 | for (i = 0; i < 80; i++) { |
| 311 | uint32_t f, k, temp; |
| 312 | |
| 313 | if (i < 20) { |
| 314 | f = (b & c) | ((~b) & d); |
| 315 | k = 0x5a827999; |
| 316 | } else if (i < 40) { |
| 317 | f = b ^ c ^ d; |
| 318 | k = 0x6ed9eba1; |
| 319 | } else if (i < 60) { |
| 320 | f = (b & c) | (b & d) | (c & d); |
| 321 | k = 0x8f1bbcdc; |
| 322 | } else { |
| 323 | f = b ^ c ^ d; |
| 324 | k = 0xca62c1d6; |
| 325 | } |
| 326 | |
| 327 | temp = rol32(a, 5) + f + e + k + w[i]; |
| 328 | e = d; |
| 329 | d = c; |
| 330 | c = rol32(b, 30); |
| 331 | b = a; |
| 332 | a = temp; |
| 333 | } |
| 334 | |
| 335 | orig[0] += a; |
| 336 | orig[1] += b; |
| 337 | orig[2] += c; |
| 338 | orig[3] += d; |
| 339 | orig[4] += e; |
| 340 | for (i = 0; i < 5; i++) { |
| 341 | octeon_hsh_set32(crypto->hsh_iv, i, orig[i]); |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | static void octeon_sha256_transform(MIPSOcteonCryptoState *crypto) |
| 346 | { |
| 347 | static const uint32_t k[64] = { |
| 348 | 0x428a2f98U, 0x71374491U, 0xb5c0fbcfU, 0xe9b5dba5U, |
| 349 | 0x3956c25bU, 0x59f111f1U, 0x923f82a4U, 0xab1c5ed5U, |
| 350 | 0xd807aa98U, 0x12835b01U, 0x243185beU, 0x550c7dc3U, |
| 351 | 0x72be5d74U, 0x80deb1feU, 0x9bdc06a7U, 0xc19bf174U, |
| 352 | 0xe49b69c1U, 0xefbe4786U, 0x0fc19dc6U, 0x240ca1ccU, |
| 353 | 0x2de92c6fU, 0x4a7484aaU, 0x5cb0a9dcU, 0x76f988daU, |
| 354 | 0x983e5152U, 0xa831c66dU, 0xb00327c8U, 0xbf597fc7U, |
| 355 | 0xc6e00bf3U, 0xd5a79147U, 0x06ca6351U, 0x14292967U, |
| 356 | 0x27b70a85U, 0x2e1b2138U, 0x4d2c6dfcU, 0x53380d13U, |
| 357 | 0x650a7354U, 0x766a0abbU, 0x81c2c92eU, 0x92722c85U, |
| 358 | 0xa2bfe8a1U, 0xa81a664bU, 0xc24b8b70U, 0xc76c51a3U, |
| 359 | 0xd192e819U, 0xd6990624U, 0xf40e3585U, 0x106aa070U, |
| 360 | 0x19a4c116U, 0x1e376c08U, 0x2748774cU, 0x34b0bcb5U, |
| 361 | 0x391c0cb3U, 0x4ed8aa4aU, 0x5b9cca4fU, 0x682e6ff3U, |
| 362 | 0x748f82eeU, 0x78a5636fU, 0x84c87814U, 0x8cc70208U, |
| 363 | 0x90befffaU, 0xa4506cebU, 0xbef9a3f7U, 0xc67178f2U, |
| 364 | }; |
| 365 | uint32_t w[64]; |
| 366 | uint32_t a, b, c, d, e, f, g, h; |
| 367 | uint32_t orig[8]; |
| 368 | int i; |
| 369 | |
| 370 | for (i = 0; i < 16; i++) { |
| 371 | w[i] = octeon_hsh_get32(crypto->hsh_dat, i); |
| 372 | } |
| 373 | for (i = 16; i < 64; i++) { |
| 374 | uint32_t s0 = ror32(w[i - 15], 7) ^ |
| 375 | ror32(w[i - 15], 18) ^ |
| 376 | (w[i - 15] >> 3); |
| 377 | uint32_t s1 = ror32(w[i - 2], 17) ^ |
| 378 | ror32(w[i - 2], 19) ^ |
| 379 | (w[i - 2] >> 10); |
| 380 | w[i] = w[i - 16] + s0 + w[i - 7] + s1; |
| 381 | } |
| 382 | |
| 383 | for (i = 0; i < 8; i++) { |
| 384 | orig[i] = octeon_hsh_get32(crypto->hsh_iv, i); |
| 385 | } |
| 386 | a = orig[0]; |
| 387 | b = orig[1]; |
| 388 | c = orig[2]; |
| 389 | d = orig[3]; |
| 390 | e = orig[4]; |
| 391 | f = orig[5]; |
| 392 | g = orig[6]; |
| 393 | h = orig[7]; |
| 394 | |
| 395 | for (i = 0; i < 64; i++) { |
| 396 | uint32_t s1 = ror32(e, 6) ^ |
| 397 | ror32(e, 11) ^ |
| 398 | ror32(e, 25); |
| 399 | uint32_t ch = (e & f) ^ ((~e) & g); |
| 400 | uint32_t temp1 = h + s1 + ch + k[i] + w[i]; |
| 401 | uint32_t s0 = ror32(a, 2) ^ |
| 402 | ror32(a, 13) ^ |
| 403 | ror32(a, 22); |
| 404 | uint32_t maj = (a & b) ^ (a & c) ^ (b & c); |
| 405 | uint32_t temp2 = s0 + maj; |
| 406 | |
| 407 | h = g; |
| 408 | g = f; |
| 409 | f = e; |
| 410 | e = d + temp1; |
| 411 | d = c; |
| 412 | c = b; |
| 413 | b = a; |
| 414 | a = temp1 + temp2; |
| 415 | } |
| 416 | |
| 417 | orig[0] += a; |
| 418 | orig[1] += b; |
| 419 | orig[2] += c; |
| 420 | orig[3] += d; |
| 421 | orig[4] += e; |
| 422 | orig[5] += f; |
| 423 | orig[6] += g; |
| 424 | orig[7] += h; |
| 425 | for (i = 0; i < 8; i++) { |
| 426 | octeon_hsh_set32(crypto->hsh_iv, i, orig[i]); |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | static void octeon_sha512_transform(MIPSOcteonCryptoState *crypto) |
| 431 | { |
| 432 | static const uint64_t k[80] = { |
| 433 | 0x428a2f98d728ae22ULL, 0x7137449123ef65cdULL, |
| 434 | 0xb5c0fbcfec4d3b2fULL, 0xe9b5dba58189dbbcULL, |
| 435 | 0x3956c25bf348b538ULL, 0x59f111f1b605d019ULL, |
| 436 | 0x923f82a4af194f9bULL, 0xab1c5ed5da6d8118ULL, |
| 437 | 0xd807aa98a3030242ULL, 0x12835b0145706fbeULL, |
| 438 | 0x243185be4ee4b28cULL, 0x550c7dc3d5ffb4e2ULL, |
| 439 | 0x72be5d74f27b896fULL, 0x80deb1fe3b1696b1ULL, |
| 440 | 0x9bdc06a725c71235ULL, 0xc19bf174cf692694ULL, |
| 441 | 0xe49b69c19ef14ad2ULL, 0xefbe4786384f25e3ULL, |
| 442 | 0x0fc19dc68b8cd5b5ULL, 0x240ca1cc77ac9c65ULL, |
| 443 | 0x2de92c6f592b0275ULL, 0x4a7484aa6ea6e483ULL, |
| 444 | 0x5cb0a9dcbd41fbd4ULL, 0x76f988da831153b5ULL, |
| 445 | 0x983e5152ee66dfabULL, 0xa831c66d2db43210ULL, |
| 446 | 0xb00327c898fb213fULL, 0xbf597fc7beef0ee4ULL, |
| 447 | 0xc6e00bf33da88fc2ULL, 0xd5a79147930aa725ULL, |
| 448 | 0x06ca6351e003826fULL, 0x142929670a0e6e70ULL, |
| 449 | 0x27b70a8546d22ffcULL, 0x2e1b21385c26c926ULL, |
| 450 | 0x4d2c6dfc5ac42aedULL, 0x53380d139d95b3dfULL, |
| 451 | 0x650a73548baf63deULL, 0x766a0abb3c77b2a8ULL, |
| 452 | 0x81c2c92e47edaee6ULL, 0x92722c851482353bULL, |
| 453 | 0xa2bfe8a14cf10364ULL, 0xa81a664bbc423001ULL, |
| 454 | 0xc24b8b70d0f89791ULL, 0xc76c51a30654be30ULL, |
| 455 | 0xd192e819d6ef5218ULL, 0xd69906245565a910ULL, |
| 456 | 0xf40e35855771202aULL, 0x106aa07032bbd1b8ULL, |
| 457 | 0x19a4c116b8d2d0c8ULL, 0x1e376c085141ab53ULL, |
| 458 | 0x2748774cdf8eeb99ULL, 0x34b0bcb5e19b48a8ULL, |
| 459 | 0x391c0cb3c5c95a63ULL, 0x4ed8aa4ae3418acbULL, |
| 460 | 0x5b9cca4f7763e373ULL, 0x682e6ff3d6b2b8a3ULL, |
| 461 | 0x748f82ee5defb2fcULL, 0x78a5636f43172f60ULL, |
| 462 | 0x84c87814a1f0ab72ULL, 0x8cc702081a6439ecULL, |
| 463 | 0x90befffa23631e28ULL, 0xa4506cebde82bde9ULL, |
| 464 | 0xbef9a3f7b2c67915ULL, 0xc67178f2e372532bULL, |
| 465 | 0xca273eceea26619cULL, 0xd186b8c721c0c207ULL, |
| 466 | 0xeada7dd6cde0eb1eULL, 0xf57d4f7fee6ed178ULL, |
| 467 | 0x06f067aa72176fbaULL, 0x0a637dc5a2c898a6ULL, |
| 468 | 0x113f9804bef90daeULL, 0x1b710b35131c471bULL, |
| 469 | 0x28db77f523047d84ULL, 0x32caab7b40c72493ULL, |
| 470 | 0x3c9ebe0a15c9bebcULL, 0x431d67c49c100d4cULL, |
| 471 | 0x4cc5d4becb3e42b6ULL, 0x597f299cfc657e2aULL, |
| 472 | 0x5fcb6fab3ad6faecULL, 0x6c44198c4a475817ULL, |
| 473 | }; |
| 474 | uint64_t w[80]; |
| 475 | uint64_t a, b, c, d, e, f, g, h; |
| 476 | int i; |
| 477 | |
| 478 | for (i = 0; i < 16; i++) { |
| 479 | w[i] = crypto->hsh_dat[i]; |
| 480 | } |
| 481 | for (i = 16; i < 80; i++) { |
| 482 | uint64_t s0 = ror64(w[i - 15], 1) ^ |
| 483 | ror64(w[i - 15], 8) ^ |
| 484 | (w[i - 15] >> 7); |
| 485 | uint64_t s1 = ror64(w[i - 2], 19) ^ |
| 486 | ror64(w[i - 2], 61) ^ |
| 487 | (w[i - 2] >> 6); |
| 488 | w[i] = w[i - 16] + s0 + w[i - 7] + s1; |
| 489 | } |
| 490 | |
| 491 | a = crypto->hsh_iv[0]; |
| 492 | b = crypto->hsh_iv[1]; |
| 493 | c = crypto->hsh_iv[2]; |
| 494 | d = crypto->hsh_iv[3]; |
| 495 | e = crypto->hsh_iv[4]; |
| 496 | f = crypto->hsh_iv[5]; |
| 497 | g = crypto->hsh_iv[6]; |
| 498 | h = crypto->hsh_iv[7]; |
| 499 | |
| 500 | for (i = 0; i < 80; i++) { |
| 501 | uint64_t s0 = ror64(a, 28) ^ |
| 502 | ror64(a, 34) ^ |
| 503 | ror64(a, 39); |
| 504 | uint64_t s1 = ror64(e, 14) ^ |
| 505 | ror64(e, 18) ^ |
| 506 | ror64(e, 41); |
| 507 | uint64_t ch = (e & f) ^ ((~e) & g); |
| 508 | uint64_t maj = (a & b) ^ (a & c) ^ (b & c); |
| 509 | uint64_t temp1 = h + s1 + ch + k[i] + w[i]; |
| 510 | uint64_t temp2 = s0 + maj; |
| 511 | |
| 512 | h = g; |
| 513 | g = f; |
| 514 | f = e; |
| 515 | e = d + temp1; |
| 516 | d = c; |
| 517 | c = b; |
| 518 | b = a; |
| 519 | a = temp1 + temp2; |
| 520 | } |
| 521 | |
| 522 | crypto->hsh_iv[0] += a; |
| 523 | crypto->hsh_iv[1] += b; |
| 524 | crypto->hsh_iv[2] += c; |
| 525 | crypto->hsh_iv[3] += d; |
| 526 | crypto->hsh_iv[4] += e; |
| 527 | crypto->hsh_iv[5] += f; |
| 528 | crypto->hsh_iv[6] += g; |
| 529 | crypto->hsh_iv[7] += h; |
| 530 | } |
| 531 | |
| 532 | static const uint64_t octeon_sha3_round_constants[24] = { |
| 533 | 0x0000000000000001ULL, 0x0000000000008082ULL, |
| 534 | 0x800000000000808aULL, 0x8000000080008000ULL, |
| 535 | 0x000000000000808bULL, 0x0000000080000001ULL, |
| 536 | 0x8000000080008081ULL, 0x8000000000008009ULL, |
| 537 | 0x000000000000008aULL, 0x0000000000000088ULL, |
| 538 | 0x0000000080008009ULL, 0x000000008000000aULL, |
| 539 | 0x000000008000808bULL, 0x800000000000008bULL, |
| 540 | 0x8000000000008089ULL, 0x8000000000008003ULL, |
| 541 | 0x8000000000008002ULL, 0x8000000000000080ULL, |
| 542 | 0x000000000000800aULL, 0x800000008000000aULL, |
| 543 | 0x8000000080008081ULL, 0x8000000000008080ULL, |
| 544 | 0x0000000080000001ULL, 0x8000000080008008ULL, |
| 545 | }; |
| 546 | |
| 547 | static const uint8_t octeon_sha3_rotation_constants[24] = { |
| 548 | 1, 3, 6, 10, 15, 21, 28, 36, 45, 55, 2, 14, |
| 549 | 27, 41, 56, 8, 25, 43, 62, 18, 39, 61, 20, 44, |
| 550 | }; |
| 551 | |
| 552 | static const uint8_t octeon_sha3_pi_lanes[24] = { |
| 553 | 10, 7, 11, 17, 18, 3, 5, 16, 8, 21, 24, 4, |
| 554 | 15, 23, 19, 13, 12, 2, 20, 14, 22, 9, 6, 1, |
| 555 | }; |
| 556 | |
| 557 | static uint64_t octeon_sha3_reg_to_lane(uint64_t value) |
| 558 | { |
| 559 | /* |
| 560 | * The COP2 register interface is consumed by big-endian MIPS code as |
| 561 | * 64-bit register values, while Keccak lanes are byte-little-endian. |
| 562 | */ |
| 563 | return bswap64(value); |
| 564 | } |
| 565 | |
| 566 | static uint64_t octeon_sha3_lane_to_reg(uint64_t value) |
| 567 | { |
| 568 | return bswap64(value); |
| 569 | } |
| 570 | |
| 571 | static void octeon_sha3_permute(MIPSOcteonCryptoState *crypto) |
| 572 | { |
| 573 | uint64_t state[25]; |
| 574 | |
| 575 | for (int i = 0; i < 25; i++) { |
| 576 | state[i] = octeon_sha3_reg_to_lane(crypto->sha3_dat[i]); |
| 577 | } |
| 578 | |
| 579 | for (int round = 0; round < 24; round++) { |
| 580 | uint64_t bc[5]; |
| 581 | uint64_t temp; |
| 582 | |
| 583 | for (int x = 0; x < 5; x++) { |
| 584 | bc[x] = state[x] ^ state[5 + x] ^ state[10 + x] ^ |
| 585 | state[15 + x] ^ state[20 + x]; |
| 586 | } |
| 587 | for (int x = 0; x < 5; x++) { |
| 588 | temp = bc[(x + 4) % 5] ^ rol64(bc[(x + 1) % 5], 1); |
| 589 | for (int y = 0; y < 25; y += 5) { |
| 590 | state[y + x] ^= temp; |
| 591 | } |
| 592 | } |
| 593 | |
| 594 | temp = state[1]; |
| 595 | for (int i = 0; i < 24; i++) { |
| 596 | uint64_t next = state[octeon_sha3_pi_lanes[i]]; |
| 597 | |
| 598 | state[octeon_sha3_pi_lanes[i]] = |
| 599 | rol64(temp, octeon_sha3_rotation_constants[i]); |
| 600 | temp = next; |
| 601 | } |
| 602 | |
| 603 | for (int y = 0; y < 25; y += 5) { |
| 604 | for (int x = 0; x < 5; x++) { |
| 605 | bc[x] = state[y + x]; |
| 606 | } |
| 607 | for (int x = 0; x < 5; x++) { |
| 608 | state[y + x] = bc[x] ^ ((~bc[(x + 1) % 5]) & bc[(x + 2) % 5]); |
| 609 | } |
| 610 | } |
| 611 | |
| 612 | state[0] ^= octeon_sha3_round_constants[round]; |
| 613 | } |
| 614 | |
| 615 | for (int i = 0; i < 25; i++) { |
| 616 | crypto->sha3_dat[i] = octeon_sha3_lane_to_reg(state[i]); |
| 617 | } |
| 618 | } |
| 619 | |
| 620 | static uint32_t octeon_crypto_hi32(uint64_t value) |
| 621 | { |
| 622 | return value >> 32; |
| 623 | } |
| 624 | |
| 625 | static uint32_t octeon_crypto_lo32(uint64_t value) |
| 626 | { |
| 627 | return value; |
| 628 | } |
| 629 | |
| 630 | static uint64_t octeon_crypto_pack32(uint32_t hi, uint32_t lo) |
| 631 | { |
| 632 | return ((uint64_t)hi << 32) | lo; |
| 633 | } |
| 634 | |
| 635 | static const uint8_t octeon_zuc_s0[256] = { |
| 636 | 0x3e, 0x72, 0x5b, 0x47, 0xca, 0xe0, 0x00, 0x33, |
| 637 | 0x04, 0xd1, 0x54, 0x98, 0x09, 0xb9, 0x6d, 0xcb, |
| 638 | 0x7b, 0x1b, 0xf9, 0x32, 0xaf, 0x9d, 0x6a, 0xa5, |
| 639 | 0xb8, 0x2d, 0xfc, 0x1d, 0x08, 0x53, 0x03, 0x90, |
| 640 | 0x4d, 0x4e, 0x84, 0x99, 0xe4, 0xce, 0xd9, 0x91, |
| 641 | 0xdd, 0xb6, 0x85, 0x48, 0x8b, 0x29, 0x6e, 0xac, |
| 642 | 0xcd, 0xc1, 0xf8, 0x1e, 0x73, 0x43, 0x69, 0xc6, |
| 643 | 0xb5, 0xbd, 0xfd, 0x39, 0x63, 0x20, 0xd4, 0x38, |
| 644 | 0x76, 0x7d, 0xb2, 0xa7, 0xcf, 0xed, 0x57, 0xc5, |
| 645 | 0xf3, 0x2c, 0xbb, 0x14, 0x21, 0x06, 0x55, 0x9b, |
| 646 | 0xe3, 0xef, 0x5e, 0x31, 0x4f, 0x7f, 0x5a, 0xa4, |
| 647 | 0x0d, 0x82, 0x51, 0x49, 0x5f, 0xba, 0x58, 0x1c, |
| 648 | 0x4a, 0x16, 0xd5, 0x17, 0xa8, 0x92, 0x24, 0x1f, |
| 649 | 0x8c, 0xff, 0xd8, 0xae, 0x2e, 0x01, 0xd3, 0xad, |
| 650 | 0x3b, 0x4b, 0xda, 0x46, 0xeb, 0xc9, 0xde, 0x9a, |
| 651 | 0x8f, 0x87, 0xd7, 0x3a, 0x80, 0x6f, 0x2f, 0xc8, |
| 652 | 0xb1, 0xb4, 0x37, 0xf7, 0x0a, 0x22, 0x13, 0x28, |
| 653 | 0x7c, 0xcc, 0x3c, 0x89, 0xc7, 0xc3, 0x96, 0x56, |
| 654 | 0x07, 0xbf, 0x7e, 0xf0, 0x0b, 0x2b, 0x97, 0x52, |
| 655 | 0x35, 0x41, 0x79, 0x61, 0xa6, 0x4c, 0x10, 0xfe, |
| 656 | 0xbc, 0x26, 0x95, 0x88, 0x8a, 0xb0, 0xa3, 0xfb, |
| 657 | 0xc0, 0x18, 0x94, 0xf2, 0xe1, 0xe5, 0xe9, 0x5d, |
| 658 | 0xd0, 0xdc, 0x11, 0x66, 0x64, 0x5c, 0xec, 0x59, |
| 659 | 0x42, 0x75, 0x12, 0xf5, 0x74, 0x9c, 0xaa, 0x23, |
| 660 | 0x0e, 0x86, 0xab, 0xbe, 0x2a, 0x02, 0xe7, 0x67, |
| 661 | 0xe6, 0x44, 0xa2, 0x6c, 0xc2, 0x93, 0x9f, 0xf1, |
| 662 | 0xf6, 0xfa, 0x36, 0xd2, 0x50, 0x68, 0x9e, 0x62, |
| 663 | 0x71, 0x15, 0x3d, 0xd6, 0x40, 0xc4, 0xe2, 0x0f, |
| 664 | 0x8e, 0x83, 0x77, 0x6b, 0x25, 0x05, 0x3f, 0x0c, |
| 665 | 0x30, 0xea, 0x70, 0xb7, 0xa1, 0xe8, 0xa9, 0x65, |
| 666 | 0x8d, 0x27, 0x1a, 0xdb, 0x81, 0xb3, 0xa0, 0xf4, |
| 667 | 0x45, 0x7a, 0x19, 0xdf, 0xee, 0x78, 0x34, 0x60, |
| 668 | }; |
| 669 | |
| 670 | static const uint8_t octeon_zuc_s1[256] = { |
| 671 | 0x55, 0xc2, 0x63, 0x71, 0x3b, 0xc8, 0x47, 0x86, |
| 672 | 0x9f, 0x3c, 0xda, 0x5b, 0x29, 0xaa, 0xfd, 0x77, |
| 673 | 0x8c, 0xc5, 0x94, 0x0c, 0xa6, 0x1a, 0x13, 0x00, |
| 674 | 0xe3, 0xa8, 0x16, 0x72, 0x40, 0xf9, 0xf8, 0x42, |
| 675 | 0x44, 0x26, 0x68, 0x96, 0x81, 0xd9, 0x45, 0x3e, |
| 676 | 0x10, 0x76, 0xc6, 0xa7, 0x8b, 0x39, 0x43, 0xe1, |
| 677 | 0x3a, 0xb5, 0x56, 0x2a, 0xc0, 0x6d, 0xb3, 0x05, |
| 678 | 0x22, 0x66, 0xbf, 0xdc, 0x0b, 0xfa, 0x62, 0x48, |
| 679 | 0xdd, 0x20, 0x11, 0x06, 0x36, 0xc9, 0xc1, 0xcf, |
| 680 | 0xf6, 0x27, 0x52, 0xbb, 0x69, 0xf5, 0xd4, 0x87, |
| 681 | 0x7f, 0x84, 0x4c, 0xd2, 0x9c, 0x57, 0xa4, 0xbc, |
| 682 | 0x4f, 0x9a, 0xdf, 0xfe, 0xd6, 0x8d, 0x7a, 0xeb, |
| 683 | 0x2b, 0x53, 0xd8, 0x5c, 0xa1, 0x14, 0x17, 0xfb, |
| 684 | 0x23, 0xd5, 0x7d, 0x30, 0x67, 0x73, 0x08, 0x09, |
| 685 | 0xee, 0xb7, 0x70, 0x3f, 0x61, 0xb2, 0x19, 0x8e, |
| 686 | 0x4e, 0xe5, 0x4b, 0x93, 0x8f, 0x5d, 0xdb, 0xa9, |
| 687 | 0xad, 0xf1, 0xae, 0x2e, 0xcb, 0x0d, 0xfc, 0xf4, |
| 688 | 0x2d, 0x46, 0x6e, 0x1d, 0x97, 0xe8, 0xd1, 0xe9, |
| 689 | 0x4d, 0x37, 0xa5, 0x75, 0x5e, 0x83, 0x9e, 0xab, |
| 690 | 0x82, 0x9d, 0xb9, 0x1c, 0xe0, 0xcd, 0x49, 0x89, |
| 691 | 0x01, 0xb6, 0xbd, 0x58, 0x24, 0xa2, 0x5f, 0x38, |
| 692 | 0x78, 0x99, 0x15, 0x90, 0x50, 0xb8, 0x95, 0xe4, |
| 693 | 0xd0, 0x91, 0xc7, 0xce, 0xed, 0x0f, 0xb4, 0x6f, |
| 694 | 0xa0, 0xcc, 0xf0, 0x02, 0x4a, 0x79, 0xc3, 0xde, |
| 695 | 0xa3, 0xef, 0xea, 0x51, 0xe6, 0x6b, 0x18, 0xec, |
| 696 | 0x1b, 0x2c, 0x80, 0xf7, 0x74, 0xe7, 0xff, 0x21, |
| 697 | 0x5a, 0x6a, 0x54, 0x1e, 0x41, 0x31, 0x92, 0x35, |
| 698 | 0xc4, 0x33, 0x07, 0x0a, 0xba, 0x7e, 0x0e, 0x34, |
| 699 | 0x88, 0xb1, 0x98, 0x7c, 0xf3, 0x3d, 0x60, 0x6c, |
| 700 | 0x7b, 0xca, 0xd3, 0x1f, 0x32, 0x65, 0x04, 0x28, |
| 701 | 0x64, 0xbe, 0x85, 0x9b, 0x2f, 0x59, 0x8a, 0xd7, |
| 702 | 0xb0, 0x25, 0xac, 0xaf, 0x12, 0x03, 0xe2, 0xf2, |
| 703 | }; |
| 704 | |
| 705 | static uint32_t octeon_zuc_addm(uint32_t a, uint32_t b) |
| 706 | { |
| 707 | uint32_t c = a + b; |
| 708 | |
| 709 | c = (c & 0x7fffffffU) + (c >> 31); |
| 710 | return c ? c : 0x7fffffffU; |
| 711 | } |
| 712 | |
| 713 | static uint32_t octeon_zuc_mul_by_pow2(uint32_t v, unsigned int shift) |
| 714 | { |
| 715 | return ((v << shift) | (v >> (31 - shift))) & 0x7fffffffU; |
| 716 | } |
| 717 | |
| 718 | static uint32_t octeon_zuc_make_u32(uint8_t a, uint8_t b, uint8_t c, uint8_t d) |
| 719 | { |
| 720 | return ((uint32_t)a << 24) | ((uint32_t)b << 16) | |
| 721 | ((uint32_t)c << 8) | d; |
| 722 | } |
| 723 | |
| 724 | static uint64_t octeon_zuc_pack_pair(uint32_t hi, uint32_t lo) |
| 725 | { |
| 726 | return ((uint64_t)hi << 32) | lo; |
| 727 | } |
| 728 | |
| 729 | static uint32_t octeon_zuc_lfsr(const MIPSOcteonCryptoState *crypto, |
| 730 | unsigned int index) |
| 731 | { |
| 732 | uint64_t pair = crypto->hsh_dat[index / 2]; |
| 733 | |
| 734 | return index & 1 ? octeon_crypto_lo32(pair) : octeon_crypto_hi32(pair); |
| 735 | } |
| 736 | |
| 737 | static void octeon_zuc_set_lfsr(MIPSOcteonCryptoState *crypto, |
| 738 | unsigned int index, uint32_t value) |
| 739 | { |
| 740 | uint32_t hi = octeon_crypto_hi32(crypto->hsh_dat[index / 2]); |
| 741 | uint32_t lo = octeon_crypto_lo32(crypto->hsh_dat[index / 2]); |
| 742 | |
| 743 | value &= 0x7fffffffU; |
| 744 | if (index & 1) { |
| 745 | lo = value; |
| 746 | } else { |
| 747 | hi = value; |
| 748 | } |
| 749 | crypto->hsh_dat[index / 2] = octeon_zuc_pack_pair(hi, lo); |
| 750 | } |
| 751 | |
| 752 | static uint32_t octeon_zuc_fsm(const MIPSOcteonCryptoState *crypto, |
| 753 | unsigned int index) |
| 754 | { |
| 755 | g_assert(index < 2); |
| 756 | return crypto->hsh_iv[1 + index]; |
| 757 | } |
| 758 | |
| 759 | static void octeon_zuc_set_fsm(MIPSOcteonCryptoState *crypto, |
| 760 | unsigned int index, uint32_t value) |
| 761 | { |
| 762 | g_assert(index < 2); |
| 763 | crypto->hsh_iv[1 + index] = value; |
| 764 | } |
| 765 | |
| 766 | static uint32_t octeon_zuc_window(const MIPSOcteonCryptoState *crypto, |
| 767 | unsigned int index) |
| 768 | { |
| 769 | uint64_t pair = crypto->hsh_iv[0]; |
| 770 | |
| 771 | switch (index) { |
| 772 | case 0: |
| 773 | return octeon_crypto_hi32(pair); |
| 774 | case 1: |
| 775 | return octeon_crypto_lo32(pair); |
| 776 | default: |
| 777 | g_assert_not_reached(); |
| 778 | } |
| 779 | } |
| 780 | |
| 781 | static void octeon_zuc_set_window_pair(MIPSOcteonCryptoState *crypto, |
| 782 | uint32_t hi, uint32_t lo) |
| 783 | { |
| 784 | crypto->hsh_iv[0] = octeon_zuc_pack_pair(hi, lo); |
| 785 | } |
| 786 | |
| 787 | static uint32_t octeon_zuc_tresult(const MIPSOcteonCryptoState *crypto) |
| 788 | { |
| 789 | return crypto->hsh_iv[3]; |
| 790 | } |
| 791 | |
| 792 | static void octeon_zuc_set_tresult(MIPSOcteonCryptoState *crypto, |
| 793 | uint32_t value) |
| 794 | { |
| 795 | crypto->hsh_iv[3] = value; |
| 796 | } |
| 797 | |
| 798 | static void octeon_zuc_bit_reorganization(const MIPSOcteonCryptoState *crypto, |
| 799 | uint32_t x[4]) |
| 800 | { |
| 801 | x[0] = ((octeon_zuc_lfsr(crypto, 15) & 0x7fff8000U) << 1) | |
| 802 | (octeon_zuc_lfsr(crypto, 14) & 0xffffU); |
| 803 | x[1] = ((octeon_zuc_lfsr(crypto, 11) & 0xffffU) << 16) | |
| 804 | (octeon_zuc_lfsr(crypto, 9) >> 15); |
| 805 | x[2] = ((octeon_zuc_lfsr(crypto, 7) & 0xffffU) << 16) | |
| 806 | (octeon_zuc_lfsr(crypto, 5) >> 15); |
| 807 | x[3] = ((octeon_zuc_lfsr(crypto, 2) & 0xffffU) << 16) | |
| 808 | (octeon_zuc_lfsr(crypto, 0) >> 15); |
| 809 | } |
| 810 | |
| 811 | static uint32_t octeon_zuc_l1(uint32_t x) |
| 812 | { |
| 813 | return x ^ rol32(x, 2) ^ rol32(x, 10) ^ rol32(x, 18) ^ rol32(x, 24); |
| 814 | } |
| 815 | |
| 816 | static uint32_t octeon_zuc_l2(uint32_t x) |
| 817 | { |
| 818 | return x ^ rol32(x, 8) ^ rol32(x, 14) ^ rol32(x, 22) ^ rol32(x, 30); |
| 819 | } |
| 820 | |
| 821 | static uint32_t octeon_zuc_f(MIPSOcteonCryptoState *crypto, const uint32_t x[4]) |
| 822 | { |
| 823 | uint32_t fsm0 = octeon_zuc_fsm(crypto, 0); |
| 824 | uint32_t fsm1 = octeon_zuc_fsm(crypto, 1); |
| 825 | uint32_t w = (x[0] ^ fsm0) + fsm1; |
| 826 | uint32_t w1 = fsm0 + x[1]; |
| 827 | uint32_t w2 = fsm1 ^ x[2]; |
| 828 | uint32_t u = octeon_zuc_l1((w1 << 16) | (w2 >> 16)); |
| 829 | uint32_t v = octeon_zuc_l2((w2 << 16) | (w1 >> 16)); |
| 830 | |
| 831 | octeon_zuc_set_fsm(crypto, 0, |
| 832 | octeon_zuc_make_u32(octeon_zuc_s0[u >> 24], |
| 833 | octeon_zuc_s1[(uint8_t)(u >> 16)], |
| 834 | octeon_zuc_s0[(uint8_t)(u >> 8)], |
| 835 | octeon_zuc_s1[(uint8_t)u])); |
| 836 | octeon_zuc_set_fsm(crypto, 1, |
| 837 | octeon_zuc_make_u32(octeon_zuc_s0[v >> 24], |
| 838 | octeon_zuc_s1[(uint8_t)(v >> 16)], |
| 839 | octeon_zuc_s0[(uint8_t)(v >> 8)], |
| 840 | octeon_zuc_s1[(uint8_t)v])); |
| 841 | return w; |
| 842 | } |
| 843 | |
| 844 | static void octeon_zuc_lfsr_step(MIPSOcteonCryptoState *crypto, |
| 845 | bool init_mode, uint32_t u) |
| 846 | { |
| 847 | uint32_t lfsr[16]; |
| 848 | uint32_t f; |
| 849 | |
| 850 | for (int i = 0; i < 16; i++) { |
| 851 | lfsr[i] = octeon_zuc_lfsr(crypto, i); |
| 852 | } |
| 853 | |
| 854 | f = lfsr[0]; |
| 855 | f = octeon_zuc_addm(f, octeon_zuc_mul_by_pow2(lfsr[0], 8)); |
| 856 | f = octeon_zuc_addm(f, octeon_zuc_mul_by_pow2(lfsr[4], 20)); |
| 857 | f = octeon_zuc_addm(f, octeon_zuc_mul_by_pow2(lfsr[10], 21)); |
| 858 | f = octeon_zuc_addm(f, octeon_zuc_mul_by_pow2(lfsr[13], 17)); |
| 859 | f = octeon_zuc_addm(f, octeon_zuc_mul_by_pow2(lfsr[15], 15)); |
| 860 | if (init_mode) { |
| 861 | f = octeon_zuc_addm(f, u); |
| 862 | } |
| 863 | |
| 864 | for (int i = 0; i < 15; i++) { |
| 865 | octeon_zuc_set_lfsr(crypto, i, lfsr[i + 1]); |
| 866 | } |
| 867 | octeon_zuc_set_lfsr(crypto, 15, f); |
| 868 | } |
| 869 | |
| 870 | static uint32_t octeon_zuc_generate_word(MIPSOcteonCryptoState *crypto) |
| 871 | { |
| 872 | uint32_t x[4]; |
| 873 | uint32_t z; |
| 874 | |
| 875 | octeon_zuc_bit_reorganization(crypto, x); |
| 876 | z = octeon_zuc_f(crypto, x) ^ x[3]; |
| 877 | octeon_zuc_lfsr_step(crypto, false, 0); |
| 878 | return z; |
| 879 | } |
| 880 | |
| 881 | static void octeon_zuc_fill_window_pair(MIPSOcteonCryptoState *crypto) |
| 882 | { |
| 883 | uint32_t z0 = octeon_zuc_generate_word(crypto); |
| 884 | uint32_t z1 = octeon_zuc_generate_word(crypto); |
| 885 | |
| 886 | octeon_zuc_set_window_pair(crypto, z0, z1); |
| 887 | } |
| 888 | |
| 889 | static uint32_t |
| 890 | octeon_zuc_window_word(const MIPSOcteonCryptoState *crypto, unsigned int bit, |
| 891 | uint32_t z2) |
| 892 | { |
| 893 | if (bit == 0) { |
| 894 | return octeon_zuc_window(crypto, 0); |
| 895 | } |
| 896 | if (bit < 32) { |
| 897 | return (octeon_zuc_window(crypto, 0) << bit) | |
| 898 | (octeon_zuc_window(crypto, 1) >> (32 - bit)); |
| 899 | } |
| 900 | if (bit == 32) { |
| 901 | return octeon_zuc_window(crypto, 1); |
| 902 | } |
| 903 | return (octeon_zuc_window(crypto, 1) << (bit - 32)) | |
| 904 | (z2 >> (64 - bit)); |
| 905 | } |
| 906 | |
| 907 | static void octeon_zuc_advance_window(MIPSOcteonCryptoState *crypto, |
| 908 | uint32_t z2) |
| 909 | { |
| 910 | uint32_t z3 = octeon_zuc_generate_word(crypto); |
| 911 | |
| 912 | octeon_zuc_set_window_pair(crypto, z2, z3); |
| 913 | } |
| 914 | |
| 915 | static void octeon_zuc_start(MIPSOcteonCryptoState *crypto, uint64_t data) |
| 916 | { |
| 917 | uint32_t x[4]; |
| 918 | |
| 919 | for (int i = 0; i < 14; i++) { |
| 920 | octeon_zuc_set_lfsr(crypto, i, octeon_zuc_lfsr(crypto, i)); |
| 921 | } |
| 922 | octeon_zuc_set_lfsr(crypto, 14, data >> 32); |
| 923 | octeon_zuc_set_lfsr(crypto, 15, data); |
| 924 | octeon_zuc_set_fsm(crypto, 0, 0); |
| 925 | octeon_zuc_set_fsm(crypto, 1, 0); |
| 926 | octeon_zuc_set_tresult(crypto, 0); |
| 927 | |
| 928 | for (int i = 0; i < 32; i++) { |
| 929 | octeon_zuc_bit_reorganization(crypto, x); |
| 930 | octeon_zuc_lfsr_step(crypto, true, octeon_zuc_f(crypto, x) >> 1); |
| 931 | } |
| 932 | |
| 933 | octeon_zuc_bit_reorganization(crypto, x); |
| 934 | (void)octeon_zuc_f(crypto, x); |
| 935 | octeon_zuc_lfsr_step(crypto, false, 0); |
| 936 | octeon_zuc_fill_window_pair(crypto); |
| 937 | } |
| 938 | |
| 939 | static void octeon_zuc_more(MIPSOcteonCryptoState *crypto, uint64_t data) |
| 940 | { |
| 941 | uint32_t t = octeon_zuc_tresult(crypto); |
| 942 | uint32_t z2 = octeon_zuc_generate_word(crypto); |
| 943 | |
| 944 | for (unsigned int bit = 0; bit < 64; bit++) { |
| 945 | if ((data >> (63 - bit)) & 1) { |
| 946 | t ^= octeon_zuc_window_word(crypto, bit, z2); |
| 947 | } |
| 948 | } |
| 949 | octeon_zuc_set_tresult(crypto, t); |
| 950 | octeon_zuc_advance_window(crypto, z2); |
| 951 | } |
| 952 | |
| 953 | static const uint8_t octeon_snow3g_sr[256] = { |
| 954 | 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, |
| 955 | 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76, |
| 956 | 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, |
| 957 | 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0, |
| 958 | 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, |
| 959 | 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15, |
| 960 | 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, |
| 961 | 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75, |
| 962 | 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, |
| 963 | 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84, |
| 964 | 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, |
| 965 | 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf, |
| 966 | 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, |
| 967 | 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8, |
| 968 | 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, |
| 969 | 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2, |
| 970 | 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, |
| 971 | 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73, |
| 972 | 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, |
| 973 | 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb, |
| 974 | 0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, |
| 975 | 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79, |
| 976 | 0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, |
| 977 | 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08, |
| 978 | 0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, |
| 979 | 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a, |
| 980 | 0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, |
| 981 | 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e, |
| 982 | 0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, |
| 983 | 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf, |
| 984 | 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, |
| 985 | 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16, |
| 986 | }; |
| 987 | |
| 988 | static const uint8_t octeon_snow3g_sq[256] = { |
| 989 | 0x25, 0x24, 0x73, 0x67, 0xd7, 0xae, 0x5c, 0x30, |
| 990 | 0xa4, 0xee, 0x6e, 0xcb, 0x7d, 0xb5, 0x82, 0xdb, |
| 991 | 0xe4, 0x8e, 0x48, 0x49, 0x4f, 0x5d, 0x6a, 0x78, |
| 992 | 0x70, 0x88, 0xe8, 0x5f, 0x5e, 0x84, 0x65, 0xe2, |
| 993 | 0xd8, 0xe9, 0xcc, 0xed, 0x40, 0x2f, 0x11, 0x28, |
| 994 | 0x57, 0xd2, 0xac, 0xe3, 0x4a, 0x15, 0x1b, 0xb9, |
| 995 | 0xb2, 0x80, 0x85, 0xa6, 0x2e, 0x02, 0x47, 0x29, |
| 996 | 0x07, 0x4b, 0x0e, 0xc1, 0x51, 0xaa, 0x89, 0xd4, |
| 997 | 0xca, 0x01, 0x46, 0xb3, 0xef, 0xdd, 0x44, 0x7b, |
| 998 | 0xc2, 0x7f, 0xbe, 0xc3, 0x9f, 0x20, 0x4c, 0x64, |
| 999 | 0x83, 0xa2, 0x68, 0x42, 0x13, 0xb4, 0x41, 0xcd, |
| 1000 | 0xba, 0xc6, 0xbb, 0x6d, 0x4d, 0x71, 0x21, 0xf4, |
| 1001 | 0x8d, 0xb0, 0xe5, 0x93, 0xfe, 0x8f, 0xe6, 0xcf, |
| 1002 | 0x43, 0x45, 0x31, 0x22, 0x37, 0x36, 0x96, 0xfa, |
| 1003 | 0xbc, 0x0f, 0x08, 0x52, 0x1d, 0x55, 0x1a, 0xc5, |
| 1004 | 0x4e, 0x23, 0x69, 0x7a, 0x92, 0xff, 0x5b, 0x5a, |
| 1005 | 0xeb, 0x9a, 0x1c, 0xa9, 0xd1, 0x7e, 0x0d, 0xfc, |
| 1006 | 0x50, 0x8a, 0xb6, 0x62, 0xf5, 0x0a, 0xf8, 0xdc, |
| 1007 | 0x03, 0x3c, 0x0c, 0x39, 0xf1, 0xb8, 0xf3, 0x3d, |
| 1008 | 0xf2, 0xd5, 0x97, 0x66, 0x81, 0x32, 0xa0, 0x00, |
| 1009 | 0x06, 0xce, 0xf6, 0xea, 0xb7, 0x17, 0xf7, 0x8c, |
| 1010 | 0x79, 0xd6, 0xa7, 0xbf, 0x8b, 0x3f, 0x1f, 0x53, |
| 1011 | 0x63, 0x75, 0x35, 0x2c, 0x60, 0xfd, 0x27, 0xd3, |
| 1012 | 0x94, 0xa5, 0x7c, 0xa1, 0x05, 0x58, 0x2d, 0xbd, |
| 1013 | 0xd9, 0xc7, 0xaf, 0x6b, 0x54, 0x0b, 0xe0, 0x38, |
| 1014 | 0x04, 0xc8, 0x9d, 0xe7, 0x14, 0xb1, 0x87, 0x9c, |
| 1015 | 0xdf, 0x6f, 0xf9, 0xda, 0x2a, 0xc4, 0x59, 0x16, |
| 1016 | 0x74, 0x91, 0xab, 0x26, 0x61, 0x76, 0x34, 0x2b, |
| 1017 | 0xad, 0x99, 0xfb, 0x72, 0xec, 0x33, 0x12, 0xde, |
| 1018 | 0x98, 0x3b, 0xc0, 0x9b, 0x3e, 0x18, 0x10, 0x3a, |
| 1019 | 0x56, 0xe1, 0x77, 0xc9, 0x1e, 0x9e, 0x95, 0xa3, |
| 1020 | 0x90, 0x19, 0xa8, 0x6c, 0x09, 0xd0, 0xf0, 0x86, |
| 1021 | }; |
| 1022 | |
| 1023 | static uint8_t octeon_snow3g_mulx(uint8_t v, uint8_t c) |
| 1024 | { |
| 1025 | return (v & 0x80) ? ((v << 1) ^ c) : (v << 1); |
| 1026 | } |
| 1027 | |
| 1028 | static uint8_t octeon_snow3g_mulxpow(uint8_t v, unsigned int n, uint8_t c) |
| 1029 | { |
| 1030 | while (n-- > 0) { |
| 1031 | v = octeon_snow3g_mulx(v, c); |
| 1032 | } |
| 1033 | return v; |
| 1034 | } |
| 1035 | |
| 1036 | static uint32_t octeon_snow3g_pack32(uint8_t b0, uint8_t b1, |
| 1037 | uint8_t b2, uint8_t b3) |
| 1038 | { |
| 1039 | return ((uint32_t)b0 << 24) |
| 1040 | | ((uint32_t)b1 << 16) |
| 1041 | | ((uint32_t)b2 << 8) |
| 1042 | | b3; |
| 1043 | } |
| 1044 | |
| 1045 | static uint32_t octeon_snow3g_mulalpha(uint8_t c) |
| 1046 | { |
| 1047 | return octeon_snow3g_pack32(octeon_snow3g_mulxpow(c, 23, 0xa9), |
| 1048 | octeon_snow3g_mulxpow(c, 245, 0xa9), |
| 1049 | octeon_snow3g_mulxpow(c, 48, 0xa9), |
| 1050 | octeon_snow3g_mulxpow(c, 239, 0xa9)); |
| 1051 | } |
| 1052 | |
| 1053 | static uint32_t octeon_snow3g_divalpha(uint8_t c) |
| 1054 | { |
| 1055 | return octeon_snow3g_pack32(octeon_snow3g_mulxpow(c, 16, 0xa9), |
| 1056 | octeon_snow3g_mulxpow(c, 39, 0xa9), |
| 1057 | octeon_snow3g_mulxpow(c, 6, 0xa9), |
| 1058 | octeon_snow3g_mulxpow(c, 64, 0xa9)); |
| 1059 | } |
| 1060 | |
| 1061 | static uint32_t octeon_snow3g_s1(uint32_t w) |
| 1062 | { |
| 1063 | uint8_t x0 = octeon_snow3g_sr[w >> 24]; |
| 1064 | uint8_t x1 = octeon_snow3g_sr[(uint8_t)(w >> 16)]; |
| 1065 | uint8_t x2 = octeon_snow3g_sr[(uint8_t)(w >> 8)]; |
| 1066 | uint8_t x3 = octeon_snow3g_sr[(uint8_t)w]; |
| 1067 | uint8_t r0 = octeon_snow3g_mulx(x0, 0x1b) ^ x1 ^ x2 ^ |
| 1068 | octeon_snow3g_mulx(x3, 0x1b) ^ x3; |
| 1069 | uint8_t r1 = octeon_snow3g_mulx(x0, 0x1b) ^ x0 ^ |
| 1070 | octeon_snow3g_mulx(x1, 0x1b) ^ x2 ^ x3; |
| 1071 | uint8_t r2 = x0 ^ octeon_snow3g_mulx(x1, 0x1b) ^ x1 ^ |
| 1072 | octeon_snow3g_mulx(x2, 0x1b) ^ x3; |
| 1073 | uint8_t r3 = x0 ^ x1 ^ octeon_snow3g_mulx(x2, 0x1b) ^ x2 ^ |
| 1074 | octeon_snow3g_mulx(x3, 0x1b); |
| 1075 | |
| 1076 | return octeon_snow3g_pack32(r0, r1, r2, r3); |
| 1077 | } |
| 1078 | |
| 1079 | static uint32_t octeon_snow3g_s2(uint32_t w) |
| 1080 | { |
| 1081 | uint8_t x0 = octeon_snow3g_sq[w >> 24]; |
| 1082 | uint8_t x1 = octeon_snow3g_sq[(uint8_t)(w >> 16)]; |
| 1083 | uint8_t x2 = octeon_snow3g_sq[(uint8_t)(w >> 8)]; |
| 1084 | uint8_t x3 = octeon_snow3g_sq[(uint8_t)w]; |
| 1085 | uint8_t r0 = octeon_snow3g_mulx(x0, 0x69) ^ x1 ^ x2 ^ |
| 1086 | octeon_snow3g_mulx(x3, 0x69) ^ x3; |
| 1087 | uint8_t r1 = octeon_snow3g_mulx(x0, 0x69) ^ x0 ^ |
| 1088 | octeon_snow3g_mulx(x1, 0x69) ^ x2 ^ x3; |
| 1089 | uint8_t r2 = x0 ^ octeon_snow3g_mulx(x1, 0x69) ^ x1 ^ |
| 1090 | octeon_snow3g_mulx(x2, 0x69) ^ x3; |
| 1091 | uint8_t r3 = x0 ^ x1 ^ octeon_snow3g_mulx(x2, 0x69) ^ x2 ^ |
| 1092 | octeon_snow3g_mulx(x3, 0x69); |
| 1093 | |
| 1094 | return octeon_snow3g_pack32(r0, r1, r2, r3); |
| 1095 | } |
| 1096 | |
| 1097 | static uint32_t octeon_snow3g_lfsr(const MIPSOcteonCryptoState *crypto, |
| 1098 | unsigned int index) |
| 1099 | { |
| 1100 | uint64_t pair = crypto->hsh_dat[index / 2]; |
| 1101 | |
| 1102 | return index & 1 ? octeon_crypto_lo32(pair) : octeon_crypto_hi32(pair); |
| 1103 | } |
| 1104 | |
| 1105 | static void octeon_snow3g_set_lfsr(MIPSOcteonCryptoState *crypto, |
| 1106 | unsigned int index, uint32_t value) |
| 1107 | { |
| 1108 | uint32_t hi = octeon_crypto_hi32(crypto->hsh_dat[index / 2]); |
| 1109 | uint32_t lo = octeon_crypto_lo32(crypto->hsh_dat[index / 2]); |
| 1110 | |
| 1111 | if (index & 1) { |
| 1112 | lo = value; |
| 1113 | } else { |
| 1114 | hi = value; |
| 1115 | } |
| 1116 | crypto->hsh_dat[index / 2] = octeon_crypto_pack32(hi, lo); |
| 1117 | } |
| 1118 | |
| 1119 | static uint32_t octeon_snow3g_fsm(const MIPSOcteonCryptoState *crypto, |
| 1120 | unsigned int index) |
| 1121 | { |
| 1122 | return crypto->hsh_iv[1 + index]; |
| 1123 | } |
| 1124 | |
| 1125 | static void octeon_snow3g_set_fsm(MIPSOcteonCryptoState *crypto, |
| 1126 | unsigned int index, uint32_t value) |
| 1127 | { |
| 1128 | crypto->hsh_iv[1 + index] = value; |
| 1129 | } |
| 1130 | |
| 1131 | static uint32_t octeon_snow3g_clock_fsm(MIPSOcteonCryptoState *crypto) |
| 1132 | { |
| 1133 | uint32_t fsm0 = octeon_snow3g_fsm(crypto, 0); |
| 1134 | uint32_t fsm1 = octeon_snow3g_fsm(crypto, 1); |
| 1135 | uint32_t fsm2 = octeon_snow3g_fsm(crypto, 2); |
| 1136 | uint32_t f = (uint32_t)(octeon_snow3g_lfsr(crypto, 15) + fsm0) ^ fsm1; |
| 1137 | uint32_t r = (uint32_t)(fsm1 + (fsm2 ^ octeon_snow3g_lfsr(crypto, 5))); |
| 1138 | |
| 1139 | octeon_snow3g_set_fsm(crypto, 2, octeon_snow3g_s2(fsm1)); |
| 1140 | octeon_snow3g_set_fsm(crypto, 1, octeon_snow3g_s1(fsm0)); |
| 1141 | octeon_snow3g_set_fsm(crypto, 0, r); |
| 1142 | return f; |
| 1143 | } |
| 1144 | |
| 1145 | static void octeon_snow3g_clock_lfsr(MIPSOcteonCryptoState *crypto, |
| 1146 | bool init_mode, uint32_t f) |
| 1147 | { |
| 1148 | uint32_t lfsr[16]; |
| 1149 | uint32_t s0; |
| 1150 | uint32_t s11; |
| 1151 | uint32_t v; |
| 1152 | int i; |
| 1153 | |
| 1154 | for (i = 0; i < 16; i++) { |
| 1155 | lfsr[i] = octeon_snow3g_lfsr(crypto, i); |
| 1156 | } |
| 1157 | |
| 1158 | s0 = lfsr[0]; |
| 1159 | s11 = lfsr[11]; |
| 1160 | v = (s0 << 8) ^ octeon_snow3g_mulalpha(s0 >> 24) ^ |
| 1161 | lfsr[2] ^ (s11 >> 8) ^ octeon_snow3g_divalpha((uint8_t)s11); |
| 1162 | |
| 1163 | if (init_mode) { |
| 1164 | v ^= f; |
| 1165 | } |
| 1166 | |
| 1167 | for (i = 0; i < 15; i++) { |
| 1168 | octeon_snow3g_set_lfsr(crypto, i, lfsr[i + 1]); |
| 1169 | } |
| 1170 | octeon_snow3g_set_lfsr(crypto, 15, v); |
| 1171 | } |
| 1172 | |
| 1173 | static uint32_t octeon_snow3g_generate_word(MIPSOcteonCryptoState *crypto) |
| 1174 | { |
| 1175 | uint32_t f = octeon_snow3g_clock_fsm(crypto); |
| 1176 | uint32_t z = f ^ octeon_snow3g_lfsr(crypto, 0); |
| 1177 | |
| 1178 | octeon_snow3g_clock_lfsr(crypto, false, 0); |
| 1179 | return z; |
| 1180 | } |
| 1181 | |
| 1182 | static void octeon_snow3g_queue_result(MIPSOcteonCryptoState *crypto) |
| 1183 | { |
| 1184 | uint32_t z0 = octeon_snow3g_generate_word(crypto); |
| 1185 | uint32_t z1 = octeon_snow3g_generate_word(crypto); |
| 1186 | |
| 1187 | crypto->hsh_iv[0] = octeon_crypto_pack32(z0, z1); |
| 1188 | } |
| 1189 | |
| 1190 | static void octeon_snow3g_start(MIPSOcteonCryptoState *crypto, uint64_t data) |
| 1191 | { |
| 1192 | int i; |
| 1193 | |
| 1194 | for (i = 0; i < 14; i++) { |
| 1195 | octeon_snow3g_set_lfsr(crypto, i, octeon_snow3g_lfsr(crypto, i)); |
| 1196 | } |
| 1197 | octeon_snow3g_set_lfsr(crypto, 14, data >> 32); |
| 1198 | octeon_snow3g_set_lfsr(crypto, 15, data); |
| 1199 | for (i = 0; i < 3; i++) { |
| 1200 | octeon_snow3g_set_fsm(crypto, i, 0); |
| 1201 | } |
| 1202 | |
| 1203 | for (i = 0; i < 32; i++) { |
| 1204 | uint32_t f = octeon_snow3g_clock_fsm(crypto); |
| 1205 | |
| 1206 | octeon_snow3g_clock_lfsr(crypto, true, f); |
| 1207 | } |
| 1208 | |
| 1209 | (void)octeon_snow3g_clock_fsm(crypto); |
| 1210 | octeon_snow3g_clock_lfsr(crypto, false, 0); |
| 1211 | octeon_snow3g_queue_result(crypto); |
| 1212 | } |
| 1213 | |
| 1214 | static void octeon_snow3g_more(MIPSOcteonCryptoState *crypto) |
| 1215 | { |
| 1216 | octeon_snow3g_queue_result(crypto); |
| 1217 | } |
| 1218 | |
| 1219 | static int octeon_aes_key_bits(const MIPSOcteonCryptoState *crypto) |
| 1220 | { |
| 1221 | enum { |
| 1222 | OCTEON_AES_KEYLEN_128 = 1, |
| 1223 | OCTEON_AES_KEYLEN_192 = 2, |
| 1224 | OCTEON_AES_KEYLEN_256 = 3, |
| 1225 | }; |
| 1226 | |
| 1227 | switch (crypto->aes_keylen) { |
| 1228 | case OCTEON_AES_KEYLEN_128: |
| 1229 | return 128; |
| 1230 | case OCTEON_AES_KEYLEN_192: |
| 1231 | return 192; |
| 1232 | case OCTEON_AES_KEYLEN_256: |
| 1233 | return 256; |
| 1234 | default: |
| 1235 | return 0; |
| 1236 | } |
| 1237 | } |
| 1238 | |
| 1239 | static void octeon_aes_load_key(const MIPSOcteonCryptoState *crypto, |
| 1240 | uint8_t *key, size_t keylen) |
| 1241 | { |
| 1242 | stq_be_p(key, crypto->aes_key[0]); |
| 1243 | stq_be_p(key + 8, crypto->aes_key[1]); |
| 1244 | if (keylen > 16) { |
| 1245 | stq_be_p(key + 16, crypto->aes_key[2]); |
| 1246 | } |
| 1247 | if (keylen > 24) { |
| 1248 | stq_be_p(key + 24, crypto->aes_key[3]); |
| 1249 | } |
| 1250 | } |
| 1251 | |
| 1252 | static void octeon_aes_load_block(const uint64_t regs[2], uint8_t *block) |
| 1253 | { |
| 1254 | stq_be_p(block, regs[0]); |
| 1255 | stq_be_p(block + 8, regs[1]); |
| 1256 | } |
| 1257 | |
| 1258 | static void octeon_aes_store_block(uint64_t regs[2], const uint8_t *block) |
| 1259 | { |
| 1260 | regs[0] = ldq_be_p(block); |
| 1261 | regs[1] = ldq_be_p(block + 8); |
| 1262 | } |
| 1263 | |
| 1264 | static void octeon_aes_encrypt_common(MIPSOcteonCryptoState *crypto, bool cbc) |
| 1265 | { |
| 1266 | AES_KEY key; |
| 1267 | uint8_t in[16]; |
| 1268 | uint8_t out[16]; |
| 1269 | uint8_t iv[16]; |
| 1270 | uint8_t raw_key[32] = {}; |
| 1271 | int bits = octeon_aes_key_bits(crypto); |
| 1272 | |
| 1273 | if (!bits) { |
| 1274 | return; |
| 1275 | } |
| 1276 | |
| 1277 | octeon_aes_load_key(crypto, raw_key, bits / 8); |
| 1278 | octeon_aes_load_block(crypto->aes_resinp, in); |
| 1279 | if (cbc) { |
| 1280 | int i; |
| 1281 | |
| 1282 | octeon_aes_load_block(crypto->aes_iv, iv); |
| 1283 | for (i = 0; i < sizeof(in); i++) { |
| 1284 | in[i] ^= iv[i]; |
| 1285 | } |
| 1286 | } |
| 1287 | |
| 1288 | AES_set_encrypt_key(raw_key, bits, &key); |
| 1289 | AES_encrypt(in, out, &key); |
| 1290 | octeon_aes_store_block(crypto->aes_resinp, out); |
| 1291 | if (cbc) { |
| 1292 | octeon_aes_store_block(crypto->aes_iv, out); |
| 1293 | } |
| 1294 | } |
| 1295 | |
| 1296 | static void octeon_aes_decrypt_common(MIPSOcteonCryptoState *crypto, bool cbc) |
| 1297 | { |
| 1298 | AES_KEY key; |
| 1299 | uint8_t in[16]; |
| 1300 | uint8_t out[16]; |
| 1301 | uint8_t iv[16]; |
| 1302 | uint8_t next_iv[16]; |
| 1303 | uint8_t raw_key[32] = {}; |
| 1304 | int bits = octeon_aes_key_bits(crypto); |
| 1305 | int i; |
| 1306 | |
| 1307 | if (!bits) { |
| 1308 | return; |
| 1309 | } |
| 1310 | |
| 1311 | octeon_aes_load_key(crypto, raw_key, bits / 8); |
| 1312 | octeon_aes_load_block(crypto->aes_resinp, in); |
| 1313 | if (cbc) { |
| 1314 | memcpy(next_iv, in, sizeof(next_iv)); |
| 1315 | octeon_aes_load_block(crypto->aes_iv, iv); |
| 1316 | } |
| 1317 | |
| 1318 | AES_set_decrypt_key(raw_key, bits, &key); |
| 1319 | AES_decrypt(in, out, &key); |
| 1320 | if (cbc) { |
| 1321 | for (i = 0; i < sizeof(out); i++) { |
| 1322 | out[i] ^= iv[i]; |
| 1323 | } |
| 1324 | } |
| 1325 | |
| 1326 | octeon_aes_store_block(crypto->aes_resinp, out); |
| 1327 | if (cbc) { |
| 1328 | octeon_aes_store_block(crypto->aes_iv, next_iv); |
| 1329 | } |
| 1330 | } |
| 1331 | |
| 1332 | void helper_octeon_cp2_mt_aes_enc_cbc1(CPUMIPSState *env, uint64_t value) |
| 1333 | { |
| 1334 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1335 | |
| 1336 | crypto->aes_resinp[1] = value; |
| 1337 | octeon_aes_encrypt_common(crypto, true); |
| 1338 | } |
| 1339 | |
| 1340 | void helper_octeon_cp2_mt_aes_enc1(CPUMIPSState *env, uint64_t value) |
| 1341 | { |
| 1342 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1343 | |
| 1344 | crypto->aes_resinp[1] = value; |
| 1345 | octeon_aes_encrypt_common(crypto, false); |
| 1346 | } |
| 1347 | |
| 1348 | void helper_octeon_cp2_mt_aes_dec_cbc1(CPUMIPSState *env, uint64_t value) |
| 1349 | { |
| 1350 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1351 | |
| 1352 | crypto->aes_resinp[1] = value; |
| 1353 | octeon_aes_decrypt_common(crypto, true); |
| 1354 | } |
| 1355 | |
| 1356 | void helper_octeon_cp2_mt_aes_dec1(CPUMIPSState *env, uint64_t value) |
| 1357 | { |
| 1358 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1359 | |
| 1360 | crypto->aes_resinp[1] = value; |
| 1361 | octeon_aes_decrypt_common(crypto, false); |
| 1362 | } |
| 1363 | |
| 1364 | static uint32_t octeon_sms4_t(uint32_t x) |
| 1365 | { |
| 1366 | x = sm4_subword(x); |
| 1367 | return x ^ rol32(x, 2) ^ rol32(x, 10) ^ rol32(x, 18) ^ rol32(x, 24); |
| 1368 | } |
| 1369 | |
| 1370 | static uint32_t octeon_sms4_t_key(uint32_t x) |
| 1371 | { |
| 1372 | x = sm4_subword(x); |
| 1373 | return x ^ rol32(x, 13) ^ rol32(x, 23); |
| 1374 | } |
| 1375 | |
| 1376 | static void octeon_sms4_expand_key(const uint8_t *key, uint32_t round_keys[32]) |
| 1377 | { |
| 1378 | static const uint32_t fk[4] = { |
| 1379 | 0xa3b1bac6U, 0x56aa3350U, 0x677d9197U, 0xb27022dcU, |
| 1380 | }; |
| 1381 | uint32_t k[36]; |
| 1382 | |
| 1383 | for (int i = 0; i < 4; i++) { |
| 1384 | k[i] = ldl_be_p(key + i * 4) ^ fk[i]; |
| 1385 | } |
| 1386 | for (int i = 0; i < 32; i++) { |
| 1387 | k[i + 4] = k[i] ^ octeon_sms4_t_key(k[i + 1] ^ k[i + 2] ^ |
| 1388 | k[i + 3] ^ sm4_ck[i]); |
| 1389 | round_keys[i] = k[i + 4]; |
| 1390 | } |
| 1391 | } |
| 1392 | |
| 1393 | static void octeon_sms4_crypt_block(const uint8_t *in, uint8_t *out, |
| 1394 | const uint32_t round_keys[32], |
| 1395 | bool encrypt) |
| 1396 | { |
| 1397 | uint32_t x[36]; |
| 1398 | |
| 1399 | for (int i = 0; i < 4; i++) { |
| 1400 | x[i] = ldl_be_p(in + i * 4); |
| 1401 | } |
| 1402 | for (int i = 0; i < 32; i++) { |
| 1403 | uint32_t rk = round_keys[encrypt ? i : 31 - i]; |
| 1404 | |
| 1405 | x[i + 4] = x[i] ^ octeon_sms4_t(x[i + 1] ^ x[i + 2] ^ |
| 1406 | x[i + 3] ^ rk); |
| 1407 | } |
| 1408 | stl_be_p(out, x[35]); |
| 1409 | stl_be_p(out + 4, x[34]); |
| 1410 | stl_be_p(out + 8, x[33]); |
| 1411 | stl_be_p(out + 12, x[32]); |
| 1412 | } |
| 1413 | |
| 1414 | static void octeon_sms4_crypt_common(MIPSOcteonCryptoState *crypto, |
| 1415 | bool encrypt, bool cbc) |
| 1416 | { |
| 1417 | uint8_t key[16]; |
| 1418 | uint8_t in[16]; |
| 1419 | uint8_t out[16]; |
| 1420 | uint8_t iv[16]; |
| 1421 | uint8_t next_iv[16]; |
| 1422 | uint32_t round_keys[32]; |
| 1423 | |
| 1424 | /* |
| 1425 | * SMS4 aliases the AES state onto the RESINP, IV, and KEY banks, |
| 1426 | * with only the operation selectors remaining distinct. |
| 1427 | */ |
| 1428 | octeon_aes_load_key(crypto, key, sizeof(key)); |
| 1429 | octeon_aes_load_block(crypto->aes_resinp, in); |
| 1430 | if (cbc) { |
| 1431 | octeon_aes_load_block(crypto->aes_iv, iv); |
| 1432 | if (encrypt) { |
| 1433 | for (int i = 0; i < sizeof(in); i++) { |
| 1434 | in[i] ^= iv[i]; |
| 1435 | } |
| 1436 | } else { |
| 1437 | memcpy(next_iv, in, sizeof(next_iv)); |
| 1438 | } |
| 1439 | } |
| 1440 | |
| 1441 | octeon_sms4_expand_key(key, round_keys); |
| 1442 | octeon_sms4_crypt_block(in, out, round_keys, encrypt); |
| 1443 | if (cbc && !encrypt) { |
| 1444 | for (int i = 0; i < sizeof(out); i++) { |
| 1445 | out[i] ^= iv[i]; |
| 1446 | } |
| 1447 | } |
| 1448 | |
| 1449 | octeon_aes_store_block(crypto->aes_resinp, out); |
| 1450 | if (cbc) { |
| 1451 | octeon_aes_store_block(crypto->aes_iv, encrypt ? out : next_iv); |
| 1452 | } |
| 1453 | } |
| 1454 | |
| 1455 | void helper_octeon_cp2_mt_sms4_enc_cbc1(CPUMIPSState *env, uint64_t value) |
| 1456 | { |
| 1457 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1458 | |
| 1459 | crypto->aes_resinp[1] = value; |
| 1460 | octeon_sms4_crypt_common(crypto, true, true); |
| 1461 | } |
| 1462 | |
| 1463 | void helper_octeon_cp2_mt_sms4_enc1(CPUMIPSState *env, uint64_t value) |
| 1464 | { |
| 1465 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1466 | |
| 1467 | crypto->aes_resinp[1] = value; |
| 1468 | octeon_sms4_crypt_common(crypto, true, false); |
| 1469 | } |
| 1470 | |
| 1471 | void helper_octeon_cp2_mt_sms4_dec_cbc1(CPUMIPSState *env, uint64_t value) |
| 1472 | { |
| 1473 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1474 | |
| 1475 | crypto->aes_resinp[1] = value; |
| 1476 | octeon_sms4_crypt_common(crypto, false, true); |
| 1477 | } |
| 1478 | |
| 1479 | void helper_octeon_cp2_mt_sms4_dec1(CPUMIPSState *env, uint64_t value) |
| 1480 | { |
| 1481 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 1482 | |
| 1483 | crypto->aes_resinp[1] = value; |
| 1484 | octeon_sms4_crypt_common(crypto, false, false); |
| 1485 | } |
| 1486 | |
| 1487 | static const uint8_t octeon_des_ip[64] = { |
| 1488 | 58, 50, 42, 34, 26, 18, 10, 2, |
| 1489 | 60, 52, 44, 36, 28, 20, 12, 4, |
| 1490 | 62, 54, 46, 38, 30, 22, 14, 6, |
| 1491 | 64, 56, 48, 40, 32, 24, 16, 8, |
| 1492 | 57, 49, 41, 33, 25, 17, 9, 1, |
| 1493 | 59, 51, 43, 35, 27, 19, 11, 3, |
| 1494 | 61, 53, 45, 37, 29, 21, 13, 5, |
| 1495 | 63, 55, 47, 39, 31, 23, 15, 7, |
| 1496 | }; |
| 1497 | |
| 1498 | static const uint8_t octeon_des_fp[64] = { |
| 1499 | 40, 8, 48, 16, 56, 24, 64, 32, |
| 1500 | 39, 7, 47, 15, 55, 23, 63, 31, |
| 1501 | 38, 6, 46, 14, 54, 22, 62, 30, |
| 1502 | 37, 5, 45, 13, 53, 21, 61, 29, |
| 1503 | 36, 4, 44, 12, 52, 20, 60, 28, |
| 1504 | 35, 3, 43, 11, 51, 19, 59, 27, |
| 1505 | 34, 2, 42, 10, 50, 18, 58, 26, |
| 1506 | 33, 1, 41, 9, 49, 17, 57, 25, |
| 1507 | }; |
| 1508 | |
| 1509 | static const uint8_t octeon_des_e[48] = { |
| 1510 | 32, 1, 2, 3, 4, 5, |
| 1511 | 4, 5, 6, 7, 8, 9, |
| 1512 | 8, 9, 10, 11, 12, 13, |
| 1513 | 12, 13, 14, 15, 16, 17, |
| 1514 | 16, 17, 18, 19, 20, 21, |
| 1515 | 20, 21, 22, 23, 24, 25, |
| 1516 | 24, 25, 26, 27, 28, 29, |
| 1517 | 28, 29, 30, 31, 32, 1, |
| 1518 | }; |
| 1519 | |
| 1520 | static const uint8_t octeon_des_p[32] = { |
| 1521 | 16, 7, 20, 21, 29, 12, 28, 17, |
| 1522 | 1, 15, 23, 26, 5, 18, 31, 10, |
| 1523 | 2, 8, 24, 14, 32, 27, 3, 9, |
| 1524 | 19, 13, 30, 6, 22, 11, 4, 25, |
| 1525 | }; |
| 1526 | |
| 1527 | static const uint8_t octeon_des_pc1[56] = { |
| 1528 | 57, 49, 41, 33, 25, 17, 9, |
| 1529 | 1, 58, 50, 42, 34, 26, 18, |
| 1530 | 10, 2, 59, 51, 43, 35, 27, |
| 1531 | 19, 11, 3, 60, 52, 44, 36, |
| 1532 | 63, 55, 47, 39, 31, 23, 15, |
| 1533 | 7, 62, 54, 46, 38, 30, 22, |
| 1534 | 14, 6, 61, 53, 45, 37, 29, |
| 1535 | 21, 13, 5, 28, 20, 12, 4, |
| 1536 | }; |
| 1537 | |
| 1538 | static const uint8_t octeon_des_pc2[48] = { |
| 1539 | 14, 17, 11, 24, 1, 5, |
| 1540 | 3, 28, 15, 6, 21, 10, |
| 1541 | 23, 19, 12, 4, 26, 8, |
| 1542 | 16, 7, 27, 20, 13, 2, |
| 1543 | 41, 52, 31, 37, 47, 55, |
| 1544 | 30, 40, 51, 45, 33, 48, |
| 1545 | 44, 49, 39, 56, 34, 53, |
| 1546 | 46, 42, 50, 36, 29, 32, |
| 1547 | }; |
| 1548 | |
| 1549 | static const uint8_t octeon_des_rotations[16] = { |
| 1550 | 1, 1, 2, 2, 2, 2, 2, 2, |
| 1551 | 1, 2, 2, 2, 2, 2, 2, 1, |
| 1552 | }; |
| 1553 | |
| 1554 | static const uint8_t octeon_des_sboxes[8][64] = { |
| 1555 | { |
| 1556 | 14, 4, 13, 1, 2, 15, 11, 8, 3, 10, 6, 12, 5, 9, 0, 7, |
| 1557 | 0, 15, 7, 4, 14, 2, 13, 1, 10, 6, 12, 11, 9, 5, 3, 8, |
| 1558 | 4, 1, 14, 8, 13, 6, 2, 11, 15, 12, 9, 7, 3, 10, 5, 0, |
| 1559 | 15, 12, 8, 2, 4, 9, 1, 7, 5, 11, 3, 14, 10, 0, 6, 13, |
| 1560 | }, |
| 1561 | { |
| 1562 | 15, 1, 8, 14, 6, 11, 3, 4, 9, 7, 2, 13, 12, 0, 5, 10, |
| 1563 | 3, 13, 4, 7, 15, 2, 8, 14, 12, 0, 1, 10, 6, 9, 11, 5, |
| 1564 | 0, 14, 7, 11, 10, 4, 13, 1, 5, 8, 12, 6, 9, 3, 2, 15, |
| 1565 | 13, 8, 10, 1, 3, 15, 4, 2, 11, 6, 7, 12, 0, 5, 14, 9, |
| 1566 | }, |
| 1567 | { |
| 1568 | 10, 0, 9, 14, 6, 3, 15, 5, 1, 13, 12, 7, 11, 4, 2, 8, |
| 1569 | 13, 7, 0, 9, 3, 4, 6, 10, 2, 8, 5, 14, 12, 11, 15, 1, |
| 1570 | 13, 6, 4, 9, 8, 15, 3, 0, 11, 1, 2, 12, 5, 10, 14, 7, |
| 1571 | 1, 10, 13, 0, 6, 9, 8, 7, 4, 15, 14, 3, 11, 5, 2, 12, |
| 1572 | }, |
| 1573 | { |
| 1574 | 7, 13, 14, 3, 0, 6, 9, 10, 1, 2, 8, 5, 11, 12, 4, 15, |
| 1575 | 13, 8, 11, 5, 6, 15, 0, 3, 4, 7, 2, 12, 1, 10, 14, 9, |
| 1576 | 10, 6, 9, 0, 12, 11, 7, 13, 15, 1, 3, 14, 5, 2, 8, 4, |
| 1577 | 3, 15, 0, 6, 10, 1, 13, 8, 9, 4, 5, 11, 12, 7, 2, 14, |
| 1578 | }, |
| 1579 | { |
| 1580 | 2, 12, 4, 1, 7, 10, 11, 6, 8, 5, 3, 15, 13, 0, 14, 9, |
| 1581 | 14, 11, 2, 12, 4, 7, 13, 1, 5, 0, 15, 10, 3, 9, 8, 6, |
| 1582 | 4, 2, 1, 11, 10, 13, 7, 8, 15, 9, 12, 5, 6, 3, 0, 14, |
| 1583 | 11, 8, 12, 7, 1, 14, 2, 13, 6, 15, 0, 9, 10, 4, 5, 3, |
| 1584 | }, |
| 1585 | { |
| 1586 | 12, 1, 10, 15, 9, 2, 6, 8, 0, 13, 3, 4, 14, 7, 5, 11, |
| 1587 | 10, 15, 4, 2, 7, 12, 9, 5, 6, 1, 13, 14, 0, 11, 3, 8, |
| 1588 | 9, 14, 15, 5, 2, 8, 12, 3, 7, 0, 4, 10, 1, 13, 11, 6, |
| 1589 | 4, 3, 2, 12, 9, 5, 15, 10, 11, 14, 1, 7, 6, 0, 8, 13, |
| 1590 | }, |
| 1591 | { |
| 1592 | 4, 11, 2, 14, 15, 0, 8, 13, 3, 12, 9, 7, 5, 10, 6, 1, |
| 1593 | 13, 0, 11, 7, 4, 9, 1, 10, 14, 3, 5, 12, 2, 15, 8, 6, |
| 1594 | 1, 4, 11, 13, 12, 3, 7, 14, 10, 15, 6, 8, 0, 5, 9, 2, |
| 1595 | 6, 11, 13, 8, 1, 4, 10, 7, 9, 5, 0, 15, 14, 2, 3, 12, |
| 1596 | }, |
| 1597 | { |
| 1598 | 13, 2, 8, 4, 6, 15, 11, 1, 10, 9, 3, 14, 5, 0, 12, 7, |
| 1599 | 1, 15, 13, 8, 10, 3, 7, 4, 12, 5, 6, 11, 0, 14, 9, 2, |
| 1600 | 7, 11, 4, 1, 9, 12, 14, 2, 0, 6, 10, 13, 15, 3, 5, 8, |
| 1601 | 2, 1, 14, 7, 4, 10, 8, 13, 15, 12, 9, 0, 3, 5, 6, 11, |
| 1602 | }, |
| 1603 | }; |
| 1604 | |
| 1605 | static const uint8_t octeon_kasumi_s7[128] = { |
| 1606 | 54, 50, 62, 56, 22, 34, 94, 96, 38, 6, 63, 93, 2, 18, |
| 1607 | 123, 33, 55, 113, 39, 114, 21, 67, 65, 12, 47, 73, 46, 27, |
| 1608 | 25, 111, 124, 81, 53, 9, 121, 79, 52, 60, 58, 48, 101, 127, |
| 1609 | 40, 120, 104, 70, 71, 43, 20, 122, 72, 61, 23, 109, 13, 100, |
| 1610 | 77, 1, 16, 7, 82, 10, 105, 98, 117, 116, 76, 11, 89, 106, |
| 1611 | 0, 125, 118, 99, 86, 69, 30, 57, 126, 87, 112, 51, 17, 5, |
| 1612 | 95, 14, 90, 84, 91, 8, 35, 103, 32, 97, 28, 66, 102, 31, |
| 1613 | 26, 45, 75, 4, 85, 92, 37, 74, 80, 49, 68, 29, 115, 44, |
| 1614 | 64, 107, 108, 24, 110, 83, 36, 78, 42, 19, 15, 41, 88, 119, |
| 1615 | 59, 3, |
| 1616 | }; |
| 1617 | |
| 1618 | static const uint16_t octeon_kasumi_s9[512] = { |
| 1619 | 167, 239, 161, 379, 391, 334, 9, 338, 38, 226, 48, 358, 452, 385, |
| 1620 | 90, 397, 183, 253, 147, 331, 415, 340, 51, 362, 306, 500, 262, 82, |
| 1621 | 216, 159, 356, 177, 175, 241, 489, 37, 206, 17, 0, 333, 44, 254, |
| 1622 | 378, 58, 143, 220, 81, 400, 95, 3, 315, 245, 54, 235, 218, 405, |
| 1623 | 472, 264, 172, 494, 371, 290, 399, 76, 165, 197, 395, 121, 257, 480, |
| 1624 | 423, 212, 240, 28, 462, 176, 406, 507, 288, 223, 501, 407, 249, 265, |
| 1625 | 89, 186, 221, 428, 164, 74, 440, 196, 458, 421, 350, 163, 232, 158, |
| 1626 | 134, 354, 13, 250, 491, 142, 191, 69, 193, 425, 152, 227, 366, 135, |
| 1627 | 344, 300, 276, 242, 437, 320, 113, 278, 11, 243, 87, 317, 36, 93, |
| 1628 | 496, 27, 487, 446, 482, 41, 68, 156, 457, 131, 326, 403, 339, 20, |
| 1629 | 39, 115, 442, 124, 475, 384, 508, 53, 112, 170, 479, 151, 126, 169, |
| 1630 | 73, 268, 279, 321, 168, 364, 363, 292, 46, 499, 393, 327, 324, 24, |
| 1631 | 456, 267, 157, 460, 488, 426, 309, 229, 439, 506, 208, 271, 349, 401, |
| 1632 | 434, 236, 16, 209, 359, 52, 56, 120, 199, 277, 465, 416, 252, 287, |
| 1633 | 246, 6, 83, 305, 420, 345, 153, 502, 65, 61, 244, 282, 173, 222, |
| 1634 | 418, 67, 386, 368, 261, 101, 476, 291, 195, 430, 49, 79, 166, 330, |
| 1635 | 280, 383, 373, 128, 382, 408, 155, 495, 367, 388, 274, 107, 459, 417, |
| 1636 | 62, 454, 132, 225, 203, 316, 234, 14, 301, 91, 503, 286, 424, 211, |
| 1637 | 347, 307, 140, 374, 35, 103, 125, 427, 19, 214, 453, 146, 498, 314, |
| 1638 | 444, 230, 256, 329, 198, 285, 50, 116, 78, 410, 10, 205, 510, 171, |
| 1639 | 231, 45, 139, 467, 29, 86, 505, 32, 72, 26, 342, 150, 313, 490, |
| 1640 | 431, 238, 411, 325, 149, 473, 40, 119, 174, 355, 185, 233, 389, 71, |
| 1641 | 448, 273, 372, 55, 110, 178, 322, 12, 469, 392, 369, 190, 1, 109, |
| 1642 | 375, 137, 181, 88, 75, 308, 260, 484, 98, 272, 370, 275, 412, 111, |
| 1643 | 336, 318, 4, 504, 492, 259, 304, 77, 337, 435, 21, 357, 303, 332, |
| 1644 | 483, 18, 47, 85, 25, 497, 474, 289, 100, 269, 296, 478, 270, 106, |
| 1645 | 31, 104, 433, 84, 414, 486, 394, 96, 99, 154, 511, 148, 413, 361, |
| 1646 | 409, 255, 162, 215, 302, 201, 266, 351, 343, 144, 441, 365, 108, 298, |
| 1647 | 251, 34, 182, 509, 138, 210, 335, 133, 311, 352, 328, 141, 396, 346, |
| 1648 | 123, 319, 450, 281, 429, 228, 443, 481, 92, 404, 485, 422, 248, 297, |
| 1649 | 23, 213, 130, 466, 22, 217, 283, 70, 294, 360, 419, 127, 312, 377, |
| 1650 | 7, 468, 194, 2, 117, 295, 463, 258, 224, 447, 247, 187, 80, 398, |
| 1651 | 284, 353, 105, 390, 299, 471, 470, 184, 57, 200, 348, 63, 204, 188, |
| 1652 | 33, 451, 97, 30, 310, 219, 94, 160, 129, 493, 64, 179, 263, 102, |
| 1653 | 189, 207, 114, 402, 438, 477, 387, 122, 192, 42, 381, 5, 145, 118, |
| 1654 | 180, 449, 293, 323, 136, 380, 43, 66, 60, 455, 341, 445, 202, 432, |
| 1655 | 8, 237, 15, 376, 436, 464, 59, 461, |
| 1656 | }; |
| 1657 | |
| 1658 | static const uint16_t octeon_kasumi_constants[8] = { |
| 1659 | 0x0123, 0x4567, 0x89ab, 0xcdef, 0xfedc, 0xba98, 0x7654, 0x3210, |
| 1660 | }; |
| 1661 | |
| 1662 | typedef struct OcteonKasumiSubkeys { |
| 1663 | uint16_t kli1[8]; |
| 1664 | uint16_t kli2[8]; |
| 1665 | uint16_t koi1[8]; |
| 1666 | uint16_t koi2[8]; |
| 1667 | uint16_t koi3[8]; |
| 1668 | uint16_t kii1[8]; |
| 1669 | uint16_t kii2[8]; |
| 1670 | uint16_t kii3[8]; |
| 1671 | } OcteonKasumiSubkeys; |
| 1672 | |
| 1673 | static uint64_t octeon_des_permute(uint64_t input, const uint8_t *table, |
| 1674 | size_t output_bits, size_t input_bits) |
| 1675 | { |
| 1676 | uint64_t out = 0; |
| 1677 | |
| 1678 | for (size_t i = 0; i < output_bits; i++) { |
| 1679 | unsigned src = table[i] - 1; |
| 1680 | |
| 1681 | out = (out << 1) | ((input >> (input_bits - 1 - src)) & 1); |
| 1682 | } |
| 1683 | return out; |
| 1684 | } |
| 1685 | |
| 1686 | static uint32_t octeon_des_rotate28(uint32_t v, unsigned shift) |
| 1687 | { |
| 1688 | return ((v << shift) | (v >> (28 - shift))) & 0x0fffffffU; |
| 1689 | } |
| 1690 | |
| 1691 | static void octeon_des_expand_subkeys(uint64_t key, uint64_t subkeys[16]) |
| 1692 | { |
| 1693 | uint64_t permuted = octeon_des_permute(key, octeon_des_pc1, |
| 1694 | ARRAY_SIZE(octeon_des_pc1), 64); |
| 1695 | uint32_t c = (permuted >> 28) & 0x0fffffffU; |
| 1696 | uint32_t d = permuted & 0x0fffffffU; |
| 1697 | |
| 1698 | for (int i = 0; i < 16; i++) { |
| 1699 | c = octeon_des_rotate28(c, octeon_des_rotations[i]); |
| 1700 | d = octeon_des_rotate28(d, octeon_des_rotations[i]); |
| 1701 | subkeys[i] = octeon_des_permute(((uint64_t)c << 28) | d, |
| 1702 | octeon_des_pc2, |
| 1703 | ARRAY_SIZE(octeon_des_pc2), 56); |
| 1704 | } |
| 1705 | } |
| 1706 | |
| 1707 | static uint32_t octeon_des_f(uint32_t r, uint64_t subkey) |
| 1708 | { |
| 1709 | uint64_t expanded = octeon_des_permute(r, octeon_des_e, |
| 1710 | ARRAY_SIZE(octeon_des_e), 32); |
| 1711 | uint32_t out = 0; |
| 1712 | |
| 1713 | expanded ^= subkey; |
| 1714 | for (int i = 0; i < 8; i++) { |
| 1715 | uint8_t sextet = (expanded >> (42 - i * 6)) & 0x3f; |
| 1716 | uint8_t row = ((sextet & 0x20) >> 4) | (sextet & 0x01); |
| 1717 | uint8_t col = (sextet >> 1) & 0x0f; |
| 1718 | |
| 1719 | out = (out << 4) | octeon_des_sboxes[i][row * 16 + col]; |
| 1720 | } |
| 1721 | |
| 1722 | return octeon_des_permute(out, octeon_des_p, ARRAY_SIZE(octeon_des_p), 32); |
| 1723 | } |
| 1724 | |
| 1725 | static uint64_t octeon_des_block_crypt(uint64_t block, uint64_t key, |
| 1726 | bool encrypt) |
| 1727 | { |
| 1728 | uint64_t subkeys[16]; |
| 1729 | uint64_t permuted = octeon_des_permute(block, octeon_des_ip, |
| 1730 | ARRAY_SIZE(octeon_des_ip), 64); |
| 1731 | uint32_t l = permuted >> 32; |
| 1732 | uint32_t r = permuted; |
| 1733 | |
| 1734 | octeon_des_expand_subkeys(key, subkeys); |
| 1735 | |
| 1736 | for (int i = 0; i < 16; i++) { |
| 1737 | uint32_t next = l ^ octeon_des_f(r, subkeys[encrypt ? i : 15 - i]); |
| 1738 | |
| 1739 | l = r; |
| 1740 | r = next; |
| 1741 | } |
| 1742 | |
| 1743 | return octeon_des_permute(((uint64_t)r << 32) | l, |
| 1744 | octeon_des_fp, ARRAY_SIZE(octeon_des_fp), 64); |
| 1745 | } |
| 1746 | |
| 1747 | static uint64_t octeon_3des_block_crypt(uint64_t block, const uint64_t keys[3], |
| 1748 | bool encrypt) |
| 1749 | { |
| 1750 | if (encrypt) { |
| 1751 | block = octeon_des_block_crypt(block, keys[0], true); |
| 1752 | block = octeon_des_block_crypt(block, keys[1], false); |
| 1753 | block = octeon_des_block_crypt(block, keys[2], true); |
| 1754 | } else { |
| 1755 | block = octeon_des_block_crypt(block, keys[2], false); |
| 1756 | block = octeon_des_block_crypt(block, keys[1], true); |
| 1757 | block = octeon_des_block_crypt(block, keys[0], false); |
| 1758 | } |
| 1759 | return block; |
| 1760 | } |
| 1761 | |
| 1762 | static void octeon_3des_crypt_common(MIPSOcteonCryptoState *crypto, |
| 1763 | uint64_t input_reg, |
| 1764 | bool encrypt, bool cbc) |
| 1765 | { |
| 1766 | const uint64_t keys[3] = { |
| 1767 | crypto->des3_key[0], |
| 1768 | crypto->des3_key[1], |
| 1769 | crypto->des3_key[2], |
| 1770 | }; |
| 1771 | uint64_t block = input_reg; |
| 1772 | |
| 1773 | if (cbc) { |
| 1774 | if (encrypt) { |
| 1775 | block ^= crypto->des3_iv; |
| 1776 | block = octeon_3des_block_crypt(block, keys, true); |
| 1777 | crypto->des3_iv = block; |
| 1778 | } else { |
| 1779 | block = octeon_3des_block_crypt(block, keys, false); |
| 1780 | block ^= crypto->des3_iv; |
| 1781 | crypto->des3_iv = input_reg; |
| 1782 | } |
| 1783 | } else { |
| 1784 | block = octeon_3des_block_crypt(block, keys, encrypt); |
| 1785 | } |
| 1786 | |
| 1787 | crypto->des3_result = block; |
| 1788 | } |
| 1789 | |
| 1790 | static uint16_t octeon_rol16(uint16_t value, unsigned int bits) |
| 1791 | { |
| 1792 | return (value << bits) | (value >> (16 - bits)); |
| 1793 | } |
| 1794 | |
| 1795 | static void octeon_kasumi_key_schedule(const uint64_t key_regs[2], |
| 1796 | OcteonKasumiSubkeys *subkeys) |
| 1797 | { |
| 1798 | uint16_t key[8]; |
| 1799 | uint16_t key_prime[8]; |
| 1800 | |
| 1801 | key[0] = key_regs[0] >> 48; |
| 1802 | key[1] = key_regs[0] >> 32; |
| 1803 | key[2] = key_regs[0] >> 16; |
| 1804 | key[3] = key_regs[0]; |
| 1805 | key[4] = key_regs[1] >> 48; |
| 1806 | key[5] = key_regs[1] >> 32; |
| 1807 | key[6] = key_regs[1] >> 16; |
| 1808 | key[7] = key_regs[1]; |
| 1809 | |
| 1810 | for (int i = 0; i < 8; i++) { |
| 1811 | key_prime[i] = key[i] ^ octeon_kasumi_constants[i]; |
| 1812 | } |
| 1813 | |
| 1814 | for (int i = 0; i < 8; i++) { |
| 1815 | subkeys->kli1[i] = octeon_rol16(key[i], 1); |
| 1816 | subkeys->kli2[i] = key_prime[(i + 2) & 7]; |
| 1817 | subkeys->koi1[i] = octeon_rol16(key[(i + 1) & 7], 5); |
| 1818 | subkeys->koi2[i] = octeon_rol16(key[(i + 5) & 7], 8); |
| 1819 | subkeys->koi3[i] = octeon_rol16(key[(i + 6) & 7], 13); |
| 1820 | subkeys->kii1[i] = key_prime[(i + 4) & 7]; |
| 1821 | subkeys->kii2[i] = key_prime[(i + 3) & 7]; |
| 1822 | subkeys->kii3[i] = key_prime[(i + 7) & 7]; |
| 1823 | } |
| 1824 | } |
| 1825 | |
| 1826 | static uint16_t octeon_kasumi_fi(uint16_t in, uint16_t subkey) |
| 1827 | { |
| 1828 | uint16_t nine = in >> 7; |
| 1829 | uint16_t seven = in & 0x7f; |
| 1830 | |
| 1831 | nine = octeon_kasumi_s9[nine] ^ seven; |
| 1832 | seven = octeon_kasumi_s7[seven] ^ (nine & 0x7f); |
| 1833 | seven ^= subkey >> 9; |
| 1834 | nine ^= subkey & 0x1ff; |
| 1835 | nine = octeon_kasumi_s9[nine] ^ seven; |
| 1836 | seven = octeon_kasumi_s7[seven] ^ (nine & 0x7f); |
| 1837 | return (seven << 9) | nine; |
| 1838 | } |
| 1839 | |
| 1840 | static uint32_t octeon_kasumi_fo(uint32_t in, int index, |
| 1841 | const OcteonKasumiSubkeys *subkeys) |
| 1842 | { |
| 1843 | uint16_t left = in >> 16; |
| 1844 | uint16_t right = in; |
| 1845 | |
| 1846 | left ^= subkeys->koi1[index]; |
| 1847 | left = octeon_kasumi_fi(left, subkeys->kii1[index]); |
| 1848 | left ^= right; |
| 1849 | right ^= subkeys->koi2[index]; |
| 1850 | right = octeon_kasumi_fi(right, subkeys->kii2[index]); |
| 1851 | right ^= left; |
| 1852 | left ^= subkeys->koi3[index]; |
| 1853 | left = octeon_kasumi_fi(left, subkeys->kii3[index]); |
| 1854 | left ^= right; |
| 1855 | |
| 1856 | return ((uint32_t)right << 16) | left; |
| 1857 | } |
| 1858 | |
| 1859 | static uint32_t octeon_kasumi_fl(uint32_t in, int index, |
| 1860 | const OcteonKasumiSubkeys *subkeys) |
| 1861 | { |
| 1862 | uint16_t left = in >> 16; |
| 1863 | uint16_t right = in; |
| 1864 | uint16_t a = left & subkeys->kli1[index]; |
| 1865 | uint16_t b; |
| 1866 | |
| 1867 | right ^= octeon_rol16(a, 1); |
| 1868 | b = right | subkeys->kli2[index]; |
| 1869 | left ^= octeon_rol16(b, 1); |
| 1870 | return ((uint32_t)left << 16) | right; |
| 1871 | } |
| 1872 | |
| 1873 | static uint64_t octeon_kasumi_block_encrypt(uint64_t block, |
| 1874 | const uint64_t key_regs[2]) |
| 1875 | { |
| 1876 | OcteonKasumiSubkeys subkeys; |
| 1877 | uint32_t left = block >> 32; |
| 1878 | uint32_t right = block; |
| 1879 | |
| 1880 | octeon_kasumi_key_schedule(key_regs, &subkeys); |
| 1881 | |
| 1882 | for (int i = 0; i < 8; ) { |
| 1883 | uint32_t temp = octeon_kasumi_fl(left, i, &subkeys); |
| 1884 | |
| 1885 | temp = octeon_kasumi_fo(temp, i++, &subkeys); |
| 1886 | right ^= temp; |
| 1887 | temp = octeon_kasumi_fo(right, i, &subkeys); |
| 1888 | temp = octeon_kasumi_fl(temp, i++, &subkeys); |
| 1889 | left ^= temp; |
| 1890 | } |
| 1891 | |
| 1892 | return ((uint64_t)left << 32) | right; |
| 1893 | } |
| 1894 | |
| 1895 | static void octeon_kasumi_crypt_common(MIPSOcteonCryptoState *crypto, |
| 1896 | uint64_t input_reg, bool cbc) |
| 1897 | { |
| 1898 | const uint64_t key_regs[2] = { |
| 1899 | crypto->des3_key[0], |
| 1900 | crypto->des3_key[1], |
| 1901 | }; |
| 1902 | uint64_t block = input_reg; |
| 1903 | |
| 1904 | if (cbc) { |
| 1905 | block ^= crypto->des3_iv; |
| 1906 | } |
| 1907 | |
| 1908 | block = octeon_kasumi_block_encrypt(block, key_regs); |
| 1909 | if (cbc) { |
| 1910 | crypto->des3_iv = block; |
| 1911 | } |
| 1912 | crypto->des3_result = block; |
| 1913 | } |
| 1914 | |
| 1915 | void helper_octeon_cp2_mt_des3_enc_cbc(CPUMIPSState *env, uint64_t value) |
| 1916 | { |
| 1917 | octeon_3des_crypt_common(&env->octeon_crypto, value, true, true); |
| 1918 | } |
| 1919 | |
| 1920 | void helper_octeon_cp2_mt_kas_enc_cbc(CPUMIPSState *env, uint64_t value) |
| 1921 | { |
| 1922 | octeon_kasumi_crypt_common(&env->octeon_crypto, value, true); |
| 1923 | } |
| 1924 | |
| 1925 | void helper_octeon_cp2_mt_des3_enc(CPUMIPSState *env, uint64_t value) |
| 1926 | { |
| 1927 | octeon_3des_crypt_common(&env->octeon_crypto, value, true, false); |
| 1928 | } |
| 1929 | |
| 1930 | void helper_octeon_cp2_mt_kas_enc(CPUMIPSState *env, uint64_t value) |
| 1931 | { |
| 1932 | octeon_kasumi_crypt_common(&env->octeon_crypto, value, false); |
| 1933 | } |
| 1934 | |
| 1935 | void helper_octeon_cp2_mt_des3_dec_cbc(CPUMIPSState *env, uint64_t value) |
| 1936 | { |
| 1937 | octeon_3des_crypt_common(&env->octeon_crypto, value, false, true); |
| 1938 | } |
| 1939 | |
| 1940 | void helper_octeon_cp2_mt_des3_dec(CPUMIPSState *env, uint64_t value) |
| 1941 | { |
| 1942 | octeon_3des_crypt_common(&env->octeon_crypto, value, false, false); |
| 1943 | } |
| 1944 | |
| 1945 | static const uint8_t camellia_sbox1[256] = { |
| 1946 | 112, 130, 44, 236, 179, 39, 192, 229, 228, 133, 87, 53, 234, 12, |
| 1947 | 174, 65, 35, 239, 107, 147, 69, 25, 165, 33, 237, 14, 79, 78, |
| 1948 | 29, 101, 146, 189, 134, 184, 175, 143, 124, 235, 31, 206, 62, 48, |
| 1949 | 220, 95, 94, 197, 11, 26, 166, 225, 57, 202, 213, 71, 93, 61, |
| 1950 | 217, 1, 90, 214, 81, 86, 108, 77, 139, 13, 154, 102, 251, 204, |
| 1951 | 176, 45, 116, 18, 43, 32, 240, 177, 132, 153, 223, 76, 203, 194, |
| 1952 | 52, 126, 118, 5, 109, 183, 169, 49, 209, 23, 4, 215, 20, 88, |
| 1953 | 58, 97, 222, 27, 17, 28, 50, 15, 156, 22, 83, 24, 242, 34, |
| 1954 | 254, 68, 207, 178, 195, 181, 122, 145, 36, 8, 232, 168, 96, 252, |
| 1955 | 105, 80, 170, 208, 160, 125, 161, 137, 98, 151, 84, 91, 30, 149, |
| 1956 | 224, 255, 100, 210, 16, 196, 0, 72, 163, 247, 117, 219, 138, 3, |
| 1957 | 230, 218, 9, 63, 221, 148, 135, 92, 131, 2, 205, 74, 144, 51, |
| 1958 | 115, 103, 246, 243, 157, 127, 191, 226, 82, 155, 216, 38, 200, 55, |
| 1959 | 198, 59, 129, 150, 111, 75, 19, 190, 99, 46, 233, 121, 167, 140, |
| 1960 | 159, 110, 188, 142, 41, 245, 249, 182, 47, 253, 180, 89, 120, 152, |
| 1961 | 6, 106, 231, 70, 113, 186, 212, 37, 171, 66, 136, 162, 141, 250, |
| 1962 | 114, 7, 185, 85, 248, 238, 172, 10, 54, 73, 42, 104, 60, 56, |
| 1963 | 241, 164, 64, 40, 211, 123, 187, 201, 67, 193, 21, 227, 173, 244, |
| 1964 | 119, 199, 128, 158, |
| 1965 | }; |
| 1966 | |
| 1967 | static uint8_t camellia_rotl8(uint8_t v, unsigned int shift) |
| 1968 | { |
| 1969 | return (v << shift) | (v >> (8 - shift)); |
| 1970 | } |
| 1971 | |
| 1972 | static uint8_t camellia_sbox2(uint8_t x) |
| 1973 | { |
| 1974 | return camellia_rotl8(camellia_sbox1[x], 1); |
| 1975 | } |
| 1976 | |
| 1977 | static uint8_t camellia_sbox3(uint8_t x) |
| 1978 | { |
| 1979 | return camellia_rotl8(camellia_sbox1[x], 7); |
| 1980 | } |
| 1981 | |
| 1982 | static uint8_t camellia_sbox4(uint8_t x) |
| 1983 | { |
| 1984 | return camellia_sbox1[camellia_rotl8(x, 1)]; |
| 1985 | } |
| 1986 | |
| 1987 | static uint64_t camellia_f(uint64_t input, uint64_t key) |
| 1988 | { |
| 1989 | uint64_t x = input ^ key; |
| 1990 | uint8_t t1 = camellia_sbox1[x >> 56]; |
| 1991 | uint8_t t2 = camellia_sbox2((x >> 48) & 0xff); |
| 1992 | uint8_t t3 = camellia_sbox3((x >> 40) & 0xff); |
| 1993 | uint8_t t4 = camellia_sbox4((x >> 32) & 0xff); |
| 1994 | uint8_t t5 = camellia_sbox2((x >> 24) & 0xff); |
| 1995 | uint8_t t6 = camellia_sbox3((x >> 16) & 0xff); |
| 1996 | uint8_t t7 = camellia_sbox4((x >> 8) & 0xff); |
| 1997 | uint8_t t8 = camellia_sbox1[x & 0xff]; |
| 1998 | uint8_t y1 = t1 ^ t3 ^ t4 ^ t6 ^ t7 ^ t8; |
| 1999 | uint8_t y2 = t1 ^ t2 ^ t4 ^ t5 ^ t7 ^ t8; |
| 2000 | uint8_t y3 = t1 ^ t2 ^ t3 ^ t5 ^ t6 ^ t8; |
| 2001 | uint8_t y4 = t2 ^ t3 ^ t4 ^ t5 ^ t6 ^ t7; |
| 2002 | uint8_t y5 = t1 ^ t2 ^ t6 ^ t7 ^ t8; |
| 2003 | uint8_t y6 = t2 ^ t3 ^ t5 ^ t7 ^ t8; |
| 2004 | uint8_t y7 = t3 ^ t4 ^ t5 ^ t6 ^ t8; |
| 2005 | uint8_t y8 = t1 ^ t4 ^ t5 ^ t6 ^ t7; |
| 2006 | |
| 2007 | return ((uint64_t)y1 << 56) | ((uint64_t)y2 << 48) | |
| 2008 | ((uint64_t)y3 << 40) | ((uint64_t)y4 << 32) | |
| 2009 | ((uint64_t)y5 << 24) | ((uint64_t)y6 << 16) | |
| 2010 | ((uint64_t)y7 << 8) | y8; |
| 2011 | } |
| 2012 | |
| 2013 | static uint64_t camellia_fl(uint64_t input, uint64_t key) |
| 2014 | { |
| 2015 | uint32_t x1 = input >> 32; |
| 2016 | uint32_t x2 = input; |
| 2017 | uint32_t k1 = key >> 32; |
| 2018 | uint32_t k2 = key; |
| 2019 | |
| 2020 | x2 ^= rol32(x1 & k1, 1); |
| 2021 | x1 ^= x2 | k2; |
| 2022 | return ((uint64_t)x1 << 32) | x2; |
| 2023 | } |
| 2024 | |
| 2025 | static uint64_t camellia_flinv(uint64_t input, uint64_t key) |
| 2026 | { |
| 2027 | uint32_t y1 = input >> 32; |
| 2028 | uint32_t y2 = input; |
| 2029 | uint32_t k1 = key >> 32; |
| 2030 | uint32_t k2 = key; |
| 2031 | |
| 2032 | y1 ^= y2 | k2; |
| 2033 | y2 ^= rol32(y1 & k1, 1); |
| 2034 | return ((uint64_t)y1 << 32) | y2; |
| 2035 | } |
| 2036 | |
| 2037 | static void octeon_camellia_round(MIPSOcteonCryptoState *crypto, uint64_t key) |
| 2038 | { |
| 2039 | uint64_t left = crypto->aes_resinp[0]; |
| 2040 | uint64_t right = crypto->aes_resinp[1]; |
| 2041 | |
| 2042 | crypto->aes_resinp[0] = right ^ camellia_f(left, key); |
| 2043 | crypto->aes_resinp[1] = left; |
| 2044 | } |
| 2045 | |
| 2046 | static void octeon_camellia_fl_layer(MIPSOcteonCryptoState *crypto, |
| 2047 | uint64_t key, bool inverse) |
| 2048 | { |
| 2049 | uint64_t state = crypto->aes_resinp[inverse ? 1 : 0]; |
| 2050 | |
| 2051 | crypto->aes_resinp[inverse ? 1 : 0] = inverse ? |
| 2052 | camellia_flinv(state, key) : |
| 2053 | camellia_fl(state, key); |
| 2054 | } |
| 2055 | |
| 2056 | void helper_octeon_cp2_mt_camellia_fl(CPUMIPSState *env, uint64_t value) |
| 2057 | { |
| 2058 | octeon_camellia_fl_layer(&env->octeon_crypto, value, false); |
| 2059 | } |
| 2060 | |
| 2061 | void helper_octeon_cp2_mt_camellia_flinv(CPUMIPSState *env, uint64_t value) |
| 2062 | { |
| 2063 | octeon_camellia_fl_layer(&env->octeon_crypto, value, true); |
| 2064 | } |
| 2065 | |
| 2066 | void helper_octeon_cp2_mt_camellia_round(CPUMIPSState *env, uint64_t value) |
| 2067 | { |
| 2068 | octeon_camellia_round(&env->octeon_crypto, value); |
| 2069 | } |
| 2070 | |
| 2071 | void helper_octeon_cp2_mt_snow3g_start(CPUMIPSState *env, uint64_t value) |
| 2072 | { |
| 2073 | octeon_snow3g_start(&env->octeon_crypto, value); |
| 2074 | } |
| 2075 | |
| 2076 | void helper_octeon_cp2_mt_snow3g_more(CPUMIPSState *env, uint64_t value) |
| 2077 | { |
| 2078 | (void)value; |
| 2079 | octeon_snow3g_more(&env->octeon_crypto); |
| 2080 | } |
| 2081 | |
| 2082 | void helper_octeon_cp2_mt_zuc_start(CPUMIPSState *env, uint64_t value) |
| 2083 | { |
| 2084 | octeon_zuc_start(&env->octeon_crypto, value); |
| 2085 | } |
| 2086 | |
| 2087 | void helper_octeon_cp2_mt_zuc_more(CPUMIPSState *env, uint64_t value) |
| 2088 | { |
| 2089 | octeon_zuc_more(&env->octeon_crypto, value); |
| 2090 | } |
| 2091 | |
| 2092 | void helper_octeon_cp2_mt_hsh_startsha1_compat(CPUMIPSState *env, |
| 2093 | uint64_t value) |
| 2094 | { |
| 2095 | octeon_hsh_set_pair(env->octeon_crypto.hsh_dat, 7, value); |
| 2096 | octeon_sha1_transform(&env->octeon_crypto); |
| 2097 | } |
| 2098 | |
| 2099 | void helper_octeon_cp2_mt_hsh_startmd5(CPUMIPSState *env, uint64_t value) |
| 2100 | { |
| 2101 | octeon_hsh_set_pair(env->octeon_crypto.hsh_dat, 7, value); |
| 2102 | octeon_md5_transform(&env->octeon_crypto); |
| 2103 | } |
| 2104 | |
| 2105 | void helper_octeon_cp2_mt_hsh_startsha256(CPUMIPSState *env, uint64_t value) |
| 2106 | { |
| 2107 | octeon_hsh_set_pair(env->octeon_crypto.hsh_dat, 7, value); |
| 2108 | octeon_sha256_transform(&env->octeon_crypto); |
| 2109 | } |
| 2110 | |
| 2111 | void helper_octeon_cp2_mt_hsh_startsha(CPUMIPSState *env, uint64_t value) |
| 2112 | { |
| 2113 | octeon_hsh_set_pair(env->octeon_crypto.hsh_dat, 7, value); |
| 2114 | octeon_sha1_transform(&env->octeon_crypto); |
| 2115 | } |
| 2116 | |
| 2117 | void helper_octeon_cp2_mt_hsh_startsha512(CPUMIPSState *env, uint64_t value) |
| 2118 | { |
| 2119 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 2120 | |
| 2121 | crypto->hsh_dat[15] = value; |
| 2122 | octeon_sha512_transform(crypto); |
| 2123 | } |
| 2124 | |
| 2125 | uint64_t helper_octeon_cp2_mf_crc_iv_reflect(CPUMIPSState *env) |
| 2126 | { |
| 2127 | return octeon_crc_reflect32_by_byte(env->octeon_crypto.crc_iv); |
| 2128 | } |
| 2129 | |
| 2130 | static void octeon_gfm_xormul1_common(MIPSOcteonCryptoState *crypto, |
| 2131 | uint64_t value) |
| 2132 | { |
| 2133 | crypto->gfm_resinp[1] ^= value; |
| 2134 | if (crypto->gfm_poly <= 0xff && crypto->gfm_mul[1] == 0 && |
| 2135 | crypto->gfm_resinp[0] == 0) { |
| 2136 | octeon_gfm_mul64_uia2(crypto->gfm_resinp, crypto->gfm_mul, |
| 2137 | crypto->gfm_poly, crypto->gfm_resinp); |
| 2138 | } else { |
| 2139 | octeon_gfm_mul(crypto->gfm_resinp, crypto->gfm_mul, crypto->gfm_poly, |
| 2140 | crypto->gfm_resinp); |
| 2141 | } |
| 2142 | } |
| 2143 | |
| 2144 | void helper_octeon_cp2_mt_gfm_xormul1_reflect(CPUMIPSState *env, |
| 2145 | uint64_t value) |
| 2146 | { |
| 2147 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 2148 | |
| 2149 | octeon_gfm_xormul1_common(crypto, revbit64(value)); |
| 2150 | } |
| 2151 | |
| 2152 | void helper_octeon_cp2_mt_gfm_xormul1(CPUMIPSState *env, uint64_t value) |
| 2153 | { |
| 2154 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 2155 | |
| 2156 | octeon_gfm_xormul1_common(crypto, value); |
| 2157 | } |
| 2158 | |
| 2159 | void helper_octeon_cp2_mt_sha3_startop(CPUMIPSState *env) |
| 2160 | { |
| 2161 | octeon_sha3_permute(&env->octeon_crypto); |
| 2162 | } |
| 2163 | |
| 2164 | void helper_octeon_cp2_mt_crc_write_iv_reflect(CPUMIPSState *env, |
| 2165 | uint64_t value) |
| 2166 | { |
| 2167 | env->octeon_crypto.crc_iv = |
| 2168 | octeon_crc_reflect32_by_byte((uint32_t)value); |
| 2169 | } |
| 2170 | |
| 2171 | void helper_octeon_cp2_mt_crc_write_polynomial_reflect(CPUMIPSState *env, |
| 2172 | uint64_t value) |
| 2173 | { |
| 2174 | env->octeon_crypto.crc_poly = |
| 2175 | octeon_crc_reflect32_by_byte((uint32_t)value); |
| 2176 | } |
| 2177 | |
| 2178 | void helper_octeon_cp2_mt_crc_write_byte(CPUMIPSState *env, uint64_t value) |
| 2179 | { |
| 2180 | octeon_crc_update_normal(&env->octeon_crypto, value, 1); |
| 2181 | } |
| 2182 | |
| 2183 | void helper_octeon_cp2_mt_crc_write_half(CPUMIPSState *env, uint64_t value) |
| 2184 | { |
| 2185 | octeon_crc_update_normal(&env->octeon_crypto, value, 2); |
| 2186 | } |
| 2187 | |
| 2188 | void helper_octeon_cp2_mt_crc_write_word(CPUMIPSState *env, uint64_t value) |
| 2189 | { |
| 2190 | octeon_crc_update_normal(&env->octeon_crypto, value, 4); |
| 2191 | } |
| 2192 | |
| 2193 | void helper_octeon_cp2_mt_crc_write_dword(CPUMIPSState *env, uint64_t value) |
| 2194 | { |
| 2195 | octeon_crc_update_normal(&env->octeon_crypto, value, 8); |
| 2196 | } |
| 2197 | |
| 2198 | void helper_octeon_cp2_mt_crc_write_var(CPUMIPSState *env, uint64_t value) |
| 2199 | { |
| 2200 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 2201 | |
| 2202 | octeon_crc_update_normal(crypto, value, MIN(8U, crypto->crc_len & 0xf)); |
| 2203 | } |
| 2204 | |
| 2205 | void helper_octeon_cp2_mt_crc_write_byte_reflect(CPUMIPSState *env, |
| 2206 | uint64_t value) |
| 2207 | { |
| 2208 | octeon_crc_update_reflect(&env->octeon_crypto, value, 1); |
| 2209 | } |
| 2210 | |
| 2211 | void helper_octeon_cp2_mt_crc_write_half_reflect(CPUMIPSState *env, |
| 2212 | uint64_t value) |
| 2213 | { |
| 2214 | octeon_crc_update_reflect(&env->octeon_crypto, value, 2); |
| 2215 | } |
| 2216 | |
| 2217 | void helper_octeon_cp2_mt_crc_write_word_reflect(CPUMIPSState *env, |
| 2218 | uint64_t value) |
| 2219 | { |
| 2220 | octeon_crc_update_reflect(&env->octeon_crypto, value, 4); |
| 2221 | } |
| 2222 | |
| 2223 | void helper_octeon_cp2_mt_crc_write_dword_reflect(CPUMIPSState *env, |
| 2224 | uint64_t value) |
| 2225 | { |
| 2226 | octeon_crc_update_reflect(&env->octeon_crypto, value, 8); |
| 2227 | } |
| 2228 | |
| 2229 | void helper_octeon_cp2_mt_crc_write_var_reflect(CPUMIPSState *env, |
| 2230 | uint64_t value) |
| 2231 | { |
| 2232 | MIPSOcteonCryptoState *crypto = &env->octeon_crypto; |
| 2233 | |
| 2234 | octeon_crc_update_reflect(crypto, value, MIN(8U, crypto->crc_len & 0xf)); |
| 2235 | } |
| 2236 | |
| 2237 | void helper_octeon_cp2_mt_llm_read_addr0(CPUMIPSState *env, uint64_t value) |
| 2238 | { |
| 2239 | octeon_llm_read(&env->octeon_crypto, 0, value, false); |
| 2240 | } |
| 2241 | |
| 2242 | void helper_octeon_cp2_mt_llm_write_addr0(CPUMIPSState *env, uint64_t value) |
| 2243 | { |
| 2244 | octeon_llm_write(&env->octeon_crypto, 0, value, false); |
| 2245 | } |
| 2246 | |
| 2247 | void helper_octeon_cp2_mt_llm_read64_addr0(CPUMIPSState *env, uint64_t value) |
| 2248 | { |
| 2249 | octeon_llm_read(&env->octeon_crypto, 0, value, true); |
| 2250 | } |
| 2251 | |
| 2252 | void helper_octeon_cp2_mt_llm_write64_addr0(CPUMIPSState *env, uint64_t value) |
| 2253 | { |
| 2254 | octeon_llm_write(&env->octeon_crypto, 0, value, true); |
| 2255 | } |
| 2256 | |
| 2257 | void helper_octeon_cp2_mt_llm_read_addr1(CPUMIPSState *env, uint64_t value) |
| 2258 | { |
| 2259 | octeon_llm_read(&env->octeon_crypto, 1, value, false); |
| 2260 | } |
| 2261 | |
| 2262 | void helper_octeon_cp2_mt_llm_write_addr1(CPUMIPSState *env, uint64_t value) |
| 2263 | { |
| 2264 | octeon_llm_write(&env->octeon_crypto, 1, value, false); |
| 2265 | } |
| 2266 | |
| 2267 | void helper_octeon_cp2_mt_llm_read64_addr1(CPUMIPSState *env, uint64_t value) |
| 2268 | { |
| 2269 | octeon_llm_read(&env->octeon_crypto, 1, value, true); |
| 2270 | } |
| 2271 | |
| 2272 | void helper_octeon_cp2_mt_llm_write64_addr1(CPUMIPSState *env, uint64_t value) |
| 2273 | { |
| 2274 | octeon_llm_write(&env->octeon_crypto, 1, value, true); |
| 2275 | } |