Eric Biggers | 7931411 | 2018-02-14 10:42:21 -0800 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * NEON-accelerated implementation of Speck128-XTS and Speck64-XTS |
| 4 | * |
| 5 | * Copyright (c) 2018 Google, Inc |
| 6 | * |
| 7 | * Note: the NIST recommendation for XTS only specifies a 128-bit block size, |
| 8 | * but a 64-bit version (needed for Speck64) is fairly straightforward; the math |
| 9 | * is just done in GF(2^64) instead of GF(2^128), with the reducing polynomial |
| 10 | * x^64 + x^4 + x^3 + x + 1 from the original XEX paper (Rogaway, 2004: |
| 11 | * "Efficient Instantiations of Tweakable Blockciphers and Refinements to Modes |
| 12 | * OCB and PMAC"), represented as 0x1B. |
| 13 | */ |
| 14 | |
| 15 | #include <asm/hwcap.h> |
| 16 | #include <asm/neon.h> |
| 17 | #include <asm/simd.h> |
| 18 | #include <crypto/algapi.h> |
| 19 | #include <crypto/gf128mul.h> |
| 20 | #include <crypto/internal/skcipher.h> |
| 21 | #include <crypto/speck.h> |
| 22 | #include <crypto/xts.h> |
| 23 | #include <linux/kernel.h> |
| 24 | #include <linux/module.h> |
| 25 | |
| 26 | /* The assembly functions only handle multiples of 128 bytes */ |
| 27 | #define SPECK_NEON_CHUNK_SIZE 128 |
| 28 | |
| 29 | /* Speck128 */ |
| 30 | |
| 31 | struct speck128_xts_tfm_ctx { |
| 32 | struct speck128_tfm_ctx main_key; |
| 33 | struct speck128_tfm_ctx tweak_key; |
| 34 | }; |
| 35 | |
| 36 | asmlinkage void speck128_xts_encrypt_neon(const u64 *round_keys, int nrounds, |
| 37 | void *dst, const void *src, |
| 38 | unsigned int nbytes, void *tweak); |
| 39 | |
| 40 | asmlinkage void speck128_xts_decrypt_neon(const u64 *round_keys, int nrounds, |
| 41 | void *dst, const void *src, |
| 42 | unsigned int nbytes, void *tweak); |
| 43 | |
| 44 | typedef void (*speck128_crypt_one_t)(const struct speck128_tfm_ctx *, |
| 45 | u8 *, const u8 *); |
| 46 | typedef void (*speck128_xts_crypt_many_t)(const u64 *, int, void *, |
| 47 | const void *, unsigned int, void *); |
| 48 | |
| 49 | static __always_inline int |
| 50 | __speck128_xts_crypt(struct skcipher_request *req, |
| 51 | speck128_crypt_one_t crypt_one, |
| 52 | speck128_xts_crypt_many_t crypt_many) |
| 53 | { |
| 54 | struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); |
| 55 | const struct speck128_xts_tfm_ctx *ctx = crypto_skcipher_ctx(tfm); |
| 56 | struct skcipher_walk walk; |
| 57 | le128 tweak; |
| 58 | int err; |
| 59 | |
| 60 | err = skcipher_walk_virt(&walk, req, true); |
| 61 | |
| 62 | crypto_speck128_encrypt(&ctx->tweak_key, (u8 *)&tweak, walk.iv); |
| 63 | |
| 64 | while (walk.nbytes > 0) { |
| 65 | unsigned int nbytes = walk.nbytes; |
| 66 | u8 *dst = walk.dst.virt.addr; |
| 67 | const u8 *src = walk.src.virt.addr; |
| 68 | |
| 69 | if (nbytes >= SPECK_NEON_CHUNK_SIZE && may_use_simd()) { |
| 70 | unsigned int count; |
| 71 | |
| 72 | count = round_down(nbytes, SPECK_NEON_CHUNK_SIZE); |
| 73 | kernel_neon_begin(); |
| 74 | (*crypt_many)(ctx->main_key.round_keys, |
| 75 | ctx->main_key.nrounds, |
| 76 | dst, src, count, &tweak); |
| 77 | kernel_neon_end(); |
| 78 | dst += count; |
| 79 | src += count; |
| 80 | nbytes -= count; |
| 81 | } |
| 82 | |
| 83 | /* Handle any remainder with generic code */ |
| 84 | while (nbytes >= sizeof(tweak)) { |
| 85 | le128_xor((le128 *)dst, (const le128 *)src, &tweak); |
| 86 | (*crypt_one)(&ctx->main_key, dst, dst); |
| 87 | le128_xor((le128 *)dst, (const le128 *)dst, &tweak); |
| 88 | gf128mul_x_ble(&tweak, &tweak); |
| 89 | |
| 90 | dst += sizeof(tweak); |
| 91 | src += sizeof(tweak); |
| 92 | nbytes -= sizeof(tweak); |
| 93 | } |
| 94 | err = skcipher_walk_done(&walk, nbytes); |
| 95 | } |
| 96 | |
| 97 | return err; |
| 98 | } |
| 99 | |
| 100 | static int speck128_xts_encrypt(struct skcipher_request *req) |
| 101 | { |
| 102 | return __speck128_xts_crypt(req, crypto_speck128_encrypt, |
| 103 | speck128_xts_encrypt_neon); |
| 104 | } |
| 105 | |
| 106 | static int speck128_xts_decrypt(struct skcipher_request *req) |
| 107 | { |
| 108 | return __speck128_xts_crypt(req, crypto_speck128_decrypt, |
| 109 | speck128_xts_decrypt_neon); |
| 110 | } |
| 111 | |
| 112 | static int speck128_xts_setkey(struct crypto_skcipher *tfm, const u8 *key, |
| 113 | unsigned int keylen) |
| 114 | { |
| 115 | struct speck128_xts_tfm_ctx *ctx = crypto_skcipher_ctx(tfm); |
| 116 | int err; |
| 117 | |
| 118 | err = xts_verify_key(tfm, key, keylen); |
| 119 | if (err) |
| 120 | return err; |
| 121 | |
| 122 | keylen /= 2; |
| 123 | |
| 124 | err = crypto_speck128_setkey(&ctx->main_key, key, keylen); |
| 125 | if (err) |
| 126 | return err; |
| 127 | |
| 128 | return crypto_speck128_setkey(&ctx->tweak_key, key + keylen, keylen); |
| 129 | } |
| 130 | |
| 131 | /* Speck64 */ |
| 132 | |
| 133 | struct speck64_xts_tfm_ctx { |
| 134 | struct speck64_tfm_ctx main_key; |
| 135 | struct speck64_tfm_ctx tweak_key; |
| 136 | }; |
| 137 | |
| 138 | asmlinkage void speck64_xts_encrypt_neon(const u32 *round_keys, int nrounds, |
| 139 | void *dst, const void *src, |
| 140 | unsigned int nbytes, void *tweak); |
| 141 | |
| 142 | asmlinkage void speck64_xts_decrypt_neon(const u32 *round_keys, int nrounds, |
| 143 | void *dst, const void *src, |
| 144 | unsigned int nbytes, void *tweak); |
| 145 | |
| 146 | typedef void (*speck64_crypt_one_t)(const struct speck64_tfm_ctx *, |
| 147 | u8 *, const u8 *); |
| 148 | typedef void (*speck64_xts_crypt_many_t)(const u32 *, int, void *, |
| 149 | const void *, unsigned int, void *); |
| 150 | |
| 151 | static __always_inline int |
| 152 | __speck64_xts_crypt(struct skcipher_request *req, speck64_crypt_one_t crypt_one, |
| 153 | speck64_xts_crypt_many_t crypt_many) |
| 154 | { |
| 155 | struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); |
| 156 | const struct speck64_xts_tfm_ctx *ctx = crypto_skcipher_ctx(tfm); |
| 157 | struct skcipher_walk walk; |
| 158 | __le64 tweak; |
| 159 | int err; |
| 160 | |
| 161 | err = skcipher_walk_virt(&walk, req, true); |
| 162 | |
| 163 | crypto_speck64_encrypt(&ctx->tweak_key, (u8 *)&tweak, walk.iv); |
| 164 | |
| 165 | while (walk.nbytes > 0) { |
| 166 | unsigned int nbytes = walk.nbytes; |
| 167 | u8 *dst = walk.dst.virt.addr; |
| 168 | const u8 *src = walk.src.virt.addr; |
| 169 | |
| 170 | if (nbytes >= SPECK_NEON_CHUNK_SIZE && may_use_simd()) { |
| 171 | unsigned int count; |
| 172 | |
| 173 | count = round_down(nbytes, SPECK_NEON_CHUNK_SIZE); |
| 174 | kernel_neon_begin(); |
| 175 | (*crypt_many)(ctx->main_key.round_keys, |
| 176 | ctx->main_key.nrounds, |
| 177 | dst, src, count, &tweak); |
| 178 | kernel_neon_end(); |
| 179 | dst += count; |
| 180 | src += count; |
| 181 | nbytes -= count; |
| 182 | } |
| 183 | |
| 184 | /* Handle any remainder with generic code */ |
| 185 | while (nbytes >= sizeof(tweak)) { |
| 186 | *(__le64 *)dst = *(__le64 *)src ^ tweak; |
| 187 | (*crypt_one)(&ctx->main_key, dst, dst); |
| 188 | *(__le64 *)dst ^= tweak; |
| 189 | tweak = cpu_to_le64((le64_to_cpu(tweak) << 1) ^ |
| 190 | ((tweak & cpu_to_le64(1ULL << 63)) ? |
| 191 | 0x1B : 0)); |
| 192 | dst += sizeof(tweak); |
| 193 | src += sizeof(tweak); |
| 194 | nbytes -= sizeof(tweak); |
| 195 | } |
| 196 | err = skcipher_walk_done(&walk, nbytes); |
| 197 | } |
| 198 | |
| 199 | return err; |
| 200 | } |
| 201 | |
| 202 | static int speck64_xts_encrypt(struct skcipher_request *req) |
| 203 | { |
| 204 | return __speck64_xts_crypt(req, crypto_speck64_encrypt, |
| 205 | speck64_xts_encrypt_neon); |
| 206 | } |
| 207 | |
| 208 | static int speck64_xts_decrypt(struct skcipher_request *req) |
| 209 | { |
| 210 | return __speck64_xts_crypt(req, crypto_speck64_decrypt, |
| 211 | speck64_xts_decrypt_neon); |
| 212 | } |
| 213 | |
| 214 | static int speck64_xts_setkey(struct crypto_skcipher *tfm, const u8 *key, |
| 215 | unsigned int keylen) |
| 216 | { |
| 217 | struct speck64_xts_tfm_ctx *ctx = crypto_skcipher_ctx(tfm); |
| 218 | int err; |
| 219 | |
| 220 | err = xts_verify_key(tfm, key, keylen); |
| 221 | if (err) |
| 222 | return err; |
| 223 | |
| 224 | keylen /= 2; |
| 225 | |
| 226 | err = crypto_speck64_setkey(&ctx->main_key, key, keylen); |
| 227 | if (err) |
| 228 | return err; |
| 229 | |
| 230 | return crypto_speck64_setkey(&ctx->tweak_key, key + keylen, keylen); |
| 231 | } |
| 232 | |
| 233 | static struct skcipher_alg speck_algs[] = { |
| 234 | { |
| 235 | .base.cra_name = "xts(speck128)", |
| 236 | .base.cra_driver_name = "xts-speck128-neon", |
| 237 | .base.cra_priority = 300, |
| 238 | .base.cra_blocksize = SPECK128_BLOCK_SIZE, |
| 239 | .base.cra_ctxsize = sizeof(struct speck128_xts_tfm_ctx), |
| 240 | .base.cra_alignmask = 7, |
| 241 | .base.cra_module = THIS_MODULE, |
| 242 | .min_keysize = 2 * SPECK128_128_KEY_SIZE, |
| 243 | .max_keysize = 2 * SPECK128_256_KEY_SIZE, |
| 244 | .ivsize = SPECK128_BLOCK_SIZE, |
| 245 | .walksize = SPECK_NEON_CHUNK_SIZE, |
| 246 | .setkey = speck128_xts_setkey, |
| 247 | .encrypt = speck128_xts_encrypt, |
| 248 | .decrypt = speck128_xts_decrypt, |
| 249 | }, { |
| 250 | .base.cra_name = "xts(speck64)", |
| 251 | .base.cra_driver_name = "xts-speck64-neon", |
| 252 | .base.cra_priority = 300, |
| 253 | .base.cra_blocksize = SPECK64_BLOCK_SIZE, |
| 254 | .base.cra_ctxsize = sizeof(struct speck64_xts_tfm_ctx), |
| 255 | .base.cra_alignmask = 7, |
| 256 | .base.cra_module = THIS_MODULE, |
| 257 | .min_keysize = 2 * SPECK64_96_KEY_SIZE, |
| 258 | .max_keysize = 2 * SPECK64_128_KEY_SIZE, |
| 259 | .ivsize = SPECK64_BLOCK_SIZE, |
| 260 | .walksize = SPECK_NEON_CHUNK_SIZE, |
| 261 | .setkey = speck64_xts_setkey, |
| 262 | .encrypt = speck64_xts_encrypt, |
| 263 | .decrypt = speck64_xts_decrypt, |
| 264 | } |
| 265 | }; |
| 266 | |
| 267 | static int __init speck_neon_module_init(void) |
| 268 | { |
| 269 | if (!(elf_hwcap & HWCAP_NEON)) |
| 270 | return -ENODEV; |
| 271 | return crypto_register_skciphers(speck_algs, ARRAY_SIZE(speck_algs)); |
| 272 | } |
| 273 | |
| 274 | static void __exit speck_neon_module_exit(void) |
| 275 | { |
| 276 | crypto_unregister_skciphers(speck_algs, ARRAY_SIZE(speck_algs)); |
| 277 | } |
| 278 | |
| 279 | module_init(speck_neon_module_init); |
| 280 | module_exit(speck_neon_module_exit); |
| 281 | |
| 282 | MODULE_DESCRIPTION("Speck block cipher (NEON-accelerated)"); |
| 283 | MODULE_LICENSE("GPL"); |
| 284 | MODULE_AUTHOR("Eric Biggers <ebiggers@google.com>"); |
| 285 | MODULE_ALIAS_CRYPTO("xts(speck128)"); |
| 286 | MODULE_ALIAS_CRYPTO("xts-speck128-neon"); |
| 287 | MODULE_ALIAS_CRYPTO("xts(speck64)"); |
| 288 | MODULE_ALIAS_CRYPTO("xts-speck64-neon"); |