1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * AMD Cryptographic Coprocessor (CCP) driver 4 * 5 * Copyright (C) 2013,2018 Advanced Micro Devices, Inc. 6 * 7 * Author: Tom Lendacky <thomas.lendacky@amd.com> 8 * Author: Gary R Hook <gary.hook@amd.com> 9 */ 10 11 #include <linux/module.h> 12 #include <linux/kernel.h> 13 #include <linux/pci.h> 14 #include <linux/interrupt.h> 15 #include <crypto/scatterwalk.h> 16 #include <crypto/des.h> 17 #include <linux/ccp.h> 18 19 #include "ccp-dev.h" 20 21 /* SHA initial context values */ 22 static const __be32 ccp_sha1_init[SHA1_DIGEST_SIZE / sizeof(__be32)] = { 23 cpu_to_be32(SHA1_H0), cpu_to_be32(SHA1_H1), 24 cpu_to_be32(SHA1_H2), cpu_to_be32(SHA1_H3), 25 cpu_to_be32(SHA1_H4), 26 }; 27 28 static const __be32 ccp_sha224_init[SHA256_DIGEST_SIZE / sizeof(__be32)] = { 29 cpu_to_be32(SHA224_H0), cpu_to_be32(SHA224_H1), 30 cpu_to_be32(SHA224_H2), cpu_to_be32(SHA224_H3), 31 cpu_to_be32(SHA224_H4), cpu_to_be32(SHA224_H5), 32 cpu_to_be32(SHA224_H6), cpu_to_be32(SHA224_H7), 33 }; 34 35 static const __be32 ccp_sha256_init[SHA256_DIGEST_SIZE / sizeof(__be32)] = { 36 cpu_to_be32(SHA256_H0), cpu_to_be32(SHA256_H1), 37 cpu_to_be32(SHA256_H2), cpu_to_be32(SHA256_H3), 38 cpu_to_be32(SHA256_H4), cpu_to_be32(SHA256_H5), 39 cpu_to_be32(SHA256_H6), cpu_to_be32(SHA256_H7), 40 }; 41 42 static const __be64 ccp_sha384_init[SHA512_DIGEST_SIZE / sizeof(__be64)] = { 43 cpu_to_be64(SHA384_H0), cpu_to_be64(SHA384_H1), 44 cpu_to_be64(SHA384_H2), cpu_to_be64(SHA384_H3), 45 cpu_to_be64(SHA384_H4), cpu_to_be64(SHA384_H5), 46 cpu_to_be64(SHA384_H6), cpu_to_be64(SHA384_H7), 47 }; 48 49 static const __be64 ccp_sha512_init[SHA512_DIGEST_SIZE / sizeof(__be64)] = { 50 cpu_to_be64(SHA512_H0), cpu_to_be64(SHA512_H1), 51 cpu_to_be64(SHA512_H2), cpu_to_be64(SHA512_H3), 52 cpu_to_be64(SHA512_H4), cpu_to_be64(SHA512_H5), 53 cpu_to_be64(SHA512_H6), cpu_to_be64(SHA512_H7), 54 }; 55 56 #define CCP_NEW_JOBID(ccp) ((ccp->vdata->version == CCP_VERSION(3, 0)) ? \ 57 ccp_gen_jobid(ccp) : 0) 58 59 static u32 ccp_gen_jobid(struct ccp_device *ccp) 60 { 61 return atomic_inc_return(&ccp->current_id) & CCP_JOBID_MASK; 62 } 63 64 static void ccp_sg_free(struct ccp_sg_workarea *wa) 65 { 66 if (wa->dma_count) 67 dma_unmap_sg(wa->dma_dev, wa->dma_sg, wa->nents, wa->dma_dir); 68 69 wa->dma_count = 0; 70 } 71 72 static int ccp_init_sg_workarea(struct ccp_sg_workarea *wa, struct device *dev, 73 struct scatterlist *sg, u64 len, 74 enum dma_data_direction dma_dir) 75 { 76 memset(wa, 0, sizeof(*wa)); 77 78 wa->sg = sg; 79 if (!sg) 80 return 0; 81 82 wa->nents = sg_nents_for_len(sg, len); 83 if (wa->nents < 0) 84 return wa->nents; 85 86 wa->bytes_left = len; 87 wa->sg_used = 0; 88 89 if (len == 0) 90 return 0; 91 92 if (dma_dir == DMA_NONE) 93 return 0; 94 95 wa->dma_sg = sg; 96 wa->dma_dev = dev; 97 wa->dma_dir = dma_dir; 98 wa->dma_count = dma_map_sg(dev, sg, wa->nents, dma_dir); 99 if (!wa->dma_count) 100 return -ENOMEM; 101 102 return 0; 103 } 104 105 static void ccp_update_sg_workarea(struct ccp_sg_workarea *wa, unsigned int len) 106 { 107 unsigned int nbytes = min_t(u64, len, wa->bytes_left); 108 109 if (!wa->sg) 110 return; 111 112 wa->sg_used += nbytes; 113 wa->bytes_left -= nbytes; 114 if (wa->sg_used == wa->sg->length) { 115 wa->sg = sg_next(wa->sg); 116 wa->sg_used = 0; 117 } 118 } 119 120 static void ccp_dm_free(struct ccp_dm_workarea *wa) 121 { 122 if (wa->length <= CCP_DMAPOOL_MAX_SIZE) { 123 if (wa->address) 124 dma_pool_free(wa->dma_pool, wa->address, 125 wa->dma.address); 126 } else { 127 if (wa->dma.address) 128 dma_unmap_single(wa->dev, wa->dma.address, wa->length, 129 wa->dma.dir); 130 kfree(wa->address); 131 } 132 133 wa->address = NULL; 134 wa->dma.address = 0; 135 } 136 137 static int ccp_init_dm_workarea(struct ccp_dm_workarea *wa, 138 struct ccp_cmd_queue *cmd_q, 139 unsigned int len, 140 enum dma_data_direction dir) 141 { 142 memset(wa, 0, sizeof(*wa)); 143 144 if (!len) 145 return 0; 146 147 wa->dev = cmd_q->ccp->dev; 148 wa->length = len; 149 150 if (len <= CCP_DMAPOOL_MAX_SIZE) { 151 wa->dma_pool = cmd_q->dma_pool; 152 153 wa->address = dma_pool_alloc(wa->dma_pool, GFP_KERNEL, 154 &wa->dma.address); 155 if (!wa->address) 156 return -ENOMEM; 157 158 wa->dma.length = CCP_DMAPOOL_MAX_SIZE; 159 160 memset(wa->address, 0, CCP_DMAPOOL_MAX_SIZE); 161 } else { 162 wa->address = kzalloc(len, GFP_KERNEL); 163 if (!wa->address) 164 return -ENOMEM; 165 166 wa->dma.address = dma_map_single(wa->dev, wa->address, len, 167 dir); 168 if (dma_mapping_error(wa->dev, wa->dma.address)) 169 return -ENOMEM; 170 171 wa->dma.length = len; 172 } 173 wa->dma.dir = dir; 174 175 return 0; 176 } 177 178 static int ccp_set_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset, 179 struct scatterlist *sg, unsigned int sg_offset, 180 unsigned int len) 181 { 182 WARN_ON(!wa->address); 183 184 if (len > (wa->length - wa_offset)) 185 return -EINVAL; 186 187 scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len, 188 0); 189 return 0; 190 } 191 192 static void ccp_get_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset, 193 struct scatterlist *sg, unsigned int sg_offset, 194 unsigned int len) 195 { 196 WARN_ON(!wa->address); 197 198 scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len, 199 1); 200 } 201 202 static int ccp_reverse_set_dm_area(struct ccp_dm_workarea *wa, 203 unsigned int wa_offset, 204 struct scatterlist *sg, 205 unsigned int sg_offset, 206 unsigned int len) 207 { 208 u8 *p, *q; 209 int rc; 210 211 rc = ccp_set_dm_area(wa, wa_offset, sg, sg_offset, len); 212 if (rc) 213 return rc; 214 215 p = wa->address + wa_offset; 216 q = p + len - 1; 217 while (p < q) { 218 *p = *p ^ *q; 219 *q = *p ^ *q; 220 *p = *p ^ *q; 221 p++; 222 q--; 223 } 224 return 0; 225 } 226 227 static void ccp_reverse_get_dm_area(struct ccp_dm_workarea *wa, 228 unsigned int wa_offset, 229 struct scatterlist *sg, 230 unsigned int sg_offset, 231 unsigned int len) 232 { 233 u8 *p, *q; 234 235 p = wa->address + wa_offset; 236 q = p + len - 1; 237 while (p < q) { 238 *p = *p ^ *q; 239 *q = *p ^ *q; 240 *p = *p ^ *q; 241 p++; 242 q--; 243 } 244 245 ccp_get_dm_area(wa, wa_offset, sg, sg_offset, len); 246 } 247 248 static void ccp_free_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q) 249 { 250 ccp_dm_free(&data->dm_wa); 251 ccp_sg_free(&data->sg_wa); 252 } 253 254 static int ccp_init_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q, 255 struct scatterlist *sg, u64 sg_len, 256 unsigned int dm_len, 257 enum dma_data_direction dir) 258 { 259 int ret; 260 261 memset(data, 0, sizeof(*data)); 262 263 ret = ccp_init_sg_workarea(&data->sg_wa, cmd_q->ccp->dev, sg, sg_len, 264 dir); 265 if (ret) 266 goto e_err; 267 268 ret = ccp_init_dm_workarea(&data->dm_wa, cmd_q, dm_len, dir); 269 if (ret) 270 goto e_err; 271 272 return 0; 273 274 e_err: 275 ccp_free_data(data, cmd_q); 276 277 return ret; 278 } 279 280 static unsigned int ccp_queue_buf(struct ccp_data *data, unsigned int from) 281 { 282 struct ccp_sg_workarea *sg_wa = &data->sg_wa; 283 struct ccp_dm_workarea *dm_wa = &data->dm_wa; 284 unsigned int buf_count, nbytes; 285 286 /* Clear the buffer if setting it */ 287 if (!from) 288 memset(dm_wa->address, 0, dm_wa->length); 289 290 if (!sg_wa->sg) 291 return 0; 292 293 /* Perform the copy operation 294 * nbytes will always be <= UINT_MAX because dm_wa->length is 295 * an unsigned int 296 */ 297 nbytes = min_t(u64, sg_wa->bytes_left, dm_wa->length); 298 scatterwalk_map_and_copy(dm_wa->address, sg_wa->sg, sg_wa->sg_used, 299 nbytes, from); 300 301 /* Update the structures and generate the count */ 302 buf_count = 0; 303 while (sg_wa->bytes_left && (buf_count < dm_wa->length)) { 304 nbytes = min(sg_wa->sg->length - sg_wa->sg_used, 305 dm_wa->length - buf_count); 306 nbytes = min_t(u64, sg_wa->bytes_left, nbytes); 307 308 buf_count += nbytes; 309 ccp_update_sg_workarea(sg_wa, nbytes); 310 } 311 312 return buf_count; 313 } 314 315 static unsigned int ccp_fill_queue_buf(struct ccp_data *data) 316 { 317 return ccp_queue_buf(data, 0); 318 } 319 320 static unsigned int ccp_empty_queue_buf(struct ccp_data *data) 321 { 322 return ccp_queue_buf(data, 1); 323 } 324 325 static void ccp_prepare_data(struct ccp_data *src, struct ccp_data *dst, 326 struct ccp_op *op, unsigned int block_size, 327 bool blocksize_op) 328 { 329 unsigned int sg_src_len, sg_dst_len, op_len; 330 331 /* The CCP can only DMA from/to one address each per operation. This 332 * requires that we find the smallest DMA area between the source 333 * and destination. The resulting len values will always be <= UINT_MAX 334 * because the dma length is an unsigned int. 335 */ 336 sg_src_len = sg_dma_len(src->sg_wa.sg) - src->sg_wa.sg_used; 337 sg_src_len = min_t(u64, src->sg_wa.bytes_left, sg_src_len); 338 339 if (dst) { 340 sg_dst_len = sg_dma_len(dst->sg_wa.sg) - dst->sg_wa.sg_used; 341 sg_dst_len = min_t(u64, src->sg_wa.bytes_left, sg_dst_len); 342 op_len = min(sg_src_len, sg_dst_len); 343 } else { 344 op_len = sg_src_len; 345 } 346 347 /* The data operation length will be at least block_size in length 348 * or the smaller of available sg room remaining for the source or 349 * the destination 350 */ 351 op_len = max(op_len, block_size); 352 353 /* Unless we have to buffer data, there's no reason to wait */ 354 op->soc = 0; 355 356 if (sg_src_len < block_size) { 357 /* Not enough data in the sg element, so it 358 * needs to be buffered into a blocksize chunk 359 */ 360 int cp_len = ccp_fill_queue_buf(src); 361 362 op->soc = 1; 363 op->src.u.dma.address = src->dm_wa.dma.address; 364 op->src.u.dma.offset = 0; 365 op->src.u.dma.length = (blocksize_op) ? block_size : cp_len; 366 } else { 367 /* Enough data in the sg element, but we need to 368 * adjust for any previously copied data 369 */ 370 op->src.u.dma.address = sg_dma_address(src->sg_wa.sg); 371 op->src.u.dma.offset = src->sg_wa.sg_used; 372 op->src.u.dma.length = op_len & ~(block_size - 1); 373 374 ccp_update_sg_workarea(&src->sg_wa, op->src.u.dma.length); 375 } 376 377 if (dst) { 378 if (sg_dst_len < block_size) { 379 /* Not enough room in the sg element or we're on the 380 * last piece of data (when using padding), so the 381 * output needs to be buffered into a blocksize chunk 382 */ 383 op->soc = 1; 384 op->dst.u.dma.address = dst->dm_wa.dma.address; 385 op->dst.u.dma.offset = 0; 386 op->dst.u.dma.length = op->src.u.dma.length; 387 } else { 388 /* Enough room in the sg element, but we need to 389 * adjust for any previously used area 390 */ 391 op->dst.u.dma.address = sg_dma_address(dst->sg_wa.sg); 392 op->dst.u.dma.offset = dst->sg_wa.sg_used; 393 op->dst.u.dma.length = op->src.u.dma.length; 394 } 395 } 396 } 397 398 static void ccp_process_data(struct ccp_data *src, struct ccp_data *dst, 399 struct ccp_op *op) 400 { 401 op->init = 0; 402 403 if (dst) { 404 if (op->dst.u.dma.address == dst->dm_wa.dma.address) 405 ccp_empty_queue_buf(dst); 406 else 407 ccp_update_sg_workarea(&dst->sg_wa, 408 op->dst.u.dma.length); 409 } 410 } 411 412 static int ccp_copy_to_from_sb(struct ccp_cmd_queue *cmd_q, 413 struct ccp_dm_workarea *wa, u32 jobid, u32 sb, 414 u32 byte_swap, bool from) 415 { 416 struct ccp_op op; 417 418 memset(&op, 0, sizeof(op)); 419 420 op.cmd_q = cmd_q; 421 op.jobid = jobid; 422 op.eom = 1; 423 424 if (from) { 425 op.soc = 1; 426 op.src.type = CCP_MEMTYPE_SB; 427 op.src.u.sb = sb; 428 op.dst.type = CCP_MEMTYPE_SYSTEM; 429 op.dst.u.dma.address = wa->dma.address; 430 op.dst.u.dma.length = wa->length; 431 } else { 432 op.src.type = CCP_MEMTYPE_SYSTEM; 433 op.src.u.dma.address = wa->dma.address; 434 op.src.u.dma.length = wa->length; 435 op.dst.type = CCP_MEMTYPE_SB; 436 op.dst.u.sb = sb; 437 } 438 439 op.u.passthru.byte_swap = byte_swap; 440 441 return cmd_q->ccp->vdata->perform->passthru(&op); 442 } 443 444 static int ccp_copy_to_sb(struct ccp_cmd_queue *cmd_q, 445 struct ccp_dm_workarea *wa, u32 jobid, u32 sb, 446 u32 byte_swap) 447 { 448 return ccp_copy_to_from_sb(cmd_q, wa, jobid, sb, byte_swap, false); 449 } 450 451 static int ccp_copy_from_sb(struct ccp_cmd_queue *cmd_q, 452 struct ccp_dm_workarea *wa, u32 jobid, u32 sb, 453 u32 byte_swap) 454 { 455 return ccp_copy_to_from_sb(cmd_q, wa, jobid, sb, byte_swap, true); 456 } 457 458 static int ccp_run_aes_cmac_cmd(struct ccp_cmd_queue *cmd_q, 459 struct ccp_cmd *cmd) 460 { 461 struct ccp_aes_engine *aes = &cmd->u.aes; 462 struct ccp_dm_workarea key, ctx; 463 struct ccp_data src; 464 struct ccp_op op; 465 unsigned int dm_offset; 466 int ret; 467 468 if (!((aes->key_len == AES_KEYSIZE_128) || 469 (aes->key_len == AES_KEYSIZE_192) || 470 (aes->key_len == AES_KEYSIZE_256))) 471 return -EINVAL; 472 473 if (aes->src_len & (AES_BLOCK_SIZE - 1)) 474 return -EINVAL; 475 476 if (aes->iv_len != AES_BLOCK_SIZE) 477 return -EINVAL; 478 479 if (!aes->key || !aes->iv || !aes->src) 480 return -EINVAL; 481 482 if (aes->cmac_final) { 483 if (aes->cmac_key_len != AES_BLOCK_SIZE) 484 return -EINVAL; 485 486 if (!aes->cmac_key) 487 return -EINVAL; 488 } 489 490 BUILD_BUG_ON(CCP_AES_KEY_SB_COUNT != 1); 491 BUILD_BUG_ON(CCP_AES_CTX_SB_COUNT != 1); 492 493 ret = -EIO; 494 memset(&op, 0, sizeof(op)); 495 op.cmd_q = cmd_q; 496 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 497 op.sb_key = cmd_q->sb_key; 498 op.sb_ctx = cmd_q->sb_ctx; 499 op.init = 1; 500 op.u.aes.type = aes->type; 501 op.u.aes.mode = aes->mode; 502 op.u.aes.action = aes->action; 503 504 /* All supported key sizes fit in a single (32-byte) SB entry 505 * and must be in little endian format. Use the 256-bit byte 506 * swap passthru option to convert from big endian to little 507 * endian. 508 */ 509 ret = ccp_init_dm_workarea(&key, cmd_q, 510 CCP_AES_KEY_SB_COUNT * CCP_SB_BYTES, 511 DMA_TO_DEVICE); 512 if (ret) 513 return ret; 514 515 dm_offset = CCP_SB_BYTES - aes->key_len; 516 ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len); 517 if (ret) 518 goto e_key; 519 ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, 520 CCP_PASSTHRU_BYTESWAP_256BIT); 521 if (ret) { 522 cmd->engine_error = cmd_q->cmd_error; 523 goto e_key; 524 } 525 526 /* The AES context fits in a single (32-byte) SB entry and 527 * must be in little endian format. Use the 256-bit byte swap 528 * passthru option to convert from big endian to little endian. 529 */ 530 ret = ccp_init_dm_workarea(&ctx, cmd_q, 531 CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES, 532 DMA_BIDIRECTIONAL); 533 if (ret) 534 goto e_key; 535 536 dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE; 537 ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 538 if (ret) 539 goto e_ctx; 540 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 541 CCP_PASSTHRU_BYTESWAP_256BIT); 542 if (ret) { 543 cmd->engine_error = cmd_q->cmd_error; 544 goto e_ctx; 545 } 546 547 /* Send data to the CCP AES engine */ 548 ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len, 549 AES_BLOCK_SIZE, DMA_TO_DEVICE); 550 if (ret) 551 goto e_ctx; 552 553 while (src.sg_wa.bytes_left) { 554 ccp_prepare_data(&src, NULL, &op, AES_BLOCK_SIZE, true); 555 if (aes->cmac_final && !src.sg_wa.bytes_left) { 556 op.eom = 1; 557 558 /* Push the K1/K2 key to the CCP now */ 559 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, 560 op.sb_ctx, 561 CCP_PASSTHRU_BYTESWAP_256BIT); 562 if (ret) { 563 cmd->engine_error = cmd_q->cmd_error; 564 goto e_src; 565 } 566 567 ret = ccp_set_dm_area(&ctx, 0, aes->cmac_key, 0, 568 aes->cmac_key_len); 569 if (ret) 570 goto e_src; 571 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 572 CCP_PASSTHRU_BYTESWAP_256BIT); 573 if (ret) { 574 cmd->engine_error = cmd_q->cmd_error; 575 goto e_src; 576 } 577 } 578 579 ret = cmd_q->ccp->vdata->perform->aes(&op); 580 if (ret) { 581 cmd->engine_error = cmd_q->cmd_error; 582 goto e_src; 583 } 584 585 ccp_process_data(&src, NULL, &op); 586 } 587 588 /* Retrieve the AES context - convert from LE to BE using 589 * 32-byte (256-bit) byteswapping 590 */ 591 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 592 CCP_PASSTHRU_BYTESWAP_256BIT); 593 if (ret) { 594 cmd->engine_error = cmd_q->cmd_error; 595 goto e_src; 596 } 597 598 /* ...but we only need AES_BLOCK_SIZE bytes */ 599 dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE; 600 ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 601 602 e_src: 603 ccp_free_data(&src, cmd_q); 604 605 e_ctx: 606 ccp_dm_free(&ctx); 607 608 e_key: 609 ccp_dm_free(&key); 610 611 return ret; 612 } 613 614 static int ccp_run_aes_gcm_cmd(struct ccp_cmd_queue *cmd_q, 615 struct ccp_cmd *cmd) 616 { 617 struct ccp_aes_engine *aes = &cmd->u.aes; 618 struct ccp_dm_workarea key, ctx, final_wa, tag; 619 struct ccp_data src, dst; 620 struct ccp_data aad; 621 struct ccp_op op; 622 623 unsigned long long *final; 624 unsigned int dm_offset; 625 unsigned int ilen; 626 bool in_place = true; /* Default value */ 627 int ret; 628 629 struct scatterlist *p_inp, sg_inp[2]; 630 struct scatterlist *p_tag, sg_tag[2]; 631 struct scatterlist *p_outp, sg_outp[2]; 632 struct scatterlist *p_aad; 633 634 if (!aes->iv) 635 return -EINVAL; 636 637 if (!((aes->key_len == AES_KEYSIZE_128) || 638 (aes->key_len == AES_KEYSIZE_192) || 639 (aes->key_len == AES_KEYSIZE_256))) 640 return -EINVAL; 641 642 if (!aes->key) /* Gotta have a key SGL */ 643 return -EINVAL; 644 645 /* First, decompose the source buffer into AAD & PT, 646 * and the destination buffer into AAD, CT & tag, or 647 * the input into CT & tag. 648 * It is expected that the input and output SGs will 649 * be valid, even if the AAD and input lengths are 0. 650 */ 651 p_aad = aes->src; 652 p_inp = scatterwalk_ffwd(sg_inp, aes->src, aes->aad_len); 653 p_outp = scatterwalk_ffwd(sg_outp, aes->dst, aes->aad_len); 654 if (aes->action == CCP_AES_ACTION_ENCRYPT) { 655 ilen = aes->src_len; 656 p_tag = scatterwalk_ffwd(sg_tag, p_outp, ilen); 657 } else { 658 /* Input length for decryption includes tag */ 659 ilen = aes->src_len - AES_BLOCK_SIZE; 660 p_tag = scatterwalk_ffwd(sg_tag, p_inp, ilen); 661 } 662 663 memset(&op, 0, sizeof(op)); 664 op.cmd_q = cmd_q; 665 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 666 op.sb_key = cmd_q->sb_key; /* Pre-allocated */ 667 op.sb_ctx = cmd_q->sb_ctx; /* Pre-allocated */ 668 op.init = 1; 669 op.u.aes.type = aes->type; 670 671 /* Copy the key to the LSB */ 672 ret = ccp_init_dm_workarea(&key, cmd_q, 673 CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES, 674 DMA_TO_DEVICE); 675 if (ret) 676 return ret; 677 678 dm_offset = CCP_SB_BYTES - aes->key_len; 679 ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len); 680 if (ret) 681 goto e_key; 682 ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, 683 CCP_PASSTHRU_BYTESWAP_256BIT); 684 if (ret) { 685 cmd->engine_error = cmd_q->cmd_error; 686 goto e_key; 687 } 688 689 /* Copy the context (IV) to the LSB. 690 * There is an assumption here that the IV is 96 bits in length, plus 691 * a nonce of 32 bits. If no IV is present, use a zeroed buffer. 692 */ 693 ret = ccp_init_dm_workarea(&ctx, cmd_q, 694 CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES, 695 DMA_BIDIRECTIONAL); 696 if (ret) 697 goto e_key; 698 699 dm_offset = CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES - aes->iv_len; 700 ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 701 if (ret) 702 goto e_ctx; 703 704 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 705 CCP_PASSTHRU_BYTESWAP_256BIT); 706 if (ret) { 707 cmd->engine_error = cmd_q->cmd_error; 708 goto e_ctx; 709 } 710 711 op.init = 1; 712 if (aes->aad_len > 0) { 713 /* Step 1: Run a GHASH over the Additional Authenticated Data */ 714 ret = ccp_init_data(&aad, cmd_q, p_aad, aes->aad_len, 715 AES_BLOCK_SIZE, 716 DMA_TO_DEVICE); 717 if (ret) 718 goto e_ctx; 719 720 op.u.aes.mode = CCP_AES_MODE_GHASH; 721 op.u.aes.action = CCP_AES_GHASHAAD; 722 723 while (aad.sg_wa.bytes_left) { 724 ccp_prepare_data(&aad, NULL, &op, AES_BLOCK_SIZE, true); 725 726 ret = cmd_q->ccp->vdata->perform->aes(&op); 727 if (ret) { 728 cmd->engine_error = cmd_q->cmd_error; 729 goto e_aad; 730 } 731 732 ccp_process_data(&aad, NULL, &op); 733 op.init = 0; 734 } 735 } 736 737 op.u.aes.mode = CCP_AES_MODE_GCTR; 738 op.u.aes.action = aes->action; 739 740 if (ilen > 0) { 741 /* Step 2: Run a GCTR over the plaintext */ 742 in_place = (sg_virt(p_inp) == sg_virt(p_outp)) ? true : false; 743 744 ret = ccp_init_data(&src, cmd_q, p_inp, ilen, 745 AES_BLOCK_SIZE, 746 in_place ? DMA_BIDIRECTIONAL 747 : DMA_TO_DEVICE); 748 if (ret) 749 goto e_ctx; 750 751 if (in_place) { 752 dst = src; 753 } else { 754 ret = ccp_init_data(&dst, cmd_q, p_outp, ilen, 755 AES_BLOCK_SIZE, DMA_FROM_DEVICE); 756 if (ret) 757 goto e_src; 758 } 759 760 op.soc = 0; 761 op.eom = 0; 762 op.init = 1; 763 while (src.sg_wa.bytes_left) { 764 ccp_prepare_data(&src, &dst, &op, AES_BLOCK_SIZE, true); 765 if (!src.sg_wa.bytes_left) { 766 unsigned int nbytes = aes->src_len 767 % AES_BLOCK_SIZE; 768 769 if (nbytes) { 770 op.eom = 1; 771 op.u.aes.size = (nbytes * 8) - 1; 772 } 773 } 774 775 ret = cmd_q->ccp->vdata->perform->aes(&op); 776 if (ret) { 777 cmd->engine_error = cmd_q->cmd_error; 778 goto e_dst; 779 } 780 781 ccp_process_data(&src, &dst, &op); 782 op.init = 0; 783 } 784 } 785 786 /* Step 3: Update the IV portion of the context with the original IV */ 787 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 788 CCP_PASSTHRU_BYTESWAP_256BIT); 789 if (ret) { 790 cmd->engine_error = cmd_q->cmd_error; 791 goto e_dst; 792 } 793 794 ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 795 if (ret) 796 goto e_dst; 797 798 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 799 CCP_PASSTHRU_BYTESWAP_256BIT); 800 if (ret) { 801 cmd->engine_error = cmd_q->cmd_error; 802 goto e_dst; 803 } 804 805 /* Step 4: Concatenate the lengths of the AAD and source, and 806 * hash that 16 byte buffer. 807 */ 808 ret = ccp_init_dm_workarea(&final_wa, cmd_q, AES_BLOCK_SIZE, 809 DMA_BIDIRECTIONAL); 810 if (ret) 811 goto e_dst; 812 final = (unsigned long long *) final_wa.address; 813 final[0] = cpu_to_be64(aes->aad_len * 8); 814 final[1] = cpu_to_be64(ilen * 8); 815 816 op.u.aes.mode = CCP_AES_MODE_GHASH; 817 op.u.aes.action = CCP_AES_GHASHFINAL; 818 op.src.type = CCP_MEMTYPE_SYSTEM; 819 op.src.u.dma.address = final_wa.dma.address; 820 op.src.u.dma.length = AES_BLOCK_SIZE; 821 op.dst.type = CCP_MEMTYPE_SYSTEM; 822 op.dst.u.dma.address = final_wa.dma.address; 823 op.dst.u.dma.length = AES_BLOCK_SIZE; 824 op.eom = 1; 825 op.u.aes.size = 0; 826 ret = cmd_q->ccp->vdata->perform->aes(&op); 827 if (ret) 828 goto e_dst; 829 830 if (aes->action == CCP_AES_ACTION_ENCRYPT) { 831 /* Put the ciphered tag after the ciphertext. */ 832 ccp_get_dm_area(&final_wa, 0, p_tag, 0, AES_BLOCK_SIZE); 833 } else { 834 /* Does this ciphered tag match the input? */ 835 ret = ccp_init_dm_workarea(&tag, cmd_q, AES_BLOCK_SIZE, 836 DMA_BIDIRECTIONAL); 837 if (ret) 838 goto e_tag; 839 ret = ccp_set_dm_area(&tag, 0, p_tag, 0, AES_BLOCK_SIZE); 840 if (ret) 841 goto e_tag; 842 843 ret = memcmp(tag.address, final_wa.address, AES_BLOCK_SIZE); 844 ccp_dm_free(&tag); 845 } 846 847 e_tag: 848 ccp_dm_free(&final_wa); 849 850 e_dst: 851 if (aes->src_len && !in_place) 852 ccp_free_data(&dst, cmd_q); 853 854 e_src: 855 if (aes->src_len) 856 ccp_free_data(&src, cmd_q); 857 858 e_aad: 859 if (aes->aad_len) 860 ccp_free_data(&aad, cmd_q); 861 862 e_ctx: 863 ccp_dm_free(&ctx); 864 865 e_key: 866 ccp_dm_free(&key); 867 868 return ret; 869 } 870 871 static int ccp_run_aes_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 872 { 873 struct ccp_aes_engine *aes = &cmd->u.aes; 874 struct ccp_dm_workarea key, ctx; 875 struct ccp_data src, dst; 876 struct ccp_op op; 877 unsigned int dm_offset; 878 bool in_place = false; 879 int ret; 880 881 if (aes->mode == CCP_AES_MODE_CMAC) 882 return ccp_run_aes_cmac_cmd(cmd_q, cmd); 883 884 if (aes->mode == CCP_AES_MODE_GCM) 885 return ccp_run_aes_gcm_cmd(cmd_q, cmd); 886 887 if (!((aes->key_len == AES_KEYSIZE_128) || 888 (aes->key_len == AES_KEYSIZE_192) || 889 (aes->key_len == AES_KEYSIZE_256))) 890 return -EINVAL; 891 892 if (((aes->mode == CCP_AES_MODE_ECB) || 893 (aes->mode == CCP_AES_MODE_CBC) || 894 (aes->mode == CCP_AES_MODE_CFB)) && 895 (aes->src_len & (AES_BLOCK_SIZE - 1))) 896 return -EINVAL; 897 898 if (!aes->key || !aes->src || !aes->dst) 899 return -EINVAL; 900 901 if (aes->mode != CCP_AES_MODE_ECB) { 902 if (aes->iv_len != AES_BLOCK_SIZE) 903 return -EINVAL; 904 905 if (!aes->iv) 906 return -EINVAL; 907 } 908 909 BUILD_BUG_ON(CCP_AES_KEY_SB_COUNT != 1); 910 BUILD_BUG_ON(CCP_AES_CTX_SB_COUNT != 1); 911 912 ret = -EIO; 913 memset(&op, 0, sizeof(op)); 914 op.cmd_q = cmd_q; 915 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 916 op.sb_key = cmd_q->sb_key; 917 op.sb_ctx = cmd_q->sb_ctx; 918 op.init = (aes->mode == CCP_AES_MODE_ECB) ? 0 : 1; 919 op.u.aes.type = aes->type; 920 op.u.aes.mode = aes->mode; 921 op.u.aes.action = aes->action; 922 923 /* All supported key sizes fit in a single (32-byte) SB entry 924 * and must be in little endian format. Use the 256-bit byte 925 * swap passthru option to convert from big endian to little 926 * endian. 927 */ 928 ret = ccp_init_dm_workarea(&key, cmd_q, 929 CCP_AES_KEY_SB_COUNT * CCP_SB_BYTES, 930 DMA_TO_DEVICE); 931 if (ret) 932 return ret; 933 934 dm_offset = CCP_SB_BYTES - aes->key_len; 935 ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len); 936 if (ret) 937 goto e_key; 938 ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, 939 CCP_PASSTHRU_BYTESWAP_256BIT); 940 if (ret) { 941 cmd->engine_error = cmd_q->cmd_error; 942 goto e_key; 943 } 944 945 /* The AES context fits in a single (32-byte) SB entry and 946 * must be in little endian format. Use the 256-bit byte swap 947 * passthru option to convert from big endian to little endian. 948 */ 949 ret = ccp_init_dm_workarea(&ctx, cmd_q, 950 CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES, 951 DMA_BIDIRECTIONAL); 952 if (ret) 953 goto e_key; 954 955 if (aes->mode != CCP_AES_MODE_ECB) { 956 /* Load the AES context - convert to LE */ 957 dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE; 958 ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 959 if (ret) 960 goto e_ctx; 961 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 962 CCP_PASSTHRU_BYTESWAP_256BIT); 963 if (ret) { 964 cmd->engine_error = cmd_q->cmd_error; 965 goto e_ctx; 966 } 967 } 968 switch (aes->mode) { 969 case CCP_AES_MODE_CFB: /* CFB128 only */ 970 case CCP_AES_MODE_CTR: 971 op.u.aes.size = AES_BLOCK_SIZE * BITS_PER_BYTE - 1; 972 break; 973 default: 974 op.u.aes.size = 0; 975 } 976 977 /* Prepare the input and output data workareas. For in-place 978 * operations we need to set the dma direction to BIDIRECTIONAL 979 * and copy the src workarea to the dst workarea. 980 */ 981 if (sg_virt(aes->src) == sg_virt(aes->dst)) 982 in_place = true; 983 984 ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len, 985 AES_BLOCK_SIZE, 986 in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); 987 if (ret) 988 goto e_ctx; 989 990 if (in_place) { 991 dst = src; 992 } else { 993 ret = ccp_init_data(&dst, cmd_q, aes->dst, aes->src_len, 994 AES_BLOCK_SIZE, DMA_FROM_DEVICE); 995 if (ret) 996 goto e_src; 997 } 998 999 /* Send data to the CCP AES engine */ 1000 while (src.sg_wa.bytes_left) { 1001 ccp_prepare_data(&src, &dst, &op, AES_BLOCK_SIZE, true); 1002 if (!src.sg_wa.bytes_left) { 1003 op.eom = 1; 1004 1005 /* Since we don't retrieve the AES context in ECB 1006 * mode we have to wait for the operation to complete 1007 * on the last piece of data 1008 */ 1009 if (aes->mode == CCP_AES_MODE_ECB) 1010 op.soc = 1; 1011 } 1012 1013 ret = cmd_q->ccp->vdata->perform->aes(&op); 1014 if (ret) { 1015 cmd->engine_error = cmd_q->cmd_error; 1016 goto e_dst; 1017 } 1018 1019 ccp_process_data(&src, &dst, &op); 1020 } 1021 1022 if (aes->mode != CCP_AES_MODE_ECB) { 1023 /* Retrieve the AES context - convert from LE to BE using 1024 * 32-byte (256-bit) byteswapping 1025 */ 1026 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1027 CCP_PASSTHRU_BYTESWAP_256BIT); 1028 if (ret) { 1029 cmd->engine_error = cmd_q->cmd_error; 1030 goto e_dst; 1031 } 1032 1033 /* ...but we only need AES_BLOCK_SIZE bytes */ 1034 dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE; 1035 ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len); 1036 } 1037 1038 e_dst: 1039 if (!in_place) 1040 ccp_free_data(&dst, cmd_q); 1041 1042 e_src: 1043 ccp_free_data(&src, cmd_q); 1044 1045 e_ctx: 1046 ccp_dm_free(&ctx); 1047 1048 e_key: 1049 ccp_dm_free(&key); 1050 1051 return ret; 1052 } 1053 1054 static int ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, 1055 struct ccp_cmd *cmd) 1056 { 1057 struct ccp_xts_aes_engine *xts = &cmd->u.xts; 1058 struct ccp_dm_workarea key, ctx; 1059 struct ccp_data src, dst; 1060 struct ccp_op op; 1061 unsigned int unit_size, dm_offset; 1062 bool in_place = false; 1063 unsigned int sb_count; 1064 enum ccp_aes_type aestype; 1065 int ret; 1066 1067 switch (xts->unit_size) { 1068 case CCP_XTS_AES_UNIT_SIZE_16: 1069 unit_size = 16; 1070 break; 1071 case CCP_XTS_AES_UNIT_SIZE_512: 1072 unit_size = 512; 1073 break; 1074 case CCP_XTS_AES_UNIT_SIZE_1024: 1075 unit_size = 1024; 1076 break; 1077 case CCP_XTS_AES_UNIT_SIZE_2048: 1078 unit_size = 2048; 1079 break; 1080 case CCP_XTS_AES_UNIT_SIZE_4096: 1081 unit_size = 4096; 1082 break; 1083 1084 default: 1085 return -EINVAL; 1086 } 1087 1088 if (xts->key_len == AES_KEYSIZE_128) 1089 aestype = CCP_AES_TYPE_128; 1090 else if (xts->key_len == AES_KEYSIZE_256) 1091 aestype = CCP_AES_TYPE_256; 1092 else 1093 return -EINVAL; 1094 1095 if (!xts->final && (xts->src_len & (AES_BLOCK_SIZE - 1))) 1096 return -EINVAL; 1097 1098 if (xts->iv_len != AES_BLOCK_SIZE) 1099 return -EINVAL; 1100 1101 if (!xts->key || !xts->iv || !xts->src || !xts->dst) 1102 return -EINVAL; 1103 1104 BUILD_BUG_ON(CCP_XTS_AES_KEY_SB_COUNT != 1); 1105 BUILD_BUG_ON(CCP_XTS_AES_CTX_SB_COUNT != 1); 1106 1107 ret = -EIO; 1108 memset(&op, 0, sizeof(op)); 1109 op.cmd_q = cmd_q; 1110 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 1111 op.sb_key = cmd_q->sb_key; 1112 op.sb_ctx = cmd_q->sb_ctx; 1113 op.init = 1; 1114 op.u.xts.type = aestype; 1115 op.u.xts.action = xts->action; 1116 op.u.xts.unit_size = xts->unit_size; 1117 1118 /* A version 3 device only supports 128-bit keys, which fits into a 1119 * single SB entry. A version 5 device uses a 512-bit vector, so two 1120 * SB entries. 1121 */ 1122 if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) 1123 sb_count = CCP_XTS_AES_KEY_SB_COUNT; 1124 else 1125 sb_count = CCP5_XTS_AES_KEY_SB_COUNT; 1126 ret = ccp_init_dm_workarea(&key, cmd_q, 1127 sb_count * CCP_SB_BYTES, 1128 DMA_TO_DEVICE); 1129 if (ret) 1130 return ret; 1131 1132 if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) { 1133 /* All supported key sizes must be in little endian format. 1134 * Use the 256-bit byte swap passthru option to convert from 1135 * big endian to little endian. 1136 */ 1137 dm_offset = CCP_SB_BYTES - AES_KEYSIZE_128; 1138 ret = ccp_set_dm_area(&key, dm_offset, xts->key, 0, xts->key_len); 1139 if (ret) 1140 goto e_key; 1141 ret = ccp_set_dm_area(&key, 0, xts->key, xts->key_len, xts->key_len); 1142 if (ret) 1143 goto e_key; 1144 } else { 1145 /* Version 5 CCPs use a 512-bit space for the key: each portion 1146 * occupies 256 bits, or one entire slot, and is zero-padded. 1147 */ 1148 unsigned int pad; 1149 1150 dm_offset = CCP_SB_BYTES; 1151 pad = dm_offset - xts->key_len; 1152 ret = ccp_set_dm_area(&key, pad, xts->key, 0, xts->key_len); 1153 if (ret) 1154 goto e_key; 1155 ret = ccp_set_dm_area(&key, dm_offset + pad, xts->key, 1156 xts->key_len, xts->key_len); 1157 if (ret) 1158 goto e_key; 1159 } 1160 ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, 1161 CCP_PASSTHRU_BYTESWAP_256BIT); 1162 if (ret) { 1163 cmd->engine_error = cmd_q->cmd_error; 1164 goto e_key; 1165 } 1166 1167 /* The AES context fits in a single (32-byte) SB entry and 1168 * for XTS is already in little endian format so no byte swapping 1169 * is needed. 1170 */ 1171 ret = ccp_init_dm_workarea(&ctx, cmd_q, 1172 CCP_XTS_AES_CTX_SB_COUNT * CCP_SB_BYTES, 1173 DMA_BIDIRECTIONAL); 1174 if (ret) 1175 goto e_key; 1176 1177 ret = ccp_set_dm_area(&ctx, 0, xts->iv, 0, xts->iv_len); 1178 if (ret) 1179 goto e_ctx; 1180 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1181 CCP_PASSTHRU_BYTESWAP_NOOP); 1182 if (ret) { 1183 cmd->engine_error = cmd_q->cmd_error; 1184 goto e_ctx; 1185 } 1186 1187 /* Prepare the input and output data workareas. For in-place 1188 * operations we need to set the dma direction to BIDIRECTIONAL 1189 * and copy the src workarea to the dst workarea. 1190 */ 1191 if (sg_virt(xts->src) == sg_virt(xts->dst)) 1192 in_place = true; 1193 1194 ret = ccp_init_data(&src, cmd_q, xts->src, xts->src_len, 1195 unit_size, 1196 in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); 1197 if (ret) 1198 goto e_ctx; 1199 1200 if (in_place) { 1201 dst = src; 1202 } else { 1203 ret = ccp_init_data(&dst, cmd_q, xts->dst, xts->src_len, 1204 unit_size, DMA_FROM_DEVICE); 1205 if (ret) 1206 goto e_src; 1207 } 1208 1209 /* Send data to the CCP AES engine */ 1210 while (src.sg_wa.bytes_left) { 1211 ccp_prepare_data(&src, &dst, &op, unit_size, true); 1212 if (!src.sg_wa.bytes_left) 1213 op.eom = 1; 1214 1215 ret = cmd_q->ccp->vdata->perform->xts_aes(&op); 1216 if (ret) { 1217 cmd->engine_error = cmd_q->cmd_error; 1218 goto e_dst; 1219 } 1220 1221 ccp_process_data(&src, &dst, &op); 1222 } 1223 1224 /* Retrieve the AES context - convert from LE to BE using 1225 * 32-byte (256-bit) byteswapping 1226 */ 1227 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1228 CCP_PASSTHRU_BYTESWAP_256BIT); 1229 if (ret) { 1230 cmd->engine_error = cmd_q->cmd_error; 1231 goto e_dst; 1232 } 1233 1234 /* ...but we only need AES_BLOCK_SIZE bytes */ 1235 dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE; 1236 ccp_get_dm_area(&ctx, dm_offset, xts->iv, 0, xts->iv_len); 1237 1238 e_dst: 1239 if (!in_place) 1240 ccp_free_data(&dst, cmd_q); 1241 1242 e_src: 1243 ccp_free_data(&src, cmd_q); 1244 1245 e_ctx: 1246 ccp_dm_free(&ctx); 1247 1248 e_key: 1249 ccp_dm_free(&key); 1250 1251 return ret; 1252 } 1253 1254 static int ccp_run_des3_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 1255 { 1256 struct ccp_des3_engine *des3 = &cmd->u.des3; 1257 1258 struct ccp_dm_workarea key, ctx; 1259 struct ccp_data src, dst; 1260 struct ccp_op op; 1261 unsigned int dm_offset; 1262 unsigned int len_singlekey; 1263 bool in_place = false; 1264 int ret; 1265 1266 /* Error checks */ 1267 if (!cmd_q->ccp->vdata->perform->des3) 1268 return -EINVAL; 1269 1270 if (des3->key_len != DES3_EDE_KEY_SIZE) 1271 return -EINVAL; 1272 1273 if (((des3->mode == CCP_DES3_MODE_ECB) || 1274 (des3->mode == CCP_DES3_MODE_CBC)) && 1275 (des3->src_len & (DES3_EDE_BLOCK_SIZE - 1))) 1276 return -EINVAL; 1277 1278 if (!des3->key || !des3->src || !des3->dst) 1279 return -EINVAL; 1280 1281 if (des3->mode != CCP_DES3_MODE_ECB) { 1282 if (des3->iv_len != DES3_EDE_BLOCK_SIZE) 1283 return -EINVAL; 1284 1285 if (!des3->iv) 1286 return -EINVAL; 1287 } 1288 1289 ret = -EIO; 1290 /* Zero out all the fields of the command desc */ 1291 memset(&op, 0, sizeof(op)); 1292 1293 /* Set up the Function field */ 1294 op.cmd_q = cmd_q; 1295 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 1296 op.sb_key = cmd_q->sb_key; 1297 1298 op.init = (des3->mode == CCP_DES3_MODE_ECB) ? 0 : 1; 1299 op.u.des3.type = des3->type; 1300 op.u.des3.mode = des3->mode; 1301 op.u.des3.action = des3->action; 1302 1303 /* 1304 * All supported key sizes fit in a single (32-byte) KSB entry and 1305 * (like AES) must be in little endian format. Use the 256-bit byte 1306 * swap passthru option to convert from big endian to little endian. 1307 */ 1308 ret = ccp_init_dm_workarea(&key, cmd_q, 1309 CCP_DES3_KEY_SB_COUNT * CCP_SB_BYTES, 1310 DMA_TO_DEVICE); 1311 if (ret) 1312 return ret; 1313 1314 /* 1315 * The contents of the key triplet are in the reverse order of what 1316 * is required by the engine. Copy the 3 pieces individually to put 1317 * them where they belong. 1318 */ 1319 dm_offset = CCP_SB_BYTES - des3->key_len; /* Basic offset */ 1320 1321 len_singlekey = des3->key_len / 3; 1322 ret = ccp_set_dm_area(&key, dm_offset + 2 * len_singlekey, 1323 des3->key, 0, len_singlekey); 1324 if (ret) 1325 goto e_key; 1326 ret = ccp_set_dm_area(&key, dm_offset + len_singlekey, 1327 des3->key, len_singlekey, len_singlekey); 1328 if (ret) 1329 goto e_key; 1330 ret = ccp_set_dm_area(&key, dm_offset, 1331 des3->key, 2 * len_singlekey, len_singlekey); 1332 if (ret) 1333 goto e_key; 1334 1335 /* Copy the key to the SB */ 1336 ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key, 1337 CCP_PASSTHRU_BYTESWAP_256BIT); 1338 if (ret) { 1339 cmd->engine_error = cmd_q->cmd_error; 1340 goto e_key; 1341 } 1342 1343 /* 1344 * The DES3 context fits in a single (32-byte) KSB entry and 1345 * must be in little endian format. Use the 256-bit byte swap 1346 * passthru option to convert from big endian to little endian. 1347 */ 1348 if (des3->mode != CCP_DES3_MODE_ECB) { 1349 u32 load_mode; 1350 1351 op.sb_ctx = cmd_q->sb_ctx; 1352 1353 ret = ccp_init_dm_workarea(&ctx, cmd_q, 1354 CCP_DES3_CTX_SB_COUNT * CCP_SB_BYTES, 1355 DMA_BIDIRECTIONAL); 1356 if (ret) 1357 goto e_key; 1358 1359 /* Load the context into the LSB */ 1360 dm_offset = CCP_SB_BYTES - des3->iv_len; 1361 ret = ccp_set_dm_area(&ctx, dm_offset, des3->iv, 0, 1362 des3->iv_len); 1363 if (ret) 1364 goto e_ctx; 1365 1366 if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) 1367 load_mode = CCP_PASSTHRU_BYTESWAP_NOOP; 1368 else 1369 load_mode = CCP_PASSTHRU_BYTESWAP_256BIT; 1370 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1371 load_mode); 1372 if (ret) { 1373 cmd->engine_error = cmd_q->cmd_error; 1374 goto e_ctx; 1375 } 1376 } 1377 1378 /* 1379 * Prepare the input and output data workareas. For in-place 1380 * operations we need to set the dma direction to BIDIRECTIONAL 1381 * and copy the src workarea to the dst workarea. 1382 */ 1383 if (sg_virt(des3->src) == sg_virt(des3->dst)) 1384 in_place = true; 1385 1386 ret = ccp_init_data(&src, cmd_q, des3->src, des3->src_len, 1387 DES3_EDE_BLOCK_SIZE, 1388 in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); 1389 if (ret) 1390 goto e_ctx; 1391 1392 if (in_place) 1393 dst = src; 1394 else { 1395 ret = ccp_init_data(&dst, cmd_q, des3->dst, des3->src_len, 1396 DES3_EDE_BLOCK_SIZE, DMA_FROM_DEVICE); 1397 if (ret) 1398 goto e_src; 1399 } 1400 1401 /* Send data to the CCP DES3 engine */ 1402 while (src.sg_wa.bytes_left) { 1403 ccp_prepare_data(&src, &dst, &op, DES3_EDE_BLOCK_SIZE, true); 1404 if (!src.sg_wa.bytes_left) { 1405 op.eom = 1; 1406 1407 /* Since we don't retrieve the context in ECB mode 1408 * we have to wait for the operation to complete 1409 * on the last piece of data 1410 */ 1411 op.soc = 0; 1412 } 1413 1414 ret = cmd_q->ccp->vdata->perform->des3(&op); 1415 if (ret) { 1416 cmd->engine_error = cmd_q->cmd_error; 1417 goto e_dst; 1418 } 1419 1420 ccp_process_data(&src, &dst, &op); 1421 } 1422 1423 if (des3->mode != CCP_DES3_MODE_ECB) { 1424 /* Retrieve the context and make BE */ 1425 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1426 CCP_PASSTHRU_BYTESWAP_256BIT); 1427 if (ret) { 1428 cmd->engine_error = cmd_q->cmd_error; 1429 goto e_dst; 1430 } 1431 1432 /* ...but we only need the last DES3_EDE_BLOCK_SIZE bytes */ 1433 if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) 1434 dm_offset = CCP_SB_BYTES - des3->iv_len; 1435 else 1436 dm_offset = 0; 1437 ccp_get_dm_area(&ctx, dm_offset, des3->iv, 0, 1438 DES3_EDE_BLOCK_SIZE); 1439 } 1440 e_dst: 1441 if (!in_place) 1442 ccp_free_data(&dst, cmd_q); 1443 1444 e_src: 1445 ccp_free_data(&src, cmd_q); 1446 1447 e_ctx: 1448 if (des3->mode != CCP_DES3_MODE_ECB) 1449 ccp_dm_free(&ctx); 1450 1451 e_key: 1452 ccp_dm_free(&key); 1453 1454 return ret; 1455 } 1456 1457 static int ccp_run_sha_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 1458 { 1459 struct ccp_sha_engine *sha = &cmd->u.sha; 1460 struct ccp_dm_workarea ctx; 1461 struct ccp_data src; 1462 struct ccp_op op; 1463 unsigned int ioffset, ooffset; 1464 unsigned int digest_size; 1465 int sb_count; 1466 const void *init; 1467 u64 block_size; 1468 int ctx_size; 1469 int ret; 1470 1471 switch (sha->type) { 1472 case CCP_SHA_TYPE_1: 1473 if (sha->ctx_len < SHA1_DIGEST_SIZE) 1474 return -EINVAL; 1475 block_size = SHA1_BLOCK_SIZE; 1476 break; 1477 case CCP_SHA_TYPE_224: 1478 if (sha->ctx_len < SHA224_DIGEST_SIZE) 1479 return -EINVAL; 1480 block_size = SHA224_BLOCK_SIZE; 1481 break; 1482 case CCP_SHA_TYPE_256: 1483 if (sha->ctx_len < SHA256_DIGEST_SIZE) 1484 return -EINVAL; 1485 block_size = SHA256_BLOCK_SIZE; 1486 break; 1487 case CCP_SHA_TYPE_384: 1488 if (cmd_q->ccp->vdata->version < CCP_VERSION(4, 0) 1489 || sha->ctx_len < SHA384_DIGEST_SIZE) 1490 return -EINVAL; 1491 block_size = SHA384_BLOCK_SIZE; 1492 break; 1493 case CCP_SHA_TYPE_512: 1494 if (cmd_q->ccp->vdata->version < CCP_VERSION(4, 0) 1495 || sha->ctx_len < SHA512_DIGEST_SIZE) 1496 return -EINVAL; 1497 block_size = SHA512_BLOCK_SIZE; 1498 break; 1499 default: 1500 return -EINVAL; 1501 } 1502 1503 if (!sha->ctx) 1504 return -EINVAL; 1505 1506 if (!sha->final && (sha->src_len & (block_size - 1))) 1507 return -EINVAL; 1508 1509 /* The version 3 device can't handle zero-length input */ 1510 if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) { 1511 1512 if (!sha->src_len) { 1513 unsigned int digest_len; 1514 const u8 *sha_zero; 1515 1516 /* Not final, just return */ 1517 if (!sha->final) 1518 return 0; 1519 1520 /* CCP can't do a zero length sha operation so the 1521 * caller must buffer the data. 1522 */ 1523 if (sha->msg_bits) 1524 return -EINVAL; 1525 1526 /* The CCP cannot perform zero-length sha operations 1527 * so the caller is required to buffer data for the 1528 * final operation. However, a sha operation for a 1529 * message with a total length of zero is valid so 1530 * known values are required to supply the result. 1531 */ 1532 switch (sha->type) { 1533 case CCP_SHA_TYPE_1: 1534 sha_zero = sha1_zero_message_hash; 1535 digest_len = SHA1_DIGEST_SIZE; 1536 break; 1537 case CCP_SHA_TYPE_224: 1538 sha_zero = sha224_zero_message_hash; 1539 digest_len = SHA224_DIGEST_SIZE; 1540 break; 1541 case CCP_SHA_TYPE_256: 1542 sha_zero = sha256_zero_message_hash; 1543 digest_len = SHA256_DIGEST_SIZE; 1544 break; 1545 default: 1546 return -EINVAL; 1547 } 1548 1549 scatterwalk_map_and_copy((void *)sha_zero, sha->ctx, 0, 1550 digest_len, 1); 1551 1552 return 0; 1553 } 1554 } 1555 1556 /* Set variables used throughout */ 1557 switch (sha->type) { 1558 case CCP_SHA_TYPE_1: 1559 digest_size = SHA1_DIGEST_SIZE; 1560 init = (void *) ccp_sha1_init; 1561 ctx_size = SHA1_DIGEST_SIZE; 1562 sb_count = 1; 1563 if (cmd_q->ccp->vdata->version != CCP_VERSION(3, 0)) 1564 ooffset = ioffset = CCP_SB_BYTES - SHA1_DIGEST_SIZE; 1565 else 1566 ooffset = ioffset = 0; 1567 break; 1568 case CCP_SHA_TYPE_224: 1569 digest_size = SHA224_DIGEST_SIZE; 1570 init = (void *) ccp_sha224_init; 1571 ctx_size = SHA256_DIGEST_SIZE; 1572 sb_count = 1; 1573 ioffset = 0; 1574 if (cmd_q->ccp->vdata->version != CCP_VERSION(3, 0)) 1575 ooffset = CCP_SB_BYTES - SHA224_DIGEST_SIZE; 1576 else 1577 ooffset = 0; 1578 break; 1579 case CCP_SHA_TYPE_256: 1580 digest_size = SHA256_DIGEST_SIZE; 1581 init = (void *) ccp_sha256_init; 1582 ctx_size = SHA256_DIGEST_SIZE; 1583 sb_count = 1; 1584 ooffset = ioffset = 0; 1585 break; 1586 case CCP_SHA_TYPE_384: 1587 digest_size = SHA384_DIGEST_SIZE; 1588 init = (void *) ccp_sha384_init; 1589 ctx_size = SHA512_DIGEST_SIZE; 1590 sb_count = 2; 1591 ioffset = 0; 1592 ooffset = 2 * CCP_SB_BYTES - SHA384_DIGEST_SIZE; 1593 break; 1594 case CCP_SHA_TYPE_512: 1595 digest_size = SHA512_DIGEST_SIZE; 1596 init = (void *) ccp_sha512_init; 1597 ctx_size = SHA512_DIGEST_SIZE; 1598 sb_count = 2; 1599 ooffset = ioffset = 0; 1600 break; 1601 default: 1602 ret = -EINVAL; 1603 goto e_data; 1604 } 1605 1606 /* For zero-length plaintext the src pointer is ignored; 1607 * otherwise both parts must be valid 1608 */ 1609 if (sha->src_len && !sha->src) 1610 return -EINVAL; 1611 1612 memset(&op, 0, sizeof(op)); 1613 op.cmd_q = cmd_q; 1614 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 1615 op.sb_ctx = cmd_q->sb_ctx; /* Pre-allocated */ 1616 op.u.sha.type = sha->type; 1617 op.u.sha.msg_bits = sha->msg_bits; 1618 1619 /* For SHA1/224/256 the context fits in a single (32-byte) SB entry; 1620 * SHA384/512 require 2 adjacent SB slots, with the right half in the 1621 * first slot, and the left half in the second. Each portion must then 1622 * be in little endian format: use the 256-bit byte swap option. 1623 */ 1624 ret = ccp_init_dm_workarea(&ctx, cmd_q, sb_count * CCP_SB_BYTES, 1625 DMA_BIDIRECTIONAL); 1626 if (ret) 1627 return ret; 1628 if (sha->first) { 1629 switch (sha->type) { 1630 case CCP_SHA_TYPE_1: 1631 case CCP_SHA_TYPE_224: 1632 case CCP_SHA_TYPE_256: 1633 memcpy(ctx.address + ioffset, init, ctx_size); 1634 break; 1635 case CCP_SHA_TYPE_384: 1636 case CCP_SHA_TYPE_512: 1637 memcpy(ctx.address + ctx_size / 2, init, 1638 ctx_size / 2); 1639 memcpy(ctx.address, init + ctx_size / 2, 1640 ctx_size / 2); 1641 break; 1642 default: 1643 ret = -EINVAL; 1644 goto e_ctx; 1645 } 1646 } else { 1647 /* Restore the context */ 1648 ret = ccp_set_dm_area(&ctx, 0, sha->ctx, 0, 1649 sb_count * CCP_SB_BYTES); 1650 if (ret) 1651 goto e_ctx; 1652 } 1653 1654 ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1655 CCP_PASSTHRU_BYTESWAP_256BIT); 1656 if (ret) { 1657 cmd->engine_error = cmd_q->cmd_error; 1658 goto e_ctx; 1659 } 1660 1661 if (sha->src) { 1662 /* Send data to the CCP SHA engine; block_size is set above */ 1663 ret = ccp_init_data(&src, cmd_q, sha->src, sha->src_len, 1664 block_size, DMA_TO_DEVICE); 1665 if (ret) 1666 goto e_ctx; 1667 1668 while (src.sg_wa.bytes_left) { 1669 ccp_prepare_data(&src, NULL, &op, block_size, false); 1670 if (sha->final && !src.sg_wa.bytes_left) 1671 op.eom = 1; 1672 1673 ret = cmd_q->ccp->vdata->perform->sha(&op); 1674 if (ret) { 1675 cmd->engine_error = cmd_q->cmd_error; 1676 goto e_data; 1677 } 1678 1679 ccp_process_data(&src, NULL, &op); 1680 } 1681 } else { 1682 op.eom = 1; 1683 ret = cmd_q->ccp->vdata->perform->sha(&op); 1684 if (ret) { 1685 cmd->engine_error = cmd_q->cmd_error; 1686 goto e_data; 1687 } 1688 } 1689 1690 /* Retrieve the SHA context - convert from LE to BE using 1691 * 32-byte (256-bit) byteswapping to BE 1692 */ 1693 ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx, 1694 CCP_PASSTHRU_BYTESWAP_256BIT); 1695 if (ret) { 1696 cmd->engine_error = cmd_q->cmd_error; 1697 goto e_data; 1698 } 1699 1700 if (sha->final) { 1701 /* Finishing up, so get the digest */ 1702 switch (sha->type) { 1703 case CCP_SHA_TYPE_1: 1704 case CCP_SHA_TYPE_224: 1705 case CCP_SHA_TYPE_256: 1706 ccp_get_dm_area(&ctx, ooffset, 1707 sha->ctx, 0, 1708 digest_size); 1709 break; 1710 case CCP_SHA_TYPE_384: 1711 case CCP_SHA_TYPE_512: 1712 ccp_get_dm_area(&ctx, 0, 1713 sha->ctx, LSB_ITEM_SIZE - ooffset, 1714 LSB_ITEM_SIZE); 1715 ccp_get_dm_area(&ctx, LSB_ITEM_SIZE + ooffset, 1716 sha->ctx, 0, 1717 LSB_ITEM_SIZE - ooffset); 1718 break; 1719 default: 1720 ret = -EINVAL; 1721 goto e_ctx; 1722 } 1723 } else { 1724 /* Stash the context */ 1725 ccp_get_dm_area(&ctx, 0, sha->ctx, 0, 1726 sb_count * CCP_SB_BYTES); 1727 } 1728 1729 if (sha->final && sha->opad) { 1730 /* HMAC operation, recursively perform final SHA */ 1731 struct ccp_cmd hmac_cmd; 1732 struct scatterlist sg; 1733 u8 *hmac_buf; 1734 1735 if (sha->opad_len != block_size) { 1736 ret = -EINVAL; 1737 goto e_data; 1738 } 1739 1740 hmac_buf = kmalloc(block_size + digest_size, GFP_KERNEL); 1741 if (!hmac_buf) { 1742 ret = -ENOMEM; 1743 goto e_data; 1744 } 1745 sg_init_one(&sg, hmac_buf, block_size + digest_size); 1746 1747 scatterwalk_map_and_copy(hmac_buf, sha->opad, 0, block_size, 0); 1748 switch (sha->type) { 1749 case CCP_SHA_TYPE_1: 1750 case CCP_SHA_TYPE_224: 1751 case CCP_SHA_TYPE_256: 1752 memcpy(hmac_buf + block_size, 1753 ctx.address + ooffset, 1754 digest_size); 1755 break; 1756 case CCP_SHA_TYPE_384: 1757 case CCP_SHA_TYPE_512: 1758 memcpy(hmac_buf + block_size, 1759 ctx.address + LSB_ITEM_SIZE + ooffset, 1760 LSB_ITEM_SIZE); 1761 memcpy(hmac_buf + block_size + 1762 (LSB_ITEM_SIZE - ooffset), 1763 ctx.address, 1764 LSB_ITEM_SIZE); 1765 break; 1766 default: 1767 ret = -EINVAL; 1768 goto e_ctx; 1769 } 1770 1771 memset(&hmac_cmd, 0, sizeof(hmac_cmd)); 1772 hmac_cmd.engine = CCP_ENGINE_SHA; 1773 hmac_cmd.u.sha.type = sha->type; 1774 hmac_cmd.u.sha.ctx = sha->ctx; 1775 hmac_cmd.u.sha.ctx_len = sha->ctx_len; 1776 hmac_cmd.u.sha.src = &sg; 1777 hmac_cmd.u.sha.src_len = block_size + digest_size; 1778 hmac_cmd.u.sha.opad = NULL; 1779 hmac_cmd.u.sha.opad_len = 0; 1780 hmac_cmd.u.sha.first = 1; 1781 hmac_cmd.u.sha.final = 1; 1782 hmac_cmd.u.sha.msg_bits = (block_size + digest_size) << 3; 1783 1784 ret = ccp_run_sha_cmd(cmd_q, &hmac_cmd); 1785 if (ret) 1786 cmd->engine_error = hmac_cmd.engine_error; 1787 1788 kfree(hmac_buf); 1789 } 1790 1791 e_data: 1792 if (sha->src) 1793 ccp_free_data(&src, cmd_q); 1794 1795 e_ctx: 1796 ccp_dm_free(&ctx); 1797 1798 return ret; 1799 } 1800 1801 static int ccp_run_rsa_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 1802 { 1803 struct ccp_rsa_engine *rsa = &cmd->u.rsa; 1804 struct ccp_dm_workarea exp, src, dst; 1805 struct ccp_op op; 1806 unsigned int sb_count, i_len, o_len; 1807 int ret; 1808 1809 /* Check against the maximum allowable size, in bits */ 1810 if (rsa->key_size > cmd_q->ccp->vdata->rsamax) 1811 return -EINVAL; 1812 1813 if (!rsa->exp || !rsa->mod || !rsa->src || !rsa->dst) 1814 return -EINVAL; 1815 1816 memset(&op, 0, sizeof(op)); 1817 op.cmd_q = cmd_q; 1818 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 1819 1820 /* The RSA modulus must precede the message being acted upon, so 1821 * it must be copied to a DMA area where the message and the 1822 * modulus can be concatenated. Therefore the input buffer 1823 * length required is twice the output buffer length (which 1824 * must be a multiple of 256-bits). Compute o_len, i_len in bytes. 1825 * Buffer sizes must be a multiple of 32 bytes; rounding up may be 1826 * required. 1827 */ 1828 o_len = 32 * ((rsa->key_size + 255) / 256); 1829 i_len = o_len * 2; 1830 1831 sb_count = 0; 1832 if (cmd_q->ccp->vdata->version < CCP_VERSION(5, 0)) { 1833 /* sb_count is the number of storage block slots required 1834 * for the modulus. 1835 */ 1836 sb_count = o_len / CCP_SB_BYTES; 1837 op.sb_key = cmd_q->ccp->vdata->perform->sballoc(cmd_q, 1838 sb_count); 1839 if (!op.sb_key) 1840 return -EIO; 1841 } else { 1842 /* A version 5 device allows a modulus size that will not fit 1843 * in the LSB, so the command will transfer it from memory. 1844 * Set the sb key to the default, even though it's not used. 1845 */ 1846 op.sb_key = cmd_q->sb_key; 1847 } 1848 1849 /* The RSA exponent must be in little endian format. Reverse its 1850 * byte order. 1851 */ 1852 ret = ccp_init_dm_workarea(&exp, cmd_q, o_len, DMA_TO_DEVICE); 1853 if (ret) 1854 goto e_sb; 1855 1856 ret = ccp_reverse_set_dm_area(&exp, 0, rsa->exp, 0, rsa->exp_len); 1857 if (ret) 1858 goto e_exp; 1859 1860 if (cmd_q->ccp->vdata->version < CCP_VERSION(5, 0)) { 1861 /* Copy the exponent to the local storage block, using 1862 * as many 32-byte blocks as were allocated above. It's 1863 * already little endian, so no further change is required. 1864 */ 1865 ret = ccp_copy_to_sb(cmd_q, &exp, op.jobid, op.sb_key, 1866 CCP_PASSTHRU_BYTESWAP_NOOP); 1867 if (ret) { 1868 cmd->engine_error = cmd_q->cmd_error; 1869 goto e_exp; 1870 } 1871 } else { 1872 /* The exponent can be retrieved from memory via DMA. */ 1873 op.exp.u.dma.address = exp.dma.address; 1874 op.exp.u.dma.offset = 0; 1875 } 1876 1877 /* Concatenate the modulus and the message. Both the modulus and 1878 * the operands must be in little endian format. Since the input 1879 * is in big endian format it must be converted. 1880 */ 1881 ret = ccp_init_dm_workarea(&src, cmd_q, i_len, DMA_TO_DEVICE); 1882 if (ret) 1883 goto e_exp; 1884 1885 ret = ccp_reverse_set_dm_area(&src, 0, rsa->mod, 0, rsa->mod_len); 1886 if (ret) 1887 goto e_src; 1888 ret = ccp_reverse_set_dm_area(&src, o_len, rsa->src, 0, rsa->src_len); 1889 if (ret) 1890 goto e_src; 1891 1892 /* Prepare the output area for the operation */ 1893 ret = ccp_init_dm_workarea(&dst, cmd_q, o_len, DMA_FROM_DEVICE); 1894 if (ret) 1895 goto e_src; 1896 1897 op.soc = 1; 1898 op.src.u.dma.address = src.dma.address; 1899 op.src.u.dma.offset = 0; 1900 op.src.u.dma.length = i_len; 1901 op.dst.u.dma.address = dst.dma.address; 1902 op.dst.u.dma.offset = 0; 1903 op.dst.u.dma.length = o_len; 1904 1905 op.u.rsa.mod_size = rsa->key_size; 1906 op.u.rsa.input_len = i_len; 1907 1908 ret = cmd_q->ccp->vdata->perform->rsa(&op); 1909 if (ret) { 1910 cmd->engine_error = cmd_q->cmd_error; 1911 goto e_dst; 1912 } 1913 1914 ccp_reverse_get_dm_area(&dst, 0, rsa->dst, 0, rsa->mod_len); 1915 1916 e_dst: 1917 ccp_dm_free(&dst); 1918 1919 e_src: 1920 ccp_dm_free(&src); 1921 1922 e_exp: 1923 ccp_dm_free(&exp); 1924 1925 e_sb: 1926 if (sb_count) 1927 cmd_q->ccp->vdata->perform->sbfree(cmd_q, op.sb_key, sb_count); 1928 1929 return ret; 1930 } 1931 1932 static int ccp_run_passthru_cmd(struct ccp_cmd_queue *cmd_q, 1933 struct ccp_cmd *cmd) 1934 { 1935 struct ccp_passthru_engine *pt = &cmd->u.passthru; 1936 struct ccp_dm_workarea mask; 1937 struct ccp_data src, dst; 1938 struct ccp_op op; 1939 bool in_place = false; 1940 unsigned int i; 1941 int ret = 0; 1942 1943 if (!pt->final && (pt->src_len & (CCP_PASSTHRU_BLOCKSIZE - 1))) 1944 return -EINVAL; 1945 1946 if (!pt->src || !pt->dst) 1947 return -EINVAL; 1948 1949 if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { 1950 if (pt->mask_len != CCP_PASSTHRU_MASKSIZE) 1951 return -EINVAL; 1952 if (!pt->mask) 1953 return -EINVAL; 1954 } 1955 1956 BUILD_BUG_ON(CCP_PASSTHRU_SB_COUNT != 1); 1957 1958 memset(&op, 0, sizeof(op)); 1959 op.cmd_q = cmd_q; 1960 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 1961 1962 if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { 1963 /* Load the mask */ 1964 op.sb_key = cmd_q->sb_key; 1965 1966 ret = ccp_init_dm_workarea(&mask, cmd_q, 1967 CCP_PASSTHRU_SB_COUNT * 1968 CCP_SB_BYTES, 1969 DMA_TO_DEVICE); 1970 if (ret) 1971 return ret; 1972 1973 ret = ccp_set_dm_area(&mask, 0, pt->mask, 0, pt->mask_len); 1974 if (ret) 1975 goto e_mask; 1976 ret = ccp_copy_to_sb(cmd_q, &mask, op.jobid, op.sb_key, 1977 CCP_PASSTHRU_BYTESWAP_NOOP); 1978 if (ret) { 1979 cmd->engine_error = cmd_q->cmd_error; 1980 goto e_mask; 1981 } 1982 } 1983 1984 /* Prepare the input and output data workareas. For in-place 1985 * operations we need to set the dma direction to BIDIRECTIONAL 1986 * and copy the src workarea to the dst workarea. 1987 */ 1988 if (sg_virt(pt->src) == sg_virt(pt->dst)) 1989 in_place = true; 1990 1991 ret = ccp_init_data(&src, cmd_q, pt->src, pt->src_len, 1992 CCP_PASSTHRU_MASKSIZE, 1993 in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE); 1994 if (ret) 1995 goto e_mask; 1996 1997 if (in_place) { 1998 dst = src; 1999 } else { 2000 ret = ccp_init_data(&dst, cmd_q, pt->dst, pt->src_len, 2001 CCP_PASSTHRU_MASKSIZE, DMA_FROM_DEVICE); 2002 if (ret) 2003 goto e_src; 2004 } 2005 2006 /* Send data to the CCP Passthru engine 2007 * Because the CCP engine works on a single source and destination 2008 * dma address at a time, each entry in the source scatterlist 2009 * (after the dma_map_sg call) must be less than or equal to the 2010 * (remaining) length in the destination scatterlist entry and the 2011 * length must be a multiple of CCP_PASSTHRU_BLOCKSIZE 2012 */ 2013 dst.sg_wa.sg_used = 0; 2014 for (i = 1; i <= src.sg_wa.dma_count; i++) { 2015 if (!dst.sg_wa.sg || 2016 (dst.sg_wa.sg->length < src.sg_wa.sg->length)) { 2017 ret = -EINVAL; 2018 goto e_dst; 2019 } 2020 2021 if (i == src.sg_wa.dma_count) { 2022 op.eom = 1; 2023 op.soc = 1; 2024 } 2025 2026 op.src.type = CCP_MEMTYPE_SYSTEM; 2027 op.src.u.dma.address = sg_dma_address(src.sg_wa.sg); 2028 op.src.u.dma.offset = 0; 2029 op.src.u.dma.length = sg_dma_len(src.sg_wa.sg); 2030 2031 op.dst.type = CCP_MEMTYPE_SYSTEM; 2032 op.dst.u.dma.address = sg_dma_address(dst.sg_wa.sg); 2033 op.dst.u.dma.offset = dst.sg_wa.sg_used; 2034 op.dst.u.dma.length = op.src.u.dma.length; 2035 2036 ret = cmd_q->ccp->vdata->perform->passthru(&op); 2037 if (ret) { 2038 cmd->engine_error = cmd_q->cmd_error; 2039 goto e_dst; 2040 } 2041 2042 dst.sg_wa.sg_used += src.sg_wa.sg->length; 2043 if (dst.sg_wa.sg_used == dst.sg_wa.sg->length) { 2044 dst.sg_wa.sg = sg_next(dst.sg_wa.sg); 2045 dst.sg_wa.sg_used = 0; 2046 } 2047 src.sg_wa.sg = sg_next(src.sg_wa.sg); 2048 } 2049 2050 e_dst: 2051 if (!in_place) 2052 ccp_free_data(&dst, cmd_q); 2053 2054 e_src: 2055 ccp_free_data(&src, cmd_q); 2056 2057 e_mask: 2058 if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) 2059 ccp_dm_free(&mask); 2060 2061 return ret; 2062 } 2063 2064 static int ccp_run_passthru_nomap_cmd(struct ccp_cmd_queue *cmd_q, 2065 struct ccp_cmd *cmd) 2066 { 2067 struct ccp_passthru_nomap_engine *pt = &cmd->u.passthru_nomap; 2068 struct ccp_dm_workarea mask; 2069 struct ccp_op op; 2070 int ret; 2071 2072 if (!pt->final && (pt->src_len & (CCP_PASSTHRU_BLOCKSIZE - 1))) 2073 return -EINVAL; 2074 2075 if (!pt->src_dma || !pt->dst_dma) 2076 return -EINVAL; 2077 2078 if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { 2079 if (pt->mask_len != CCP_PASSTHRU_MASKSIZE) 2080 return -EINVAL; 2081 if (!pt->mask) 2082 return -EINVAL; 2083 } 2084 2085 BUILD_BUG_ON(CCP_PASSTHRU_SB_COUNT != 1); 2086 2087 memset(&op, 0, sizeof(op)); 2088 op.cmd_q = cmd_q; 2089 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 2090 2091 if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) { 2092 /* Load the mask */ 2093 op.sb_key = cmd_q->sb_key; 2094 2095 mask.length = pt->mask_len; 2096 mask.dma.address = pt->mask; 2097 mask.dma.length = pt->mask_len; 2098 2099 ret = ccp_copy_to_sb(cmd_q, &mask, op.jobid, op.sb_key, 2100 CCP_PASSTHRU_BYTESWAP_NOOP); 2101 if (ret) { 2102 cmd->engine_error = cmd_q->cmd_error; 2103 return ret; 2104 } 2105 } 2106 2107 /* Send data to the CCP Passthru engine */ 2108 op.eom = 1; 2109 op.soc = 1; 2110 2111 op.src.type = CCP_MEMTYPE_SYSTEM; 2112 op.src.u.dma.address = pt->src_dma; 2113 op.src.u.dma.offset = 0; 2114 op.src.u.dma.length = pt->src_len; 2115 2116 op.dst.type = CCP_MEMTYPE_SYSTEM; 2117 op.dst.u.dma.address = pt->dst_dma; 2118 op.dst.u.dma.offset = 0; 2119 op.dst.u.dma.length = pt->src_len; 2120 2121 ret = cmd_q->ccp->vdata->perform->passthru(&op); 2122 if (ret) 2123 cmd->engine_error = cmd_q->cmd_error; 2124 2125 return ret; 2126 } 2127 2128 static int ccp_run_ecc_mm_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 2129 { 2130 struct ccp_ecc_engine *ecc = &cmd->u.ecc; 2131 struct ccp_dm_workarea src, dst; 2132 struct ccp_op op; 2133 int ret; 2134 u8 *save; 2135 2136 if (!ecc->u.mm.operand_1 || 2137 (ecc->u.mm.operand_1_len > CCP_ECC_MODULUS_BYTES)) 2138 return -EINVAL; 2139 2140 if (ecc->function != CCP_ECC_FUNCTION_MINV_384BIT) 2141 if (!ecc->u.mm.operand_2 || 2142 (ecc->u.mm.operand_2_len > CCP_ECC_MODULUS_BYTES)) 2143 return -EINVAL; 2144 2145 if (!ecc->u.mm.result || 2146 (ecc->u.mm.result_len < CCP_ECC_MODULUS_BYTES)) 2147 return -EINVAL; 2148 2149 memset(&op, 0, sizeof(op)); 2150 op.cmd_q = cmd_q; 2151 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 2152 2153 /* Concatenate the modulus and the operands. Both the modulus and 2154 * the operands must be in little endian format. Since the input 2155 * is in big endian format it must be converted and placed in a 2156 * fixed length buffer. 2157 */ 2158 ret = ccp_init_dm_workarea(&src, cmd_q, CCP_ECC_SRC_BUF_SIZE, 2159 DMA_TO_DEVICE); 2160 if (ret) 2161 return ret; 2162 2163 /* Save the workarea address since it is updated in order to perform 2164 * the concatenation 2165 */ 2166 save = src.address; 2167 2168 /* Copy the ECC modulus */ 2169 ret = ccp_reverse_set_dm_area(&src, 0, ecc->mod, 0, ecc->mod_len); 2170 if (ret) 2171 goto e_src; 2172 src.address += CCP_ECC_OPERAND_SIZE; 2173 2174 /* Copy the first operand */ 2175 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.mm.operand_1, 0, 2176 ecc->u.mm.operand_1_len); 2177 if (ret) 2178 goto e_src; 2179 src.address += CCP_ECC_OPERAND_SIZE; 2180 2181 if (ecc->function != CCP_ECC_FUNCTION_MINV_384BIT) { 2182 /* Copy the second operand */ 2183 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.mm.operand_2, 0, 2184 ecc->u.mm.operand_2_len); 2185 if (ret) 2186 goto e_src; 2187 src.address += CCP_ECC_OPERAND_SIZE; 2188 } 2189 2190 /* Restore the workarea address */ 2191 src.address = save; 2192 2193 /* Prepare the output area for the operation */ 2194 ret = ccp_init_dm_workarea(&dst, cmd_q, CCP_ECC_DST_BUF_SIZE, 2195 DMA_FROM_DEVICE); 2196 if (ret) 2197 goto e_src; 2198 2199 op.soc = 1; 2200 op.src.u.dma.address = src.dma.address; 2201 op.src.u.dma.offset = 0; 2202 op.src.u.dma.length = src.length; 2203 op.dst.u.dma.address = dst.dma.address; 2204 op.dst.u.dma.offset = 0; 2205 op.dst.u.dma.length = dst.length; 2206 2207 op.u.ecc.function = cmd->u.ecc.function; 2208 2209 ret = cmd_q->ccp->vdata->perform->ecc(&op); 2210 if (ret) { 2211 cmd->engine_error = cmd_q->cmd_error; 2212 goto e_dst; 2213 } 2214 2215 ecc->ecc_result = le16_to_cpup( 2216 (const __le16 *)(dst.address + CCP_ECC_RESULT_OFFSET)); 2217 if (!(ecc->ecc_result & CCP_ECC_RESULT_SUCCESS)) { 2218 ret = -EIO; 2219 goto e_dst; 2220 } 2221 2222 /* Save the ECC result */ 2223 ccp_reverse_get_dm_area(&dst, 0, ecc->u.mm.result, 0, 2224 CCP_ECC_MODULUS_BYTES); 2225 2226 e_dst: 2227 ccp_dm_free(&dst); 2228 2229 e_src: 2230 ccp_dm_free(&src); 2231 2232 return ret; 2233 } 2234 2235 static int ccp_run_ecc_pm_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 2236 { 2237 struct ccp_ecc_engine *ecc = &cmd->u.ecc; 2238 struct ccp_dm_workarea src, dst; 2239 struct ccp_op op; 2240 int ret; 2241 u8 *save; 2242 2243 if (!ecc->u.pm.point_1.x || 2244 (ecc->u.pm.point_1.x_len > CCP_ECC_MODULUS_BYTES) || 2245 !ecc->u.pm.point_1.y || 2246 (ecc->u.pm.point_1.y_len > CCP_ECC_MODULUS_BYTES)) 2247 return -EINVAL; 2248 2249 if (ecc->function == CCP_ECC_FUNCTION_PADD_384BIT) { 2250 if (!ecc->u.pm.point_2.x || 2251 (ecc->u.pm.point_2.x_len > CCP_ECC_MODULUS_BYTES) || 2252 !ecc->u.pm.point_2.y || 2253 (ecc->u.pm.point_2.y_len > CCP_ECC_MODULUS_BYTES)) 2254 return -EINVAL; 2255 } else { 2256 if (!ecc->u.pm.domain_a || 2257 (ecc->u.pm.domain_a_len > CCP_ECC_MODULUS_BYTES)) 2258 return -EINVAL; 2259 2260 if (ecc->function == CCP_ECC_FUNCTION_PMUL_384BIT) 2261 if (!ecc->u.pm.scalar || 2262 (ecc->u.pm.scalar_len > CCP_ECC_MODULUS_BYTES)) 2263 return -EINVAL; 2264 } 2265 2266 if (!ecc->u.pm.result.x || 2267 (ecc->u.pm.result.x_len < CCP_ECC_MODULUS_BYTES) || 2268 !ecc->u.pm.result.y || 2269 (ecc->u.pm.result.y_len < CCP_ECC_MODULUS_BYTES)) 2270 return -EINVAL; 2271 2272 memset(&op, 0, sizeof(op)); 2273 op.cmd_q = cmd_q; 2274 op.jobid = CCP_NEW_JOBID(cmd_q->ccp); 2275 2276 /* Concatenate the modulus and the operands. Both the modulus and 2277 * the operands must be in little endian format. Since the input 2278 * is in big endian format it must be converted and placed in a 2279 * fixed length buffer. 2280 */ 2281 ret = ccp_init_dm_workarea(&src, cmd_q, CCP_ECC_SRC_BUF_SIZE, 2282 DMA_TO_DEVICE); 2283 if (ret) 2284 return ret; 2285 2286 /* Save the workarea address since it is updated in order to perform 2287 * the concatenation 2288 */ 2289 save = src.address; 2290 2291 /* Copy the ECC modulus */ 2292 ret = ccp_reverse_set_dm_area(&src, 0, ecc->mod, 0, ecc->mod_len); 2293 if (ret) 2294 goto e_src; 2295 src.address += CCP_ECC_OPERAND_SIZE; 2296 2297 /* Copy the first point X and Y coordinate */ 2298 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.pm.point_1.x, 0, 2299 ecc->u.pm.point_1.x_len); 2300 if (ret) 2301 goto e_src; 2302 src.address += CCP_ECC_OPERAND_SIZE; 2303 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.pm.point_1.y, 0, 2304 ecc->u.pm.point_1.y_len); 2305 if (ret) 2306 goto e_src; 2307 src.address += CCP_ECC_OPERAND_SIZE; 2308 2309 /* Set the first point Z coordinate to 1 */ 2310 *src.address = 0x01; 2311 src.address += CCP_ECC_OPERAND_SIZE; 2312 2313 if (ecc->function == CCP_ECC_FUNCTION_PADD_384BIT) { 2314 /* Copy the second point X and Y coordinate */ 2315 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.pm.point_2.x, 0, 2316 ecc->u.pm.point_2.x_len); 2317 if (ret) 2318 goto e_src; 2319 src.address += CCP_ECC_OPERAND_SIZE; 2320 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.pm.point_2.y, 0, 2321 ecc->u.pm.point_2.y_len); 2322 if (ret) 2323 goto e_src; 2324 src.address += CCP_ECC_OPERAND_SIZE; 2325 2326 /* Set the second point Z coordinate to 1 */ 2327 *src.address = 0x01; 2328 src.address += CCP_ECC_OPERAND_SIZE; 2329 } else { 2330 /* Copy the Domain "a" parameter */ 2331 ret = ccp_reverse_set_dm_area(&src, 0, ecc->u.pm.domain_a, 0, 2332 ecc->u.pm.domain_a_len); 2333 if (ret) 2334 goto e_src; 2335 src.address += CCP_ECC_OPERAND_SIZE; 2336 2337 if (ecc->function == CCP_ECC_FUNCTION_PMUL_384BIT) { 2338 /* Copy the scalar value */ 2339 ret = ccp_reverse_set_dm_area(&src, 0, 2340 ecc->u.pm.scalar, 0, 2341 ecc->u.pm.scalar_len); 2342 if (ret) 2343 goto e_src; 2344 src.address += CCP_ECC_OPERAND_SIZE; 2345 } 2346 } 2347 2348 /* Restore the workarea address */ 2349 src.address = save; 2350 2351 /* Prepare the output area for the operation */ 2352 ret = ccp_init_dm_workarea(&dst, cmd_q, CCP_ECC_DST_BUF_SIZE, 2353 DMA_FROM_DEVICE); 2354 if (ret) 2355 goto e_src; 2356 2357 op.soc = 1; 2358 op.src.u.dma.address = src.dma.address; 2359 op.src.u.dma.offset = 0; 2360 op.src.u.dma.length = src.length; 2361 op.dst.u.dma.address = dst.dma.address; 2362 op.dst.u.dma.offset = 0; 2363 op.dst.u.dma.length = dst.length; 2364 2365 op.u.ecc.function = cmd->u.ecc.function; 2366 2367 ret = cmd_q->ccp->vdata->perform->ecc(&op); 2368 if (ret) { 2369 cmd->engine_error = cmd_q->cmd_error; 2370 goto e_dst; 2371 } 2372 2373 ecc->ecc_result = le16_to_cpup( 2374 (const __le16 *)(dst.address + CCP_ECC_RESULT_OFFSET)); 2375 if (!(ecc->ecc_result & CCP_ECC_RESULT_SUCCESS)) { 2376 ret = -EIO; 2377 goto e_dst; 2378 } 2379 2380 /* Save the workarea address since it is updated as we walk through 2381 * to copy the point math result 2382 */ 2383 save = dst.address; 2384 2385 /* Save the ECC result X and Y coordinates */ 2386 ccp_reverse_get_dm_area(&dst, 0, ecc->u.pm.result.x, 0, 2387 CCP_ECC_MODULUS_BYTES); 2388 dst.address += CCP_ECC_OUTPUT_SIZE; 2389 ccp_reverse_get_dm_area(&dst, 0, ecc->u.pm.result.y, 0, 2390 CCP_ECC_MODULUS_BYTES); 2391 dst.address += CCP_ECC_OUTPUT_SIZE; 2392 2393 /* Restore the workarea address */ 2394 dst.address = save; 2395 2396 e_dst: 2397 ccp_dm_free(&dst); 2398 2399 e_src: 2400 ccp_dm_free(&src); 2401 2402 return ret; 2403 } 2404 2405 static int ccp_run_ecc_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 2406 { 2407 struct ccp_ecc_engine *ecc = &cmd->u.ecc; 2408 2409 ecc->ecc_result = 0; 2410 2411 if (!ecc->mod || 2412 (ecc->mod_len > CCP_ECC_MODULUS_BYTES)) 2413 return -EINVAL; 2414 2415 switch (ecc->function) { 2416 case CCP_ECC_FUNCTION_MMUL_384BIT: 2417 case CCP_ECC_FUNCTION_MADD_384BIT: 2418 case CCP_ECC_FUNCTION_MINV_384BIT: 2419 return ccp_run_ecc_mm_cmd(cmd_q, cmd); 2420 2421 case CCP_ECC_FUNCTION_PADD_384BIT: 2422 case CCP_ECC_FUNCTION_PMUL_384BIT: 2423 case CCP_ECC_FUNCTION_PDBL_384BIT: 2424 return ccp_run_ecc_pm_cmd(cmd_q, cmd); 2425 2426 default: 2427 return -EINVAL; 2428 } 2429 } 2430 2431 int ccp_run_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd) 2432 { 2433 int ret; 2434 2435 cmd->engine_error = 0; 2436 cmd_q->cmd_error = 0; 2437 cmd_q->int_rcvd = 0; 2438 cmd_q->free_slots = cmd_q->ccp->vdata->perform->get_free_slots(cmd_q); 2439 2440 switch (cmd->engine) { 2441 case CCP_ENGINE_AES: 2442 ret = ccp_run_aes_cmd(cmd_q, cmd); 2443 break; 2444 case CCP_ENGINE_XTS_AES_128: 2445 ret = ccp_run_xts_aes_cmd(cmd_q, cmd); 2446 break; 2447 case CCP_ENGINE_DES3: 2448 ret = ccp_run_des3_cmd(cmd_q, cmd); 2449 break; 2450 case CCP_ENGINE_SHA: 2451 ret = ccp_run_sha_cmd(cmd_q, cmd); 2452 break; 2453 case CCP_ENGINE_RSA: 2454 ret = ccp_run_rsa_cmd(cmd_q, cmd); 2455 break; 2456 case CCP_ENGINE_PASSTHRU: 2457 if (cmd->flags & CCP_CMD_PASSTHRU_NO_DMA_MAP) 2458 ret = ccp_run_passthru_nomap_cmd(cmd_q, cmd); 2459 else 2460 ret = ccp_run_passthru_cmd(cmd_q, cmd); 2461 break; 2462 case CCP_ENGINE_ECC: 2463 ret = ccp_run_ecc_cmd(cmd_q, cmd); 2464 break; 2465 default: 2466 ret = -EINVAL; 2467 } 2468 2469 return ret; 2470 } 2471