Ard Biesheuvel | f1e866b | 2015-03-10 09:47:48 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Accelerated GHASH implementation with ARMv8 vmull.p64 instructions. |
| 3 | * |
| 4 | * Copyright (C) 2015 Linaro Ltd. <ard.biesheuvel@linaro.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify it |
| 7 | * under the terms of the GNU General Public License version 2 as published |
| 8 | * by the Free Software Foundation. |
| 9 | */ |
| 10 | |
| 11 | #include <asm/hwcap.h> |
| 12 | #include <asm/neon.h> |
| 13 | #include <asm/simd.h> |
| 14 | #include <asm/unaligned.h> |
| 15 | #include <crypto/cryptd.h> |
| 16 | #include <crypto/internal/hash.h> |
| 17 | #include <crypto/gf128mul.h> |
| 18 | #include <linux/crypto.h> |
| 19 | #include <linux/module.h> |
| 20 | |
| 21 | MODULE_DESCRIPTION("GHASH secure hash using ARMv8 Crypto Extensions"); |
| 22 | MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>"); |
| 23 | MODULE_LICENSE("GPL v2"); |
| 24 | |
| 25 | #define GHASH_BLOCK_SIZE 16 |
| 26 | #define GHASH_DIGEST_SIZE 16 |
| 27 | |
| 28 | struct ghash_key { |
| 29 | u64 a; |
| 30 | u64 b; |
| 31 | }; |
| 32 | |
| 33 | struct ghash_desc_ctx { |
| 34 | u64 digest[GHASH_DIGEST_SIZE/sizeof(u64)]; |
| 35 | u8 buf[GHASH_BLOCK_SIZE]; |
| 36 | u32 count; |
| 37 | }; |
| 38 | |
| 39 | struct ghash_async_ctx { |
| 40 | struct cryptd_ahash *cryptd_tfm; |
| 41 | }; |
| 42 | |
| 43 | asmlinkage void pmull_ghash_update(int blocks, u64 dg[], const char *src, |
| 44 | struct ghash_key const *k, const char *head); |
| 45 | |
| 46 | static int ghash_init(struct shash_desc *desc) |
| 47 | { |
| 48 | struct ghash_desc_ctx *ctx = shash_desc_ctx(desc); |
| 49 | |
| 50 | *ctx = (struct ghash_desc_ctx){}; |
| 51 | return 0; |
| 52 | } |
| 53 | |
| 54 | static int ghash_update(struct shash_desc *desc, const u8 *src, |
| 55 | unsigned int len) |
| 56 | { |
| 57 | struct ghash_desc_ctx *ctx = shash_desc_ctx(desc); |
| 58 | unsigned int partial = ctx->count % GHASH_BLOCK_SIZE; |
| 59 | |
| 60 | ctx->count += len; |
| 61 | |
| 62 | if ((partial + len) >= GHASH_BLOCK_SIZE) { |
| 63 | struct ghash_key *key = crypto_shash_ctx(desc->tfm); |
| 64 | int blocks; |
| 65 | |
| 66 | if (partial) { |
| 67 | int p = GHASH_BLOCK_SIZE - partial; |
| 68 | |
| 69 | memcpy(ctx->buf + partial, src, p); |
| 70 | src += p; |
| 71 | len -= p; |
| 72 | } |
| 73 | |
| 74 | blocks = len / GHASH_BLOCK_SIZE; |
| 75 | len %= GHASH_BLOCK_SIZE; |
| 76 | |
| 77 | kernel_neon_begin(); |
| 78 | pmull_ghash_update(blocks, ctx->digest, src, key, |
| 79 | partial ? ctx->buf : NULL); |
| 80 | kernel_neon_end(); |
| 81 | src += blocks * GHASH_BLOCK_SIZE; |
| 82 | partial = 0; |
| 83 | } |
| 84 | if (len) |
| 85 | memcpy(ctx->buf + partial, src, len); |
| 86 | return 0; |
| 87 | } |
| 88 | |
| 89 | static int ghash_final(struct shash_desc *desc, u8 *dst) |
| 90 | { |
| 91 | struct ghash_desc_ctx *ctx = shash_desc_ctx(desc); |
| 92 | unsigned int partial = ctx->count % GHASH_BLOCK_SIZE; |
| 93 | |
| 94 | if (partial) { |
| 95 | struct ghash_key *key = crypto_shash_ctx(desc->tfm); |
| 96 | |
| 97 | memset(ctx->buf + partial, 0, GHASH_BLOCK_SIZE - partial); |
| 98 | kernel_neon_begin(); |
| 99 | pmull_ghash_update(1, ctx->digest, ctx->buf, key, NULL); |
| 100 | kernel_neon_end(); |
| 101 | } |
| 102 | put_unaligned_be64(ctx->digest[1], dst); |
| 103 | put_unaligned_be64(ctx->digest[0], dst + 8); |
| 104 | |
| 105 | *ctx = (struct ghash_desc_ctx){}; |
| 106 | return 0; |
| 107 | } |
| 108 | |
| 109 | static int ghash_setkey(struct crypto_shash *tfm, |
| 110 | const u8 *inkey, unsigned int keylen) |
| 111 | { |
| 112 | struct ghash_key *key = crypto_shash_ctx(tfm); |
| 113 | u64 a, b; |
| 114 | |
| 115 | if (keylen != GHASH_BLOCK_SIZE) { |
| 116 | crypto_shash_set_flags(tfm, CRYPTO_TFM_RES_BAD_KEY_LEN); |
| 117 | return -EINVAL; |
| 118 | } |
| 119 | |
| 120 | /* perform multiplication by 'x' in GF(2^128) */ |
| 121 | b = get_unaligned_be64(inkey); |
| 122 | a = get_unaligned_be64(inkey + 8); |
| 123 | |
| 124 | key->a = (a << 1) | (b >> 63); |
| 125 | key->b = (b << 1) | (a >> 63); |
| 126 | |
| 127 | if (b >> 63) |
| 128 | key->b ^= 0xc200000000000000UL; |
| 129 | |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | static struct shash_alg ghash_alg = { |
| 134 | .digestsize = GHASH_DIGEST_SIZE, |
| 135 | .init = ghash_init, |
| 136 | .update = ghash_update, |
| 137 | .final = ghash_final, |
| 138 | .setkey = ghash_setkey, |
| 139 | .descsize = sizeof(struct ghash_desc_ctx), |
| 140 | .base = { |
| 141 | .cra_name = "ghash", |
| 142 | .cra_driver_name = "__driver-ghash-ce", |
| 143 | .cra_priority = 0, |
Stephan Mueller | 4b3f4e3 | 2015-03-30 22:02:36 +0200 | [diff] [blame] | 144 | .cra_flags = CRYPTO_ALG_TYPE_SHASH | CRYPTO_ALG_INTERNAL, |
Ard Biesheuvel | f1e866b | 2015-03-10 09:47:48 +0100 | [diff] [blame] | 145 | .cra_blocksize = GHASH_BLOCK_SIZE, |
| 146 | .cra_ctxsize = sizeof(struct ghash_key), |
| 147 | .cra_module = THIS_MODULE, |
| 148 | }, |
| 149 | }; |
| 150 | |
| 151 | static int ghash_async_init(struct ahash_request *req) |
| 152 | { |
| 153 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
| 154 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); |
| 155 | struct ahash_request *cryptd_req = ahash_request_ctx(req); |
| 156 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; |
| 157 | |
| 158 | if (!may_use_simd()) { |
| 159 | memcpy(cryptd_req, req, sizeof(*req)); |
| 160 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); |
| 161 | return crypto_ahash_init(cryptd_req); |
| 162 | } else { |
| 163 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); |
| 164 | struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm); |
| 165 | |
| 166 | desc->tfm = child; |
| 167 | desc->flags = req->base.flags; |
| 168 | return crypto_shash_init(desc); |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | static int ghash_async_update(struct ahash_request *req) |
| 173 | { |
| 174 | struct ahash_request *cryptd_req = ahash_request_ctx(req); |
| 175 | |
| 176 | if (!may_use_simd()) { |
| 177 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
| 178 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); |
| 179 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; |
| 180 | |
| 181 | memcpy(cryptd_req, req, sizeof(*req)); |
| 182 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); |
| 183 | return crypto_ahash_update(cryptd_req); |
| 184 | } else { |
| 185 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); |
| 186 | return shash_ahash_update(req, desc); |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | static int ghash_async_final(struct ahash_request *req) |
| 191 | { |
| 192 | struct ahash_request *cryptd_req = ahash_request_ctx(req); |
| 193 | |
| 194 | if (!may_use_simd()) { |
| 195 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
| 196 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); |
| 197 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; |
| 198 | |
| 199 | memcpy(cryptd_req, req, sizeof(*req)); |
| 200 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); |
| 201 | return crypto_ahash_final(cryptd_req); |
| 202 | } else { |
| 203 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); |
| 204 | return crypto_shash_final(desc, req->result); |
| 205 | } |
| 206 | } |
| 207 | |
| 208 | static int ghash_async_digest(struct ahash_request *req) |
| 209 | { |
| 210 | struct crypto_ahash *tfm = crypto_ahash_reqtfm(req); |
| 211 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); |
| 212 | struct ahash_request *cryptd_req = ahash_request_ctx(req); |
| 213 | struct cryptd_ahash *cryptd_tfm = ctx->cryptd_tfm; |
| 214 | |
| 215 | if (!may_use_simd()) { |
| 216 | memcpy(cryptd_req, req, sizeof(*req)); |
| 217 | ahash_request_set_tfm(cryptd_req, &cryptd_tfm->base); |
| 218 | return crypto_ahash_digest(cryptd_req); |
| 219 | } else { |
| 220 | struct shash_desc *desc = cryptd_shash_desc(cryptd_req); |
| 221 | struct crypto_shash *child = cryptd_ahash_child(cryptd_tfm); |
| 222 | |
| 223 | desc->tfm = child; |
| 224 | desc->flags = req->base.flags; |
| 225 | return shash_ahash_digest(req, desc); |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | static int ghash_async_setkey(struct crypto_ahash *tfm, const u8 *key, |
| 230 | unsigned int keylen) |
| 231 | { |
| 232 | struct ghash_async_ctx *ctx = crypto_ahash_ctx(tfm); |
| 233 | struct crypto_ahash *child = &ctx->cryptd_tfm->base; |
| 234 | int err; |
| 235 | |
| 236 | crypto_ahash_clear_flags(child, CRYPTO_TFM_REQ_MASK); |
| 237 | crypto_ahash_set_flags(child, crypto_ahash_get_flags(tfm) |
| 238 | & CRYPTO_TFM_REQ_MASK); |
| 239 | err = crypto_ahash_setkey(child, key, keylen); |
| 240 | crypto_ahash_set_flags(tfm, crypto_ahash_get_flags(child) |
| 241 | & CRYPTO_TFM_RES_MASK); |
| 242 | |
| 243 | return err; |
| 244 | } |
| 245 | |
| 246 | static int ghash_async_init_tfm(struct crypto_tfm *tfm) |
| 247 | { |
| 248 | struct cryptd_ahash *cryptd_tfm; |
| 249 | struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm); |
| 250 | |
Stephan Mueller | 4b3f4e3 | 2015-03-30 22:02:36 +0200 | [diff] [blame] | 251 | cryptd_tfm = cryptd_alloc_ahash("__driver-ghash-ce", |
| 252 | CRYPTO_ALG_INTERNAL, |
| 253 | CRYPTO_ALG_INTERNAL); |
Ard Biesheuvel | f1e866b | 2015-03-10 09:47:48 +0100 | [diff] [blame] | 254 | if (IS_ERR(cryptd_tfm)) |
| 255 | return PTR_ERR(cryptd_tfm); |
| 256 | ctx->cryptd_tfm = cryptd_tfm; |
| 257 | crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm), |
| 258 | sizeof(struct ahash_request) + |
| 259 | crypto_ahash_reqsize(&cryptd_tfm->base)); |
| 260 | |
| 261 | return 0; |
| 262 | } |
| 263 | |
| 264 | static void ghash_async_exit_tfm(struct crypto_tfm *tfm) |
| 265 | { |
| 266 | struct ghash_async_ctx *ctx = crypto_tfm_ctx(tfm); |
| 267 | |
| 268 | cryptd_free_ahash(ctx->cryptd_tfm); |
| 269 | } |
| 270 | |
| 271 | static struct ahash_alg ghash_async_alg = { |
| 272 | .init = ghash_async_init, |
| 273 | .update = ghash_async_update, |
| 274 | .final = ghash_async_final, |
| 275 | .setkey = ghash_async_setkey, |
| 276 | .digest = ghash_async_digest, |
| 277 | .halg.digestsize = GHASH_DIGEST_SIZE, |
| 278 | .halg.base = { |
| 279 | .cra_name = "ghash", |
| 280 | .cra_driver_name = "ghash-ce", |
| 281 | .cra_priority = 300, |
| 282 | .cra_flags = CRYPTO_ALG_TYPE_AHASH | CRYPTO_ALG_ASYNC, |
| 283 | .cra_blocksize = GHASH_BLOCK_SIZE, |
| 284 | .cra_type = &crypto_ahash_type, |
| 285 | .cra_ctxsize = sizeof(struct ghash_async_ctx), |
| 286 | .cra_module = THIS_MODULE, |
| 287 | .cra_init = ghash_async_init_tfm, |
| 288 | .cra_exit = ghash_async_exit_tfm, |
| 289 | }, |
| 290 | }; |
| 291 | |
| 292 | static int __init ghash_ce_mod_init(void) |
| 293 | { |
| 294 | int err; |
| 295 | |
| 296 | if (!(elf_hwcap2 & HWCAP2_PMULL)) |
| 297 | return -ENODEV; |
| 298 | |
| 299 | err = crypto_register_shash(&ghash_alg); |
| 300 | if (err) |
| 301 | return err; |
| 302 | err = crypto_register_ahash(&ghash_async_alg); |
| 303 | if (err) |
| 304 | goto err_shash; |
| 305 | |
| 306 | return 0; |
| 307 | |
| 308 | err_shash: |
| 309 | crypto_unregister_shash(&ghash_alg); |
| 310 | return err; |
| 311 | } |
| 312 | |
| 313 | static void __exit ghash_ce_mod_exit(void) |
| 314 | { |
| 315 | crypto_unregister_ahash(&ghash_async_alg); |
| 316 | crypto_unregister_shash(&ghash_alg); |
| 317 | } |
| 318 | |
| 319 | module_init(ghash_ce_mod_init); |
| 320 | module_exit(ghash_ce_mod_exit); |