?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] sata_sil24: add suspend/sleep support
[linux-2.6/kmemtrace.git] / drivers / crypto / padlock-aes.c
blob17ee684144f9d2d54aae8ff95b16f77bdd091965
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 struct {
64 struct cword encrypt;
65 struct cword decrypt;
66 } cword;
67 u32 *D;
68 int key_length;
69 u32 E[AES_EXTENDED_KEY_SIZE];
70 u32 d_data[AES_EXTENDED_KEY_SIZE];
71 };
72
73 /* ====== Key management routines ====== */
74
75 static inline uint32_t
76 generic_rotr32 (const uint32_t x, const unsigned bits)
77 {
78 const unsigned n = bits % 32;
79 return (x >> n) | (x << (32 - n));
80 }
81
82 static inline uint32_t
83 generic_rotl32 (const uint32_t x, const unsigned bits)
84 {
85 const unsigned n = bits % 32;
86 return (x << n) | (x >> (32 - n));
87 }
88
89 #define rotl generic_rotl32
90 #define rotr generic_rotr32
91
92 /*
93 * #define byte(x, nr) ((unsigned char)((x) >> (nr*8)))
94 */
95 static inline uint8_t
96 byte(const uint32_t x, const unsigned n)
97 {
98 return x >> (n << 3);
99 }
100
101 #define E_KEY ctx->E
102 #define D_KEY ctx->D
103
104 static uint8_t pow_tab[256];
105 static uint8_t log_tab[256];
106 static uint8_t sbx_tab[256];
107 static uint8_t isb_tab[256];
108 static uint32_t rco_tab[10];
109 static uint32_t ft_tab[4][256];
110 static uint32_t it_tab[4][256];
111
112 static uint32_t fl_tab[4][256];
113 static uint32_t il_tab[4][256];
114
115 static inline uint8_t
116 f_mult (uint8_t a, uint8_t b)
117 {
118 uint8_t aa = log_tab[a], cc = aa + log_tab[b];
119
120 return pow_tab[cc + (cc < aa ? 1 : 0)];
121 }
122
123 #define ff_mult(a,b) (a && b ? f_mult(a, b) : 0)
124
125 #define f_rn(bo, bi, n, k) \
126 bo[n] = ft_tab[0][byte(bi[n],0)] ^ \
127 ft_tab[1][byte(bi[(n + 1) & 3],1)] ^ \
128 ft_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
129 ft_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n)
130
131 #define i_rn(bo, bi, n, k) \
132 bo[n] = it_tab[0][byte(bi[n],0)] ^ \
133 it_tab[1][byte(bi[(n + 3) & 3],1)] ^ \
134 it_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
135 it_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n)
136
137 #define ls_box(x) \
138 ( fl_tab[0][byte(x, 0)] ^ \
139 fl_tab[1][byte(x, 1)] ^ \
140 fl_tab[2][byte(x, 2)] ^ \
141 fl_tab[3][byte(x, 3)] )
142
143 #define f_rl(bo, bi, n, k) \
144 bo[n] = fl_tab[0][byte(bi[n],0)] ^ \
145 fl_tab[1][byte(bi[(n + 1) & 3],1)] ^ \
146 fl_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
147 fl_tab[3][byte(bi[(n + 3) & 3],3)] ^ *(k + n)
148
149 #define i_rl(bo, bi, n, k) \
150 bo[n] = il_tab[0][byte(bi[n],0)] ^ \
151 il_tab[1][byte(bi[(n + 3) & 3],1)] ^ \
152 il_tab[2][byte(bi[(n + 2) & 3],2)] ^ \
153 il_tab[3][byte(bi[(n + 1) & 3],3)] ^ *(k + n)
154
155 static void
156 gen_tabs (void)
157 {
158 uint32_t i, t;
159 uint8_t p, q;
160
161 /* log and power tables for GF(2**8) finite field with
162 0x011b as modular polynomial - the simplest prmitive
163 root is 0x03, used here to generate the tables */
164
165 for (i = 0, p = 1; i < 256; ++i) {
166 pow_tab[i] = (uint8_t) p;
167 log_tab[p] = (uint8_t) i;
168
169 p ^= (p << 1) ^ (p & 0x80 ? 0x01b : 0);
170 }
171
172 log_tab[1] = 0;
173
174 for (i = 0, p = 1; i < 10; ++i) {
175 rco_tab[i] = p;
176
177 p = (p << 1) ^ (p & 0x80 ? 0x01b : 0);
178 }
179
180 for (i = 0; i < 256; ++i) {
181 p = (i ? pow_tab[255 - log_tab[i]] : 0);
182 q = ((p >> 7) | (p << 1)) ^ ((p >> 6) | (p << 2));
183 p ^= 0x63 ^ q ^ ((q >> 6) | (q << 2));
184 sbx_tab[i] = p;
185 isb_tab[p] = (uint8_t) i;
186 }
187
188 for (i = 0; i < 256; ++i) {
189 p = sbx_tab[i];
190
191 t = p;
192 fl_tab[0][i] = t;
193 fl_tab[1][i] = rotl (t, 8);
194 fl_tab[2][i] = rotl (t, 16);
195 fl_tab[3][i] = rotl (t, 24);
196
197 t = ((uint32_t) ff_mult (2, p)) |
198 ((uint32_t) p << 8) |
199 ((uint32_t) p << 16) | ((uint32_t) ff_mult (3, p) << 24);
200
201 ft_tab[0][i] = t;
202 ft_tab[1][i] = rotl (t, 8);
203 ft_tab[2][i] = rotl (t, 16);
204 ft_tab[3][i] = rotl (t, 24);
205
206 p = isb_tab[i];
207
208 t = p;
209 il_tab[0][i] = t;
210 il_tab[1][i] = rotl (t, 8);
211 il_tab[2][i] = rotl (t, 16);
212 il_tab[3][i] = rotl (t, 24);
213
214 t = ((uint32_t) ff_mult (14, p)) |
215 ((uint32_t) ff_mult (9, p) << 8) |
216 ((uint32_t) ff_mult (13, p) << 16) |
217 ((uint32_t) ff_mult (11, p) << 24);
218
219 it_tab[0][i] = t;
220 it_tab[1][i] = rotl (t, 8);
221 it_tab[2][i] = rotl (t, 16);
222 it_tab[3][i] = rotl (t, 24);
223 }
224 }
225
226 #define star_x(x) (((x) & 0x7f7f7f7f) << 1) ^ ((((x) & 0x80808080) >> 7) * 0x1b)
227
228 #define imix_col(y,x) \
229 u = star_x(x); \
230 v = star_x(u); \
231 w = star_x(v); \
232 t = w ^ (x); \
233 (y) = u ^ v ^ w; \
234 (y) ^= rotr(u ^ t, 8) ^ \
235 rotr(v ^ t, 16) ^ \
236 rotr(t,24)
237
238 /* initialise the key schedule from the user supplied key */
239
240 #define loop4(i) \
241 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \
242 t ^= E_KEY[4 * i]; E_KEY[4 * i + 4] = t; \
243 t ^= E_KEY[4 * i + 1]; E_KEY[4 * i + 5] = t; \
244 t ^= E_KEY[4 * i + 2]; E_KEY[4 * i + 6] = t; \
245 t ^= E_KEY[4 * i + 3]; E_KEY[4 * i + 7] = t; \
246 }
247
248 #define loop6(i) \
249 { t = rotr(t, 8); t = ls_box(t) ^ rco_tab[i]; \
250 t ^= E_KEY[6 * i]; E_KEY[6 * i + 6] = t; \
251 t ^= E_KEY[6 * i + 1]; E_KEY[6 * i + 7] = t; \
252 t ^= E_KEY[6 * i + 2]; E_KEY[6 * i + 8] = t; \
253 t ^= E_KEY[6 * i + 3]; E_KEY[6 * i + 9] = t; \
254 t ^= E_KEY[6 * i + 4]; E_KEY[6 * i + 10] = t; \
255 t ^= E_KEY[6 * i + 5]; E_KEY[6 * i + 11] = t; \
256 }
257
258 #define loop8(i) \
259 { t = rotr(t, 8); ; t = ls_box(t) ^ rco_tab[i]; \
260 t ^= E_KEY[8 * i]; E_KEY[8 * i + 8] = t; \
261 t ^= E_KEY[8 * i + 1]; E_KEY[8 * i + 9] = t; \
262 t ^= E_KEY[8 * i + 2]; E_KEY[8 * i + 10] = t; \
263 t ^= E_KEY[8 * i + 3]; E_KEY[8 * i + 11] = t; \
264 t = E_KEY[8 * i + 4] ^ ls_box(t); \
265 E_KEY[8 * i + 12] = t; \
266 t ^= E_KEY[8 * i + 5]; E_KEY[8 * i + 13] = t; \
267 t ^= E_KEY[8 * i + 6]; E_KEY[8 * i + 14] = t; \
268 t ^= E_KEY[8 * i + 7]; E_KEY[8 * i + 15] = t; \
269 }
270
271 /* Tells whether the ACE is capable to generate
272 the extended key for a given key_len. */
273 static inline int
274 aes_hw_extkey_available(uint8_t key_len)
275 {
276 /* TODO: We should check the actual CPU model/stepping
277 as it's possible that the capability will be
278 added in the next CPU revisions. */
279 if (key_len == 16)
280 return 1;
281 return 0;
282 }
283
284 static inline struct aes_ctx *aes_ctx(struct crypto_tfm *tfm)
285 {
286 unsigned long addr = (unsigned long)crypto_tfm_ctx(tfm);
287 unsigned long align = PADLOCK_ALIGNMENT;
288
289 if (align <= crypto_tfm_ctx_alignment())
290 align = 1;
291 return (struct aes_ctx *)ALIGN(addr, align);
292 }
293
294 static int aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
295 unsigned int key_len, u32 *flags)
296 {
297 struct aes_ctx *ctx = aes_ctx(tfm);
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->D = ctx->E;
316
317 E_KEY[0] = le32_to_cpu(key[0]);
318 E_KEY[1] = le32_to_cpu(key[1]);
319 E_KEY[2] = le32_to_cpu(key[2]);
320 E_KEY[3] = le32_to_cpu(key[3]);
321
322 /* Prepare control words. */
323 memset(&ctx->cword, 0, sizeof(ctx->cword));
324
325 ctx->cword.decrypt.encdec = 1;
326 ctx->cword.encrypt.rounds = 10 + (key_len - 16) / 4;
327 ctx->cword.decrypt.rounds = ctx->cword.encrypt.rounds;
328 ctx->cword.encrypt.ksize = (key_len - 16) / 8;
329 ctx->cword.decrypt.ksize = ctx->cword.encrypt.ksize;
330
331 /* Don't generate extended keys if the hardware can do it. */
332 if (aes_hw_extkey_available(key_len))
333 return 0;
334
335 ctx->D = ctx->d_data;
336 ctx->cword.encrypt.keygen = 1;
337 ctx->cword.decrypt.keygen = 1;
338
339 switch (key_len) {
340 case 16:
341 t = E_KEY[3];
342 for (i = 0; i < 10; ++i)
343 loop4 (i);
344 break;
345
346 case 24:
347 E_KEY[4] = le32_to_cpu(key[4]);
348 t = E_KEY[5] = le32_to_cpu(key[5]);
349 for (i = 0; i < 8; ++i)
350 loop6 (i);
351 break;
352
353 case 32:
354 E_KEY[4] = le32_to_cpu(key[4]);
355 E_KEY[5] = le32_to_cpu(key[5]);
356 E_KEY[6] = le32_to_cpu(key[6]);
357 t = E_KEY[7] = le32_to_cpu(key[7]);
358 for (i = 0; i < 7; ++i)
359 loop8 (i);
360 break;
361 }
362
363 D_KEY[0] = E_KEY[0];
364 D_KEY[1] = E_KEY[1];
365 D_KEY[2] = E_KEY[2];
366 D_KEY[3] = E_KEY[3];
367
368 for (i = 4; i < key_len + 24; ++i) {
369 imix_col (D_KEY[i], E_KEY[i]);
370 }
371
372 /* PadLock needs a different format of the decryption key. */
373 rounds = 10 + (key_len - 16) / 4;
374
375 for (i = 0; i < rounds; i++) {
376 P[((i + 1) * 4) + 0] = D_KEY[((rounds - i - 1) * 4) + 0];
377 P[((i + 1) * 4) + 1] = D_KEY[((rounds - i - 1) * 4) + 1];
378 P[((i + 1) * 4) + 2] = D_KEY[((rounds - i - 1) * 4) + 2];
379 P[((i + 1) * 4) + 3] = D_KEY[((rounds - i - 1) * 4) + 3];
380 }
381
382 P[0] = E_KEY[(rounds * 4) + 0];
383 P[1] = E_KEY[(rounds * 4) + 1];
384 P[2] = E_KEY[(rounds * 4) + 2];
385 P[3] = E_KEY[(rounds * 4) + 3];
386
387 memcpy(D_KEY, P, AES_EXTENDED_KEY_SIZE_B);
388
389 return 0;
390 }
391
392 /* ====== Encryption/decryption routines ====== */
393
394 /* These are the real call to PadLock. */
395 static inline void padlock_xcrypt_ecb(const u8 *input, u8 *output, void *key,
396 void *control_word, u32 count)
397 {
398 asm volatile ("pushfl; popfl"); /* enforce key reload. */
399 asm volatile (".byte 0xf3,0x0f,0xa7,0xc8" /* rep xcryptecb */
400 : "+S"(input), "+D"(output)
401 : "d"(control_word), "b"(key), "c"(count));
402 }
403
404 static inline u8 *padlock_xcrypt_cbc(const u8 *input, u8 *output, void *key,
405 u8 *iv, void *control_word, u32 count)
406 {
407 /* Enforce key reload. */
408 asm volatile ("pushfl; popfl");
409 /* rep xcryptcbc */
410 asm volatile (".byte 0xf3,0x0f,0xa7,0xd0"
411 : "+S" (input), "+D" (output), "+a" (iv)
412 : "d" (control_word), "b" (key), "c" (count));
413 return iv;
414 }
415
416 static void aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
417 {
418 struct aes_ctx *ctx = aes_ctx(tfm);
419 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt, 1);
420 }
421
422 static void aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
423 {
424 struct aes_ctx *ctx = aes_ctx(tfm);
425 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt, 1);
426 }
427
428 static unsigned int aes_encrypt_ecb(const struct cipher_desc *desc, u8 *out,
429 const u8 *in, unsigned int nbytes)
430 {
431 struct aes_ctx *ctx = aes_ctx(desc->tfm);
432 padlock_xcrypt_ecb(in, out, ctx->E, &ctx->cword.encrypt,
433 nbytes / AES_BLOCK_SIZE);
434 return nbytes & ~(AES_BLOCK_SIZE - 1);
435 }
436
437 static unsigned int aes_decrypt_ecb(const struct cipher_desc *desc, u8 *out,
438 const u8 *in, unsigned int nbytes)
439 {
440 struct aes_ctx *ctx = aes_ctx(desc->tfm);
441 padlock_xcrypt_ecb(in, out, ctx->D, &ctx->cword.decrypt,
442 nbytes / AES_BLOCK_SIZE);
443 return nbytes & ~(AES_BLOCK_SIZE - 1);
444 }
445
446 static unsigned int aes_encrypt_cbc(const struct cipher_desc *desc, u8 *out,
447 const u8 *in, unsigned int nbytes)
448 {
449 struct aes_ctx *ctx = aes_ctx(desc->tfm);
450 u8 *iv;
451
452 iv = padlock_xcrypt_cbc(in, out, ctx->E, desc->info,
453 &ctx->cword.encrypt, nbytes / AES_BLOCK_SIZE);
454 memcpy(desc->info, iv, AES_BLOCK_SIZE);
455
456 return nbytes & ~(AES_BLOCK_SIZE - 1);
457 }
458
459 static unsigned int aes_decrypt_cbc(const struct cipher_desc *desc, u8 *out,
460 const u8 *in, unsigned int nbytes)
461 {
462 struct aes_ctx *ctx = aes_ctx(desc->tfm);
463 padlock_xcrypt_cbc(in, out, ctx->D, desc->info, &ctx->cword.decrypt,
464 nbytes / AES_BLOCK_SIZE);
465 return nbytes & ~(AES_BLOCK_SIZE - 1);
466 }
467
468 static struct crypto_alg aes_alg = {
469 .cra_name = "aes",
470 .cra_driver_name = "aes-padlock",
471 .cra_priority = 300,
472 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
473 .cra_blocksize = AES_BLOCK_SIZE,
474 .cra_ctxsize = sizeof(struct aes_ctx),
475 .cra_alignmask = PADLOCK_ALIGNMENT - 1,
476 .cra_module = THIS_MODULE,
477 .cra_list = LIST_HEAD_INIT(aes_alg.cra_list),
478 .cra_u = {
479 .cipher = {
480 .cia_min_keysize = AES_MIN_KEY_SIZE,
481 .cia_max_keysize = AES_MAX_KEY_SIZE,
482 .cia_setkey = aes_set_key,
483 .cia_encrypt = aes_encrypt,
484 .cia_decrypt = aes_decrypt,
485 .cia_encrypt_ecb = aes_encrypt_ecb,
486 .cia_decrypt_ecb = aes_decrypt_ecb,
487 .cia_encrypt_cbc = aes_encrypt_cbc,
488 .cia_decrypt_cbc = aes_decrypt_cbc,
489 }
490 }
491 };
492
493 int __init padlock_init_aes(void)
494 {
495 printk(KERN_NOTICE PFX "Using VIA PadLock ACE for AES algorithm.\n");
496
497 gen_tabs();
498 return crypto_register_alg(&aes_alg);
499 }
500
501 void __exit padlock_fini_aes(void)
502 {
503 crypto_unregister_alg(&aes_alg);
504 }
kmemtrace - Kernel memory tracer