1 /* 2 * Cryptographic API. 3 * 4 * Support for VIA PadLock hardware crypto engine. 5 * 6 * Copyright (c) 2004 Michal Ludvig <michal@logix.cz> 7 * 8 * Key expansion routine taken from crypto/aes.c 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of the GNU General Public License as published by 12 * the Free Software Foundation; either version 2 of the License, or 13 * (at your option) any later version. 14 * 15 * --------------------------------------------------------------------------- 16 * Copyright (c) 2002, Dr Brian Gladman <brg@gladman.me.uk>, Worcester, UK. 17 * All rights reserved. 18 * 19 * LICENSE TERMS 20 * 21 * The free distribution and use of this software in both source and binary 22 * form is allowed (with or without changes) provided that: 23 * 24 * 1. distributions of this source code include the above copyright 25 * notice, this list of conditions and the following disclaimer; 26 * 27 * 2. distributions in binary form include the above copyright 28 * notice, this list of conditions and the following disclaimer 29 * in the documentation and/or other associated materials; 30 * 31 * 3. the copyright holder's name is not used to endorse products 32 * built using this software without specific written permission. 33 * 34 * ALTERNATIVELY, provided that this notice is retained in full, this product 35 * may be distributed under the terms of the GNU General Public License (GPL), 36 * in which case the provisions of the GPL apply INSTEAD OF those given above. 37 * 38 * DISCLAIMER 39 * 40 * This software is provided 'as is' with no explicit or implied warranties 41 * in respect of its properties, including, but not limited to, correctness 42 * and/or fitness for purpose. 43 * --------------------------------------------------------------------------- 44 */ 45 46 #include <linux/module.h> 47 #include <linux/init.h> 48 #include <linux/types.h> 49 #include <linux/errno.h> 50 #include <linux/crypto.h> 51 #include <linux/interrupt.h> 52 #include <linux/kernel.h> 53 #include <asm/byteorder.h> 54 #include "padlock.h" 55 56 #define AES_MIN_KEY_SIZE 16 /* in uint8_t units */ 57 #define AES_MAX_KEY_SIZE 32 /* ditto */ 58 #define AES_BLOCK_SIZE 16 /* ditto */ 59 #define AES_EXTENDED_KEY_SIZE 64 /* in uint32_t units */ 60 #define AES_EXTENDED_KEY_SIZE_B (AES_EXTENDED_KEY_SIZE * sizeof(uint32_t)) 61 62 /* Whenever making any changes to the following 63 * structure *make sure* you keep E, d_data 64 * and cword aligned on 16 Bytes boundaries!!! */ 65 struct aes_ctx { 66 struct { 67 struct cword encrypt; 68 struct cword decrypt; 69 } cword; 70 u32 *D; 71 int key_length; 72 u32 E[AES_EXTENDED_KEY_SIZE] 73 __attribute__ ((__aligned__(PADLOCK_ALIGNMENT))); 74 u32 d_data[AES_EXTENDED_KEY_SIZE] 75 __attribute__ ((__aligned__(PADLOCK_ALIGNMENT))); 76 }; 77 78 /* ====== Key management routines ====== */ 79 80 static inline uint32_t 81 generic_rotr32 (const uint32_t x, const unsigned bits) 82 { 83 const unsigned n = bits % 32; 84 return (x >> n) | (x << (32 - n)); 85 } 86 87 static inline uint32_t 88 generic_rotl32 (const uint32_t x, const unsigned bits) 89 { 90 const unsigned n = bits % 32; 91 return (x << n) | (x >> (32 - n)); 92 } 93 94 #define rotl generic_rotl32 95 #define rotr generic_rotr32 96 97 /* 98 * #define byte(x, nr) ((unsigned char)((x) >> (nr*8))) 99 */ 100 static inline uint8_t 101 byte(const uint32_t x, const unsigned n) 102 { 103 return x >> (n << 3); 104 } 105 106 #define E_KEY ctx->E 107 #define D_KEY ctx->D 108 109 static uint8_t pow_tab[256]; 110 static uint8_t log_tab[256]; 111 static uint8_t sbx_tab[256]; 112 static uint8_t isb_tab[256]; 113 static uint32_t rco_tab[10]; 114 static uint32_t ft_tab[4][256]; 115 static uint32_t it_tab[4][256]; 116 117 static uint32_t fl_tab[4][256]; 118 static uint32_t il_tab[4][256]; 119 120 static inline uint8_t 121 f_mult (uint8_t a, uint8_t b) 122 { 123 uint8_t aa = log_tab[a], cc = aa + log_tab[b]; 124 125 return pow_tab[cc + (cc < aa ? 1 : 0)]; 126 } 127 128 #define ff_mult(a,b) (a && b ? f_mult(a, b) : 0) 129 130 #define f_rn(bo, bi, n, k) \ 131 bo[n] = ft_tab[0][byte(bi[n],0)] ^ \ 132 ft_tab[1][byte(bi[(n + 1) & 3],1)] ^ \ 133 ft_tab[2][byte(bi[(n + 2) & 3],2)] ^ \ 134 ft_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n) 135 136 #define i_rn(bo, bi, n, k) \ 137 bo[n] = it_tab[0][byte(bi[n],0)] ^ \ 138 it_tab[1][byte(bi[(n + 3) & 3],1)] ^ \ 139 it_tab[2][byte(bi[(n + 2) & 3],2)] ^ \ 140 it_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n) 141 142 #define ls_box(x) \ 143 ( fl_tab[0][byte(x, 0)] ^ \ 144 fl_tab[1][byte(x, 1)] ^ \ 145 fl_tab[2][byte(x, 2)] ^ \ 146 fl_tab[3][byte(x, 3)] ) 147 148 #define f_rl(bo, bi, n, k) \ 149 bo[n] = fl_tab[0][byte(bi[n],0)] ^ \ 150 fl_tab[1][byte(bi[(n + 1) & 3],1)] ^ \ 151 fl_tab[2][byte(bi[(n + 2) & 3],2)] ^ \ 152 fl_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n) 153 154 #define i_rl(bo, bi, n, k) \ 155 bo[n] = il_tab[0][byte(bi[n],0)] ^ \ 156 il_tab[1][byte(bi[(n + 3) & 3],1)] ^ \ 157 il_tab[2][byte(bi[(n + 2) & 3],2)] ^ \ 158 il_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n) 159 160 static void 161 gen_tabs (void) 162 { 163 uint32_t i, t; 164 uint8_t p, q; 165 166 /* log and power tables for GF(2**8) finite field with 167 0x011b as modular polynomial - the simplest prmitive 168 root is 0x03, used here to generate the tables */ 169 170 for (i = 0, p = 1; i < 256; ++i) { 171 pow_tab[i] = (uint8_t) p; 172 log_tab[p] = (uint8_t) i; 173 174 p ^= (p << 1) ^ (p & 0x80 ? 0x01b : 0); 175 } 176 177 log_tab[1] = 0; 178 179 for (i = 0, p = 1; i < 10; ++i) { 180 rco_tab[i] = p; 181 182 p = (p << 1) ^ (p & 0x80 ? 0x01b : 0); 183 } 184 185 for (i = 0; i < 256; ++i) { 186 p = (i ? pow_tab[255 - log_tab[i]] : 0); 187 q = ((p >> 7) | (p << 1)) ^ ((p >> 6) | (p << 2)); 188 p ^= 0x63 ^ q ^ ((q >> 6) | (q << 2)); 189 sbx_tab[i] = p; 190 isb_tab[p] = (uint8_t) i; 191 } 192 193 for (i = 0; i < 256; ++i) { 194 p = sbx_tab[i]; 195 196 t = p; 197 fl_tab[0][i] = t; 198 fl_tab[1][i] = rotl (t, 8); 199 fl_tab[2][i] = rotl (t, 16); 200 fl_tab[3][i] = rotl (t, 24); 201 202 t = ((uint32_t) ff_mult (2, p)) | 203 ((uint32_t) p << 8) | 204 ((uint32_t) p << 16) | ((uint32_t) ff_mult (3, p) << 24); 205 206 ft_tab[0][i] = t; 207 ft_tab[1][i] = rotl (t, 8); 208 ft_tab[2][i] = rotl (t, 16); 209 ft_tab[3][i] = rotl (t, 24); 210 211 p = isb_tab[i]; 212 213 t = p; 214 il_tab[0][i] = t; 215 il_tab[1][i] = rotl (t, 8); 216 il_tab[2][i] = rotl (t, 16); 217 il_tab[3][i] = rotl (t, 24); 218 219 t = ((uint32_t) ff_mult (14, p)) | 220 ((uint32_t) ff_mult (9, p) << 8) | 221 ((uint32_t) ff_mult (13, p) << 16) | 222 ((uint32_t) ff_mult (11, p) << 24); 223 224 it_tab[0][i] = t; 225 it_tab[1][i] = rotl (t, 8); 226 it_tab[2][i] = rotl (t, 16); 227 it_tab[3][i] = rotl (t, 24); 228 } 229 } 230 231 #define star_x(x) (((x) & 0x7f7f7f7f) << 1) ^ ((((x) & 0x80808080) >> 7) * 0x1b) 232 233 #define imix_col(y,x) \ 234 u = star_x(x); \ 235 v = star_x(u); \ 236 w = star_x(v); \ 237 t = w ^ (x); \ 238 (y) = u ^ v ^ w; \ 239 (y) ^= rotr(u ^ t, 8) ^ \ 240 rotr(v ^ t, 16) ^ \ 241 rotr(t,24) 242 243 /* initialise the key schedule from the user supplied key */ 244 245 #define loop4(i) \ 246 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \ 247 t ^= E_KEY[4 * i]; E_KEY[4 * i + 4] = t; \ 248 t ^= E_KEY[4 * i + 1]; E_KEY[4 * i + 5] = t; \ 249 t ^= E_KEY[4 * i + 2]; E_KEY[4 * i + 6] = t; \ 250 t ^= E_KEY[4 * i + 3]; E_KEY[4 * i + 7] = t; \ 251 } 252 253 #define loop6(i) \ 254 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \ 255 t ^= E_KEY[6 * i]; E_KEY[6 * i + 6] = t; \ 256 t ^= E_KEY[6 * i + 1]; E_KEY[6 * i + 7] = t; \ 257 t ^= E_KEY[6 * i + 2]; E_KEY[6 * i + 8] = t; \ 258 t ^= E_KEY[6 * i + 3]; E_KEY[6 * i + 9] = t; \ 259 t ^= E_KEY[6 * i + 4]; E_KEY[6 * i + 10] = t; \ 260 t ^= E_KEY[6 * i + 5]; E_KEY[6 * i + 11] = t; \ 261 } 262 263 #define loop8(i) \ 264 { t = rotr(t, 8); ; t = ls_box(t) ^ rco_tab[i]; \ 265 t ^= E_KEY[8 * i]; E_KEY[8 * i + 8] = t; \ 266 t ^= E_KEY[8 * i + 1]; E_KEY[8 * i + 9] = t; \ 267 t ^= E_KEY[8 * i + 2]; E_KEY[8 * i + 10] = t; \ 268 t ^= E_KEY[8 * i + 3]; E_KEY[8 * i + 11] = t; \ 269 t = E_KEY[8 * i + 4] ^ ls_box(t); \ 270 E_KEY[8 * i + 12] = t; \ 271 t ^= E_KEY[8 * i + 5]; E_KEY[8 * i + 13] = t; \ 272 t ^= E_KEY[8 * i + 6]; E_KEY[8 * i + 14] = t; \ 273 t ^= E_KEY[8 * i + 7]; E_KEY[8 * i + 15] = t; \ 274 } 275 276 /* Tells whether the ACE is capable to generate 277 the extended key for a given key_len. */ 278 static inline int 279 aes_hw_extkey_available(uint8_t key_len) 280 { 281 /* TODO: We should check the actual CPU model/stepping 282 as it's possible that the capability will be 283 added in the next CPU revisions. */ 284 if (key_len == 16) 285 return 1; 286 return 0; 287 } 288 289 static inline struct aes_ctx *aes_ctx(struct crypto_tfm *tfm) 290 { 291 unsigned long addr = (unsigned long)crypto_tfm_ctx(tfm); 292 unsigned long align = PADLOCK_ALIGNMENT; 293 294 if (align <= crypto_tfm_ctx_alignment()) 295 align = 1; 296 return (struct aes_ctx *)ALIGN(addr, align); 297 } 298 299 static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key, 300 unsigned int key_len, u32 *flags) 301 { 302 struct aes_ctx *ctx = aes_ctx(tfm); 303 const __le32 *key = (const __le32 *)in_key; 304 uint32_t i, t, u, v, w; 305 uint32_t P[AES_EXTENDED_KEY_SIZE]; 306 uint32_t rounds; 307 308 if (key_len != 16 && key_len != 24 && key_len != 32) { 309 *flags |= CRYPTO_TFM_RES_BAD_KEY_LEN; 310 return -EINVAL; 311 } 312 313 ctx->key_length = key_len; 314 315 /* 316 * If the hardware is capable of generating the extended key 317 * itself we must supply the plain key for both encryption 318 * and decryption. 319 */ 320 ctx->D = ctx->E; 321 322 E_KEY[0] = le32_to_cpu(key[0]); 323 E_KEY[1] = le32_to_cpu(key[1]); 324 E_KEY[2] = le32_to_cpu(key[2]); 325 E_KEY[3] = le32_to_cpu(key[3]); 326 327 /* Prepare control words. */ 328 memset(&ctx->cword, 0, sizeof(ctx->cword)); 329 330 ctx->cword.decrypt.encdec = 1; 331 ctx->cword.encrypt.rounds = 10 + (key_len - 16) / 4; 332 ctx->cword.decrypt.rounds = ctx->cword.encrypt.rounds; 333 ctx->cword.encrypt.ksize = (key_len - 16) / 8; 334 ctx->cword.decrypt.ksize = ctx->cword.encrypt.ksize; 335 336 /* Don't generate extended keys if the hardware can do it. */ 337 if (aes_hw_extkey_available(key_len)) 338 return 0; 339 340 ctx->D = ctx->d_data; 341 ctx->cword.encrypt.keygen = 1; 342 ctx->cword.decrypt.keygen = 1; 343 344 switch (key_len) { 345 case 16: 346 t = E_KEY[3]; 347 for (i = 0; i < 10; ++i) 348 loop4 (i); 349 break; 350 351 case 24: 352 E_KEY[4] = le32_to_cpu(key[4]); 353 t = E_KEY[5] = le32_to_cpu(key[5]); 354 for (i = 0; i < 8; ++i) 355 loop6 (i); 356 break; 357 358 case 32: 359 E_KEY[4] = le32_to_cpu(key[4]); 360 E_KEY[5] = le32_to_cpu(key[5]); 361 E_KEY[6] = le32_to_cpu(key[6]); 362 t = E_KEY[7] = le32_to_cpu(key[7]); 363 for (i = 0; i < 7; ++i) 364 loop8 (i); 365 break; 366 } 367 368 D_KEY[0] = E_KEY[0]; 369 D_KEY[1] = E_KEY[1]; 370 D_KEY[2] = E_KEY[2]; 371 D_KEY[3] = E_KEY[3]; 372 373 for (i = 4; i < key_len + 24; ++i) { 374 imix_col (D_KEY[i], E_KEY[i]); 375 } 376 377 /* PadLock needs a different format of the decryption key. */ 378 rounds = 10 + (key_len - 16) / 4; 379 380 for (i = 0; i < rounds; i++) { 381 P[((i + 1) * 4) + 0] = D_KEY[((rounds - i - 1) * 4) + 0]; 382 P[((i + 1) * 4) + 1] = D_KEY[((rounds - i - 1) * 4) + 1]; 383 P[((i + 1) * 4) + 2] = D_KEY[((rounds - i - 1) * 4) + 2]; 384 P[((i + 1) * 4) + 3] = D_KEY[((rounds - i - 1) * 4) + 3]; 385 } 386 387 P[0] = E_KEY[(rounds * 4) + 0]; 388 P[1] = E_KEY[(rounds * 4) + 1]; 389 P[2] = E_KEY[(rounds * 4) + 2]; 390 P[3] = E_KEY[(rounds * 4) + 3]; 391 392 memcpy(D_KEY, P, AES_EXTENDED_KEY_SIZE_B); 393 394 return 0; 395 } 396 397 /* ====== Encryption/decryption routines ====== */ 398 399 /* These are the real call to PadLock. */ 400 static inline void padlock_xcrypt_ecb(const u8 *input, u8 *output, void *key, 401 void *control_word, u32 count) 402 { 403 asm volatile ("pushfl; popfl"); /* enforce key reload. */ 404 asm volatile (".byte 0xf3,0x0f,0xa7,0xc8" /* rep xcryptecb */ 405 : "+S"(input), "+D"(output) 406 : "d"(control_word), "b"(key), "c"(count)); 407 } 408 409 static inline u8 *padlock_xcrypt_cbc(const u8 *input, u8 *output, void *key, 410 u8 *iv, void *control_word, u32 count) 411 { 412 /* Enforce key reload. */ 413 asm volatile ("pushfl; popfl"); 414 /* rep xcryptcbc */ 415 asm volatile (".byte 0xf3,0x0f,0xa7,0xd0" 416 : "+S" (input), "+D" (output), "+a" (iv) 417 : "d" (control_word), "b" (key), "c" (count)); 418 return iv; 419 } 420 421 static void aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 422 { 423 struct aes_ctx *ctx = aes_ctx(tfm); 424 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt, 1); 425 } 426 427 static void aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 428 { 429 struct aes_ctx *ctx = aes_ctx(tfm); 430 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt, 1); 431 } 432 433 static unsigned int aes_encrypt_ecb(const struct cipher_desc *desc, u8 *out, 434 const u8 *in, unsigned int nbytes) 435 { 436 struct aes_ctx *ctx = aes_ctx(desc->tfm); 437 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt, 438 nbytes / AES_BLOCK_SIZE); 439 return nbytes & ~(AES_BLOCK_SIZE - 1); 440 } 441 442 static unsigned int aes_decrypt_ecb(const struct cipher_desc *desc, u8 *out, 443 const u8 *in, unsigned int nbytes) 444 { 445 struct aes_ctx *ctx = aes_ctx(desc->tfm); 446 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt, 447 nbytes / AES_BLOCK_SIZE); 448 return nbytes & ~(AES_BLOCK_SIZE - 1); 449 } 450 451 static unsigned int aes_encrypt_cbc(const struct cipher_desc *desc, u8 *out, 452 const u8 *in, unsigned int nbytes) 453 { 454 struct aes_ctx *ctx = aes_ctx(desc->tfm); 455 u8 *iv; 456 457 iv = padlock_xcrypt_cbc(in, out, ctx->E, desc->info, 458 &ctx->cword.encrypt, nbytes / AES_BLOCK_SIZE); 459 memcpy(desc->info, iv, AES_BLOCK_SIZE); 460 461 return nbytes & ~(AES_BLOCK_SIZE - 1); 462 } 463 464 static unsigned int aes_decrypt_cbc(const struct cipher_desc *desc, u8 *out, 465 const u8 *in, unsigned int nbytes) 466 { 467 struct aes_ctx *ctx = aes_ctx(desc->tfm); 468 padlock_xcrypt_cbc(in, out, ctx->D, desc->info, &ctx->cword.decrypt, 469 nbytes / AES_BLOCK_SIZE); 470 return nbytes & ~(AES_BLOCK_SIZE - 1); 471 } 472 473 static struct crypto_alg aes_alg = { 474 .cra_name = "aes", 475 .cra_driver_name = "aes-padlock", 476 .cra_priority = 300, 477 .cra_flags = CRYPTO_ALG_TYPE_CIPHER, 478 .cra_blocksize = AES_BLOCK_SIZE, 479 .cra_ctxsize = sizeof(struct aes_ctx), 480 .cra_alignmask = PADLOCK_ALIGNMENT - 1, 481 .cra_module = THIS_MODULE, 482 .cra_list = LIST_HEAD_INIT(aes_alg.cra_list), 483 .cra_u = { 484 .cipher = { 485 .cia_min_keysize = AES_MIN_KEY_SIZE, 486 .cia_max_keysize = AES_MAX_KEY_SIZE, 487 .cia_setkey = aes_set_key, 488 .cia_encrypt = aes_encrypt, 489 .cia_decrypt = aes_decrypt, 490 .cia_encrypt_ecb = aes_encrypt_ecb, 491 .cia_decrypt_ecb = aes_decrypt_ecb, 492 .cia_encrypt_cbc = aes_encrypt_cbc, 493 .cia_decrypt_cbc = aes_decrypt_cbc, 494 } 495 } 496 }; 497 498 static int __init padlock_init(void) 499 { 500 int ret; 501 502 if (!cpu_has_xcrypt) { 503 printk(KERN_ERR PFX "VIA PadLock not detected.\n"); 504 return -ENODEV; 505 } 506 507 if (!cpu_has_xcrypt_enabled) { 508 printk(KERN_ERR PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n"); 509 return -ENODEV; 510 } 511 512 gen_tabs(); 513 if ((ret = crypto_register_alg(&aes_alg))) { 514 printk(KERN_ERR PFX "VIA PadLock AES initialization failed.\n"); 515 return ret; 516 } 517 518 printk(KERN_NOTICE PFX "Using VIA PadLock ACE for AES algorithm.\n"); 519 520 return ret; 521 } 522 523 static void __exit padlock_fini(void) 524 { 525 crypto_unregister_alg(&aes_alg); 526 } 527 528 module_init(padlock_init); 529 module_exit(padlock_fini); 530 531 MODULE_DESCRIPTION("VIA PadLock AES algorithm support"); 532 MODULE_LICENSE("GPL"); 533 MODULE_AUTHOR("Michal Ludvig"); 534 535 MODULE_ALIAS("aes-padlock"); 536