xref: /linux/crypto/aes_ti.c (revision e21f9e2e862e9eb3dd64eaddb6256b3e5098660f)
1 /*
2  * Scalar fixed time AES core transform
3  *
4  * Copyright (C) 2017 Linaro Ltd <ard.biesheuvel@linaro.org>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #include <crypto/aes.h>
12 #include <linux/crypto.h>
13 #include <linux/module.h>
14 #include <asm/unaligned.h>
15 
16 /*
17  * Emit the sbox as volatile const to prevent the compiler from doing
18  * constant folding on sbox references involving fixed indexes.
19  */
20 static volatile const u8 __cacheline_aligned __aesti_sbox[] = {
21 	0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5,
22 	0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
23 	0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0,
24 	0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
25 	0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc,
26 	0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
27 	0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a,
28 	0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
29 	0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0,
30 	0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
31 	0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b,
32 	0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
33 	0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85,
34 	0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
35 	0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5,
36 	0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
37 	0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17,
38 	0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
39 	0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88,
40 	0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
41 	0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c,
42 	0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
43 	0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9,
44 	0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
45 	0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6,
46 	0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
47 	0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e,
48 	0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
49 	0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94,
50 	0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
51 	0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68,
52 	0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
53 };
54 
55 static volatile const u8 __cacheline_aligned __aesti_inv_sbox[] = {
56 	0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38,
57 	0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
58 	0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87,
59 	0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
60 	0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d,
61 	0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
62 	0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2,
63 	0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
64 	0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16,
65 	0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
66 	0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda,
67 	0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
68 	0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a,
69 	0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
70 	0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02,
71 	0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
72 	0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea,
73 	0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
74 	0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85,
75 	0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
76 	0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89,
77 	0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
78 	0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20,
79 	0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
80 	0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31,
81 	0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
82 	0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d,
83 	0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
84 	0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0,
85 	0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
86 	0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26,
87 	0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
88 };
89 
90 static u32 mul_by_x(u32 w)
91 {
92 	u32 x = w & 0x7f7f7f7f;
93 	u32 y = w & 0x80808080;
94 
95 	/* multiply by polynomial 'x' (0b10) in GF(2^8) */
96 	return (x << 1) ^ (y >> 7) * 0x1b;
97 }
98 
99 static u32 mul_by_x2(u32 w)
100 {
101 	u32 x = w & 0x3f3f3f3f;
102 	u32 y = w & 0x80808080;
103 	u32 z = w & 0x40404040;
104 
105 	/* multiply by polynomial 'x^2' (0b100) in GF(2^8) */
106 	return (x << 2) ^ (y >> 7) * 0x36 ^ (z >> 6) * 0x1b;
107 }
108 
109 static u32 mix_columns(u32 x)
110 {
111 	/*
112 	 * Perform the following matrix multiplication in GF(2^8)
113 	 *
114 	 * | 0x2 0x3 0x1 0x1 |   | x[0] |
115 	 * | 0x1 0x2 0x3 0x1 |   | x[1] |
116 	 * | 0x1 0x1 0x2 0x3 | x | x[2] |
117 	 * | 0x3 0x1 0x1 0x2 |   | x[3] |
118 	 */
119 	u32 y = mul_by_x(x) ^ ror32(x, 16);
120 
121 	return y ^ ror32(x ^ y, 8);
122 }
123 
124 static u32 inv_mix_columns(u32 x)
125 {
126 	/*
127 	 * Perform the following matrix multiplication in GF(2^8)
128 	 *
129 	 * | 0xe 0xb 0xd 0x9 |   | x[0] |
130 	 * | 0x9 0xe 0xb 0xd |   | x[1] |
131 	 * | 0xd 0x9 0xe 0xb | x | x[2] |
132 	 * | 0xb 0xd 0x9 0xe |   | x[3] |
133 	 *
134 	 * which can conveniently be reduced to
135 	 *
136 	 * | 0x2 0x3 0x1 0x1 |   | 0x5 0x0 0x4 0x0 |   | x[0] |
137 	 * | 0x1 0x2 0x3 0x1 |   | 0x0 0x5 0x0 0x4 |   | x[1] |
138 	 * | 0x1 0x1 0x2 0x3 | x | 0x4 0x0 0x5 0x0 | x | x[2] |
139 	 * | 0x3 0x1 0x1 0x2 |   | 0x0 0x4 0x0 0x5 |   | x[3] |
140 	 */
141 	u32 y = mul_by_x2(x);
142 
143 	return mix_columns(x ^ y ^ ror32(y, 16));
144 }
145 
146 static __always_inline u32 subshift(u32 in[], int pos)
147 {
148 	return (__aesti_sbox[in[pos] & 0xff]) ^
149 	       (__aesti_sbox[(in[(pos + 1) % 4] >>  8) & 0xff] <<  8) ^
150 	       (__aesti_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^
151 	       (__aesti_sbox[(in[(pos + 3) % 4] >> 24) & 0xff] << 24);
152 }
153 
154 static __always_inline u32 inv_subshift(u32 in[], int pos)
155 {
156 	return (__aesti_inv_sbox[in[pos] & 0xff]) ^
157 	       (__aesti_inv_sbox[(in[(pos + 3) % 4] >>  8) & 0xff] <<  8) ^
158 	       (__aesti_inv_sbox[(in[(pos + 2) % 4] >> 16) & 0xff] << 16) ^
159 	       (__aesti_inv_sbox[(in[(pos + 1) % 4] >> 24) & 0xff] << 24);
160 }
161 
162 static u32 subw(u32 in)
163 {
164 	return (__aesti_sbox[in & 0xff]) ^
165 	       (__aesti_sbox[(in >>  8) & 0xff] <<  8) ^
166 	       (__aesti_sbox[(in >> 16) & 0xff] << 16) ^
167 	       (__aesti_sbox[(in >> 24) & 0xff] << 24);
168 }
169 
170 static int aesti_expand_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
171 			    unsigned int key_len)
172 {
173 	u32 kwords = key_len / sizeof(u32);
174 	u32 rc, i, j;
175 
176 	if (key_len != AES_KEYSIZE_128 &&
177 	    key_len != AES_KEYSIZE_192 &&
178 	    key_len != AES_KEYSIZE_256)
179 		return -EINVAL;
180 
181 	ctx->key_length = key_len;
182 
183 	for (i = 0; i < kwords; i++)
184 		ctx->key_enc[i] = get_unaligned_le32(in_key + i * sizeof(u32));
185 
186 	for (i = 0, rc = 1; i < 10; i++, rc = mul_by_x(rc)) {
187 		u32 *rki = ctx->key_enc + (i * kwords);
188 		u32 *rko = rki + kwords;
189 
190 		rko[0] = ror32(subw(rki[kwords - 1]), 8) ^ rc ^ rki[0];
191 		rko[1] = rko[0] ^ rki[1];
192 		rko[2] = rko[1] ^ rki[2];
193 		rko[3] = rko[2] ^ rki[3];
194 
195 		if (key_len == 24) {
196 			if (i >= 7)
197 				break;
198 			rko[4] = rko[3] ^ rki[4];
199 			rko[5] = rko[4] ^ rki[5];
200 		} else if (key_len == 32) {
201 			if (i >= 6)
202 				break;
203 			rko[4] = subw(rko[3]) ^ rki[4];
204 			rko[5] = rko[4] ^ rki[5];
205 			rko[6] = rko[5] ^ rki[6];
206 			rko[7] = rko[6] ^ rki[7];
207 		}
208 	}
209 
210 	/*
211 	 * Generate the decryption keys for the Equivalent Inverse Cipher.
212 	 * This involves reversing the order of the round keys, and applying
213 	 * the Inverse Mix Columns transformation to all but the first and
214 	 * the last one.
215 	 */
216 	ctx->key_dec[0] = ctx->key_enc[key_len + 24];
217 	ctx->key_dec[1] = ctx->key_enc[key_len + 25];
218 	ctx->key_dec[2] = ctx->key_enc[key_len + 26];
219 	ctx->key_dec[3] = ctx->key_enc[key_len + 27];
220 
221 	for (i = 4, j = key_len + 20; j > 0; i += 4, j -= 4) {
222 		ctx->key_dec[i]     = inv_mix_columns(ctx->key_enc[j]);
223 		ctx->key_dec[i + 1] = inv_mix_columns(ctx->key_enc[j + 1]);
224 		ctx->key_dec[i + 2] = inv_mix_columns(ctx->key_enc[j + 2]);
225 		ctx->key_dec[i + 3] = inv_mix_columns(ctx->key_enc[j + 3]);
226 	}
227 
228 	ctx->key_dec[i]     = ctx->key_enc[0];
229 	ctx->key_dec[i + 1] = ctx->key_enc[1];
230 	ctx->key_dec[i + 2] = ctx->key_enc[2];
231 	ctx->key_dec[i + 3] = ctx->key_enc[3];
232 
233 	return 0;
234 }
235 
236 static int aesti_set_key(struct crypto_tfm *tfm, const u8 *in_key,
237 			 unsigned int key_len)
238 {
239 	struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
240 	int err;
241 
242 	err = aesti_expand_key(ctx, in_key, key_len);
243 	if (err)
244 		return err;
245 
246 	/*
247 	 * In order to force the compiler to emit data independent Sbox lookups
248 	 * at the start of each block, xor the first round key with values at
249 	 * fixed indexes in the Sbox. This will need to be repeated each time
250 	 * the key is used, which will pull the entire Sbox into the D-cache
251 	 * before any data dependent Sbox lookups are performed.
252 	 */
253 	ctx->key_enc[0] ^= __aesti_sbox[ 0] ^ __aesti_sbox[128];
254 	ctx->key_enc[1] ^= __aesti_sbox[32] ^ __aesti_sbox[160];
255 	ctx->key_enc[2] ^= __aesti_sbox[64] ^ __aesti_sbox[192];
256 	ctx->key_enc[3] ^= __aesti_sbox[96] ^ __aesti_sbox[224];
257 
258 	ctx->key_dec[0] ^= __aesti_inv_sbox[ 0] ^ __aesti_inv_sbox[128];
259 	ctx->key_dec[1] ^= __aesti_inv_sbox[32] ^ __aesti_inv_sbox[160];
260 	ctx->key_dec[2] ^= __aesti_inv_sbox[64] ^ __aesti_inv_sbox[192];
261 	ctx->key_dec[3] ^= __aesti_inv_sbox[96] ^ __aesti_inv_sbox[224];
262 
263 	return 0;
264 }
265 
266 static void aesti_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
267 {
268 	const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
269 	const u32 *rkp = ctx->key_enc + 4;
270 	int rounds = 6 + ctx->key_length / 4;
271 	u32 st0[4], st1[4];
272 	int round;
273 
274 	st0[0] = ctx->key_enc[0] ^ get_unaligned_le32(in);
275 	st0[1] = ctx->key_enc[1] ^ get_unaligned_le32(in + 4);
276 	st0[2] = ctx->key_enc[2] ^ get_unaligned_le32(in + 8);
277 	st0[3] = ctx->key_enc[3] ^ get_unaligned_le32(in + 12);
278 
279 	st0[0] ^= __aesti_sbox[ 0] ^ __aesti_sbox[128];
280 	st0[1] ^= __aesti_sbox[32] ^ __aesti_sbox[160];
281 	st0[2] ^= __aesti_sbox[64] ^ __aesti_sbox[192];
282 	st0[3] ^= __aesti_sbox[96] ^ __aesti_sbox[224];
283 
284 	for (round = 0;; round += 2, rkp += 8) {
285 		st1[0] = mix_columns(subshift(st0, 0)) ^ rkp[0];
286 		st1[1] = mix_columns(subshift(st0, 1)) ^ rkp[1];
287 		st1[2] = mix_columns(subshift(st0, 2)) ^ rkp[2];
288 		st1[3] = mix_columns(subshift(st0, 3)) ^ rkp[3];
289 
290 		if (round == rounds - 2)
291 			break;
292 
293 		st0[0] = mix_columns(subshift(st1, 0)) ^ rkp[4];
294 		st0[1] = mix_columns(subshift(st1, 1)) ^ rkp[5];
295 		st0[2] = mix_columns(subshift(st1, 2)) ^ rkp[6];
296 		st0[3] = mix_columns(subshift(st1, 3)) ^ rkp[7];
297 	}
298 
299 	put_unaligned_le32(subshift(st1, 0) ^ rkp[4], out);
300 	put_unaligned_le32(subshift(st1, 1) ^ rkp[5], out + 4);
301 	put_unaligned_le32(subshift(st1, 2) ^ rkp[6], out + 8);
302 	put_unaligned_le32(subshift(st1, 3) ^ rkp[7], out + 12);
303 }
304 
305 static void aesti_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
306 {
307 	const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
308 	const u32 *rkp = ctx->key_dec + 4;
309 	int rounds = 6 + ctx->key_length / 4;
310 	u32 st0[4], st1[4];
311 	int round;
312 
313 	st0[0] = ctx->key_dec[0] ^ get_unaligned_le32(in);
314 	st0[1] = ctx->key_dec[1] ^ get_unaligned_le32(in + 4);
315 	st0[2] = ctx->key_dec[2] ^ get_unaligned_le32(in + 8);
316 	st0[3] = ctx->key_dec[3] ^ get_unaligned_le32(in + 12);
317 
318 	st0[0] ^= __aesti_inv_sbox[ 0] ^ __aesti_inv_sbox[128];
319 	st0[1] ^= __aesti_inv_sbox[32] ^ __aesti_inv_sbox[160];
320 	st0[2] ^= __aesti_inv_sbox[64] ^ __aesti_inv_sbox[192];
321 	st0[3] ^= __aesti_inv_sbox[96] ^ __aesti_inv_sbox[224];
322 
323 	for (round = 0;; round += 2, rkp += 8) {
324 		st1[0] = inv_mix_columns(inv_subshift(st0, 0)) ^ rkp[0];
325 		st1[1] = inv_mix_columns(inv_subshift(st0, 1)) ^ rkp[1];
326 		st1[2] = inv_mix_columns(inv_subshift(st0, 2)) ^ rkp[2];
327 		st1[3] = inv_mix_columns(inv_subshift(st0, 3)) ^ rkp[3];
328 
329 		if (round == rounds - 2)
330 			break;
331 
332 		st0[0] = inv_mix_columns(inv_subshift(st1, 0)) ^ rkp[4];
333 		st0[1] = inv_mix_columns(inv_subshift(st1, 1)) ^ rkp[5];
334 		st0[2] = inv_mix_columns(inv_subshift(st1, 2)) ^ rkp[6];
335 		st0[3] = inv_mix_columns(inv_subshift(st1, 3)) ^ rkp[7];
336 	}
337 
338 	put_unaligned_le32(inv_subshift(st1, 0) ^ rkp[4], out);
339 	put_unaligned_le32(inv_subshift(st1, 1) ^ rkp[5], out + 4);
340 	put_unaligned_le32(inv_subshift(st1, 2) ^ rkp[6], out + 8);
341 	put_unaligned_le32(inv_subshift(st1, 3) ^ rkp[7], out + 12);
342 }
343 
344 static struct crypto_alg aes_alg = {
345 	.cra_name			= "aes",
346 	.cra_driver_name		= "aes-fixed-time",
347 	.cra_priority			= 100 + 1,
348 	.cra_flags			= CRYPTO_ALG_TYPE_CIPHER,
349 	.cra_blocksize			= AES_BLOCK_SIZE,
350 	.cra_ctxsize			= sizeof(struct crypto_aes_ctx),
351 	.cra_module			= THIS_MODULE,
352 
353 	.cra_cipher.cia_min_keysize	= AES_MIN_KEY_SIZE,
354 	.cra_cipher.cia_max_keysize	= AES_MAX_KEY_SIZE,
355 	.cra_cipher.cia_setkey		= aesti_set_key,
356 	.cra_cipher.cia_encrypt		= aesti_encrypt,
357 	.cra_cipher.cia_decrypt		= aesti_decrypt
358 };
359 
360 static int __init aes_init(void)
361 {
362 	return crypto_register_alg(&aes_alg);
363 }
364 
365 static void __exit aes_fini(void)
366 {
367 	crypto_unregister_alg(&aes_alg);
368 }
369 
370 module_init(aes_init);
371 module_exit(aes_fini);
372 
373 MODULE_DESCRIPTION("Generic fixed time AES");
374 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
375 MODULE_LICENSE("GPL v2");
376