1 /* 2 * Glue code for AES implementation for SPE instructions (PPC) 3 * 4 * Based on generic implementation. The assembler module takes care 5 * about the SPE registers so it can run from interrupt context. 6 * 7 * Copyright (c) 2015 Markus Stockhausen <stockhausen@collogia.de> 8 * 9 * This program is free software; you can redistribute it and/or modify it 10 * under the terms of the GNU General Public License as published by the Free 11 * Software Foundation; either version 2 of the License, or (at your option) 12 * any later version. 13 * 14 */ 15 16 #include <crypto/aes.h> 17 #include <linux/module.h> 18 #include <linux/init.h> 19 #include <linux/types.h> 20 #include <linux/errno.h> 21 #include <linux/crypto.h> 22 #include <asm/byteorder.h> 23 #include <asm/switch_to.h> 24 #include <crypto/algapi.h> 25 26 /* 27 * MAX_BYTES defines the number of bytes that are allowed to be processed 28 * between preempt_disable() and preempt_enable(). e500 cores can issue two 29 * instructions per clock cycle using one 32/64 bit unit (SU1) and one 32 30 * bit unit (SU2). One of these can be a memory access that is executed via 31 * a single load and store unit (LSU). XTS-AES-256 takes ~780 operations per 32 * 16 byte block block or 25 cycles per byte. Thus 768 bytes of input data 33 * will need an estimated maximum of 20,000 cycles. Headroom for cache misses 34 * included. Even with the low end model clocked at 667 MHz this equals to a 35 * critical time window of less than 30us. The value has been choosen to 36 * process a 512 byte disk block in one or a large 1400 bytes IPsec network 37 * packet in two runs. 38 * 39 */ 40 #define MAX_BYTES 768 41 42 struct ppc_aes_ctx { 43 u32 key_enc[AES_MAX_KEYLENGTH_U32]; 44 u32 key_dec[AES_MAX_KEYLENGTH_U32]; 45 u32 rounds; 46 }; 47 48 struct ppc_xts_ctx { 49 u32 key_enc[AES_MAX_KEYLENGTH_U32]; 50 u32 key_dec[AES_MAX_KEYLENGTH_U32]; 51 u32 key_twk[AES_MAX_KEYLENGTH_U32]; 52 u32 rounds; 53 }; 54 55 extern void ppc_encrypt_aes(u8 *out, const u8 *in, u32 *key_enc, u32 rounds); 56 extern void ppc_decrypt_aes(u8 *out, const u8 *in, u32 *key_dec, u32 rounds); 57 extern void ppc_encrypt_ecb(u8 *out, const u8 *in, u32 *key_enc, u32 rounds, 58 u32 bytes); 59 extern void ppc_decrypt_ecb(u8 *out, const u8 *in, u32 *key_dec, u32 rounds, 60 u32 bytes); 61 extern void ppc_encrypt_cbc(u8 *out, const u8 *in, u32 *key_enc, u32 rounds, 62 u32 bytes, u8 *iv); 63 extern void ppc_decrypt_cbc(u8 *out, const u8 *in, u32 *key_dec, u32 rounds, 64 u32 bytes, u8 *iv); 65 extern void ppc_crypt_ctr (u8 *out, const u8 *in, u32 *key_enc, u32 rounds, 66 u32 bytes, u8 *iv); 67 extern void ppc_encrypt_xts(u8 *out, const u8 *in, u32 *key_enc, u32 rounds, 68 u32 bytes, u8 *iv, u32 *key_twk); 69 extern void ppc_decrypt_xts(u8 *out, const u8 *in, u32 *key_dec, u32 rounds, 70 u32 bytes, u8 *iv, u32 *key_twk); 71 72 extern void ppc_expand_key_128(u32 *key_enc, const u8 *key); 73 extern void ppc_expand_key_192(u32 *key_enc, const u8 *key); 74 extern void ppc_expand_key_256(u32 *key_enc, const u8 *key); 75 76 extern void ppc_generate_decrypt_key(u32 *key_dec,u32 *key_enc, 77 unsigned int key_len); 78 79 static void spe_begin(void) 80 { 81 /* disable preemption and save users SPE registers if required */ 82 preempt_disable(); 83 enable_kernel_spe(); 84 } 85 86 static void spe_end(void) 87 { 88 /* reenable preemption */ 89 preempt_enable(); 90 } 91 92 static int ppc_aes_setkey(struct crypto_tfm *tfm, const u8 *in_key, 93 unsigned int key_len) 94 { 95 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm); 96 97 if (key_len != AES_KEYSIZE_128 && 98 key_len != AES_KEYSIZE_192 && 99 key_len != AES_KEYSIZE_256) { 100 tfm->crt_flags |= CRYPTO_TFM_RES_BAD_KEY_LEN; 101 return -EINVAL; 102 } 103 104 switch (key_len) { 105 case AES_KEYSIZE_128: 106 ctx->rounds = 4; 107 ppc_expand_key_128(ctx->key_enc, in_key); 108 break; 109 case AES_KEYSIZE_192: 110 ctx->rounds = 5; 111 ppc_expand_key_192(ctx->key_enc, in_key); 112 break; 113 case AES_KEYSIZE_256: 114 ctx->rounds = 6; 115 ppc_expand_key_256(ctx->key_enc, in_key); 116 break; 117 } 118 119 ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len); 120 121 return 0; 122 } 123 124 static int ppc_xts_setkey(struct crypto_tfm *tfm, const u8 *in_key, 125 unsigned int key_len) 126 { 127 struct ppc_xts_ctx *ctx = crypto_tfm_ctx(tfm); 128 129 key_len >>= 1; 130 131 if (key_len != AES_KEYSIZE_128 && 132 key_len != AES_KEYSIZE_192 && 133 key_len != AES_KEYSIZE_256) { 134 tfm->crt_flags |= CRYPTO_TFM_RES_BAD_KEY_LEN; 135 return -EINVAL; 136 } 137 138 switch (key_len) { 139 case AES_KEYSIZE_128: 140 ctx->rounds = 4; 141 ppc_expand_key_128(ctx->key_enc, in_key); 142 ppc_expand_key_128(ctx->key_twk, in_key + AES_KEYSIZE_128); 143 break; 144 case AES_KEYSIZE_192: 145 ctx->rounds = 5; 146 ppc_expand_key_192(ctx->key_enc, in_key); 147 ppc_expand_key_192(ctx->key_twk, in_key + AES_KEYSIZE_192); 148 break; 149 case AES_KEYSIZE_256: 150 ctx->rounds = 6; 151 ppc_expand_key_256(ctx->key_enc, in_key); 152 ppc_expand_key_256(ctx->key_twk, in_key + AES_KEYSIZE_256); 153 break; 154 } 155 156 ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len); 157 158 return 0; 159 } 160 161 static void ppc_aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 162 { 163 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm); 164 165 spe_begin(); 166 ppc_encrypt_aes(out, in, ctx->key_enc, ctx->rounds); 167 spe_end(); 168 } 169 170 static void ppc_aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in) 171 { 172 struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm); 173 174 spe_begin(); 175 ppc_decrypt_aes(out, in, ctx->key_dec, ctx->rounds); 176 spe_end(); 177 } 178 179 static int ppc_ecb_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 180 struct scatterlist *src, unsigned int nbytes) 181 { 182 struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 183 struct blkcipher_walk walk; 184 unsigned int ubytes; 185 int err; 186 187 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 188 blkcipher_walk_init(&walk, dst, src, nbytes); 189 err = blkcipher_walk_virt(desc, &walk); 190 191 while ((nbytes = walk.nbytes)) { 192 ubytes = nbytes > MAX_BYTES ? 193 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 194 nbytes -= ubytes; 195 196 spe_begin(); 197 ppc_encrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr, 198 ctx->key_enc, ctx->rounds, nbytes); 199 spe_end(); 200 201 err = blkcipher_walk_done(desc, &walk, ubytes); 202 } 203 204 return err; 205 } 206 207 static int ppc_ecb_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 208 struct scatterlist *src, unsigned int nbytes) 209 { 210 struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 211 struct blkcipher_walk walk; 212 unsigned int ubytes; 213 int err; 214 215 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 216 blkcipher_walk_init(&walk, dst, src, nbytes); 217 err = blkcipher_walk_virt(desc, &walk); 218 219 while ((nbytes = walk.nbytes)) { 220 ubytes = nbytes > MAX_BYTES ? 221 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 222 nbytes -= ubytes; 223 224 spe_begin(); 225 ppc_decrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr, 226 ctx->key_dec, ctx->rounds, nbytes); 227 spe_end(); 228 229 err = blkcipher_walk_done(desc, &walk, ubytes); 230 } 231 232 return err; 233 } 234 235 static int ppc_cbc_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 236 struct scatterlist *src, unsigned int nbytes) 237 { 238 struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 239 struct blkcipher_walk walk; 240 unsigned int ubytes; 241 int err; 242 243 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 244 blkcipher_walk_init(&walk, dst, src, nbytes); 245 err = blkcipher_walk_virt(desc, &walk); 246 247 while ((nbytes = walk.nbytes)) { 248 ubytes = nbytes > MAX_BYTES ? 249 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 250 nbytes -= ubytes; 251 252 spe_begin(); 253 ppc_encrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr, 254 ctx->key_enc, ctx->rounds, nbytes, walk.iv); 255 spe_end(); 256 257 err = blkcipher_walk_done(desc, &walk, ubytes); 258 } 259 260 return err; 261 } 262 263 static int ppc_cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 264 struct scatterlist *src, unsigned int nbytes) 265 { 266 struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 267 struct blkcipher_walk walk; 268 unsigned int ubytes; 269 int err; 270 271 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 272 blkcipher_walk_init(&walk, dst, src, nbytes); 273 err = blkcipher_walk_virt(desc, &walk); 274 275 while ((nbytes = walk.nbytes)) { 276 ubytes = nbytes > MAX_BYTES ? 277 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 278 nbytes -= ubytes; 279 280 spe_begin(); 281 ppc_decrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr, 282 ctx->key_dec, ctx->rounds, nbytes, walk.iv); 283 spe_end(); 284 285 err = blkcipher_walk_done(desc, &walk, ubytes); 286 } 287 288 return err; 289 } 290 291 static int ppc_ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst, 292 struct scatterlist *src, unsigned int nbytes) 293 { 294 struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 295 struct blkcipher_walk walk; 296 unsigned int pbytes, ubytes; 297 int err; 298 299 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 300 blkcipher_walk_init(&walk, dst, src, nbytes); 301 err = blkcipher_walk_virt_block(desc, &walk, AES_BLOCK_SIZE); 302 303 while ((pbytes = walk.nbytes)) { 304 pbytes = pbytes > MAX_BYTES ? MAX_BYTES : pbytes; 305 pbytes = pbytes == nbytes ? 306 nbytes : pbytes & ~(AES_BLOCK_SIZE - 1); 307 ubytes = walk.nbytes - pbytes; 308 309 spe_begin(); 310 ppc_crypt_ctr(walk.dst.virt.addr, walk.src.virt.addr, 311 ctx->key_enc, ctx->rounds, pbytes , walk.iv); 312 spe_end(); 313 314 nbytes -= pbytes; 315 err = blkcipher_walk_done(desc, &walk, ubytes); 316 } 317 318 return err; 319 } 320 321 static int ppc_xts_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 322 struct scatterlist *src, unsigned int nbytes) 323 { 324 struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 325 struct blkcipher_walk walk; 326 unsigned int ubytes; 327 int err; 328 u32 *twk; 329 330 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 331 blkcipher_walk_init(&walk, dst, src, nbytes); 332 err = blkcipher_walk_virt(desc, &walk); 333 twk = ctx->key_twk; 334 335 while ((nbytes = walk.nbytes)) { 336 ubytes = nbytes > MAX_BYTES ? 337 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 338 nbytes -= ubytes; 339 340 spe_begin(); 341 ppc_encrypt_xts(walk.dst.virt.addr, walk.src.virt.addr, 342 ctx->key_enc, ctx->rounds, nbytes, walk.iv, twk); 343 spe_end(); 344 345 twk = NULL; 346 err = blkcipher_walk_done(desc, &walk, ubytes); 347 } 348 349 return err; 350 } 351 352 static int ppc_xts_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst, 353 struct scatterlist *src, unsigned int nbytes) 354 { 355 struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); 356 struct blkcipher_walk walk; 357 unsigned int ubytes; 358 int err; 359 u32 *twk; 360 361 desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP; 362 blkcipher_walk_init(&walk, dst, src, nbytes); 363 err = blkcipher_walk_virt(desc, &walk); 364 twk = ctx->key_twk; 365 366 while ((nbytes = walk.nbytes)) { 367 ubytes = nbytes > MAX_BYTES ? 368 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1); 369 nbytes -= ubytes; 370 371 spe_begin(); 372 ppc_decrypt_xts(walk.dst.virt.addr, walk.src.virt.addr, 373 ctx->key_dec, ctx->rounds, nbytes, walk.iv, twk); 374 spe_end(); 375 376 twk = NULL; 377 err = blkcipher_walk_done(desc, &walk, ubytes); 378 } 379 380 return err; 381 } 382 383 /* 384 * Algorithm definitions. Disabling alignment (cra_alignmask=0) was chosen 385 * because the e500 platform can handle unaligned reads/writes very efficently. 386 * This improves IPsec thoughput by another few percent. Additionally we assume 387 * that AES context is always aligned to at least 8 bytes because it is created 388 * with kmalloc() in the crypto infrastructure 389 * 390 */ 391 static struct crypto_alg aes_algs[] = { { 392 .cra_name = "aes", 393 .cra_driver_name = "aes-ppc-spe", 394 .cra_priority = 300, 395 .cra_flags = CRYPTO_ALG_TYPE_CIPHER, 396 .cra_blocksize = AES_BLOCK_SIZE, 397 .cra_ctxsize = sizeof(struct ppc_aes_ctx), 398 .cra_alignmask = 0, 399 .cra_module = THIS_MODULE, 400 .cra_u = { 401 .cipher = { 402 .cia_min_keysize = AES_MIN_KEY_SIZE, 403 .cia_max_keysize = AES_MAX_KEY_SIZE, 404 .cia_setkey = ppc_aes_setkey, 405 .cia_encrypt = ppc_aes_encrypt, 406 .cia_decrypt = ppc_aes_decrypt 407 } 408 } 409 }, { 410 .cra_name = "ecb(aes)", 411 .cra_driver_name = "ecb-ppc-spe", 412 .cra_priority = 300, 413 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, 414 .cra_blocksize = AES_BLOCK_SIZE, 415 .cra_ctxsize = sizeof(struct ppc_aes_ctx), 416 .cra_alignmask = 0, 417 .cra_type = &crypto_blkcipher_type, 418 .cra_module = THIS_MODULE, 419 .cra_u = { 420 .blkcipher = { 421 .min_keysize = AES_MIN_KEY_SIZE, 422 .max_keysize = AES_MAX_KEY_SIZE, 423 .ivsize = AES_BLOCK_SIZE, 424 .setkey = ppc_aes_setkey, 425 .encrypt = ppc_ecb_encrypt, 426 .decrypt = ppc_ecb_decrypt, 427 } 428 } 429 }, { 430 .cra_name = "cbc(aes)", 431 .cra_driver_name = "cbc-ppc-spe", 432 .cra_priority = 300, 433 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, 434 .cra_blocksize = AES_BLOCK_SIZE, 435 .cra_ctxsize = sizeof(struct ppc_aes_ctx), 436 .cra_alignmask = 0, 437 .cra_type = &crypto_blkcipher_type, 438 .cra_module = THIS_MODULE, 439 .cra_u = { 440 .blkcipher = { 441 .min_keysize = AES_MIN_KEY_SIZE, 442 .max_keysize = AES_MAX_KEY_SIZE, 443 .ivsize = AES_BLOCK_SIZE, 444 .setkey = ppc_aes_setkey, 445 .encrypt = ppc_cbc_encrypt, 446 .decrypt = ppc_cbc_decrypt, 447 } 448 } 449 }, { 450 .cra_name = "ctr(aes)", 451 .cra_driver_name = "ctr-ppc-spe", 452 .cra_priority = 300, 453 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, 454 .cra_blocksize = 1, 455 .cra_ctxsize = sizeof(struct ppc_aes_ctx), 456 .cra_alignmask = 0, 457 .cra_type = &crypto_blkcipher_type, 458 .cra_module = THIS_MODULE, 459 .cra_u = { 460 .blkcipher = { 461 .min_keysize = AES_MIN_KEY_SIZE, 462 .max_keysize = AES_MAX_KEY_SIZE, 463 .ivsize = AES_BLOCK_SIZE, 464 .setkey = ppc_aes_setkey, 465 .encrypt = ppc_ctr_crypt, 466 .decrypt = ppc_ctr_crypt, 467 } 468 } 469 }, { 470 .cra_name = "xts(aes)", 471 .cra_driver_name = "xts-ppc-spe", 472 .cra_priority = 300, 473 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, 474 .cra_blocksize = AES_BLOCK_SIZE, 475 .cra_ctxsize = sizeof(struct ppc_xts_ctx), 476 .cra_alignmask = 0, 477 .cra_type = &crypto_blkcipher_type, 478 .cra_module = THIS_MODULE, 479 .cra_u = { 480 .blkcipher = { 481 .min_keysize = AES_MIN_KEY_SIZE * 2, 482 .max_keysize = AES_MAX_KEY_SIZE * 2, 483 .ivsize = AES_BLOCK_SIZE, 484 .setkey = ppc_xts_setkey, 485 .encrypt = ppc_xts_encrypt, 486 .decrypt = ppc_xts_decrypt, 487 } 488 } 489 } }; 490 491 static int __init ppc_aes_mod_init(void) 492 { 493 return crypto_register_algs(aes_algs, ARRAY_SIZE(aes_algs)); 494 } 495 496 static void __exit ppc_aes_mod_fini(void) 497 { 498 crypto_unregister_algs(aes_algs, ARRAY_SIZE(aes_algs)); 499 } 500 501 module_init(ppc_aes_mod_init); 502 module_exit(ppc_aes_mod_fini); 503 504 MODULE_LICENSE("GPL"); 505 MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS, SPE optimized"); 506 507 MODULE_ALIAS_CRYPTO("aes"); 508 MODULE_ALIAS_CRYPTO("ecb(aes)"); 509 MODULE_ALIAS_CRYPTO("cbc(aes)"); 510 MODULE_ALIAS_CRYPTO("ctr(aes)"); 511 MODULE_ALIAS_CRYPTO("xts(aes)"); 512 MODULE_ALIAS_CRYPTO("aes-ppc-spe"); 513