Michal Ludvig | 6c83327 | 2006-07-12 12:29:38 +1000 | [diff] [blame^] | 1 | /* |
| 2 | * Cryptographic API. |
| 3 | * |
| 4 | * Support for VIA PadLock hardware crypto engine. |
| 5 | * |
| 6 | * Copyright (c) 2006 Michal Ludvig <michal@logix.cz> |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or modify |
| 9 | * it under the terms of the GNU General Public License as published by |
| 10 | * the Free Software Foundation; either version 2 of the License, or |
| 11 | * (at your option) any later version. |
| 12 | * |
| 13 | */ |
| 14 | |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/init.h> |
| 17 | #include <linux/errno.h> |
| 18 | #include <linux/crypto.h> |
| 19 | #include <linux/cryptohash.h> |
| 20 | #include <linux/interrupt.h> |
| 21 | #include <linux/kernel.h> |
| 22 | #include <linux/scatterlist.h> |
| 23 | #include "padlock.h" |
| 24 | |
| 25 | #define SHA1_DEFAULT_FALLBACK "sha1-generic" |
| 26 | #define SHA1_DIGEST_SIZE 20 |
| 27 | #define SHA1_HMAC_BLOCK_SIZE 64 |
| 28 | |
| 29 | #define SHA256_DEFAULT_FALLBACK "sha256-generic" |
| 30 | #define SHA256_DIGEST_SIZE 32 |
| 31 | #define SHA256_HMAC_BLOCK_SIZE 64 |
| 32 | |
| 33 | static char *sha1_fallback = SHA1_DEFAULT_FALLBACK; |
| 34 | static char *sha256_fallback = SHA256_DEFAULT_FALLBACK; |
| 35 | |
| 36 | module_param(sha1_fallback, charp, 0644); |
| 37 | module_param(sha256_fallback, charp, 0644); |
| 38 | |
| 39 | MODULE_PARM_DESC(sha1_fallback, "Fallback driver for SHA1. Default is " |
| 40 | SHA1_DEFAULT_FALLBACK); |
| 41 | MODULE_PARM_DESC(sha256_fallback, "Fallback driver for SHA256. Default is " |
| 42 | SHA256_DEFAULT_FALLBACK); |
| 43 | |
| 44 | struct padlock_sha_ctx { |
| 45 | char *data; |
| 46 | size_t used; |
| 47 | int bypass; |
| 48 | void (*f_sha_padlock)(const char *in, char *out, int count); |
| 49 | struct crypto_tfm *fallback_tfm; |
| 50 | }; |
| 51 | |
| 52 | static inline struct padlock_sha_ctx *ctx(struct crypto_tfm *tfm) |
| 53 | { |
| 54 | return (struct padlock_sha_ctx *)(crypto_tfm_ctx(tfm)); |
| 55 | } |
| 56 | |
| 57 | /* We'll need aligned address on the stack */ |
| 58 | #define NEAREST_ALIGNED(ptr) \ |
| 59 | ((void *)ALIGN((size_t)(ptr), PADLOCK_ALIGNMENT)) |
| 60 | |
| 61 | static struct crypto_alg sha1_alg, sha256_alg; |
| 62 | |
| 63 | static void padlock_sha_bypass(struct crypto_tfm *tfm) |
| 64 | { |
| 65 | if (ctx(tfm)->bypass) |
| 66 | return; |
| 67 | |
| 68 | BUG_ON(!ctx(tfm)->fallback_tfm); |
| 69 | |
| 70 | crypto_digest_init(ctx(tfm)->fallback_tfm); |
| 71 | if (ctx(tfm)->data && ctx(tfm)->used) { |
| 72 | struct scatterlist sg; |
| 73 | |
| 74 | sg_set_buf(&sg, ctx(tfm)->data, ctx(tfm)->used); |
| 75 | crypto_digest_update(ctx(tfm)->fallback_tfm, &sg, 1); |
| 76 | } |
| 77 | |
| 78 | ctx(tfm)->used = 0; |
| 79 | ctx(tfm)->bypass = 1; |
| 80 | } |
| 81 | |
| 82 | static void padlock_sha_init(struct crypto_tfm *tfm) |
| 83 | { |
| 84 | ctx(tfm)->used = 0; |
| 85 | ctx(tfm)->bypass = 0; |
| 86 | } |
| 87 | |
| 88 | static void padlock_sha_update(struct crypto_tfm *tfm, |
| 89 | const uint8_t *data, unsigned int length) |
| 90 | { |
| 91 | /* Our buffer is always one page. */ |
| 92 | if (unlikely(!ctx(tfm)->bypass && |
| 93 | (ctx(tfm)->used + length > PAGE_SIZE))) |
| 94 | padlock_sha_bypass(tfm); |
| 95 | |
| 96 | if (unlikely(ctx(tfm)->bypass)) { |
| 97 | struct scatterlist sg; |
| 98 | BUG_ON(!ctx(tfm)->fallback_tfm); |
| 99 | sg_set_buf(&sg, (uint8_t *)data, length); |
| 100 | crypto_digest_update(ctx(tfm)->fallback_tfm, &sg, 1); |
| 101 | return; |
| 102 | } |
| 103 | |
| 104 | memcpy(ctx(tfm)->data + ctx(tfm)->used, data, length); |
| 105 | ctx(tfm)->used += length; |
| 106 | } |
| 107 | |
| 108 | static inline void padlock_output_block(uint32_t *src, |
| 109 | uint32_t *dst, size_t count) |
| 110 | { |
| 111 | while (count--) |
| 112 | *dst++ = swab32(*src++); |
| 113 | } |
| 114 | |
| 115 | void padlock_do_sha1(const char *in, char *out, int count) |
| 116 | { |
| 117 | /* We can't store directly to *out as it may be unaligned. */ |
| 118 | /* BTW Don't reduce the buffer size below 128 Bytes! |
| 119 | * PadLock microcode needs it that big. */ |
| 120 | char buf[128+16]; |
| 121 | char *result = NEAREST_ALIGNED(buf); |
| 122 | |
| 123 | ((uint32_t *)result)[0] = 0x67452301; |
| 124 | ((uint32_t *)result)[1] = 0xEFCDAB89; |
| 125 | ((uint32_t *)result)[2] = 0x98BADCFE; |
| 126 | ((uint32_t *)result)[3] = 0x10325476; |
| 127 | ((uint32_t *)result)[4] = 0xC3D2E1F0; |
| 128 | |
| 129 | asm volatile (".byte 0xf3,0x0f,0xa6,0xc8" /* rep xsha1 */ |
| 130 | : "+S"(in), "+D"(result) |
| 131 | : "c"(count), "a"(0)); |
| 132 | |
| 133 | padlock_output_block((uint32_t *)result, (uint32_t *)out, 5); |
| 134 | } |
| 135 | |
| 136 | void padlock_do_sha256(const char *in, char *out, int count) |
| 137 | { |
| 138 | /* We can't store directly to *out as it may be unaligned. */ |
| 139 | /* BTW Don't reduce the buffer size below 128 Bytes! |
| 140 | * PadLock microcode needs it that big. */ |
| 141 | char buf[128+16]; |
| 142 | char *result = NEAREST_ALIGNED(buf); |
| 143 | |
| 144 | ((uint32_t *)result)[0] = 0x6A09E667; |
| 145 | ((uint32_t *)result)[1] = 0xBB67AE85; |
| 146 | ((uint32_t *)result)[2] = 0x3C6EF372; |
| 147 | ((uint32_t *)result)[3] = 0xA54FF53A; |
| 148 | ((uint32_t *)result)[4] = 0x510E527F; |
| 149 | ((uint32_t *)result)[5] = 0x9B05688C; |
| 150 | ((uint32_t *)result)[6] = 0x1F83D9AB; |
| 151 | ((uint32_t *)result)[7] = 0x5BE0CD19; |
| 152 | |
| 153 | asm volatile (".byte 0xf3,0x0f,0xa6,0xd0" /* rep xsha256 */ |
| 154 | : "+S"(in), "+D"(result) |
| 155 | : "c"(count), "a"(0)); |
| 156 | |
| 157 | padlock_output_block((uint32_t *)result, (uint32_t *)out, 8); |
| 158 | } |
| 159 | |
| 160 | static void padlock_sha_final(struct crypto_tfm *tfm, uint8_t *out) |
| 161 | { |
| 162 | if (unlikely(ctx(tfm)->bypass)) { |
| 163 | BUG_ON(!ctx(tfm)->fallback_tfm); |
| 164 | crypto_digest_final(ctx(tfm)->fallback_tfm, out); |
| 165 | ctx(tfm)->bypass = 0; |
| 166 | return; |
| 167 | } |
| 168 | |
| 169 | /* Pass the input buffer to PadLock microcode... */ |
| 170 | ctx(tfm)->f_sha_padlock(ctx(tfm)->data, out, ctx(tfm)->used); |
| 171 | |
| 172 | ctx(tfm)->used = 0; |
| 173 | } |
| 174 | |
| 175 | static int padlock_cra_init(struct crypto_tfm *tfm, const char *fallback_driver_name) |
| 176 | { |
| 177 | /* For now we'll allocate one page. This |
| 178 | * could eventually be configurable one day. */ |
| 179 | ctx(tfm)->data = (char *)__get_free_page(GFP_KERNEL); |
| 180 | if (!ctx(tfm)->data) |
| 181 | return -ENOMEM; |
| 182 | |
| 183 | /* Allocate a fallback and abort if it failed. */ |
| 184 | ctx(tfm)->fallback_tfm = crypto_alloc_tfm(fallback_driver_name, 0); |
| 185 | if (!ctx(tfm)->fallback_tfm) { |
| 186 | printk(KERN_WARNING PFX "Fallback driver '%s' could not be loaded!\n", |
| 187 | fallback_driver_name); |
| 188 | free_page((unsigned long)(ctx(tfm)->data)); |
| 189 | return -ENOENT; |
| 190 | } |
| 191 | |
| 192 | return 0; |
| 193 | } |
| 194 | |
| 195 | static int padlock_sha1_cra_init(struct crypto_tfm *tfm) |
| 196 | { |
| 197 | ctx(tfm)->f_sha_padlock = padlock_do_sha1; |
| 198 | |
| 199 | return padlock_cra_init(tfm, sha1_fallback); |
| 200 | } |
| 201 | |
| 202 | static int padlock_sha256_cra_init(struct crypto_tfm *tfm) |
| 203 | { |
| 204 | ctx(tfm)->f_sha_padlock = padlock_do_sha256; |
| 205 | |
| 206 | return padlock_cra_init(tfm, sha256_fallback); |
| 207 | } |
| 208 | |
| 209 | static void padlock_cra_exit(struct crypto_tfm *tfm) |
| 210 | { |
| 211 | if (ctx(tfm)->data) { |
| 212 | free_page((unsigned long)(ctx(tfm)->data)); |
| 213 | ctx(tfm)->data = NULL; |
| 214 | } |
| 215 | |
| 216 | BUG_ON(!ctx(tfm)->fallback_tfm); |
| 217 | crypto_free_tfm(ctx(tfm)->fallback_tfm); |
| 218 | ctx(tfm)->fallback_tfm = NULL; |
| 219 | } |
| 220 | |
| 221 | static struct crypto_alg sha1_alg = { |
| 222 | .cra_name = "sha1", |
| 223 | .cra_driver_name = "sha1-padlock", |
| 224 | .cra_priority = PADLOCK_CRA_PRIORITY, |
| 225 | .cra_flags = CRYPTO_ALG_TYPE_DIGEST, |
| 226 | .cra_blocksize = SHA1_HMAC_BLOCK_SIZE, |
| 227 | .cra_ctxsize = sizeof(struct padlock_sha_ctx), |
| 228 | .cra_module = THIS_MODULE, |
| 229 | .cra_list = LIST_HEAD_INIT(sha1_alg.cra_list), |
| 230 | .cra_init = padlock_sha1_cra_init, |
| 231 | .cra_exit = padlock_cra_exit, |
| 232 | .cra_u = { |
| 233 | .digest = { |
| 234 | .dia_digestsize = SHA1_DIGEST_SIZE, |
| 235 | .dia_init = padlock_sha_init, |
| 236 | .dia_update = padlock_sha_update, |
| 237 | .dia_final = padlock_sha_final, |
| 238 | } |
| 239 | } |
| 240 | }; |
| 241 | |
| 242 | static struct crypto_alg sha256_alg = { |
| 243 | .cra_name = "sha256", |
| 244 | .cra_driver_name = "sha256-padlock", |
| 245 | .cra_priority = PADLOCK_CRA_PRIORITY, |
| 246 | .cra_flags = CRYPTO_ALG_TYPE_DIGEST, |
| 247 | .cra_blocksize = SHA256_HMAC_BLOCK_SIZE, |
| 248 | .cra_ctxsize = sizeof(struct padlock_sha_ctx), |
| 249 | .cra_module = THIS_MODULE, |
| 250 | .cra_list = LIST_HEAD_INIT(sha256_alg.cra_list), |
| 251 | .cra_init = padlock_sha256_cra_init, |
| 252 | .cra_exit = padlock_cra_exit, |
| 253 | .cra_u = { |
| 254 | .digest = { |
| 255 | .dia_digestsize = SHA256_DIGEST_SIZE, |
| 256 | .dia_init = padlock_sha_init, |
| 257 | .dia_update = padlock_sha_update, |
| 258 | .dia_final = padlock_sha_final, |
| 259 | } |
| 260 | } |
| 261 | }; |
| 262 | |
| 263 | static void __init padlock_sha_check_fallbacks(void) |
| 264 | { |
| 265 | static struct crypto_tfm *tfm_sha1, *tfm_sha256; |
| 266 | |
| 267 | /* We'll try to allocate one TFM for each fallback |
| 268 | * to test that the modules are available. */ |
| 269 | tfm_sha1 = crypto_alloc_tfm(sha1_fallback, 0); |
| 270 | if (!tfm_sha1) { |
| 271 | printk(KERN_WARNING PFX "Couldn't load fallback module for '%s'. Tried '%s'.\n", |
| 272 | sha1_alg.cra_name, sha1_fallback); |
| 273 | } else { |
| 274 | printk(KERN_NOTICE PFX "Fallback for '%s' is driver '%s' (prio=%d)\n", sha1_alg.cra_name, |
| 275 | crypto_tfm_alg_driver_name(tfm_sha1), crypto_tfm_alg_priority(tfm_sha1)); |
| 276 | crypto_free_tfm(tfm_sha1); |
| 277 | } |
| 278 | |
| 279 | tfm_sha256 = crypto_alloc_tfm(sha256_fallback, 0); |
| 280 | if (!tfm_sha256) { |
| 281 | printk(KERN_WARNING PFX "Couldn't load fallback module for '%s'. Tried '%s'.\n", |
| 282 | sha256_alg.cra_name, sha256_fallback); |
| 283 | } else { |
| 284 | printk(KERN_NOTICE PFX "Fallback for '%s' is driver '%s' (prio=%d)\n", sha256_alg.cra_name, |
| 285 | crypto_tfm_alg_driver_name(tfm_sha256), crypto_tfm_alg_priority(tfm_sha256)); |
| 286 | crypto_free_tfm(tfm_sha256); |
| 287 | } |
| 288 | } |
| 289 | |
| 290 | static int __init padlock_init(void) |
| 291 | { |
| 292 | int rc = -ENODEV; |
| 293 | |
| 294 | if (!cpu_has_phe) { |
| 295 | printk(KERN_ERR PFX "VIA PadLock Hash Engine not detected.\n"); |
| 296 | return -ENODEV; |
| 297 | } |
| 298 | |
| 299 | if (!cpu_has_phe_enabled) { |
| 300 | printk(KERN_ERR PFX "VIA PadLock detected, but not enabled. Hmm, strange...\n"); |
| 301 | return -ENODEV; |
| 302 | } |
| 303 | |
| 304 | padlock_sha_check_fallbacks(); |
| 305 | |
| 306 | rc = crypto_register_alg(&sha1_alg); |
| 307 | if (rc) |
| 308 | goto out; |
| 309 | |
| 310 | rc = crypto_register_alg(&sha256_alg); |
| 311 | if (rc) |
| 312 | goto out_unreg1; |
| 313 | |
| 314 | printk(KERN_NOTICE PFX "Using VIA PadLock ACE for SHA1/SHA256 algorithms.\n"); |
| 315 | |
| 316 | return 0; |
| 317 | |
| 318 | out_unreg1: |
| 319 | crypto_unregister_alg(&sha1_alg); |
| 320 | out: |
| 321 | printk(KERN_ERR PFX "VIA PadLock SHA1/SHA256 initialization failed.\n"); |
| 322 | return rc; |
| 323 | } |
| 324 | |
| 325 | static void __exit padlock_fini(void) |
| 326 | { |
| 327 | crypto_unregister_alg(&sha1_alg); |
| 328 | crypto_unregister_alg(&sha256_alg); |
| 329 | } |
| 330 | |
| 331 | module_init(padlock_init); |
| 332 | module_exit(padlock_fini); |
| 333 | |
| 334 | MODULE_DESCRIPTION("VIA PadLock SHA1/SHA256 algorithms support."); |
| 335 | MODULE_LICENSE("GPL"); |
| 336 | MODULE_AUTHOR("Michal Ludvig"); |
| 337 | |
| 338 | MODULE_ALIAS("sha1-padlock"); |
| 339 | MODULE_ALIAS("sha256-padlock"); |