Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Cryptographic API. |
| 3 | * |
| 4 | * Glue code for the SHA512 Secure Hash Algorithm assembler |
| 5 | * implementation using supplemental SSE3 / AVX / AVX2 instructions. |
| 6 | * |
| 7 | * This file is based on sha512_generic.c |
| 8 | * |
| 9 | * Copyright (C) 2013 Intel Corporation |
| 10 | * Author: Tim Chen <tim.c.chen@linux.intel.com> |
| 11 | * |
| 12 | * This program is free software; you can redistribute it and/or modify it |
| 13 | * under the terms of the GNU General Public License as published by the Free |
| 14 | * Software Foundation; either version 2 of the License, or (at your option) |
| 15 | * any later version. |
| 16 | * |
| 17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 18 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
| 19 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 20 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS |
| 21 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN |
| 22 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
| 23 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 24 | * SOFTWARE. |
| 25 | * |
| 26 | */ |
| 27 | |
| 28 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 29 | |
| 30 | #include <crypto/internal/hash.h> |
| 31 | #include <linux/init.h> |
| 32 | #include <linux/module.h> |
| 33 | #include <linux/mm.h> |
| 34 | #include <linux/cryptohash.h> |
| 35 | #include <linux/types.h> |
| 36 | #include <crypto/sha.h> |
| 37 | #include <asm/byteorder.h> |
| 38 | #include <asm/i387.h> |
| 39 | #include <asm/xcr.h> |
| 40 | #include <asm/xsave.h> |
| 41 | |
| 42 | #include <linux/string.h> |
| 43 | |
| 44 | asmlinkage void sha512_transform_ssse3(const char *data, u64 *digest, |
| 45 | u64 rounds); |
| 46 | #ifdef CONFIG_AS_AVX |
| 47 | asmlinkage void sha512_transform_avx(const char *data, u64 *digest, |
| 48 | u64 rounds); |
| 49 | #endif |
| 50 | #ifdef CONFIG_AS_AVX2 |
| 51 | asmlinkage void sha512_transform_rorx(const char *data, u64 *digest, |
| 52 | u64 rounds); |
| 53 | #endif |
| 54 | |
| 55 | static asmlinkage void (*sha512_transform_asm)(const char *, u64 *, u64); |
| 56 | |
| 57 | |
| 58 | static int sha512_ssse3_init(struct shash_desc *desc) |
| 59 | { |
| 60 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 61 | |
| 62 | sctx->state[0] = SHA512_H0; |
| 63 | sctx->state[1] = SHA512_H1; |
| 64 | sctx->state[2] = SHA512_H2; |
| 65 | sctx->state[3] = SHA512_H3; |
| 66 | sctx->state[4] = SHA512_H4; |
| 67 | sctx->state[5] = SHA512_H5; |
| 68 | sctx->state[6] = SHA512_H6; |
| 69 | sctx->state[7] = SHA512_H7; |
| 70 | sctx->count[0] = sctx->count[1] = 0; |
| 71 | |
| 72 | return 0; |
| 73 | } |
| 74 | |
| 75 | static int __sha512_ssse3_update(struct shash_desc *desc, const u8 *data, |
| 76 | unsigned int len, unsigned int partial) |
| 77 | { |
| 78 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 79 | unsigned int done = 0; |
| 80 | |
| 81 | sctx->count[0] += len; |
| 82 | if (sctx->count[0] < len) |
| 83 | sctx->count[1]++; |
| 84 | |
| 85 | if (partial) { |
| 86 | done = SHA512_BLOCK_SIZE - partial; |
| 87 | memcpy(sctx->buf + partial, data, done); |
| 88 | sha512_transform_asm(sctx->buf, sctx->state, 1); |
| 89 | } |
| 90 | |
| 91 | if (len - done >= SHA512_BLOCK_SIZE) { |
| 92 | const unsigned int rounds = (len - done) / SHA512_BLOCK_SIZE; |
| 93 | |
| 94 | sha512_transform_asm(data + done, sctx->state, (u64) rounds); |
| 95 | |
| 96 | done += rounds * SHA512_BLOCK_SIZE; |
| 97 | } |
| 98 | |
| 99 | memcpy(sctx->buf, data + done, len - done); |
| 100 | |
| 101 | return 0; |
| 102 | } |
| 103 | |
| 104 | static int sha512_ssse3_update(struct shash_desc *desc, const u8 *data, |
| 105 | unsigned int len) |
| 106 | { |
| 107 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 108 | unsigned int partial = sctx->count[0] % SHA512_BLOCK_SIZE; |
| 109 | int res; |
| 110 | |
| 111 | /* Handle the fast case right here */ |
| 112 | if (partial + len < SHA512_BLOCK_SIZE) { |
| 113 | sctx->count[0] += len; |
| 114 | if (sctx->count[0] < len) |
| 115 | sctx->count[1]++; |
| 116 | memcpy(sctx->buf + partial, data, len); |
| 117 | |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | if (!irq_fpu_usable()) { |
| 122 | res = crypto_sha512_update(desc, data, len); |
| 123 | } else { |
| 124 | kernel_fpu_begin(); |
| 125 | res = __sha512_ssse3_update(desc, data, len, partial); |
| 126 | kernel_fpu_end(); |
| 127 | } |
| 128 | |
| 129 | return res; |
| 130 | } |
| 131 | |
| 132 | |
| 133 | /* Add padding and return the message digest. */ |
| 134 | static int sha512_ssse3_final(struct shash_desc *desc, u8 *out) |
| 135 | { |
| 136 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 137 | unsigned int i, index, padlen; |
| 138 | __be64 *dst = (__be64 *)out; |
| 139 | __be64 bits[2]; |
| 140 | static const u8 padding[SHA512_BLOCK_SIZE] = { 0x80, }; |
| 141 | |
| 142 | /* save number of bits */ |
| 143 | bits[1] = cpu_to_be64(sctx->count[0] << 3); |
| 144 | bits[0] = cpu_to_be64(sctx->count[1] << 3) | sctx->count[0] >> 61; |
| 145 | |
| 146 | /* Pad out to 112 mod 128 and append length */ |
| 147 | index = sctx->count[0] & 0x7f; |
| 148 | padlen = (index < 112) ? (112 - index) : ((128+112) - index); |
| 149 | |
| 150 | if (!irq_fpu_usable()) { |
| 151 | crypto_sha512_update(desc, padding, padlen); |
| 152 | crypto_sha512_update(desc, (const u8 *)&bits, sizeof(bits)); |
| 153 | } else { |
| 154 | kernel_fpu_begin(); |
| 155 | /* We need to fill a whole block for __sha512_ssse3_update() */ |
| 156 | if (padlen <= 112) { |
| 157 | sctx->count[0] += padlen; |
| 158 | if (sctx->count[0] < padlen) |
| 159 | sctx->count[1]++; |
| 160 | memcpy(sctx->buf + index, padding, padlen); |
| 161 | } else { |
| 162 | __sha512_ssse3_update(desc, padding, padlen, index); |
| 163 | } |
| 164 | __sha512_ssse3_update(desc, (const u8 *)&bits, |
| 165 | sizeof(bits), 112); |
| 166 | kernel_fpu_end(); |
| 167 | } |
| 168 | |
| 169 | /* Store state in digest */ |
| 170 | for (i = 0; i < 8; i++) |
| 171 | dst[i] = cpu_to_be64(sctx->state[i]); |
| 172 | |
| 173 | /* Wipe context */ |
| 174 | memset(sctx, 0, sizeof(*sctx)); |
| 175 | |
| 176 | return 0; |
| 177 | } |
| 178 | |
| 179 | static int sha512_ssse3_export(struct shash_desc *desc, void *out) |
| 180 | { |
| 181 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 182 | |
| 183 | memcpy(out, sctx, sizeof(*sctx)); |
| 184 | |
| 185 | return 0; |
| 186 | } |
| 187 | |
| 188 | static int sha512_ssse3_import(struct shash_desc *desc, const void *in) |
| 189 | { |
| 190 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 191 | |
| 192 | memcpy(sctx, in, sizeof(*sctx)); |
| 193 | |
| 194 | return 0; |
| 195 | } |
| 196 | |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 197 | static int sha384_ssse3_init(struct shash_desc *desc) |
| 198 | { |
| 199 | struct sha512_state *sctx = shash_desc_ctx(desc); |
| 200 | |
| 201 | sctx->state[0] = SHA384_H0; |
| 202 | sctx->state[1] = SHA384_H1; |
| 203 | sctx->state[2] = SHA384_H2; |
| 204 | sctx->state[3] = SHA384_H3; |
| 205 | sctx->state[4] = SHA384_H4; |
| 206 | sctx->state[5] = SHA384_H5; |
| 207 | sctx->state[6] = SHA384_H6; |
| 208 | sctx->state[7] = SHA384_H7; |
| 209 | |
| 210 | sctx->count[0] = sctx->count[1] = 0; |
| 211 | |
| 212 | return 0; |
| 213 | } |
| 214 | |
| 215 | static int sha384_ssse3_final(struct shash_desc *desc, u8 *hash) |
| 216 | { |
| 217 | u8 D[SHA512_DIGEST_SIZE]; |
| 218 | |
| 219 | sha512_ssse3_final(desc, D); |
| 220 | |
| 221 | memcpy(hash, D, SHA384_DIGEST_SIZE); |
| 222 | memset(D, 0, SHA512_DIGEST_SIZE); |
| 223 | |
| 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | static struct shash_alg algs[] = { { |
Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 228 | .digestsize = SHA512_DIGEST_SIZE, |
| 229 | .init = sha512_ssse3_init, |
| 230 | .update = sha512_ssse3_update, |
| 231 | .final = sha512_ssse3_final, |
| 232 | .export = sha512_ssse3_export, |
| 233 | .import = sha512_ssse3_import, |
| 234 | .descsize = sizeof(struct sha512_state), |
| 235 | .statesize = sizeof(struct sha512_state), |
| 236 | .base = { |
| 237 | .cra_name = "sha512", |
| 238 | .cra_driver_name = "sha512-ssse3", |
| 239 | .cra_priority = 150, |
| 240 | .cra_flags = CRYPTO_ALG_TYPE_SHASH, |
| 241 | .cra_blocksize = SHA512_BLOCK_SIZE, |
| 242 | .cra_module = THIS_MODULE, |
| 243 | } |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 244 | }, { |
| 245 | .digestsize = SHA384_DIGEST_SIZE, |
| 246 | .init = sha384_ssse3_init, |
| 247 | .update = sha512_ssse3_update, |
| 248 | .final = sha384_ssse3_final, |
| 249 | .export = sha512_ssse3_export, |
| 250 | .import = sha512_ssse3_import, |
| 251 | .descsize = sizeof(struct sha512_state), |
| 252 | .statesize = sizeof(struct sha512_state), |
| 253 | .base = { |
| 254 | .cra_name = "sha384", |
| 255 | .cra_driver_name = "sha384-ssse3", |
| 256 | .cra_priority = 150, |
| 257 | .cra_flags = CRYPTO_ALG_TYPE_SHASH, |
| 258 | .cra_blocksize = SHA384_BLOCK_SIZE, |
| 259 | .cra_module = THIS_MODULE, |
| 260 | } |
| 261 | } }; |
Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 262 | |
| 263 | #ifdef CONFIG_AS_AVX |
| 264 | static bool __init avx_usable(void) |
| 265 | { |
| 266 | u64 xcr0; |
| 267 | |
| 268 | if (!cpu_has_avx || !cpu_has_osxsave) |
| 269 | return false; |
| 270 | |
| 271 | xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); |
| 272 | if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) { |
| 273 | pr_info("AVX detected but unusable.\n"); |
| 274 | |
| 275 | return false; |
| 276 | } |
| 277 | |
| 278 | return true; |
| 279 | } |
| 280 | #endif |
| 281 | |
| 282 | static int __init sha512_ssse3_mod_init(void) |
| 283 | { |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 284 | /* test for SSSE3 first */ |
Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 285 | if (cpu_has_ssse3) |
| 286 | sha512_transform_asm = sha512_transform_ssse3; |
| 287 | |
| 288 | #ifdef CONFIG_AS_AVX |
| 289 | /* allow AVX to override SSSE3, it's a little faster */ |
| 290 | if (avx_usable()) { |
| 291 | #ifdef CONFIG_AS_AVX2 |
| 292 | if (boot_cpu_has(X86_FEATURE_AVX2)) |
| 293 | sha512_transform_asm = sha512_transform_rorx; |
| 294 | else |
| 295 | #endif |
| 296 | sha512_transform_asm = sha512_transform_avx; |
| 297 | } |
| 298 | #endif |
| 299 | |
| 300 | if (sha512_transform_asm) { |
| 301 | #ifdef CONFIG_AS_AVX |
| 302 | if (sha512_transform_asm == sha512_transform_avx) |
| 303 | pr_info("Using AVX optimized SHA-512 implementation\n"); |
| 304 | #ifdef CONFIG_AS_AVX2 |
| 305 | else if (sha512_transform_asm == sha512_transform_rorx) |
| 306 | pr_info("Using AVX2 optimized SHA-512 implementation\n"); |
| 307 | #endif |
| 308 | else |
| 309 | #endif |
| 310 | pr_info("Using SSSE3 optimized SHA-512 implementation\n"); |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 311 | return crypto_register_shashes(algs, ARRAY_SIZE(algs)); |
Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 312 | } |
| 313 | pr_info("Neither AVX nor SSSE3 is available/usable.\n"); |
| 314 | |
| 315 | return -ENODEV; |
| 316 | } |
| 317 | |
| 318 | static void __exit sha512_ssse3_mod_fini(void) |
| 319 | { |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 320 | crypto_unregister_shashes(algs, ARRAY_SIZE(algs)); |
Tim Chen | 87de457 | 2013-03-26 14:00:02 -0700 | [diff] [blame] | 321 | } |
| 322 | |
| 323 | module_init(sha512_ssse3_mod_init); |
| 324 | module_exit(sha512_ssse3_mod_fini); |
| 325 | |
| 326 | MODULE_LICENSE("GPL"); |
| 327 | MODULE_DESCRIPTION("SHA512 Secure Hash Algorithm, Supplemental SSE3 accelerated"); |
| 328 | |
| 329 | MODULE_ALIAS("sha512"); |
Jussi Kivilinna | 340991e | 2013-05-21 17:10:44 +0300 | [diff] [blame] | 330 | MODULE_ALIAS("sha384"); |