?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.17.7
[linux/fpc-iii.git] / drivers / crypto / padlock-aes.c
blob5158a9db4bc58faeabd16b44a7ed2221cbd141ed
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 struct aes_ctx {
63 uint32_t e_data[AES_EXTENDED_KEY_SIZE];
64 uint32_t d_data[AES_EXTENDED_KEY_SIZE];
65 struct {
66 struct cword encrypt;
67 struct cword decrypt;
68 } cword;
69 uint32_t *E;
70 uint32_t *D;
71 int key_length;
72 };
73
74 /* ====== Key management routines ====== */
75
76 static inline uint32_t
77 generic_rotr32 (const uint32_t x, const unsigned bits)
78 {
79 const unsigned n = bits % 32;
80 return (x >> n) | (x << (32 - n));
81 }
82
83 static inline uint32_t
84 generic_rotl32 (const uint32_t x, const unsigned bits)
85 {
86 const unsigned n = bits % 32;
87 return (x << n) | (x >> (32 - n));
88 }
89
90 #define rotl generic_rotl32
91 #define rotr generic_rotr32
92
93 /*
94 * #define byte(x, nr) ((unsigned char)((x) >> (nr*8)))
95 */
96 static inline uint8_t
97 byte(const uint32_t x, const unsigned n)
98 {
99 return x >> (n << 3);
100 }
101
102 #define E_KEY ctx->E
103 #define D_KEY ctx->D
104
105 static uint8_t pow_tab[256];
106 static uint8_t log_tab[256];
107 static uint8_t sbx_tab[256];
108 static uint8_t isb_tab[256];
109 static uint32_t rco_tab[10];
110 static uint32_t ft_tab[4][256];
111 static uint32_t it_tab[4][256];
112
113 static uint32_t fl_tab[4][256];
114 static uint32_t il_tab[4][256];
115
116 static inline uint8_t
117 f_mult (uint8_t a, uint8_t b)
118 {
119 uint8_t aa = log_tab[a], cc = aa + log_tab[b];
120
121 return pow_tab[cc + (cc < aa ? 1 : 0)];
122 }
123
124 #define ff_mult(a,b) (a && b ? f_mult(a, b) : 0)
125
126 #define f_rn(bo, bi, n, k) \
127 bo[n] = ft_tab[0][byte(bi[n],0)] ^ \
128 ft_tab[1][byte(bi[(n + 1) & 3],1)] ^ \
129 ft_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
130 ft_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n)
131
132 #define i_rn(bo, bi, n, k) \
133 bo[n] = it_tab[0][byte(bi[n],0)] ^ \
134 it_tab[1][byte(bi[(n + 3) & 3],1)] ^ \
135 it_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
136 it_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n)
137
138 #define ls_box(x) \
139 ( fl_tab[0][byte(x, 0)] ^ \
140 fl_tab[1][byte(x, 1)] ^ \
141 fl_tab[2][byte(x, 2)] ^ \
142 fl_tab[3][byte(x, 3)] )
143
144 #define f_rl(bo, bi, n, k) \
145 bo[n] = fl_tab[0][byte(bi[n],0)] ^ \
146 fl_tab[1][byte(bi[(n + 1) & 3],1)] ^ \
147 fl_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
148 fl_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n)
149
150 #define i_rl(bo, bi, n, k) \
151 bo[n] = il_tab[0][byte(bi[n],0)] ^ \
152 il_tab[1][byte(bi[(n + 3) & 3],1)] ^ \
153 il_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
154 il_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n)
155
156 static void
157 gen_tabs (void)
158 {
159 uint32_t i, t;
160 uint8_t p, q;
161
162 /* log and power tables for GF(2**8) finite field with
163 0x011b as modular polynomial - the simplest prmitive
164 root is 0x03, used here to generate the tables */
165
166 for (i = 0, p = 1; i < 256; ++i) {
167 pow_tab[i] = (uint8_t) p;
168 log_tab[p] = (uint8_t) i;
169
170 p ^= (p << 1) ^ (p & 0x80 ? 0x01b : 0);
171 }
172
173 log_tab[1] = 0;
174
175 for (i = 0, p = 1; i < 10; ++i) {
176 rco_tab[i] = p;
177
178 p = (p << 1) ^ (p & 0x80 ? 0x01b : 0);
179 }
180
181 for (i = 0; i < 256; ++i) {
182 p = (i ? pow_tab[255 - log_tab[i]] : 0);
183 q = ((p >> 7) | (p << 1)) ^ ((p >> 6) | (p << 2));
184 p ^= 0x63 ^ q ^ ((q >> 6) | (q << 2));
185 sbx_tab[i] = p;
186 isb_tab[p] = (uint8_t) i;
187 }
188
189 for (i = 0; i < 256; ++i) {
190 p = sbx_tab[i];
191
192 t = p;
193 fl_tab[0][i] = t;
194 fl_tab[1][i] = rotl (t, 8);
195 fl_tab[2][i] = rotl (t, 16);
196 fl_tab[3][i] = rotl (t, 24);
197
198 t = ((uint32_t) ff_mult (2, p)) |
199 ((uint32_t) p << 8) |
200 ((uint32_t) p << 16) | ((uint32_t) ff_mult (3, p) << 24);
201
202 ft_tab[0][i] = t;
203 ft_tab[1][i] = rotl (t, 8);
204 ft_tab[2][i] = rotl (t, 16);
205 ft_tab[3][i] = rotl (t, 24);
206
207 p = isb_tab[i];
208
209 t = p;
210 il_tab[0][i] = t;
211 il_tab[1][i] = rotl (t, 8);
212 il_tab[2][i] = rotl (t, 16);
213 il_tab[3][i] = rotl (t, 24);
214
215 t = ((uint32_t) ff_mult (14, p)) |
216 ((uint32_t) ff_mult (9, p) << 8) |
217 ((uint32_t) ff_mult (13, p) << 16) |
218 ((uint32_t) ff_mult (11, p) << 24);
219
220 it_tab[0][i] = t;
221 it_tab[1][i] = rotl (t, 8);
222 it_tab[2][i] = rotl (t, 16);
223 it_tab[3][i] = rotl (t, 24);
224 }
225 }
226
227 #define star_x(x) (((x) & 0x7f7f7f7f) << 1) ^ ((((x) & 0x80808080) >> 7) * 0x1b)
228
229 #define imix_col(y,x) \
230 u = star_x(x); \
231 v = star_x(u); \
232 w = star_x(v); \
233 t = w ^ (x); \
234 (y) = u ^ v ^ w; \
235 (y) ^= rotr(u ^ t, 8) ^ \
236 rotr(v ^ t, 16) ^ \
237 rotr(t,24)
238
239 /* initialise the key schedule from the user supplied key */
240
241 #define loop4(i) \
242 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \
243 t ^= E_KEY[4 * i]; E_KEY[4 * i + 4] = t; \
244 t ^= E_KEY[4 * i + 1]; E_KEY[4 * i + 5] = t; \
245 t ^= E_KEY[4 * i + 2]; E_KEY[4 * i + 6] = t; \
246 t ^= E_KEY[4 * i + 3]; E_KEY[4 * i + 7] = t; \
247 }
248
249 #define loop6(i) \
250 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \
251 t ^= E_KEY[6 * i]; E_KEY[6 * i + 6] = t; \
252 t ^= E_KEY[6 * i + 1]; E_KEY[6 * i + 7] = t; \
253 t ^= E_KEY[6 * i + 2]; E_KEY[6 * i + 8] = t; \
254 t ^= E_KEY[6 * i + 3]; E_KEY[6 * i + 9] = t; \
255 t ^= E_KEY[6 * i + 4]; E_KEY[6 * i + 10] = t; \
256 t ^= E_KEY[6 * i + 5]; E_KEY[6 * i + 11] = t; \
257 }
258
259 #define loop8(i) \
260 { t = rotr(t, 8); ; t = ls_box(t) ^ rco_tab[i]; \
261 t ^= E_KEY[8 * i]; E_KEY[8 * i + 8] = t; \
262 t ^= E_KEY[8 * i + 1]; E_KEY[8 * i + 9] = t; \
263 t ^= E_KEY[8 * i + 2]; E_KEY[8 * i + 10] = t; \
264 t ^= E_KEY[8 * i + 3]; E_KEY[8 * i + 11] = t; \
265 t = E_KEY[8 * i + 4] ^ ls_box(t); \
266 E_KEY[8 * i + 12] = t; \
267 t ^= E_KEY[8 * i + 5]; E_KEY[8 * i + 13] = t; \
268 t ^= E_KEY[8 * i + 6]; E_KEY[8 * i + 14] = t; \
269 t ^= E_KEY[8 * i + 7]; E_KEY[8 * i + 15] = t; \
270 }
271
272 /* Tells whether the ACE is capable to generate
273 the extended key for a given key_len. */
274 static inline int
275 aes_hw_extkey_available(uint8_t key_len)
276 {
277 /* TODO: We should check the actual CPU model/stepping
278 as it's possible that the capability will be
279 added in the next CPU revisions. */
280 if (key_len == 16)
281 return 1;
282 return 0;
283 }
284
285 static inline struct aes_ctx *aes_ctx(void *ctx)
286 {
287 unsigned long align = PADLOCK_ALIGNMENT;
288
289 if (align <= crypto_tfm_ctx_alignment())
290 align = 1;
291 return (struct aes_ctx *)ALIGN((unsigned long)ctx, align);
292 }
293
294 static int
295 aes_set_key(void *ctx_arg, const uint8_t *in_key, unsigned int key_len, uint32_t *flags)
296 {
297 struct aes_ctx *ctx = aes_ctx(ctx_arg);
298 const __le32 *key = (const __le32 *)in_key;
299 uint32_t i, t, u, v, w;
300 uint32_t P[AES_EXTENDED_KEY_SIZE];
301 uint32_t rounds;
302
303 if (key_len != 16 && key_len != 24 && key_len != 32) {
304 *flags |= CRYPTO_TFM_RES_BAD_KEY_LEN;
305 return -EINVAL;
306 }
307
308 ctx->key_length = key_len;
309
310 /*
311 * If the hardware is capable of generating the extended key
312 * itself we must supply the plain key for both encryption
313 * and decryption.
314 */
315 ctx->E = ctx->e_data;
316 ctx->D = ctx->e_data;
317
318 E_KEY[0] = le32_to_cpu(key[0]);
319 E_KEY[1] = le32_to_cpu(key[1]);
320 E_KEY[2] = le32_to_cpu(key[2]);
321 E_KEY[3] = le32_to_cpu(key[3]);
322
323 /* Prepare control words. */
324 memset(&ctx->cword, 0, sizeof(ctx->cword));
325
326 ctx->cword.decrypt.encdec = 1;
327 ctx->cword.encrypt.rounds = 10 + (key_len - 16) / 4;
328 ctx->cword.decrypt.rounds = ctx->cword.encrypt.rounds;
329 ctx->cword.encrypt.ksize = (key_len - 16) / 8;
330 ctx->cword.decrypt.ksize = ctx->cword.encrypt.ksize;
331
332 /* Don't generate extended keys if the hardware can do it. */
333 if (aes_hw_extkey_available(key_len))
334 return 0;
335
336 ctx->D = ctx->d_data;
337 ctx->cword.encrypt.keygen = 1;
338 ctx->cword.decrypt.keygen = 1;
339
340 switch (key_len) {
341 case 16:
342 t = E_KEY[3];
343 for (i = 0; i < 10; ++i)
344 loop4 (i);
345 break;
346
347 case 24:
348 E_KEY[4] = le32_to_cpu(key[4]);
349 t = E_KEY[5] = le32_to_cpu(key[5]);
350 for (i = 0; i < 8; ++i)
351 loop6 (i);
352 break;
353
354 case 32:
355 E_KEY[4] = le32_to_cpu(key[4]);
356 E_KEY[5] = le32_to_cpu(key[5]);
357 E_KEY[6] = le32_to_cpu(key[6]);
358 t = E_KEY[7] = le32_to_cpu(key[7]);
359 for (i = 0; i < 7; ++i)
360 loop8 (i);
361 break;
362 }
363
364 D_KEY[0] = E_KEY[0];
365 D_KEY[1] = E_KEY[1];
366 D_KEY[2] = E_KEY[2];
367 D_KEY[3] = E_KEY[3];
368
369 for (i = 4; i < key_len + 24; ++i) {
370 imix_col (D_KEY[i], E_KEY[i]);
371 }
372
373 /* PadLock needs a different format of the decryption key. */
374 rounds = 10 + (key_len - 16) / 4;
375
376 for (i = 0; i < rounds; i++) {
377 P[((i + 1) * 4) + 0] = D_KEY[((rounds - i - 1) * 4) + 0];
378 P[((i + 1) * 4) + 1] = D_KEY[((rounds - i - 1) * 4) + 1];
379 P[((i + 1) * 4) + 2] = D_KEY[((rounds - i - 1) * 4) + 2];
380 P[((i + 1) * 4) + 3] = D_KEY[((rounds - i - 1) * 4) + 3];
381 }
382
383 P[0] = E_KEY[(rounds * 4) + 0];
384 P[1] = E_KEY[(rounds * 4) + 1];
385 P[2] = E_KEY[(rounds * 4) + 2];
386 P[3] = E_KEY[(rounds * 4) + 3];
387
388 memcpy(D_KEY, P, AES_EXTENDED_KEY_SIZE_B);
389
390 return 0;
391 }
392
393 /* ====== Encryption/decryption routines ====== */
394
395 /* These are the real call to PadLock. */
396 static inline void padlock_xcrypt_ecb(const u8 *input, u8 *output, void *key,
397 void *control_word, u32 count)
398 {
399 asm volatile ("pushfl; popfl"); /* enforce key reload. */
400 asm volatile (".byte 0xf3,0x0f,0xa7,0xc8" /* rep xcryptecb */
401 : "+S"(input), "+D"(output)
402 : "d"(control_word), "b"(key), "c"(count));
403 }
404
405 static inline u8 *padlock_xcrypt_cbc(const u8 *input, u8 *output, void *key,
406 u8 *iv, void *control_word, u32 count)
407 {
408 /* Enforce key reload. */
409 asm volatile ("pushfl; popfl");
410 /* rep xcryptcbc */
411 asm volatile (".byte 0xf3,0x0f,0xa7,0xd0"
412 : "+S" (input), "+D" (output), "+a" (iv)
413 : "d" (control_word), "b" (key), "c" (count));
414 return iv;
415 }
416
417 static void
418 aes_encrypt(void *ctx_arg, uint8_t *out, const uint8_t *in)
419 {
420 struct aes_ctx *ctx = aes_ctx(ctx_arg);
421 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt, 1);
422 }
423
424 static void
425 aes_decrypt(void *ctx_arg, uint8_t *out, const uint8_t *in)
426 {
427 struct aes_ctx *ctx = aes_ctx(ctx_arg);
428 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt, 1);
429 }
430
431 static unsigned int aes_encrypt_ecb(const struct cipher_desc *desc, u8 *out,
432 const u8 *in, unsigned int nbytes)
433 {
434 struct aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(desc->tfm));
435 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt,
436 nbytes / AES_BLOCK_SIZE);
437 return nbytes & ~(AES_BLOCK_SIZE - 1);
438 }
439
440 static unsigned int aes_decrypt_ecb(const struct cipher_desc *desc, u8 *out,
441 const u8 *in, unsigned int nbytes)
442 {
443 struct aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(desc->tfm));
444 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt,
445 nbytes / AES_BLOCK_SIZE);
446 return nbytes & ~(AES_BLOCK_SIZE - 1);
447 }
448
449 static unsigned int aes_encrypt_cbc(const struct cipher_desc *desc, u8 *out,
450 const u8 *in, unsigned int nbytes)
451 {
452 struct aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(desc->tfm));
453 u8 *iv;
454
455 iv = padlock_xcrypt_cbc(in, out, ctx->E, desc->info,
456 &ctx->cword.encrypt, nbytes / AES_BLOCK_SIZE);
457 memcpy(desc->info, iv, AES_BLOCK_SIZE);
458
459 return nbytes & ~(AES_BLOCK_SIZE - 1);
460 }
461
462 static unsigned int aes_decrypt_cbc(const struct cipher_desc *desc, u8 *out,
463 const u8 *in, unsigned int nbytes)
464 {
465 struct aes_ctx *ctx = aes_ctx(crypto_tfm_ctx(desc->tfm));
466 padlock_xcrypt_cbc(in, out, ctx->D, desc->info, &ctx->cword.decrypt,
467 nbytes / AES_BLOCK_SIZE);
468 return nbytes & ~(AES_BLOCK_SIZE - 1);
469 }
470
471 static struct crypto_alg aes_alg = {
472 .cra_name = "aes",
473 .cra_driver_name = "aes-padlock",
474 .cra_priority = 300,
475 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
476 .cra_blocksize = AES_BLOCK_SIZE,
477 .cra_ctxsize = sizeof(struct aes_ctx),
478 .cra_alignmask = PADLOCK_ALIGNMENT - 1,
479 .cra_module = THIS_MODULE,
480 .cra_list = LIST_HEAD_INIT(aes_alg.cra_list),
481 .cra_u = {
482 .cipher = {
483 .cia_min_keysize = AES_MIN_KEY_SIZE,
484 .cia_max_keysize = AES_MAX_KEY_SIZE,
485 .cia_setkey = aes_set_key,
486 .cia_encrypt = aes_encrypt,
487 .cia_decrypt = aes_decrypt,
488 .cia_encrypt_ecb = aes_encrypt_ecb,
489 .cia_decrypt_ecb = aes_decrypt_ecb,
490 .cia_encrypt_cbc = aes_encrypt_cbc,
491 .cia_decrypt_cbc = aes_decrypt_cbc,
492 }
493 }
494 };
495
496 int __init padlock_init_aes(void)
497 {
498 printk(KERN_NOTICE PFX "Using VIA PadLock ACE for AES algorithm.\n");
499
500 gen_tabs();
501 return crypto_register_alg(&aes_alg);
502 }
503
504 void __exit padlock_fini_aes(void)
505 {
506 crypto_unregister_alg(&aes_alg);
507 }
Linux with added FPC-III support