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 <crypto/internal/hash.h> 16 #include <crypto/padlock.h> 17 #include <crypto/sha.h> 18 #include <linux/err.h> 19 #include <linux/module.h> 20 #include <linux/init.h> 21 #include <linux/errno.h> 22 #include <linux/interrupt.h> 23 #include <linux/kernel.h> 24 #include <linux/scatterlist.h> 25 #include <asm/cpu_device_id.h> 26 #include <asm/fpu/api.h> 27 28 struct padlock_sha_desc { 29 struct shash_desc fallback; 30 }; 31 32 struct padlock_sha_ctx { 33 struct crypto_shash *fallback; 34 }; 35 36 static int padlock_sha_init(struct shash_desc *desc) 37 { 38 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 39 struct padlock_sha_ctx *ctx = crypto_shash_ctx(desc->tfm); 40 41 dctx->fallback.tfm = ctx->fallback; 42 dctx->fallback.flags = desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP; 43 return crypto_shash_init(&dctx->fallback); 44 } 45 46 static int padlock_sha_update(struct shash_desc *desc, 47 const u8 *data, unsigned int length) 48 { 49 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 50 51 dctx->fallback.flags = desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP; 52 return crypto_shash_update(&dctx->fallback, data, length); 53 } 54 55 static int padlock_sha_export(struct shash_desc *desc, void *out) 56 { 57 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 58 59 return crypto_shash_export(&dctx->fallback, out); 60 } 61 62 static int padlock_sha_import(struct shash_desc *desc, const void *in) 63 { 64 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 65 struct padlock_sha_ctx *ctx = crypto_shash_ctx(desc->tfm); 66 67 dctx->fallback.tfm = ctx->fallback; 68 dctx->fallback.flags = desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP; 69 return crypto_shash_import(&dctx->fallback, in); 70 } 71 72 static inline void padlock_output_block(uint32_t *src, 73 uint32_t *dst, size_t count) 74 { 75 while (count--) 76 *dst++ = swab32(*src++); 77 } 78 79 static int padlock_sha1_finup(struct shash_desc *desc, const u8 *in, 80 unsigned int count, u8 *out) 81 { 82 /* We can't store directly to *out as it may be unaligned. */ 83 /* BTW Don't reduce the buffer size below 128 Bytes! 84 * PadLock microcode needs it that big. */ 85 char buf[128 + PADLOCK_ALIGNMENT - STACK_ALIGN] __attribute__ 86 ((aligned(STACK_ALIGN))); 87 char *result = PTR_ALIGN(&buf[0], PADLOCK_ALIGNMENT); 88 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 89 struct sha1_state state; 90 unsigned int space; 91 unsigned int leftover; 92 int err; 93 94 dctx->fallback.flags = desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP; 95 err = crypto_shash_export(&dctx->fallback, &state); 96 if (err) 97 goto out; 98 99 if (state.count + count > ULONG_MAX) 100 return crypto_shash_finup(&dctx->fallback, in, count, out); 101 102 leftover = ((state.count - 1) & (SHA1_BLOCK_SIZE - 1)) + 1; 103 space = SHA1_BLOCK_SIZE - leftover; 104 if (space) { 105 if (count > space) { 106 err = crypto_shash_update(&dctx->fallback, in, space) ?: 107 crypto_shash_export(&dctx->fallback, &state); 108 if (err) 109 goto out; 110 count -= space; 111 in += space; 112 } else { 113 memcpy(state.buffer + leftover, in, count); 114 in = state.buffer; 115 count += leftover; 116 state.count &= ~(SHA1_BLOCK_SIZE - 1); 117 } 118 } 119 120 memcpy(result, &state.state, SHA1_DIGEST_SIZE); 121 122 asm volatile (".byte 0xf3,0x0f,0xa6,0xc8" /* rep xsha1 */ 123 : \ 124 : "c"((unsigned long)state.count + count), \ 125 "a"((unsigned long)state.count), \ 126 "S"(in), "D"(result)); 127 128 padlock_output_block((uint32_t *)result, (uint32_t *)out, 5); 129 130 out: 131 return err; 132 } 133 134 static int padlock_sha1_final(struct shash_desc *desc, u8 *out) 135 { 136 u8 buf[4]; 137 138 return padlock_sha1_finup(desc, buf, 0, out); 139 } 140 141 static int padlock_sha256_finup(struct shash_desc *desc, const u8 *in, 142 unsigned int count, u8 *out) 143 { 144 /* We can't store directly to *out as it may be unaligned. */ 145 /* BTW Don't reduce the buffer size below 128 Bytes! 146 * PadLock microcode needs it that big. */ 147 char buf[128 + PADLOCK_ALIGNMENT - STACK_ALIGN] __attribute__ 148 ((aligned(STACK_ALIGN))); 149 char *result = PTR_ALIGN(&buf[0], PADLOCK_ALIGNMENT); 150 struct padlock_sha_desc *dctx = shash_desc_ctx(desc); 151 struct sha256_state state; 152 unsigned int space; 153 unsigned int leftover; 154 int err; 155 156 dctx->fallback.flags = desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP; 157 err = crypto_shash_export(&dctx->fallback, &state); 158 if (err) 159 goto out; 160 161 if (state.count + count > ULONG_MAX) 162 return crypto_shash_finup(&dctx->fallback, in, count, out); 163 164 leftover = ((state.count - 1) & (SHA256_BLOCK_SIZE - 1)) + 1; 165 space = SHA256_BLOCK_SIZE - leftover; 166 if (space) { 167 if (count > space) { 168 err = crypto_shash_update(&dctx->fallback, in, space) ?: 169 crypto_shash_export(&dctx->fallback, &state); 170 if (err) 171 goto out; 172 count -= space; 173 in += space; 174 } else { 175 memcpy(state.buf + leftover, in, count); 176 in = state.buf; 177 count += leftover; 178 state.count &= ~(SHA1_BLOCK_SIZE - 1); 179 } 180 } 181 182 memcpy(result, &state.state, SHA256_DIGEST_SIZE); 183 184 asm volatile (".byte 0xf3,0x0f,0xa6,0xd0" /* rep xsha256 */ 185 : \ 186 : "c"((unsigned long)state.count + count), \ 187 "a"((unsigned long)state.count), \ 188 "S"(in), "D"(result)); 189 190 padlock_output_block((uint32_t *)result, (uint32_t *)out, 8); 191 192 out: 193 return err; 194 } 195 196 static int padlock_sha256_final(struct shash_desc *desc, u8 *out) 197 { 198 u8 buf[4]; 199 200 return padlock_sha256_finup(desc, buf, 0, out); 201 } 202 203 static int padlock_cra_init(struct crypto_tfm *tfm) 204 { 205 struct crypto_shash *hash = __crypto_shash_cast(tfm); 206 const char *fallback_driver_name = crypto_tfm_alg_name(tfm); 207 struct padlock_sha_ctx *ctx = crypto_tfm_ctx(tfm); 208 struct crypto_shash *fallback_tfm; 209 int err = -ENOMEM; 210 211 /* Allocate a fallback and abort if it failed. */ 212 fallback_tfm = crypto_alloc_shash(fallback_driver_name, 0, 213 CRYPTO_ALG_NEED_FALLBACK); 214 if (IS_ERR(fallback_tfm)) { 215 printk(KERN_WARNING PFX "Fallback driver '%s' could not be loaded!\n", 216 fallback_driver_name); 217 err = PTR_ERR(fallback_tfm); 218 goto out; 219 } 220 221 ctx->fallback = fallback_tfm; 222 hash->descsize += crypto_shash_descsize(fallback_tfm); 223 return 0; 224 225 out: 226 return err; 227 } 228 229 static void padlock_cra_exit(struct crypto_tfm *tfm) 230 { 231 struct padlock_sha_ctx *ctx = crypto_tfm_ctx(tfm); 232 233 crypto_free_shash(ctx->fallback); 234 } 235 236 static struct shash_alg sha1_alg = { 237 .digestsize = SHA1_DIGEST_SIZE, 238 .init = padlock_sha_init, 239 .update = padlock_sha_update, 240 .finup = padlock_sha1_finup, 241 .final = padlock_sha1_final, 242 .export = padlock_sha_export, 243 .import = padlock_sha_import, 244 .descsize = sizeof(struct padlock_sha_desc), 245 .statesize = sizeof(struct sha1_state), 246 .base = { 247 .cra_name = "sha1", 248 .cra_driver_name = "sha1-padlock", 249 .cra_priority = PADLOCK_CRA_PRIORITY, 250 .cra_flags = CRYPTO_ALG_NEED_FALLBACK, 251 .cra_blocksize = SHA1_BLOCK_SIZE, 252 .cra_ctxsize = sizeof(struct padlock_sha_ctx), 253 .cra_module = THIS_MODULE, 254 .cra_init = padlock_cra_init, 255 .cra_exit = padlock_cra_exit, 256 } 257 }; 258 259 static struct shash_alg sha256_alg = { 260 .digestsize = SHA256_DIGEST_SIZE, 261 .init = padlock_sha_init, 262 .update = padlock_sha_update, 263 .finup = padlock_sha256_finup, 264 .final = padlock_sha256_final, 265 .export = padlock_sha_export, 266 .import = padlock_sha_import, 267 .descsize = sizeof(struct padlock_sha_desc), 268 .statesize = sizeof(struct sha256_state), 269 .base = { 270 .cra_name = "sha256", 271 .cra_driver_name = "sha256-padlock", 272 .cra_priority = PADLOCK_CRA_PRIORITY, 273 .cra_flags = CRYPTO_ALG_NEED_FALLBACK, 274 .cra_blocksize = SHA256_BLOCK_SIZE, 275 .cra_ctxsize = sizeof(struct padlock_sha_ctx), 276 .cra_module = THIS_MODULE, 277 .cra_init = padlock_cra_init, 278 .cra_exit = padlock_cra_exit, 279 } 280 }; 281 282 /* Add two shash_alg instance for hardware-implemented * 283 * multiple-parts hash supported by VIA Nano Processor.*/ 284 static int padlock_sha1_init_nano(struct shash_desc *desc) 285 { 286 struct sha1_state *sctx = shash_desc_ctx(desc); 287 288 *sctx = (struct sha1_state){ 289 .state = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 }, 290 }; 291 292 return 0; 293 } 294 295 static int padlock_sha1_update_nano(struct shash_desc *desc, 296 const u8 *data, unsigned int len) 297 { 298 struct sha1_state *sctx = shash_desc_ctx(desc); 299 unsigned int partial, done; 300 const u8 *src; 301 /*The PHE require the out buffer must 128 bytes and 16-bytes aligned*/ 302 u8 buf[128 + PADLOCK_ALIGNMENT - STACK_ALIGN] __attribute__ 303 ((aligned(STACK_ALIGN))); 304 u8 *dst = PTR_ALIGN(&buf[0], PADLOCK_ALIGNMENT); 305 306 partial = sctx->count & 0x3f; 307 sctx->count += len; 308 done = 0; 309 src = data; 310 memcpy(dst, (u8 *)(sctx->state), SHA1_DIGEST_SIZE); 311 312 if ((partial + len) >= SHA1_BLOCK_SIZE) { 313 314 /* Append the bytes in state's buffer to a block to handle */ 315 if (partial) { 316 done = -partial; 317 memcpy(sctx->buffer + partial, data, 318 done + SHA1_BLOCK_SIZE); 319 src = sctx->buffer; 320 asm volatile (".byte 0xf3,0x0f,0xa6,0xc8" 321 : "+S"(src), "+D"(dst) \ 322 : "a"((long)-1), "c"((unsigned long)1)); 323 done += SHA1_BLOCK_SIZE; 324 src = data + done; 325 } 326 327 /* Process the left bytes from the input data */ 328 if (len - done >= SHA1_BLOCK_SIZE) { 329 asm volatile (".byte 0xf3,0x0f,0xa6,0xc8" 330 : "+S"(src), "+D"(dst) 331 : "a"((long)-1), 332 "c"((unsigned long)((len - done) / SHA1_BLOCK_SIZE))); 333 done += ((len - done) - (len - done) % SHA1_BLOCK_SIZE); 334 src = data + done; 335 } 336 partial = 0; 337 } 338 memcpy((u8 *)(sctx->state), dst, SHA1_DIGEST_SIZE); 339 memcpy(sctx->buffer + partial, src, len - done); 340 341 return 0; 342 } 343 344 static int padlock_sha1_final_nano(struct shash_desc *desc, u8 *out) 345 { 346 struct sha1_state *state = (struct sha1_state *)shash_desc_ctx(desc); 347 unsigned int partial, padlen; 348 __be64 bits; 349 static const u8 padding[64] = { 0x80, }; 350 351 bits = cpu_to_be64(state->count << 3); 352 353 /* Pad out to 56 mod 64 */ 354 partial = state->count & 0x3f; 355 padlen = (partial < 56) ? (56 - partial) : ((64+56) - partial); 356 padlock_sha1_update_nano(desc, padding, padlen); 357 358 /* Append length field bytes */ 359 padlock_sha1_update_nano(desc, (const u8 *)&bits, sizeof(bits)); 360 361 /* Swap to output */ 362 padlock_output_block((uint32_t *)(state->state), (uint32_t *)out, 5); 363 364 return 0; 365 } 366 367 static int padlock_sha256_init_nano(struct shash_desc *desc) 368 { 369 struct sha256_state *sctx = shash_desc_ctx(desc); 370 371 *sctx = (struct sha256_state){ 372 .state = { SHA256_H0, SHA256_H1, SHA256_H2, SHA256_H3, \ 373 SHA256_H4, SHA256_H5, SHA256_H6, SHA256_H7}, 374 }; 375 376 return 0; 377 } 378 379 static int padlock_sha256_update_nano(struct shash_desc *desc, const u8 *data, 380 unsigned int len) 381 { 382 struct sha256_state *sctx = shash_desc_ctx(desc); 383 unsigned int partial, done; 384 const u8 *src; 385 /*The PHE require the out buffer must 128 bytes and 16-bytes aligned*/ 386 u8 buf[128 + PADLOCK_ALIGNMENT - STACK_ALIGN] __attribute__ 387 ((aligned(STACK_ALIGN))); 388 u8 *dst = PTR_ALIGN(&buf[0], PADLOCK_ALIGNMENT); 389 390 partial = sctx->count & 0x3f; 391 sctx->count += len; 392 done = 0; 393 src = data; 394 memcpy(dst, (u8 *)(sctx->state), SHA256_DIGEST_SIZE); 395 396 if ((partial + len) >= SHA256_BLOCK_SIZE) { 397 398 /* Append the bytes in state's buffer to a block to handle */ 399 if (partial) { 400 done = -partial; 401 memcpy(sctx->buf + partial, data, 402 done + SHA256_BLOCK_SIZE); 403 src = sctx->buf; 404 asm volatile (".byte 0xf3,0x0f,0xa6,0xd0" 405 : "+S"(src), "+D"(dst) 406 : "a"((long)-1), "c"((unsigned long)1)); 407 done += SHA256_BLOCK_SIZE; 408 src = data + done; 409 } 410 411 /* Process the left bytes from input data*/ 412 if (len - done >= SHA256_BLOCK_SIZE) { 413 asm volatile (".byte 0xf3,0x0f,0xa6,0xd0" 414 : "+S"(src), "+D"(dst) 415 : "a"((long)-1), 416 "c"((unsigned long)((len - done) / 64))); 417 done += ((len - done) - (len - done) % 64); 418 src = data + done; 419 } 420 partial = 0; 421 } 422 memcpy((u8 *)(sctx->state), dst, SHA256_DIGEST_SIZE); 423 memcpy(sctx->buf + partial, src, len - done); 424 425 return 0; 426 } 427 428 static int padlock_sha256_final_nano(struct shash_desc *desc, u8 *out) 429 { 430 struct sha256_state *state = 431 (struct sha256_state *)shash_desc_ctx(desc); 432 unsigned int partial, padlen; 433 __be64 bits; 434 static const u8 padding[64] = { 0x80, }; 435 436 bits = cpu_to_be64(state->count << 3); 437 438 /* Pad out to 56 mod 64 */ 439 partial = state->count & 0x3f; 440 padlen = (partial < 56) ? (56 - partial) : ((64+56) - partial); 441 padlock_sha256_update_nano(desc, padding, padlen); 442 443 /* Append length field bytes */ 444 padlock_sha256_update_nano(desc, (const u8 *)&bits, sizeof(bits)); 445 446 /* Swap to output */ 447 padlock_output_block((uint32_t *)(state->state), (uint32_t *)out, 8); 448 449 return 0; 450 } 451 452 static int padlock_sha_export_nano(struct shash_desc *desc, 453 void *out) 454 { 455 int statesize = crypto_shash_statesize(desc->tfm); 456 void *sctx = shash_desc_ctx(desc); 457 458 memcpy(out, sctx, statesize); 459 return 0; 460 } 461 462 static int padlock_sha_import_nano(struct shash_desc *desc, 463 const void *in) 464 { 465 int statesize = crypto_shash_statesize(desc->tfm); 466 void *sctx = shash_desc_ctx(desc); 467 468 memcpy(sctx, in, statesize); 469 return 0; 470 } 471 472 static struct shash_alg sha1_alg_nano = { 473 .digestsize = SHA1_DIGEST_SIZE, 474 .init = padlock_sha1_init_nano, 475 .update = padlock_sha1_update_nano, 476 .final = padlock_sha1_final_nano, 477 .export = padlock_sha_export_nano, 478 .import = padlock_sha_import_nano, 479 .descsize = sizeof(struct sha1_state), 480 .statesize = sizeof(struct sha1_state), 481 .base = { 482 .cra_name = "sha1", 483 .cra_driver_name = "sha1-padlock-nano", 484 .cra_priority = PADLOCK_CRA_PRIORITY, 485 .cra_blocksize = SHA1_BLOCK_SIZE, 486 .cra_module = THIS_MODULE, 487 } 488 }; 489 490 static struct shash_alg sha256_alg_nano = { 491 .digestsize = SHA256_DIGEST_SIZE, 492 .init = padlock_sha256_init_nano, 493 .update = padlock_sha256_update_nano, 494 .final = padlock_sha256_final_nano, 495 .export = padlock_sha_export_nano, 496 .import = padlock_sha_import_nano, 497 .descsize = sizeof(struct sha256_state), 498 .statesize = sizeof(struct sha256_state), 499 .base = { 500 .cra_name = "sha256", 501 .cra_driver_name = "sha256-padlock-nano", 502 .cra_priority = PADLOCK_CRA_PRIORITY, 503 .cra_blocksize = SHA256_BLOCK_SIZE, 504 .cra_module = THIS_MODULE, 505 } 506 }; 507 508 static const struct x86_cpu_id padlock_sha_ids[] = { 509 X86_FEATURE_MATCH(X86_FEATURE_PHE), 510 {} 511 }; 512 MODULE_DEVICE_TABLE(x86cpu, padlock_sha_ids); 513 514 static int __init padlock_init(void) 515 { 516 int rc = -ENODEV; 517 struct cpuinfo_x86 *c = &cpu_data(0); 518 struct shash_alg *sha1; 519 struct shash_alg *sha256; 520 521 if (!x86_match_cpu(padlock_sha_ids) || !boot_cpu_has(X86_FEATURE_PHE_EN)) 522 return -ENODEV; 523 524 /* Register the newly added algorithm module if on * 525 * VIA Nano processor, or else just do as before */ 526 if (c->x86_model < 0x0f) { 527 sha1 = &sha1_alg; 528 sha256 = &sha256_alg; 529 } else { 530 sha1 = &sha1_alg_nano; 531 sha256 = &sha256_alg_nano; 532 } 533 534 rc = crypto_register_shash(sha1); 535 if (rc) 536 goto out; 537 538 rc = crypto_register_shash(sha256); 539 if (rc) 540 goto out_unreg1; 541 542 printk(KERN_NOTICE PFX "Using VIA PadLock ACE for SHA1/SHA256 algorithms.\n"); 543 544 return 0; 545 546 out_unreg1: 547 crypto_unregister_shash(sha1); 548 549 out: 550 printk(KERN_ERR PFX "VIA PadLock SHA1/SHA256 initialization failed.\n"); 551 return rc; 552 } 553 554 static void __exit padlock_fini(void) 555 { 556 struct cpuinfo_x86 *c = &cpu_data(0); 557 558 if (c->x86_model >= 0x0f) { 559 crypto_unregister_shash(&sha1_alg_nano); 560 crypto_unregister_shash(&sha256_alg_nano); 561 } else { 562 crypto_unregister_shash(&sha1_alg); 563 crypto_unregister_shash(&sha256_alg); 564 } 565 } 566 567 module_init(padlock_init); 568 module_exit(padlock_fini); 569 570 MODULE_DESCRIPTION("VIA PadLock SHA1/SHA256 algorithms support."); 571 MODULE_LICENSE("GPL"); 572 MODULE_AUTHOR("Michal Ludvig"); 573 574 MODULE_ALIAS_CRYPTO("sha1-all"); 575 MODULE_ALIAS_CRYPTO("sha256-all"); 576 MODULE_ALIAS_CRYPTO("sha1-padlock"); 577 MODULE_ALIAS_CRYPTO("sha256-padlock"); 578