2 * Scalar fixed time AES core transform
4 * Copyright (C) 2017 Linaro Ltd <ard.biesheuvel@linaro.org>
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.
11 #include <crypto/aes.h>
12 #include <linux/crypto.h>
13 #include <linux/module.h>
14 #include <asm/unaligned.h>
17 * Emit the sbox as volatile const to prevent the compiler from doing
18 * constant folding on sbox references involving fixed indexes.
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,
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,
90 static u32 mul_by_x(u32 w)
92 u32 x = w & 0x7f7f7f7f;
93 u32 y = w & 0x80808080;
95 /* multiply by polynomial 'x' (0b10) in GF(2^8) */
96 return (x << 1) ^ (y >> 7) * 0x1b;
99 static u32 mul_by_x2(u32 w)
101 u32 x = w & 0x3f3f3f3f;
102 u32 y = w & 0x80808080;
103 u32 z = w & 0x40404040;
105 /* multiply by polynomial 'x^2' (0b100) in GF(2^8) */
106 return (x << 2) ^ (y >> 7) * 0x36 ^ (z >> 6) * 0x1b;
109 static u32 mix_columns(u32 x)
112 * Perform the following matrix multiplication in GF(2^8)
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] |
119 u32 y = mul_by_x(x) ^ ror32(x, 16);
121 return y ^ ror32(x ^ y, 8);
124 static u32 inv_mix_columns(u32 x)
127 * Perform the following matrix multiplication in GF(2^8)
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] |
134 * which can conveniently be reduced to
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] |
141 u32 y = mul_by_x2(x);
143 return mix_columns(x ^ y ^ ror32(y, 16));
146 static __always_inline u32 subshift(u32 in[], int pos)
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);
154 static __always_inline u32 inv_subshift(u32 in[], int pos)
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);
162 static u32 subw(u32 in)
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);
170 static int aesti_expand_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
171 unsigned int key_len)
173 u32 kwords = key_len / sizeof(u32);
176 if (key_len != AES_KEYSIZE_128 &&
177 key_len != AES_KEYSIZE_192 &&
178 key_len != AES_KEYSIZE_256)
181 ctx->key_length = key_len;
183 for (i = 0; i < kwords; i++)
184 ctx->key_enc[i] = get_unaligned_le32(in_key + i * sizeof(u32));
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;
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];
198 rko[4] = rko[3] ^ rki[4];
199 rko[5] = rko[4] ^ rki[5];
200 } else if (key_len == 32) {
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];
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
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];
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]);
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];
236 static int aesti_set_key(struct crypto_tfm *tfm, const u8 *in_key,
237 unsigned int key_len)
239 struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
242 err = aesti_expand_key(ctx, in_key, key_len);
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.
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];
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];
266 static void aesti_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
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;
275 st0[0] = ctx->key_enc[0] ^ get_unaligned_le32(in);
276 st0[1] = ctx->key_enc[1] ^ get_unaligned_le32(in + 4);
277 st0[2] = ctx->key_enc[2] ^ get_unaligned_le32(in + 8);
278 st0[3] = ctx->key_enc[3] ^ get_unaligned_le32(in + 12);
281 * Temporarily disable interrupts to avoid races where cachelines are
282 * evicted when the CPU is interrupted to do something else.
284 local_irq_save(flags);
286 st0[0] ^= __aesti_sbox[ 0] ^ __aesti_sbox[128];
287 st0[1] ^= __aesti_sbox[32] ^ __aesti_sbox[160];
288 st0[2] ^= __aesti_sbox[64] ^ __aesti_sbox[192];
289 st0[3] ^= __aesti_sbox[96] ^ __aesti_sbox[224];
291 for (round = 0;; round += 2, rkp += 8) {
292 st1[0] = mix_columns(subshift(st0, 0)) ^ rkp[0];
293 st1[1] = mix_columns(subshift(st0, 1)) ^ rkp[1];
294 st1[2] = mix_columns(subshift(st0, 2)) ^ rkp[2];
295 st1[3] = mix_columns(subshift(st0, 3)) ^ rkp[3];
297 if (round == rounds - 2)
300 st0[0] = mix_columns(subshift(st1, 0)) ^ rkp[4];
301 st0[1] = mix_columns(subshift(st1, 1)) ^ rkp[5];
302 st0[2] = mix_columns(subshift(st1, 2)) ^ rkp[6];
303 st0[3] = mix_columns(subshift(st1, 3)) ^ rkp[7];
306 put_unaligned_le32(subshift(st1, 0) ^ rkp[4], out);
307 put_unaligned_le32(subshift(st1, 1) ^ rkp[5], out + 4);
308 put_unaligned_le32(subshift(st1, 2) ^ rkp[6], out + 8);
309 put_unaligned_le32(subshift(st1, 3) ^ rkp[7], out + 12);
311 local_irq_restore(flags);
314 static void aesti_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
316 const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
317 const u32 *rkp = ctx->key_dec + 4;
318 int rounds = 6 + ctx->key_length / 4;
323 st0[0] = ctx->key_dec[0] ^ get_unaligned_le32(in);
324 st0[1] = ctx->key_dec[1] ^ get_unaligned_le32(in + 4);
325 st0[2] = ctx->key_dec[2] ^ get_unaligned_le32(in + 8);
326 st0[3] = ctx->key_dec[3] ^ get_unaligned_le32(in + 12);
329 * Temporarily disable interrupts to avoid races where cachelines are
330 * evicted when the CPU is interrupted to do something else.
332 local_irq_save(flags);
334 st0[0] ^= __aesti_inv_sbox[ 0] ^ __aesti_inv_sbox[128];
335 st0[1] ^= __aesti_inv_sbox[32] ^ __aesti_inv_sbox[160];
336 st0[2] ^= __aesti_inv_sbox[64] ^ __aesti_inv_sbox[192];
337 st0[3] ^= __aesti_inv_sbox[96] ^ __aesti_inv_sbox[224];
339 for (round = 0;; round += 2, rkp += 8) {
340 st1[0] = inv_mix_columns(inv_subshift(st0, 0)) ^ rkp[0];
341 st1[1] = inv_mix_columns(inv_subshift(st0, 1)) ^ rkp[1];
342 st1[2] = inv_mix_columns(inv_subshift(st0, 2)) ^ rkp[2];
343 st1[3] = inv_mix_columns(inv_subshift(st0, 3)) ^ rkp[3];
345 if (round == rounds - 2)
348 st0[0] = inv_mix_columns(inv_subshift(st1, 0)) ^ rkp[4];
349 st0[1] = inv_mix_columns(inv_subshift(st1, 1)) ^ rkp[5];
350 st0[2] = inv_mix_columns(inv_subshift(st1, 2)) ^ rkp[6];
351 st0[3] = inv_mix_columns(inv_subshift(st1, 3)) ^ rkp[7];
354 put_unaligned_le32(inv_subshift(st1, 0) ^ rkp[4], out);
355 put_unaligned_le32(inv_subshift(st1, 1) ^ rkp[5], out + 4);
356 put_unaligned_le32(inv_subshift(st1, 2) ^ rkp[6], out + 8);
357 put_unaligned_le32(inv_subshift(st1, 3) ^ rkp[7], out + 12);
359 local_irq_restore(flags);
362 static struct crypto_alg aes_alg = {
364 .cra_driver_name = "aes-fixed-time",
365 .cra_priority = 100 + 1,
366 .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
367 .cra_blocksize = AES_BLOCK_SIZE,
368 .cra_ctxsize = sizeof(struct crypto_aes_ctx),
369 .cra_module = THIS_MODULE,
371 .cra_cipher.cia_min_keysize = AES_MIN_KEY_SIZE,
372 .cra_cipher.cia_max_keysize = AES_MAX_KEY_SIZE,
373 .cra_cipher.cia_setkey = aesti_set_key,
374 .cra_cipher.cia_encrypt = aesti_encrypt,
375 .cra_cipher.cia_decrypt = aesti_decrypt
378 static int __init aes_init(void)
380 return crypto_register_alg(&aes_alg);
383 static void __exit aes_fini(void)
385 crypto_unregister_alg(&aes_alg);
388 module_init(aes_init);
389 module_exit(aes_fini);
391 MODULE_DESCRIPTION("Generic fixed time AES");
392 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
393 MODULE_LICENSE("GPL v2");