1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Scalar fixed time AES core transform 4 * 5 * Copyright (C) 2017 Linaro Ltd <ard.biesheuvel@linaro.org> 6 */ 7 8 #include <crypto/aes.h> 9 #include <linux/crypto.h> 10 #include <linux/module.h> 11 #include <asm/unaligned.h> 12 13 /* 14 * Emit the sbox as volatile const to prevent the compiler from doing 15 * constant folding on sbox references involving fixed indexes. 16 */ 17 static volatile const u8 __cacheline_aligned __aesti_sbox[] = { 18 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 19 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76, 20 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 21 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0, 22 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 23 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15, 24 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 25 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75, 26 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 27 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84, 28 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 29 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf, 30 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 31 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8, 32 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 33 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2, 34 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 35 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73, 36 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 37 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb, 38 0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 39 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79, 40 0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 41 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08, 42 0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 43 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a, 44 0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 45 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e, 46 0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 47 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf, 48 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 49 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16, 50 }; 51 52 static volatile const u8 __cacheline_aligned __aesti_inv_sbox[] = { 53 0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 54 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb, 55 0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 56 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb, 57 0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 58 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e, 59 0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2, 60 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25, 61 0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16, 62 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92, 63 0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda, 64 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84, 65 0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a, 66 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06, 67 0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02, 68 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b, 69 0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea, 70 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73, 71 0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85, 72 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e, 73 0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89, 74 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b, 75 0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20, 76 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4, 77 0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31, 78 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f, 79 0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d, 80 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef, 81 0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0, 82 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61, 83 0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 84 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d, 85 }; 86 87 static u32 mul_by_x(u32 w) 88 { 89 u32 x = w & 0x7f7f7f7f; 90 u32 y = w & 0x80808080; 91 92 /* multiply by polynomial 'x' (0b10) in GF(2^8) */ 93 return (x << 1) ^ (y >> 7) * 0x1b; 94 } 95 96 static u32 mul_by_x2(u32 w) 97 { 98 u32 x = w & 0x3f3f3f3f; 99 u32 y = w & 0x80808080; 100 u32 z = w & 0x40404040; 101 102 /* multiply by polynomial 'x^2' (0b100) in GF(2^8) */ 103 return (x << 2) ^ (y >> 7) * 0x36 ^ (z >> 6) * 0x1b; 104 } 105 106 static u32 mix_columns(u32 x) 107 { 108 /* 109 * Perform the following matrix multiplication in GF(2^8) 110 * 111 * | 0x2 0x3 0x1 0x1 | | x[0] | 112 * | 0x1 0x2 0x3 0x1 | | x[1] | 113 * | 0x1 0x1 0x2 0x3 | x | x[2] | 114 * | 0x3 0x1 0x1 0x2 | | x[3] | 115 */ 116 u32 y = mul_by_x(x) ^ ror32(x, 16); 117 118 return y ^ ror32(x ^ y, 8); 119 } 120 121 static u32 inv_mix_columns(u32 x) 122 { 123 /* 124 * Perform the following matrix multiplication in GF(2^8) 125 * 126 * | 0xe 0xb 0xd 0x9 | | x[0] | 127 * | 0x9 0xe 0xb 0xd | | x[1] | 128 * | 0xd 0x9 0xe 0xb | x | x[2] | 129 * | 0xb 0xd 0x9 0xe | | x[3] | 130 * 131 * which can conveniently be reduced to 132 * 133 * | 0x2 0x3 0x1 0x1 | | 0x5 0x0 0x4 0x0 | | x[0] | 134 * | 0x1 0x2 0x3 0x1 | | 0x0 0x5 0x0 0x4 | | x[1] | 135 * | 0x1 0x1 0x2 0x3 | x | 0x4 0x0 0x5 0x0 | x | x[2] | 136 * | 0x3 0x1 0x1 0x2 | | 0x0 0x4 0x0 0x5 | | x[3] | 137 */ 138 u32 y = mul_by_x2(x); 139 140 return mix_columns(x ^ y ^ ror32(y, 16)); 141 } 142 143 static __always_inline u32 subshift(u32 in[], int pos) 144 { 145 return (__aesti_sbox[in[pos] & 0xff]) ^ 146 (__aesti_sbox[(in[(pos + 1) % 4] >> 8) & 0xff] << 8) ^ 147 (__aesti_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^ 148 (__aesti_sbox[(in[(pos + 3) % 4] >> 24) & 0xff] << 24); 149 } 150 151 static __always_inline u32 inv_subshift(u32 in[], int pos) 152 { 153 return (__aesti_inv_sbox[in[pos] & 0xff]) ^ 154 (__aesti_inv_sbox[(in[(pos + 3) % 4] >> 8) & 0xff] << 8) ^ 155 (__aesti_inv_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^ 156 (__aesti_inv_sbox[(in[(pos + 1) % 4] >> 24) & 0xff] << 24); 157 } 158 159 static u32 subw(u32 in) 160 { 161 return (__aesti_sbox[in & 0xff]) ^ 162 (__aesti_sbox[(in >> 8) & 0xff] << 8) ^ 163 (__aesti_sbox[(in >> 16) & 0xff] << 16) ^ 164 (__aesti_sbox[(in >> 24) & 0xff] << 24); 165 } 166 167 static int aesti_expand_key(struct crypto_aes_ctx *ctx, const u8 *in_key, 168 unsigned int key_len) 169 { 170 u32 kwords = key_len / sizeof(u32); 171 u32 rc, i, j; 172 173 if (key_len != AES_KEYSIZE_128 && 174 key_len != AES_KEYSIZE_192 && 175 key_len != AES_KEYSIZE_256) 176 return -EINVAL; 177 178 ctx->key_length = key_len; 179 180 for (i = 0; i < kwords; i++) 181 ctx->key_enc[i] = get_unaligned_le32(in_key + i * sizeof(u32)); 182 183 for (i = 0, rc = 1; i < 10; i++, rc = mul_by_x(rc)) { 184 u32 *rki = ctx->key_enc + (i * kwords); 185 u32 *rko = rki + kwords; 186 187 rko[0] = ror32(subw(rki[kwords - 1]), 8) ^ rc ^ rki[0]; 188 rko[1] = rko[0] ^ rki[1]; 189 rko[2] = rko[1] ^ rki[2]; 190 rko[3] = rko[2] ^ rki[3]; 191 192 if (key_len == 24) { 193 if (i >= 7) 194 break; 195 rko[4] = rko[3] ^ rki[4]; 196 rko[5] = rko[4] ^ rki[5]; 197 } else if (key_len == 32) { 198 if (i >= 6) 199 break; 200 rko[4] = subw(rko[3]) ^ rki[4]; 201 rko[5] = rko[4] ^ rki[5]; 202 rko[6] = rko[5] ^ rki[6]; 203 rko[7] = rko[6] ^ rki[7]; 204 } 205 } 206 207 /* 208 * Generate the decryption keys for the Equivalent Inverse Cipher. 209 * This involves reversing the order of the round keys, and applying 210 * the Inverse Mix Columns transformation to all but the first and 211 * the last one. 212 */ 213 ctx->key_dec[0] = ctx->key_enc[key_len + 24]; 214 ctx->key_dec[1] = ctx->key_enc[key_len + 25]; 215 ctx->key_dec[2] = ctx->key_enc[key_len + 26]; 216 ctx->key_dec[3] = ctx->key_enc[key_len + 27]; 217 218 for (i = 4, j = key_len + 20; j > 0; i += 4, j -= 4) { 219 ctx->key_dec[i] = inv_mix_columns(ctx->key_enc[j]); 220 ctx->key_dec[i + 1] = inv_mix_columns(ctx->key_enc[j + 1]); 221 ctx->key_dec[i + 2] = inv_mix_columns(ctx->key_enc[j + 2]); 222 ctx->key_dec[i + 3] = inv_mix_columns(ctx->key_enc[j + 3]); 223 } 224 225 ctx->key_dec[i] = ctx->key_enc[0]; 226 ctx->key_dec[i + 1] = ctx->key_enc[1]; 227 ctx->key_dec[i + 2] = ctx->key_enc[2]; 228 ctx->key_dec[i + 3] = ctx->key_enc[3]; 229 230 return 0; 231 } 232 233 static int aesti_set_key(struct crypto_tfm *tfm, const u8 *in_key, 234 unsigned int key_len) 235 { 236 struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm); 237 int err; 238 239 err = aesti_expand_key(ctx, in_key, key_len); 240 if (err) 241 return err; 242 243 /* 244 * In order to force the compiler to emit data independent Sbox lookups 245 * at the start of each block, xor the first round key with values at 246 * fixed indexes in the Sbox. This will need to be repeated each time 247 * the key is used, which will pull the entire Sbox into the D-cache 248 * before any data dependent Sbox lookups are performed. 249 */ 250 ctx->key_enc[0] ^= __aesti_sbox[ 0] ^ __aesti_sbox[128]; 251 ctx->key_enc[1] ^= __aesti_sbox[32] ^ __aesti_sbox[160]; 252 ctx->key_enc[2] ^= __aesti_sbox[64] ^ __aesti_sbox[192]; 253 ctx->key_enc[3] ^= __aesti_sbox[96] ^ __aesti_sbox[224]; 254 255 ctx->key_dec[0] ^= __aesti_inv_sbox[ 0] ^ __aesti_inv_sbox[128]; 256 ctx->key_dec[1] ^= __aesti_inv_sbox[32] ^ __aesti_inv_sbox[160]; 257 ctx->key_dec[2] ^= __aesti_inv_sbox[64] ^ __aesti_inv_sbox[192]; 258 ctx->key_dec[3] ^= __aesti_inv_sbox[96] ^ __aesti_inv_sbox[224]; 259 260 return 0; 261 } 262 263 static void aesti_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 264 { 265 const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm); 266 const u32 *rkp = ctx->key_enc + 4; 267 int rounds = 6 + ctx->key_length / 4; 268 u32 st0[4], st1[4]; 269 unsigned long flags; 270 int round; 271 272 st0[0] = ctx->key_enc[0] ^ get_unaligned_le32(in); 273 st0[1] = ctx->key_enc[1] ^ get_unaligned_le32(in + 4); 274 st0[2] = ctx->key_enc[2] ^ get_unaligned_le32(in + 8); 275 st0[3] = ctx->key_enc[3] ^ get_unaligned_le32(in + 12); 276 277 /* 278 * Temporarily disable interrupts to avoid races where cachelines are 279 * evicted when the CPU is interrupted to do something else. 280 */ 281 local_irq_save(flags); 282 283 st0[0] ^= __aesti_sbox[ 0] ^ __aesti_sbox[128]; 284 st0[1] ^= __aesti_sbox[32] ^ __aesti_sbox[160]; 285 st0[2] ^= __aesti_sbox[64] ^ __aesti_sbox[192]; 286 st0[3] ^= __aesti_sbox[96] ^ __aesti_sbox[224]; 287 288 for (round = 0;; round += 2, rkp += 8) { 289 st1[0] = mix_columns(subshift(st0, 0)) ^ rkp[0]; 290 st1[1] = mix_columns(subshift(st0, 1)) ^ rkp[1]; 291 st1[2] = mix_columns(subshift(st0, 2)) ^ rkp[2]; 292 st1[3] = mix_columns(subshift(st0, 3)) ^ rkp[3]; 293 294 if (round == rounds - 2) 295 break; 296 297 st0[0] = mix_columns(subshift(st1, 0)) ^ rkp[4]; 298 st0[1] = mix_columns(subshift(st1, 1)) ^ rkp[5]; 299 st0[2] = mix_columns(subshift(st1, 2)) ^ rkp[6]; 300 st0[3] = mix_columns(subshift(st1, 3)) ^ rkp[7]; 301 } 302 303 put_unaligned_le32(subshift(st1, 0) ^ rkp[4], out); 304 put_unaligned_le32(subshift(st1, 1) ^ rkp[5], out + 4); 305 put_unaligned_le32(subshift(st1, 2) ^ rkp[6], out + 8); 306 put_unaligned_le32(subshift(st1, 3) ^ rkp[7], out + 12); 307 308 local_irq_restore(flags); 309 } 310 311 static void aesti_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 312 { 313 const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm); 314 const u32 *rkp = ctx->key_dec + 4; 315 int rounds = 6 + ctx->key_length / 4; 316 u32 st0[4], st1[4]; 317 unsigned long flags; 318 int round; 319 320 st0[0] = ctx->key_dec[0] ^ get_unaligned_le32(in); 321 st0[1] = ctx->key_dec[1] ^ get_unaligned_le32(in + 4); 322 st0[2] = ctx->key_dec[2] ^ get_unaligned_le32(in + 8); 323 st0[3] = ctx->key_dec[3] ^ get_unaligned_le32(in + 12); 324 325 /* 326 * Temporarily disable interrupts to avoid races where cachelines are 327 * evicted when the CPU is interrupted to do something else. 328 */ 329 local_irq_save(flags); 330 331 st0[0] ^= __aesti_inv_sbox[ 0] ^ __aesti_inv_sbox[128]; 332 st0[1] ^= __aesti_inv_sbox[32] ^ __aesti_inv_sbox[160]; 333 st0[2] ^= __aesti_inv_sbox[64] ^ __aesti_inv_sbox[192]; 334 st0[3] ^= __aesti_inv_sbox[96] ^ __aesti_inv_sbox[224]; 335 336 for (round = 0;; round += 2, rkp += 8) { 337 st1[0] = inv_mix_columns(inv_subshift(st0, 0)) ^ rkp[0]; 338 st1[1] = inv_mix_columns(inv_subshift(st0, 1)) ^ rkp[1]; 339 st1[2] = inv_mix_columns(inv_subshift(st0, 2)) ^ rkp[2]; 340 st1[3] = inv_mix_columns(inv_subshift(st0, 3)) ^ rkp[3]; 341 342 if (round == rounds - 2) 343 break; 344 345 st0[0] = inv_mix_columns(inv_subshift(st1, 0)) ^ rkp[4]; 346 st0[1] = inv_mix_columns(inv_subshift(st1, 1)) ^ rkp[5]; 347 st0[2] = inv_mix_columns(inv_subshift(st1, 2)) ^ rkp[6]; 348 st0[3] = inv_mix_columns(inv_subshift(st1, 3)) ^ rkp[7]; 349 } 350 351 put_unaligned_le32(inv_subshift(st1, 0) ^ rkp[4], out); 352 put_unaligned_le32(inv_subshift(st1, 1) ^ rkp[5], out + 4); 353 put_unaligned_le32(inv_subshift(st1, 2) ^ rkp[6], out + 8); 354 put_unaligned_le32(inv_subshift(st1, 3) ^ rkp[7], out + 12); 355 356 local_irq_restore(flags); 357 } 358 359 static struct crypto_alg aes_alg = { 360 .cra_name = "aes", 361 .cra_driver_name = "aes-fixed-time", 362 .cra_priority = 100 + 1, 363 .cra_flags = CRYPTO_ALG_TYPE_CIPHER, 364 .cra_blocksize = AES_BLOCK_SIZE, 365 .cra_ctxsize = sizeof(struct crypto_aes_ctx), 366 .cra_module = THIS_MODULE, 367 368 .cra_cipher.cia_min_keysize = AES_MIN_KEY_SIZE, 369 .cra_cipher.cia_max_keysize = AES_MAX_KEY_SIZE, 370 .cra_cipher.cia_setkey = aesti_set_key, 371 .cra_cipher.cia_encrypt = aesti_encrypt, 372 .cra_cipher.cia_decrypt = aesti_decrypt 373 }; 374 375 static int __init aes_init(void) 376 { 377 return crypto_register_alg(&aes_alg); 378 } 379 380 static void __exit aes_fini(void) 381 { 382 crypto_unregister_alg(&aes_alg); 383 } 384 385 module_init(aes_init); 386 module_exit(aes_fini); 387 388 MODULE_DESCRIPTION("Generic fixed time AES"); 389 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); 390 MODULE_LICENSE("GPL v2"); 391