xref: /openbmc/linux/drivers/crypto/ccp/ccp-ops.c (revision 23966841)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * AMD Cryptographic Coprocessor (CCP) driver
4  *
5  * Copyright (C) 2013-2019 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/interrupt.h>
14 #include <crypto/scatterwalk.h>
15 #include <crypto/des.h>
16 #include <linux/ccp.h>
17 
18 #include "ccp-dev.h"
19 
20 /* SHA initial context values */
21 static const __be32 ccp_sha1_init[SHA1_DIGEST_SIZE / sizeof(__be32)] = {
22 	cpu_to_be32(SHA1_H0), cpu_to_be32(SHA1_H1),
23 	cpu_to_be32(SHA1_H2), cpu_to_be32(SHA1_H3),
24 	cpu_to_be32(SHA1_H4),
25 };
26 
27 static const __be32 ccp_sha224_init[SHA256_DIGEST_SIZE / sizeof(__be32)] = {
28 	cpu_to_be32(SHA224_H0), cpu_to_be32(SHA224_H1),
29 	cpu_to_be32(SHA224_H2), cpu_to_be32(SHA224_H3),
30 	cpu_to_be32(SHA224_H4), cpu_to_be32(SHA224_H5),
31 	cpu_to_be32(SHA224_H6), cpu_to_be32(SHA224_H7),
32 };
33 
34 static const __be32 ccp_sha256_init[SHA256_DIGEST_SIZE / sizeof(__be32)] = {
35 	cpu_to_be32(SHA256_H0), cpu_to_be32(SHA256_H1),
36 	cpu_to_be32(SHA256_H2), cpu_to_be32(SHA256_H3),
37 	cpu_to_be32(SHA256_H4), cpu_to_be32(SHA256_H5),
38 	cpu_to_be32(SHA256_H6), cpu_to_be32(SHA256_H7),
39 };
40 
41 static const __be64 ccp_sha384_init[SHA512_DIGEST_SIZE / sizeof(__be64)] = {
42 	cpu_to_be64(SHA384_H0), cpu_to_be64(SHA384_H1),
43 	cpu_to_be64(SHA384_H2), cpu_to_be64(SHA384_H3),
44 	cpu_to_be64(SHA384_H4), cpu_to_be64(SHA384_H5),
45 	cpu_to_be64(SHA384_H6), cpu_to_be64(SHA384_H7),
46 };
47 
48 static const __be64 ccp_sha512_init[SHA512_DIGEST_SIZE / sizeof(__be64)] = {
49 	cpu_to_be64(SHA512_H0), cpu_to_be64(SHA512_H1),
50 	cpu_to_be64(SHA512_H2), cpu_to_be64(SHA512_H3),
51 	cpu_to_be64(SHA512_H4), cpu_to_be64(SHA512_H5),
52 	cpu_to_be64(SHA512_H6), cpu_to_be64(SHA512_H7),
53 };
54 
55 #define	CCP_NEW_JOBID(ccp)	((ccp->vdata->version == CCP_VERSION(3, 0)) ? \
56 					ccp_gen_jobid(ccp) : 0)
57 
58 static u32 ccp_gen_jobid(struct ccp_device *ccp)
59 {
60 	return atomic_inc_return(&ccp->current_id) & CCP_JOBID_MASK;
61 }
62 
63 static void ccp_sg_free(struct ccp_sg_workarea *wa)
64 {
65 	if (wa->dma_count)
66 		dma_unmap_sg(wa->dma_dev, wa->dma_sg, wa->nents, wa->dma_dir);
67 
68 	wa->dma_count = 0;
69 }
70 
71 static int ccp_init_sg_workarea(struct ccp_sg_workarea *wa, struct device *dev,
72 				struct scatterlist *sg, u64 len,
73 				enum dma_data_direction dma_dir)
74 {
75 	memset(wa, 0, sizeof(*wa));
76 
77 	wa->sg = sg;
78 	if (!sg)
79 		return 0;
80 
81 	wa->nents = sg_nents_for_len(sg, len);
82 	if (wa->nents < 0)
83 		return wa->nents;
84 
85 	wa->bytes_left = len;
86 	wa->sg_used = 0;
87 
88 	if (len == 0)
89 		return 0;
90 
91 	if (dma_dir == DMA_NONE)
92 		return 0;
93 
94 	wa->dma_sg = sg;
95 	wa->dma_dev = dev;
96 	wa->dma_dir = dma_dir;
97 	wa->dma_count = dma_map_sg(dev, sg, wa->nents, dma_dir);
98 	if (!wa->dma_count)
99 		return -ENOMEM;
100 
101 	return 0;
102 }
103 
104 static void ccp_update_sg_workarea(struct ccp_sg_workarea *wa, unsigned int len)
105 {
106 	unsigned int nbytes = min_t(u64, len, wa->bytes_left);
107 
108 	if (!wa->sg)
109 		return;
110 
111 	wa->sg_used += nbytes;
112 	wa->bytes_left -= nbytes;
113 	if (wa->sg_used == wa->sg->length) {
114 		wa->sg = sg_next(wa->sg);
115 		wa->sg_used = 0;
116 	}
117 }
118 
119 static void ccp_dm_free(struct ccp_dm_workarea *wa)
120 {
121 	if (wa->length <= CCP_DMAPOOL_MAX_SIZE) {
122 		if (wa->address)
123 			dma_pool_free(wa->dma_pool, wa->address,
124 				      wa->dma.address);
125 	} else {
126 		if (wa->dma.address)
127 			dma_unmap_single(wa->dev, wa->dma.address, wa->length,
128 					 wa->dma.dir);
129 		kfree(wa->address);
130 	}
131 
132 	wa->address = NULL;
133 	wa->dma.address = 0;
134 }
135 
136 static int ccp_init_dm_workarea(struct ccp_dm_workarea *wa,
137 				struct ccp_cmd_queue *cmd_q,
138 				unsigned int len,
139 				enum dma_data_direction dir)
140 {
141 	memset(wa, 0, sizeof(*wa));
142 
143 	if (!len)
144 		return 0;
145 
146 	wa->dev = cmd_q->ccp->dev;
147 	wa->length = len;
148 
149 	if (len <= CCP_DMAPOOL_MAX_SIZE) {
150 		wa->dma_pool = cmd_q->dma_pool;
151 
152 		wa->address = dma_pool_zalloc(wa->dma_pool, GFP_KERNEL,
153 					     &wa->dma.address);
154 		if (!wa->address)
155 			return -ENOMEM;
156 
157 		wa->dma.length = CCP_DMAPOOL_MAX_SIZE;
158 
159 	} else {
160 		wa->address = kzalloc(len, GFP_KERNEL);
161 		if (!wa->address)
162 			return -ENOMEM;
163 
164 		wa->dma.address = dma_map_single(wa->dev, wa->address, len,
165 						 dir);
166 		if (dma_mapping_error(wa->dev, wa->dma.address))
167 			return -ENOMEM;
168 
169 		wa->dma.length = len;
170 	}
171 	wa->dma.dir = dir;
172 
173 	return 0;
174 }
175 
176 static int ccp_set_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset,
177 			   struct scatterlist *sg, unsigned int sg_offset,
178 			   unsigned int len)
179 {
180 	WARN_ON(!wa->address);
181 
182 	if (len > (wa->length - wa_offset))
183 		return -EINVAL;
184 
185 	scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len,
186 				 0);
187 	return 0;
188 }
189 
190 static void ccp_get_dm_area(struct ccp_dm_workarea *wa, unsigned int wa_offset,
191 			    struct scatterlist *sg, unsigned int sg_offset,
192 			    unsigned int len)
193 {
194 	WARN_ON(!wa->address);
195 
196 	scatterwalk_map_and_copy(wa->address + wa_offset, sg, sg_offset, len,
197 				 1);
198 }
199 
200 static int ccp_reverse_set_dm_area(struct ccp_dm_workarea *wa,
201 				   unsigned int wa_offset,
202 				   struct scatterlist *sg,
203 				   unsigned int sg_offset,
204 				   unsigned int len)
205 {
206 	u8 *p, *q;
207 	int	rc;
208 
209 	rc = ccp_set_dm_area(wa, wa_offset, sg, sg_offset, len);
210 	if (rc)
211 		return rc;
212 
213 	p = wa->address + wa_offset;
214 	q = p + len - 1;
215 	while (p < q) {
216 		*p = *p ^ *q;
217 		*q = *p ^ *q;
218 		*p = *p ^ *q;
219 		p++;
220 		q--;
221 	}
222 	return 0;
223 }
224 
225 static void ccp_reverse_get_dm_area(struct ccp_dm_workarea *wa,
226 				    unsigned int wa_offset,
227 				    struct scatterlist *sg,
228 				    unsigned int sg_offset,
229 				    unsigned int len)
230 {
231 	u8 *p, *q;
232 
233 	p = wa->address + wa_offset;
234 	q = p + len - 1;
235 	while (p < q) {
236 		*p = *p ^ *q;
237 		*q = *p ^ *q;
238 		*p = *p ^ *q;
239 		p++;
240 		q--;
241 	}
242 
243 	ccp_get_dm_area(wa, wa_offset, sg, sg_offset, len);
244 }
245 
246 static void ccp_free_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q)
247 {
248 	ccp_dm_free(&data->dm_wa);
249 	ccp_sg_free(&data->sg_wa);
250 }
251 
252 static int ccp_init_data(struct ccp_data *data, struct ccp_cmd_queue *cmd_q,
253 			 struct scatterlist *sg, u64 sg_len,
254 			 unsigned int dm_len,
255 			 enum dma_data_direction dir)
256 {
257 	int ret;
258 
259 	memset(data, 0, sizeof(*data));
260 
261 	ret = ccp_init_sg_workarea(&data->sg_wa, cmd_q->ccp->dev, sg, sg_len,
262 				   dir);
263 	if (ret)
264 		goto e_err;
265 
266 	ret = ccp_init_dm_workarea(&data->dm_wa, cmd_q, dm_len, dir);
267 	if (ret)
268 		goto e_err;
269 
270 	return 0;
271 
272 e_err:
273 	ccp_free_data(data, cmd_q);
274 
275 	return ret;
276 }
277 
278 static unsigned int ccp_queue_buf(struct ccp_data *data, unsigned int from)
279 {
280 	struct ccp_sg_workarea *sg_wa = &data->sg_wa;
281 	struct ccp_dm_workarea *dm_wa = &data->dm_wa;
282 	unsigned int buf_count, nbytes;
283 
284 	/* Clear the buffer if setting it */
285 	if (!from)
286 		memset(dm_wa->address, 0, dm_wa->length);
287 
288 	if (!sg_wa->sg)
289 		return 0;
290 
291 	/* Perform the copy operation
292 	 *   nbytes will always be <= UINT_MAX because dm_wa->length is
293 	 *   an unsigned int
294 	 */
295 	nbytes = min_t(u64, sg_wa->bytes_left, dm_wa->length);
296 	scatterwalk_map_and_copy(dm_wa->address, sg_wa->sg, sg_wa->sg_used,
297 				 nbytes, from);
298 
299 	/* Update the structures and generate the count */
300 	buf_count = 0;
301 	while (sg_wa->bytes_left && (buf_count < dm_wa->length)) {
302 		nbytes = min(sg_wa->sg->length - sg_wa->sg_used,
303 			     dm_wa->length - buf_count);
304 		nbytes = min_t(u64, sg_wa->bytes_left, nbytes);
305 
306 		buf_count += nbytes;
307 		ccp_update_sg_workarea(sg_wa, nbytes);
308 	}
309 
310 	return buf_count;
311 }
312 
313 static unsigned int ccp_fill_queue_buf(struct ccp_data *data)
314 {
315 	return ccp_queue_buf(data, 0);
316 }
317 
318 static unsigned int ccp_empty_queue_buf(struct ccp_data *data)
319 {
320 	return ccp_queue_buf(data, 1);
321 }
322 
323 static void ccp_prepare_data(struct ccp_data *src, struct ccp_data *dst,
324 			     struct ccp_op *op, unsigned int block_size,
325 			     bool blocksize_op)
326 {
327 	unsigned int sg_src_len, sg_dst_len, op_len;
328 
329 	/* The CCP can only DMA from/to one address each per operation. This
330 	 * requires that we find the smallest DMA area between the source
331 	 * and destination. The resulting len values will always be <= UINT_MAX
332 	 * because the dma length is an unsigned int.
333 	 */
334 	sg_src_len = sg_dma_len(src->sg_wa.sg) - src->sg_wa.sg_used;
335 	sg_src_len = min_t(u64, src->sg_wa.bytes_left, sg_src_len);
336 
337 	if (dst) {
338 		sg_dst_len = sg_dma_len(dst->sg_wa.sg) - dst->sg_wa.sg_used;
339 		sg_dst_len = min_t(u64, src->sg_wa.bytes_left, sg_dst_len);
340 		op_len = min(sg_src_len, sg_dst_len);
341 	} else {
342 		op_len = sg_src_len;
343 	}
344 
345 	/* The data operation length will be at least block_size in length
346 	 * or the smaller of available sg room remaining for the source or
347 	 * the destination
348 	 */
349 	op_len = max(op_len, block_size);
350 
351 	/* Unless we have to buffer data, there's no reason to wait */
352 	op->soc = 0;
353 
354 	if (sg_src_len < block_size) {
355 		/* Not enough data in the sg element, so it
356 		 * needs to be buffered into a blocksize chunk
357 		 */
358 		int cp_len = ccp_fill_queue_buf(src);
359 
360 		op->soc = 1;
361 		op->src.u.dma.address = src->dm_wa.dma.address;
362 		op->src.u.dma.offset = 0;
363 		op->src.u.dma.length = (blocksize_op) ? block_size : cp_len;
364 	} else {
365 		/* Enough data in the sg element, but we need to
366 		 * adjust for any previously copied data
367 		 */
368 		op->src.u.dma.address = sg_dma_address(src->sg_wa.sg);
369 		op->src.u.dma.offset = src->sg_wa.sg_used;
370 		op->src.u.dma.length = op_len & ~(block_size - 1);
371 
372 		ccp_update_sg_workarea(&src->sg_wa, op->src.u.dma.length);
373 	}
374 
375 	if (dst) {
376 		if (sg_dst_len < block_size) {
377 			/* Not enough room in the sg element or we're on the
378 			 * last piece of data (when using padding), so the
379 			 * output needs to be buffered into a blocksize chunk
380 			 */
381 			op->soc = 1;
382 			op->dst.u.dma.address = dst->dm_wa.dma.address;
383 			op->dst.u.dma.offset = 0;
384 			op->dst.u.dma.length = op->src.u.dma.length;
385 		} else {
386 			/* Enough room in the sg element, but we need to
387 			 * adjust for any previously used area
388 			 */
389 			op->dst.u.dma.address = sg_dma_address(dst->sg_wa.sg);
390 			op->dst.u.dma.offset = dst->sg_wa.sg_used;
391 			op->dst.u.dma.length = op->src.u.dma.length;
392 		}
393 	}
394 }
395 
396 static void ccp_process_data(struct ccp_data *src, struct ccp_data *dst,
397 			     struct ccp_op *op)
398 {
399 	op->init = 0;
400 
401 	if (dst) {
402 		if (op->dst.u.dma.address == dst->dm_wa.dma.address)
403 			ccp_empty_queue_buf(dst);
404 		else
405 			ccp_update_sg_workarea(&dst->sg_wa,
406 					       op->dst.u.dma.length);
407 	}
408 }
409 
410 static int ccp_copy_to_from_sb(struct ccp_cmd_queue *cmd_q,
411 			       struct ccp_dm_workarea *wa, u32 jobid, u32 sb,
412 			       u32 byte_swap, bool from)
413 {
414 	struct ccp_op op;
415 
416 	memset(&op, 0, sizeof(op));
417 
418 	op.cmd_q = cmd_q;
419 	op.jobid = jobid;
420 	op.eom = 1;
421 
422 	if (from) {
423 		op.soc = 1;
424 		op.src.type = CCP_MEMTYPE_SB;
425 		op.src.u.sb = sb;
426 		op.dst.type = CCP_MEMTYPE_SYSTEM;
427 		op.dst.u.dma.address = wa->dma.address;
428 		op.dst.u.dma.length = wa->length;
429 	} else {
430 		op.src.type = CCP_MEMTYPE_SYSTEM;
431 		op.src.u.dma.address = wa->dma.address;
432 		op.src.u.dma.length = wa->length;
433 		op.dst.type = CCP_MEMTYPE_SB;
434 		op.dst.u.sb = sb;
435 	}
436 
437 	op.u.passthru.byte_swap = byte_swap;
438 
439 	return cmd_q->ccp->vdata->perform->passthru(&op);
440 }
441 
442 static int ccp_copy_to_sb(struct ccp_cmd_queue *cmd_q,
443 			  struct ccp_dm_workarea *wa, u32 jobid, u32 sb,
444 			  u32 byte_swap)
445 {
446 	return ccp_copy_to_from_sb(cmd_q, wa, jobid, sb, byte_swap, false);
447 }
448 
449 static int ccp_copy_from_sb(struct ccp_cmd_queue *cmd_q,
450 			    struct ccp_dm_workarea *wa, u32 jobid, u32 sb,
451 			    u32 byte_swap)
452 {
453 	return ccp_copy_to_from_sb(cmd_q, wa, jobid, sb, byte_swap, true);
454 }
455 
456 static noinline_for_stack int
457 ccp_run_aes_cmac_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
458 {
459 	struct ccp_aes_engine *aes = &cmd->u.aes;
460 	struct ccp_dm_workarea key, ctx;
461 	struct ccp_data src;
462 	struct ccp_op op;
463 	unsigned int dm_offset;
464 	int ret;
465 
466 	if (!((aes->key_len == AES_KEYSIZE_128) ||
467 	      (aes->key_len == AES_KEYSIZE_192) ||
468 	      (aes->key_len == AES_KEYSIZE_256)))
469 		return -EINVAL;
470 
471 	if (aes->src_len & (AES_BLOCK_SIZE - 1))
472 		return -EINVAL;
473 
474 	if (aes->iv_len != AES_BLOCK_SIZE)
475 		return -EINVAL;
476 
477 	if (!aes->key || !aes->iv || !aes->src)
478 		return -EINVAL;
479 
480 	if (aes->cmac_final) {
481 		if (aes->cmac_key_len != AES_BLOCK_SIZE)
482 			return -EINVAL;
483 
484 		if (!aes->cmac_key)
485 			return -EINVAL;
486 	}
487 
488 	BUILD_BUG_ON(CCP_AES_KEY_SB_COUNT != 1);
489 	BUILD_BUG_ON(CCP_AES_CTX_SB_COUNT != 1);
490 
491 	ret = -EIO;
492 	memset(&op, 0, sizeof(op));
493 	op.cmd_q = cmd_q;
494 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
495 	op.sb_key = cmd_q->sb_key;
496 	op.sb_ctx = cmd_q->sb_ctx;
497 	op.init = 1;
498 	op.u.aes.type = aes->type;
499 	op.u.aes.mode = aes->mode;
500 	op.u.aes.action = aes->action;
501 
502 	/* All supported key sizes fit in a single (32-byte) SB entry
503 	 * and must be in little endian format. Use the 256-bit byte
504 	 * swap passthru option to convert from big endian to little
505 	 * endian.
506 	 */
507 	ret = ccp_init_dm_workarea(&key, cmd_q,
508 				   CCP_AES_KEY_SB_COUNT * CCP_SB_BYTES,
509 				   DMA_TO_DEVICE);
510 	if (ret)
511 		return ret;
512 
513 	dm_offset = CCP_SB_BYTES - aes->key_len;
514 	ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len);
515 	if (ret)
516 		goto e_key;
517 	ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key,
518 			     CCP_PASSTHRU_BYTESWAP_256BIT);
519 	if (ret) {
520 		cmd->engine_error = cmd_q->cmd_error;
521 		goto e_key;
522 	}
523 
524 	/* The AES context fits in a single (32-byte) SB entry and
525 	 * must be in little endian format. Use the 256-bit byte swap
526 	 * passthru option to convert from big endian to little endian.
527 	 */
528 	ret = ccp_init_dm_workarea(&ctx, cmd_q,
529 				   CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES,
530 				   DMA_BIDIRECTIONAL);
531 	if (ret)
532 		goto e_key;
533 
534 	dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE;
535 	ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
536 	if (ret)
537 		goto e_ctx;
538 	ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
539 			     CCP_PASSTHRU_BYTESWAP_256BIT);
540 	if (ret) {
541 		cmd->engine_error = cmd_q->cmd_error;
542 		goto e_ctx;
543 	}
544 
545 	/* Send data to the CCP AES engine */
546 	ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len,
547 			    AES_BLOCK_SIZE, DMA_TO_DEVICE);
548 	if (ret)
549 		goto e_ctx;
550 
551 	while (src.sg_wa.bytes_left) {
552 		ccp_prepare_data(&src, NULL, &op, AES_BLOCK_SIZE, true);
553 		if (aes->cmac_final && !src.sg_wa.bytes_left) {
554 			op.eom = 1;
555 
556 			/* Push the K1/K2 key to the CCP now */
557 			ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid,
558 					       op.sb_ctx,
559 					       CCP_PASSTHRU_BYTESWAP_256BIT);
560 			if (ret) {
561 				cmd->engine_error = cmd_q->cmd_error;
562 				goto e_src;
563 			}
564 
565 			ret = ccp_set_dm_area(&ctx, 0, aes->cmac_key, 0,
566 					      aes->cmac_key_len);
567 			if (ret)
568 				goto e_src;
569 			ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
570 					     CCP_PASSTHRU_BYTESWAP_256BIT);
571 			if (ret) {
572 				cmd->engine_error = cmd_q->cmd_error;
573 				goto e_src;
574 			}
575 		}
576 
577 		ret = cmd_q->ccp->vdata->perform->aes(&op);
578 		if (ret) {
579 			cmd->engine_error = cmd_q->cmd_error;
580 			goto e_src;
581 		}
582 
583 		ccp_process_data(&src, NULL, &op);
584 	}
585 
586 	/* Retrieve the AES context - convert from LE to BE using
587 	 * 32-byte (256-bit) byteswapping
588 	 */
589 	ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
590 			       CCP_PASSTHRU_BYTESWAP_256BIT);
591 	if (ret) {
592 		cmd->engine_error = cmd_q->cmd_error;
593 		goto e_src;
594 	}
595 
596 	/* ...but we only need AES_BLOCK_SIZE bytes */
597 	dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE;
598 	ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
599 
600 e_src:
601 	ccp_free_data(&src, cmd_q);
602 
603 e_ctx:
604 	ccp_dm_free(&ctx);
605 
606 e_key:
607 	ccp_dm_free(&key);
608 
609 	return ret;
610 }
611 
612 static noinline_for_stack int
613 ccp_run_aes_gcm_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
614 {
615 	struct ccp_aes_engine *aes = &cmd->u.aes;
616 	struct ccp_dm_workarea key, ctx, final_wa, tag;
617 	struct ccp_data src, dst;
618 	struct ccp_data aad;
619 	struct ccp_op op;
620 
621 	unsigned long long *final;
622 	unsigned int dm_offset;
623 	unsigned int jobid;
624 	unsigned int ilen;
625 	bool in_place = true; /* Default value */
626 	int ret;
627 
628 	struct scatterlist *p_inp, sg_inp[2];
629 	struct scatterlist *p_tag, sg_tag[2];
630 	struct scatterlist *p_outp, sg_outp[2];
631 	struct scatterlist *p_aad;
632 
633 	if (!aes->iv)
634 		return -EINVAL;
635 
636 	if (!((aes->key_len == AES_KEYSIZE_128) ||
637 		(aes->key_len == AES_KEYSIZE_192) ||
638 		(aes->key_len == AES_KEYSIZE_256)))
639 		return -EINVAL;
640 
641 	if (!aes->key) /* Gotta have a key SGL */
642 		return -EINVAL;
643 
644 	/* First, decompose the source buffer into AAD & PT,
645 	 * and the destination buffer into AAD, CT & tag, or
646 	 * the input into CT & tag.
647 	 * It is expected that the input and output SGs will
648 	 * be valid, even if the AAD and input lengths are 0.
649 	 */
650 	p_aad = aes->src;
651 	p_inp = scatterwalk_ffwd(sg_inp, aes->src, aes->aad_len);
652 	p_outp = scatterwalk_ffwd(sg_outp, aes->dst, aes->aad_len);
653 	if (aes->action == CCP_AES_ACTION_ENCRYPT) {
654 		ilen = aes->src_len;
655 		p_tag = scatterwalk_ffwd(sg_tag, p_outp, ilen);
656 	} else {
657 		/* Input length for decryption includes tag */
658 		ilen = aes->src_len - AES_BLOCK_SIZE;
659 		p_tag = scatterwalk_ffwd(sg_tag, p_inp, ilen);
660 	}
661 
662 	jobid = CCP_NEW_JOBID(cmd_q->ccp);
663 
664 	memset(&op, 0, sizeof(op));
665 	op.cmd_q = cmd_q;
666 	op.jobid = jobid;
667 	op.sb_key = cmd_q->sb_key; /* Pre-allocated */
668 	op.sb_ctx = cmd_q->sb_ctx; /* Pre-allocated */
669 	op.init = 1;
670 	op.u.aes.type = aes->type;
671 
672 	/* Copy the key to the LSB */
673 	ret = ccp_init_dm_workarea(&key, cmd_q,
674 				   CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES,
675 				   DMA_TO_DEVICE);
676 	if (ret)
677 		return ret;
678 
679 	dm_offset = CCP_SB_BYTES - aes->key_len;
680 	ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len);
681 	if (ret)
682 		goto e_key;
683 	ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key,
684 			     CCP_PASSTHRU_BYTESWAP_256BIT);
685 	if (ret) {
686 		cmd->engine_error = cmd_q->cmd_error;
687 		goto e_key;
688 	}
689 
690 	/* Copy the context (IV) to the LSB.
691 	 * There is an assumption here that the IV is 96 bits in length, plus
692 	 * a nonce of 32 bits. If no IV is present, use a zeroed buffer.
693 	 */
694 	ret = ccp_init_dm_workarea(&ctx, cmd_q,
695 				   CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES,
696 				   DMA_BIDIRECTIONAL);
697 	if (ret)
698 		goto e_key;
699 
700 	dm_offset = CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES - aes->iv_len;
701 	ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
702 	if (ret)
703 		goto e_ctx;
704 
705 	ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
706 			     CCP_PASSTHRU_BYTESWAP_256BIT);
707 	if (ret) {
708 		cmd->engine_error = cmd_q->cmd_error;
709 		goto e_ctx;
710 	}
711 
712 	op.init = 1;
713 	if (aes->aad_len > 0) {
714 		/* Step 1: Run a GHASH over the Additional Authenticated Data */
715 		ret = ccp_init_data(&aad, cmd_q, p_aad, aes->aad_len,
716 				    AES_BLOCK_SIZE,
717 				    DMA_TO_DEVICE);
718 		if (ret)
719 			goto e_ctx;
720 
721 		op.u.aes.mode = CCP_AES_MODE_GHASH;
722 		op.u.aes.action = CCP_AES_GHASHAAD;
723 
724 		while (aad.sg_wa.bytes_left) {
725 			ccp_prepare_data(&aad, NULL, &op, AES_BLOCK_SIZE, true);
726 
727 			ret = cmd_q->ccp->vdata->perform->aes(&op);
728 			if (ret) {
729 				cmd->engine_error = cmd_q->cmd_error;
730 				goto e_aad;
731 			}
732 
733 			ccp_process_data(&aad, NULL, &op);
734 			op.init = 0;
735 		}
736 	}
737 
738 	op.u.aes.mode = CCP_AES_MODE_GCTR;
739 	op.u.aes.action = aes->action;
740 
741 	if (ilen > 0) {
742 		/* Step 2: Run a GCTR over the plaintext */
743 		in_place = (sg_virt(p_inp) == sg_virt(p_outp)) ? true : false;
744 
745 		ret = ccp_init_data(&src, cmd_q, p_inp, ilen,
746 				    AES_BLOCK_SIZE,
747 				    in_place ? DMA_BIDIRECTIONAL
748 					     : DMA_TO_DEVICE);
749 		if (ret)
750 			goto e_ctx;
751 
752 		if (in_place) {
753 			dst = src;
754 		} else {
755 			ret = ccp_init_data(&dst, cmd_q, p_outp, ilen,
756 					    AES_BLOCK_SIZE, DMA_FROM_DEVICE);
757 			if (ret)
758 				goto e_src;
759 		}
760 
761 		op.soc = 0;
762 		op.eom = 0;
763 		op.init = 1;
764 		while (src.sg_wa.bytes_left) {
765 			ccp_prepare_data(&src, &dst, &op, AES_BLOCK_SIZE, true);
766 			if (!src.sg_wa.bytes_left) {
767 				unsigned int nbytes = aes->src_len
768 						      % AES_BLOCK_SIZE;
769 
770 				if (nbytes) {
771 					op.eom = 1;
772 					op.u.aes.size = (nbytes * 8) - 1;
773 				}
774 			}
775 
776 			ret = cmd_q->ccp->vdata->perform->aes(&op);
777 			if (ret) {
778 				cmd->engine_error = cmd_q->cmd_error;
779 				goto e_dst;
780 			}
781 
782 			ccp_process_data(&src, &dst, &op);
783 			op.init = 0;
784 		}
785 	}
786 
787 	/* Step 3: Update the IV portion of the context with the original IV */
788 	ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
789 			       CCP_PASSTHRU_BYTESWAP_256BIT);
790 	if (ret) {
791 		cmd->engine_error = cmd_q->cmd_error;
792 		goto e_dst;
793 	}
794 
795 	ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
796 	if (ret)
797 		goto e_dst;
798 
799 	ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
800 			     CCP_PASSTHRU_BYTESWAP_256BIT);
801 	if (ret) {
802 		cmd->engine_error = cmd_q->cmd_error;
803 		goto e_dst;
804 	}
805 
806 	/* Step 4: Concatenate the lengths of the AAD and source, and
807 	 * hash that 16 byte buffer.
808 	 */
809 	ret = ccp_init_dm_workarea(&final_wa, cmd_q, AES_BLOCK_SIZE,
810 				   DMA_BIDIRECTIONAL);
811 	if (ret)
812 		goto e_dst;
813 	final = (unsigned long long *) final_wa.address;
814 	final[0] = cpu_to_be64(aes->aad_len * 8);
815 	final[1] = cpu_to_be64(ilen * 8);
816 
817 	memset(&op, 0, sizeof(op));
818 	op.cmd_q = cmd_q;
819 	op.jobid = jobid;
820 	op.sb_key = cmd_q->sb_key; /* Pre-allocated */
821 	op.sb_ctx = cmd_q->sb_ctx; /* Pre-allocated */
822 	op.init = 1;
823 	op.u.aes.type = aes->type;
824 	op.u.aes.mode = CCP_AES_MODE_GHASH;
825 	op.u.aes.action = CCP_AES_GHASHFINAL;
826 	op.src.type = CCP_MEMTYPE_SYSTEM;
827 	op.src.u.dma.address = final_wa.dma.address;
828 	op.src.u.dma.length = AES_BLOCK_SIZE;
829 	op.dst.type = CCP_MEMTYPE_SYSTEM;
830 	op.dst.u.dma.address = final_wa.dma.address;
831 	op.dst.u.dma.length = AES_BLOCK_SIZE;
832 	op.eom = 1;
833 	op.u.aes.size = 0;
834 	ret = cmd_q->ccp->vdata->perform->aes(&op);
835 	if (ret)
836 		goto e_dst;
837 
838 	if (aes->action == CCP_AES_ACTION_ENCRYPT) {
839 		/* Put the ciphered tag after the ciphertext. */
840 		ccp_get_dm_area(&final_wa, 0, p_tag, 0, AES_BLOCK_SIZE);
841 	} else {
842 		/* Does this ciphered tag match the input? */
843 		ret = ccp_init_dm_workarea(&tag, cmd_q, AES_BLOCK_SIZE,
844 					   DMA_BIDIRECTIONAL);
845 		if (ret)
846 			goto e_tag;
847 		ret = ccp_set_dm_area(&tag, 0, p_tag, 0, AES_BLOCK_SIZE);
848 		if (ret)
849 			goto e_tag;
850 
851 		ret = crypto_memneq(tag.address, final_wa.address,
852 				    AES_BLOCK_SIZE) ? -EBADMSG : 0;
853 		ccp_dm_free(&tag);
854 	}
855 
856 e_tag:
857 	ccp_dm_free(&final_wa);
858 
859 e_dst:
860 	if (aes->src_len && !in_place)
861 		ccp_free_data(&dst, cmd_q);
862 
863 e_src:
864 	if (aes->src_len)
865 		ccp_free_data(&src, cmd_q);
866 
867 e_aad:
868 	if (aes->aad_len)
869 		ccp_free_data(&aad, cmd_q);
870 
871 e_ctx:
872 	ccp_dm_free(&ctx);
873 
874 e_key:
875 	ccp_dm_free(&key);
876 
877 	return ret;
878 }
879 
880 static noinline_for_stack int
881 ccp_run_aes_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
882 {
883 	struct ccp_aes_engine *aes = &cmd->u.aes;
884 	struct ccp_dm_workarea key, ctx;
885 	struct ccp_data src, dst;
886 	struct ccp_op op;
887 	unsigned int dm_offset;
888 	bool in_place = false;
889 	int ret;
890 
891 	if (!((aes->key_len == AES_KEYSIZE_128) ||
892 	      (aes->key_len == AES_KEYSIZE_192) ||
893 	      (aes->key_len == AES_KEYSIZE_256)))
894 		return -EINVAL;
895 
896 	if (((aes->mode == CCP_AES_MODE_ECB) ||
897 	     (aes->mode == CCP_AES_MODE_CBC)) &&
898 	    (aes->src_len & (AES_BLOCK_SIZE - 1)))
899 		return -EINVAL;
900 
901 	if (!aes->key || !aes->src || !aes->dst)
902 		return -EINVAL;
903 
904 	if (aes->mode != CCP_AES_MODE_ECB) {
905 		if (aes->iv_len != AES_BLOCK_SIZE)
906 			return -EINVAL;
907 
908 		if (!aes->iv)
909 			return -EINVAL;
910 	}
911 
912 	BUILD_BUG_ON(CCP_AES_KEY_SB_COUNT != 1);
913 	BUILD_BUG_ON(CCP_AES_CTX_SB_COUNT != 1);
914 
915 	ret = -EIO;
916 	memset(&op, 0, sizeof(op));
917 	op.cmd_q = cmd_q;
918 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
919 	op.sb_key = cmd_q->sb_key;
920 	op.sb_ctx = cmd_q->sb_ctx;
921 	op.init = (aes->mode == CCP_AES_MODE_ECB) ? 0 : 1;
922 	op.u.aes.type = aes->type;
923 	op.u.aes.mode = aes->mode;
924 	op.u.aes.action = aes->action;
925 
926 	/* All supported key sizes fit in a single (32-byte) SB entry
927 	 * and must be in little endian format. Use the 256-bit byte
928 	 * swap passthru option to convert from big endian to little
929 	 * endian.
930 	 */
931 	ret = ccp_init_dm_workarea(&key, cmd_q,
932 				   CCP_AES_KEY_SB_COUNT * CCP_SB_BYTES,
933 				   DMA_TO_DEVICE);
934 	if (ret)
935 		return ret;
936 
937 	dm_offset = CCP_SB_BYTES - aes->key_len;
938 	ret = ccp_set_dm_area(&key, dm_offset, aes->key, 0, aes->key_len);
939 	if (ret)
940 		goto e_key;
941 	ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key,
942 			     CCP_PASSTHRU_BYTESWAP_256BIT);
943 	if (ret) {
944 		cmd->engine_error = cmd_q->cmd_error;
945 		goto e_key;
946 	}
947 
948 	/* The AES context fits in a single (32-byte) SB entry and
949 	 * must be in little endian format. Use the 256-bit byte swap
950 	 * passthru option to convert from big endian to little endian.
951 	 */
952 	ret = ccp_init_dm_workarea(&ctx, cmd_q,
953 				   CCP_AES_CTX_SB_COUNT * CCP_SB_BYTES,
954 				   DMA_BIDIRECTIONAL);
955 	if (ret)
956 		goto e_key;
957 
958 	if (aes->mode != CCP_AES_MODE_ECB) {
959 		/* Load the AES context - convert to LE */
960 		dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE;
961 		ret = ccp_set_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
962 		if (ret)
963 			goto e_ctx;
964 		ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
965 				     CCP_PASSTHRU_BYTESWAP_256BIT);
966 		if (ret) {
967 			cmd->engine_error = cmd_q->cmd_error;
968 			goto e_ctx;
969 		}
970 	}
971 	switch (aes->mode) {
972 	case CCP_AES_MODE_CFB: /* CFB128 only */
973 	case CCP_AES_MODE_CTR:
974 		op.u.aes.size = AES_BLOCK_SIZE * BITS_PER_BYTE - 1;
975 		break;
976 	default:
977 		op.u.aes.size = 0;
978 	}
979 
980 	/* Prepare the input and output data workareas. For in-place
981 	 * operations we need to set the dma direction to BIDIRECTIONAL
982 	 * and copy the src workarea to the dst workarea.
983 	 */
984 	if (sg_virt(aes->src) == sg_virt(aes->dst))
985 		in_place = true;
986 
987 	ret = ccp_init_data(&src, cmd_q, aes->src, aes->src_len,
988 			    AES_BLOCK_SIZE,
989 			    in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE);
990 	if (ret)
991 		goto e_ctx;
992 
993 	if (in_place) {
994 		dst = src;
995 	} else {
996 		ret = ccp_init_data(&dst, cmd_q, aes->dst, aes->src_len,
997 				    AES_BLOCK_SIZE, DMA_FROM_DEVICE);
998 		if (ret)
999 			goto e_src;
1000 	}
1001 
1002 	/* Send data to the CCP AES engine */
1003 	while (src.sg_wa.bytes_left) {
1004 		ccp_prepare_data(&src, &dst, &op, AES_BLOCK_SIZE, true);
1005 		if (!src.sg_wa.bytes_left) {
1006 			op.eom = 1;
1007 
1008 			/* Since we don't retrieve the AES context in ECB
1009 			 * mode we have to wait for the operation to complete
1010 			 * on the last piece of data
1011 			 */
1012 			if (aes->mode == CCP_AES_MODE_ECB)
1013 				op.soc = 1;
1014 		}
1015 
1016 		ret = cmd_q->ccp->vdata->perform->aes(&op);
1017 		if (ret) {
1018 			cmd->engine_error = cmd_q->cmd_error;
1019 			goto e_dst;
1020 		}
1021 
1022 		ccp_process_data(&src, &dst, &op);
1023 	}
1024 
1025 	if (aes->mode != CCP_AES_MODE_ECB) {
1026 		/* Retrieve the AES context - convert from LE to BE using
1027 		 * 32-byte (256-bit) byteswapping
1028 		 */
1029 		ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1030 				       CCP_PASSTHRU_BYTESWAP_256BIT);
1031 		if (ret) {
1032 			cmd->engine_error = cmd_q->cmd_error;
1033 			goto e_dst;
1034 		}
1035 
1036 		/* ...but we only need AES_BLOCK_SIZE bytes */
1037 		dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE;
1038 		ccp_get_dm_area(&ctx, dm_offset, aes->iv, 0, aes->iv_len);
1039 	}
1040 
1041 e_dst:
1042 	if (!in_place)
1043 		ccp_free_data(&dst, cmd_q);
1044 
1045 e_src:
1046 	ccp_free_data(&src, cmd_q);
1047 
1048 e_ctx:
1049 	ccp_dm_free(&ctx);
1050 
1051 e_key:
1052 	ccp_dm_free(&key);
1053 
1054 	return ret;
1055 }
1056 
1057 static noinline_for_stack int
1058 ccp_run_xts_aes_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
1059 {
1060 	struct ccp_xts_aes_engine *xts = &cmd->u.xts;
1061 	struct ccp_dm_workarea key, ctx;
1062 	struct ccp_data src, dst;
1063 	struct ccp_op op;
1064 	unsigned int unit_size, dm_offset;
1065 	bool in_place = false;
1066 	unsigned int sb_count;
1067 	enum ccp_aes_type aestype;
1068 	int ret;
1069 
1070 	switch (xts->unit_size) {
1071 	case CCP_XTS_AES_UNIT_SIZE_16:
1072 		unit_size = 16;
1073 		break;
1074 	case CCP_XTS_AES_UNIT_SIZE_512:
1075 		unit_size = 512;
1076 		break;
1077 	case CCP_XTS_AES_UNIT_SIZE_1024:
1078 		unit_size = 1024;
1079 		break;
1080 	case CCP_XTS_AES_UNIT_SIZE_2048:
1081 		unit_size = 2048;
1082 		break;
1083 	case CCP_XTS_AES_UNIT_SIZE_4096:
1084 		unit_size = 4096;
1085 		break;
1086 
1087 	default:
1088 		return -EINVAL;
1089 	}
1090 
1091 	if (xts->key_len == AES_KEYSIZE_128)
1092 		aestype = CCP_AES_TYPE_128;
1093 	else if (xts->key_len == AES_KEYSIZE_256)
1094 		aestype = CCP_AES_TYPE_256;
1095 	else
1096 		return -EINVAL;
1097 
1098 	if (!xts->final && (xts->src_len & (AES_BLOCK_SIZE - 1)))
1099 		return -EINVAL;
1100 
1101 	if (xts->iv_len != AES_BLOCK_SIZE)
1102 		return -EINVAL;
1103 
1104 	if (!xts->key || !xts->iv || !xts->src || !xts->dst)
1105 		return -EINVAL;
1106 
1107 	BUILD_BUG_ON(CCP_XTS_AES_KEY_SB_COUNT != 1);
1108 	BUILD_BUG_ON(CCP_XTS_AES_CTX_SB_COUNT != 1);
1109 
1110 	ret = -EIO;
1111 	memset(&op, 0, sizeof(op));
1112 	op.cmd_q = cmd_q;
1113 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
1114 	op.sb_key = cmd_q->sb_key;
1115 	op.sb_ctx = cmd_q->sb_ctx;
1116 	op.init = 1;
1117 	op.u.xts.type = aestype;
1118 	op.u.xts.action = xts->action;
1119 	op.u.xts.unit_size = xts->unit_size;
1120 
1121 	/* A version 3 device only supports 128-bit keys, which fits into a
1122 	 * single SB entry. A version 5 device uses a 512-bit vector, so two
1123 	 * SB entries.
1124 	 */
1125 	if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0))
1126 		sb_count = CCP_XTS_AES_KEY_SB_COUNT;
1127 	else
1128 		sb_count = CCP5_XTS_AES_KEY_SB_COUNT;
1129 	ret = ccp_init_dm_workarea(&key, cmd_q,
1130 				   sb_count * CCP_SB_BYTES,
1131 				   DMA_TO_DEVICE);
1132 	if (ret)
1133 		return ret;
1134 
1135 	if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) {
1136 		/* All supported key sizes must be in little endian format.
1137 		 * Use the 256-bit byte swap passthru option to convert from
1138 		 * big endian to little endian.
1139 		 */
1140 		dm_offset = CCP_SB_BYTES - AES_KEYSIZE_128;
1141 		ret = ccp_set_dm_area(&key, dm_offset, xts->key, 0, xts->key_len);
1142 		if (ret)
1143 			goto e_key;
1144 		ret = ccp_set_dm_area(&key, 0, xts->key, xts->key_len, xts->key_len);
1145 		if (ret)
1146 			goto e_key;
1147 	} else {
1148 		/* Version 5 CCPs use a 512-bit space for the key: each portion
1149 		 * occupies 256 bits, or one entire slot, and is zero-padded.
1150 		 */
1151 		unsigned int pad;
1152 
1153 		dm_offset = CCP_SB_BYTES;
1154 		pad = dm_offset - xts->key_len;
1155 		ret = ccp_set_dm_area(&key, pad, xts->key, 0, xts->key_len);
1156 		if (ret)
1157 			goto e_key;
1158 		ret = ccp_set_dm_area(&key, dm_offset + pad, xts->key,
1159 				      xts->key_len, xts->key_len);
1160 		if (ret)
1161 			goto e_key;
1162 	}
1163 	ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key,
1164 			     CCP_PASSTHRU_BYTESWAP_256BIT);
1165 	if (ret) {
1166 		cmd->engine_error = cmd_q->cmd_error;
1167 		goto e_key;
1168 	}
1169 
1170 	/* The AES context fits in a single (32-byte) SB entry and
1171 	 * for XTS is already in little endian format so no byte swapping
1172 	 * is needed.
1173 	 */
1174 	ret = ccp_init_dm_workarea(&ctx, cmd_q,
1175 				   CCP_XTS_AES_CTX_SB_COUNT * CCP_SB_BYTES,
1176 				   DMA_BIDIRECTIONAL);
1177 	if (ret)
1178 		goto e_key;
1179 
1180 	ret = ccp_set_dm_area(&ctx, 0, xts->iv, 0, xts->iv_len);
1181 	if (ret)
1182 		goto e_ctx;
1183 	ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1184 			     CCP_PASSTHRU_BYTESWAP_NOOP);
1185 	if (ret) {
1186 		cmd->engine_error = cmd_q->cmd_error;
1187 		goto e_ctx;
1188 	}
1189 
1190 	/* Prepare the input and output data workareas. For in-place
1191 	 * operations we need to set the dma direction to BIDIRECTIONAL
1192 	 * and copy the src workarea to the dst workarea.
1193 	 */
1194 	if (sg_virt(xts->src) == sg_virt(xts->dst))
1195 		in_place = true;
1196 
1197 	ret = ccp_init_data(&src, cmd_q, xts->src, xts->src_len,
1198 			    unit_size,
1199 			    in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE);
1200 	if (ret)
1201 		goto e_ctx;
1202 
1203 	if (in_place) {
1204 		dst = src;
1205 	} else {
1206 		ret = ccp_init_data(&dst, cmd_q, xts->dst, xts->src_len,
1207 				    unit_size, DMA_FROM_DEVICE);
1208 		if (ret)
1209 			goto e_src;
1210 	}
1211 
1212 	/* Send data to the CCP AES engine */
1213 	while (src.sg_wa.bytes_left) {
1214 		ccp_prepare_data(&src, &dst, &op, unit_size, true);
1215 		if (!src.sg_wa.bytes_left)
1216 			op.eom = 1;
1217 
1218 		ret = cmd_q->ccp->vdata->perform->xts_aes(&op);
1219 		if (ret) {
1220 			cmd->engine_error = cmd_q->cmd_error;
1221 			goto e_dst;
1222 		}
1223 
1224 		ccp_process_data(&src, &dst, &op);
1225 	}
1226 
1227 	/* Retrieve the AES context - convert from LE to BE using
1228 	 * 32-byte (256-bit) byteswapping
1229 	 */
1230 	ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1231 			       CCP_PASSTHRU_BYTESWAP_256BIT);
1232 	if (ret) {
1233 		cmd->engine_error = cmd_q->cmd_error;
1234 		goto e_dst;
1235 	}
1236 
1237 	/* ...but we only need AES_BLOCK_SIZE bytes */
1238 	dm_offset = CCP_SB_BYTES - AES_BLOCK_SIZE;
1239 	ccp_get_dm_area(&ctx, dm_offset, xts->iv, 0, xts->iv_len);
1240 
1241 e_dst:
1242 	if (!in_place)
1243 		ccp_free_data(&dst, cmd_q);
1244 
1245 e_src:
1246 	ccp_free_data(&src, cmd_q);
1247 
1248 e_ctx:
1249 	ccp_dm_free(&ctx);
1250 
1251 e_key:
1252 	ccp_dm_free(&key);
1253 
1254 	return ret;
1255 }
1256 
1257 static noinline_for_stack int
1258 ccp_run_des3_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
1259 {
1260 	struct ccp_des3_engine *des3 = &cmd->u.des3;
1261 
1262 	struct ccp_dm_workarea key, ctx;
1263 	struct ccp_data src, dst;
1264 	struct ccp_op op;
1265 	unsigned int dm_offset;
1266 	unsigned int len_singlekey;
1267 	bool in_place = false;
1268 	int ret;
1269 
1270 	/* Error checks */
1271 	if (cmd_q->ccp->vdata->version < CCP_VERSION(5, 0))
1272 		return -EINVAL;
1273 
1274 	if (!cmd_q->ccp->vdata->perform->des3)
1275 		return -EINVAL;
1276 
1277 	if (des3->key_len != DES3_EDE_KEY_SIZE)
1278 		return -EINVAL;
1279 
1280 	if (((des3->mode == CCP_DES3_MODE_ECB) ||
1281 		(des3->mode == CCP_DES3_MODE_CBC)) &&
1282 		(des3->src_len & (DES3_EDE_BLOCK_SIZE - 1)))
1283 		return -EINVAL;
1284 
1285 	if (!des3->key || !des3->src || !des3->dst)
1286 		return -EINVAL;
1287 
1288 	if (des3->mode != CCP_DES3_MODE_ECB) {
1289 		if (des3->iv_len != DES3_EDE_BLOCK_SIZE)
1290 			return -EINVAL;
1291 
1292 		if (!des3->iv)
1293 			return -EINVAL;
1294 	}
1295 
1296 	ret = -EIO;
1297 	/* Zero out all the fields of the command desc */
1298 	memset(&op, 0, sizeof(op));
1299 
1300 	/* Set up the Function field */
1301 	op.cmd_q = cmd_q;
1302 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
1303 	op.sb_key = cmd_q->sb_key;
1304 
1305 	op.init = (des3->mode == CCP_DES3_MODE_ECB) ? 0 : 1;
1306 	op.u.des3.type = des3->type;
1307 	op.u.des3.mode = des3->mode;
1308 	op.u.des3.action = des3->action;
1309 
1310 	/*
1311 	 * All supported key sizes fit in a single (32-byte) KSB entry and
1312 	 * (like AES) must be in little endian format. Use the 256-bit byte
1313 	 * swap passthru option to convert from big endian to little endian.
1314 	 */
1315 	ret = ccp_init_dm_workarea(&key, cmd_q,
1316 				   CCP_DES3_KEY_SB_COUNT * CCP_SB_BYTES,
1317 				   DMA_TO_DEVICE);
1318 	if (ret)
1319 		return ret;
1320 
1321 	/*
1322 	 * The contents of the key triplet are in the reverse order of what
1323 	 * is required by the engine. Copy the 3 pieces individually to put
1324 	 * them where they belong.
1325 	 */
1326 	dm_offset = CCP_SB_BYTES - des3->key_len; /* Basic offset */
1327 
1328 	len_singlekey = des3->key_len / 3;
1329 	ret = ccp_set_dm_area(&key, dm_offset + 2 * len_singlekey,
1330 			      des3->key, 0, len_singlekey);
1331 	if (ret)
1332 		goto e_key;
1333 	ret = ccp_set_dm_area(&key, dm_offset + len_singlekey,
1334 			      des3->key, len_singlekey, len_singlekey);
1335 	if (ret)
1336 		goto e_key;
1337 	ret = ccp_set_dm_area(&key, dm_offset,
1338 			      des3->key, 2 * len_singlekey, len_singlekey);
1339 	if (ret)
1340 		goto e_key;
1341 
1342 	/* Copy the key to the SB */
1343 	ret = ccp_copy_to_sb(cmd_q, &key, op.jobid, op.sb_key,
1344 			     CCP_PASSTHRU_BYTESWAP_256BIT);
1345 	if (ret) {
1346 		cmd->engine_error = cmd_q->cmd_error;
1347 		goto e_key;
1348 	}
1349 
1350 	/*
1351 	 * The DES3 context fits in a single (32-byte) KSB entry and
1352 	 * must be in little endian format. Use the 256-bit byte swap
1353 	 * passthru option to convert from big endian to little endian.
1354 	 */
1355 	if (des3->mode != CCP_DES3_MODE_ECB) {
1356 		op.sb_ctx = cmd_q->sb_ctx;
1357 
1358 		ret = ccp_init_dm_workarea(&ctx, cmd_q,
1359 					   CCP_DES3_CTX_SB_COUNT * CCP_SB_BYTES,
1360 					   DMA_BIDIRECTIONAL);
1361 		if (ret)
1362 			goto e_key;
1363 
1364 		/* Load the context into the LSB */
1365 		dm_offset = CCP_SB_BYTES - des3->iv_len;
1366 		ret = ccp_set_dm_area(&ctx, dm_offset, des3->iv, 0,
1367 				      des3->iv_len);
1368 		if (ret)
1369 			goto e_ctx;
1370 
1371 		ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1372 				     CCP_PASSTHRU_BYTESWAP_256BIT);
1373 		if (ret) {
1374 			cmd->engine_error = cmd_q->cmd_error;
1375 			goto e_ctx;
1376 		}
1377 	}
1378 
1379 	/*
1380 	 * Prepare the input and output data workareas. For in-place
1381 	 * operations we need to set the dma direction to BIDIRECTIONAL
1382 	 * and copy the src workarea to the dst workarea.
1383 	 */
1384 	if (sg_virt(des3->src) == sg_virt(des3->dst))
1385 		in_place = true;
1386 
1387 	ret = ccp_init_data(&src, cmd_q, des3->src, des3->src_len,
1388 			DES3_EDE_BLOCK_SIZE,
1389 			in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE);
1390 	if (ret)
1391 		goto e_ctx;
1392 
1393 	if (in_place)
1394 		dst = src;
1395 	else {
1396 		ret = ccp_init_data(&dst, cmd_q, des3->dst, des3->src_len,
1397 				DES3_EDE_BLOCK_SIZE, DMA_FROM_DEVICE);
1398 		if (ret)
1399 			goto e_src;
1400 	}
1401 
1402 	/* Send data to the CCP DES3 engine */
1403 	while (src.sg_wa.bytes_left) {
1404 		ccp_prepare_data(&src, &dst, &op, DES3_EDE_BLOCK_SIZE, true);
1405 		if (!src.sg_wa.bytes_left) {
1406 			op.eom = 1;
1407 
1408 			/* Since we don't retrieve the context in ECB mode
1409 			 * we have to wait for the operation to complete
1410 			 * on the last piece of data
1411 			 */
1412 			op.soc = 0;
1413 		}
1414 
1415 		ret = cmd_q->ccp->vdata->perform->des3(&op);
1416 		if (ret) {
1417 			cmd->engine_error = cmd_q->cmd_error;
1418 			goto e_dst;
1419 		}
1420 
1421 		ccp_process_data(&src, &dst, &op);
1422 	}
1423 
1424 	if (des3->mode != CCP_DES3_MODE_ECB) {
1425 		/* Retrieve the context and make BE */
1426 		ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1427 				       CCP_PASSTHRU_BYTESWAP_256BIT);
1428 		if (ret) {
1429 			cmd->engine_error = cmd_q->cmd_error;
1430 			goto e_dst;
1431 		}
1432 
1433 		/* ...but we only need the last DES3_EDE_BLOCK_SIZE bytes */
1434 		ccp_get_dm_area(&ctx, dm_offset, des3->iv, 0,
1435 				DES3_EDE_BLOCK_SIZE);
1436 	}
1437 e_dst:
1438 	if (!in_place)
1439 		ccp_free_data(&dst, cmd_q);
1440 
1441 e_src:
1442 	ccp_free_data(&src, cmd_q);
1443 
1444 e_ctx:
1445 	if (des3->mode != CCP_DES3_MODE_ECB)
1446 		ccp_dm_free(&ctx);
1447 
1448 e_key:
1449 	ccp_dm_free(&key);
1450 
1451 	return ret;
1452 }
1453 
1454 static noinline_for_stack int
1455 ccp_run_sha_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
1456 {
1457 	struct ccp_sha_engine *sha = &cmd->u.sha;
1458 	struct ccp_dm_workarea ctx;
1459 	struct ccp_data src;
1460 	struct ccp_op op;
1461 	unsigned int ioffset, ooffset;
1462 	unsigned int digest_size;
1463 	int sb_count;
1464 	const void *init;
1465 	u64 block_size;
1466 	int ctx_size;
1467 	int ret;
1468 
1469 	switch (sha->type) {
1470 	case CCP_SHA_TYPE_1:
1471 		if (sha->ctx_len < SHA1_DIGEST_SIZE)
1472 			return -EINVAL;
1473 		block_size = SHA1_BLOCK_SIZE;
1474 		break;
1475 	case CCP_SHA_TYPE_224:
1476 		if (sha->ctx_len < SHA224_DIGEST_SIZE)
1477 			return -EINVAL;
1478 		block_size = SHA224_BLOCK_SIZE;
1479 		break;
1480 	case CCP_SHA_TYPE_256:
1481 		if (sha->ctx_len < SHA256_DIGEST_SIZE)
1482 			return -EINVAL;
1483 		block_size = SHA256_BLOCK_SIZE;
1484 		break;
1485 	case CCP_SHA_TYPE_384:
1486 		if (cmd_q->ccp->vdata->version < CCP_VERSION(4, 0)
1487 		    || sha->ctx_len < SHA384_DIGEST_SIZE)
1488 			return -EINVAL;
1489 		block_size = SHA384_BLOCK_SIZE;
1490 		break;
1491 	case CCP_SHA_TYPE_512:
1492 		if (cmd_q->ccp->vdata->version < CCP_VERSION(4, 0)
1493 		    || sha->ctx_len < SHA512_DIGEST_SIZE)
1494 			return -EINVAL;
1495 		block_size = SHA512_BLOCK_SIZE;
1496 		break;
1497 	default:
1498 		return -EINVAL;
1499 	}
1500 
1501 	if (!sha->ctx)
1502 		return -EINVAL;
1503 
1504 	if (!sha->final && (sha->src_len & (block_size - 1)))
1505 		return -EINVAL;
1506 
1507 	/* The version 3 device can't handle zero-length input */
1508 	if (cmd_q->ccp->vdata->version == CCP_VERSION(3, 0)) {
1509 
1510 		if (!sha->src_len) {
1511 			unsigned int digest_len;
1512 			const u8 *sha_zero;
1513 
1514 			/* Not final, just return */
1515 			if (!sha->final)
1516 				return 0;
1517 
1518 			/* CCP can't do a zero length sha operation so the
1519 			 * caller must buffer the data.
1520 			 */
1521 			if (sha->msg_bits)
1522 				return -EINVAL;
1523 
1524 			/* The CCP cannot perform zero-length sha operations
1525 			 * so the caller is required to buffer data for the
1526 			 * final operation. However, a sha operation for a
1527 			 * message with a total length of zero is valid so
1528 			 * known values are required to supply the result.
1529 			 */
1530 			switch (sha->type) {
1531 			case CCP_SHA_TYPE_1:
1532 				sha_zero = sha1_zero_message_hash;
1533 				digest_len = SHA1_DIGEST_SIZE;
1534 				break;
1535 			case CCP_SHA_TYPE_224:
1536 				sha_zero = sha224_zero_message_hash;
1537 				digest_len = SHA224_DIGEST_SIZE;
1538 				break;
1539 			case CCP_SHA_TYPE_256:
1540 				sha_zero = sha256_zero_message_hash;
1541 				digest_len = SHA256_DIGEST_SIZE;
1542 				break;
1543 			default:
1544 				return -EINVAL;
1545 			}
1546 
1547 			scatterwalk_map_and_copy((void *)sha_zero, sha->ctx, 0,
1548 						 digest_len, 1);
1549 
1550 			return 0;
1551 		}
1552 	}
1553 
1554 	/* Set variables used throughout */
1555 	switch (sha->type) {
1556 	case CCP_SHA_TYPE_1:
1557 		digest_size = SHA1_DIGEST_SIZE;
1558 		init = (void *) ccp_sha1_init;
1559 		ctx_size = SHA1_DIGEST_SIZE;
1560 		sb_count = 1;
1561 		if (cmd_q->ccp->vdata->version != CCP_VERSION(3, 0))
1562 			ooffset = ioffset = CCP_SB_BYTES - SHA1_DIGEST_SIZE;
1563 		else
1564 			ooffset = ioffset = 0;
1565 		break;
1566 	case CCP_SHA_TYPE_224:
1567 		digest_size = SHA224_DIGEST_SIZE;
1568 		init = (void *) ccp_sha224_init;
1569 		ctx_size = SHA256_DIGEST_SIZE;
1570 		sb_count = 1;
1571 		ioffset = 0;
1572 		if (cmd_q->ccp->vdata->version != CCP_VERSION(3, 0))
1573 			ooffset = CCP_SB_BYTES - SHA224_DIGEST_SIZE;
1574 		else
1575 			ooffset = 0;
1576 		break;
1577 	case CCP_SHA_TYPE_256:
1578 		digest_size = SHA256_DIGEST_SIZE;
1579 		init = (void *) ccp_sha256_init;
1580 		ctx_size = SHA256_DIGEST_SIZE;
1581 		sb_count = 1;
1582 		ooffset = ioffset = 0;
1583 		break;
1584 	case CCP_SHA_TYPE_384:
1585 		digest_size = SHA384_DIGEST_SIZE;
1586 		init = (void *) ccp_sha384_init;
1587 		ctx_size = SHA512_DIGEST_SIZE;
1588 		sb_count = 2;
1589 		ioffset = 0;
1590 		ooffset = 2 * CCP_SB_BYTES - SHA384_DIGEST_SIZE;
1591 		break;
1592 	case CCP_SHA_TYPE_512:
1593 		digest_size = SHA512_DIGEST_SIZE;
1594 		init = (void *) ccp_sha512_init;
1595 		ctx_size = SHA512_DIGEST_SIZE;
1596 		sb_count = 2;
1597 		ooffset = ioffset = 0;
1598 		break;
1599 	default:
1600 		ret = -EINVAL;
1601 		goto e_data;
1602 	}
1603 
1604 	/* For zero-length plaintext the src pointer is ignored;
1605 	 * otherwise both parts must be valid
1606 	 */
1607 	if (sha->src_len && !sha->src)
1608 		return -EINVAL;
1609 
1610 	memset(&op, 0, sizeof(op));
1611 	op.cmd_q = cmd_q;
1612 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
1613 	op.sb_ctx = cmd_q->sb_ctx; /* Pre-allocated */
1614 	op.u.sha.type = sha->type;
1615 	op.u.sha.msg_bits = sha->msg_bits;
1616 
1617 	/* For SHA1/224/256 the context fits in a single (32-byte) SB entry;
1618 	 * SHA384/512 require 2 adjacent SB slots, with the right half in the
1619 	 * first slot, and the left half in the second. Each portion must then
1620 	 * be in little endian format: use the 256-bit byte swap option.
1621 	 */
1622 	ret = ccp_init_dm_workarea(&ctx, cmd_q, sb_count * CCP_SB_BYTES,
1623 				   DMA_BIDIRECTIONAL);
1624 	if (ret)
1625 		return ret;
1626 	if (sha->first) {
1627 		switch (sha->type) {
1628 		case CCP_SHA_TYPE_1:
1629 		case CCP_SHA_TYPE_224:
1630 		case CCP_SHA_TYPE_256:
1631 			memcpy(ctx.address + ioffset, init, ctx_size);
1632 			break;
1633 		case CCP_SHA_TYPE_384:
1634 		case CCP_SHA_TYPE_512:
1635 			memcpy(ctx.address + ctx_size / 2, init,
1636 			       ctx_size / 2);
1637 			memcpy(ctx.address, init + ctx_size / 2,
1638 			       ctx_size / 2);
1639 			break;
1640 		default:
1641 			ret = -EINVAL;
1642 			goto e_ctx;
1643 		}
1644 	} else {
1645 		/* Restore the context */
1646 		ret = ccp_set_dm_area(&ctx, 0, sha->ctx, 0,
1647 				      sb_count * CCP_SB_BYTES);
1648 		if (ret)
1649 			goto e_ctx;
1650 	}
1651 
1652 	ret = ccp_copy_to_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1653 			     CCP_PASSTHRU_BYTESWAP_256BIT);
1654 	if (ret) {
1655 		cmd->engine_error = cmd_q->cmd_error;
1656 		goto e_ctx;
1657 	}
1658 
1659 	if (sha->src) {
1660 		/* Send data to the CCP SHA engine; block_size is set above */
1661 		ret = ccp_init_data(&src, cmd_q, sha->src, sha->src_len,
1662 				    block_size, DMA_TO_DEVICE);
1663 		if (ret)
1664 			goto e_ctx;
1665 
1666 		while (src.sg_wa.bytes_left) {
1667 			ccp_prepare_data(&src, NULL, &op, block_size, false);
1668 			if (sha->final && !src.sg_wa.bytes_left)
1669 				op.eom = 1;
1670 
1671 			ret = cmd_q->ccp->vdata->perform->sha(&op);
1672 			if (ret) {
1673 				cmd->engine_error = cmd_q->cmd_error;
1674 				goto e_data;
1675 			}
1676 
1677 			ccp_process_data(&src, NULL, &op);
1678 		}
1679 	} else {
1680 		op.eom = 1;
1681 		ret = cmd_q->ccp->vdata->perform->sha(&op);
1682 		if (ret) {
1683 			cmd->engine_error = cmd_q->cmd_error;
1684 			goto e_data;
1685 		}
1686 	}
1687 
1688 	/* Retrieve the SHA context - convert from LE to BE using
1689 	 * 32-byte (256-bit) byteswapping to BE
1690 	 */
1691 	ret = ccp_copy_from_sb(cmd_q, &ctx, op.jobid, op.sb_ctx,
1692 			       CCP_PASSTHRU_BYTESWAP_256BIT);
1693 	if (ret) {
1694 		cmd->engine_error = cmd_q->cmd_error;
1695 		goto e_data;
1696 	}
1697 
1698 	if (sha->final) {
1699 		/* Finishing up, so get the digest */
1700 		switch (sha->type) {
1701 		case CCP_SHA_TYPE_1:
1702 		case CCP_SHA_TYPE_224:
1703 		case CCP_SHA_TYPE_256:
1704 			ccp_get_dm_area(&ctx, ooffset,
1705 					sha->ctx, 0,
1706 					digest_size);
1707 			break;
1708 		case CCP_SHA_TYPE_384:
1709 		case CCP_SHA_TYPE_512:
1710 			ccp_get_dm_area(&ctx, 0,
1711 					sha->ctx, LSB_ITEM_SIZE - ooffset,
1712 					LSB_ITEM_SIZE);
1713 			ccp_get_dm_area(&ctx, LSB_ITEM_SIZE + ooffset,
1714 					sha->ctx, 0,
1715 					LSB_ITEM_SIZE - ooffset);
1716 			break;
1717 		default:
1718 			ret = -EINVAL;
1719 			goto e_ctx;
1720 		}
1721 	} else {
1722 		/* Stash the context */
1723 		ccp_get_dm_area(&ctx, 0, sha->ctx, 0,
1724 				sb_count * CCP_SB_BYTES);
1725 	}
1726 
1727 	if (sha->final && sha->opad) {
1728 		/* HMAC operation, recursively perform final SHA */
1729 		struct ccp_cmd hmac_cmd;
1730 		struct scatterlist sg;
1731 		u8 *hmac_buf;
1732 
1733 		if (sha->opad_len != block_size) {
1734 			ret = -EINVAL;
1735 			goto e_data;
1736 		}
1737 
1738 		hmac_buf = kmalloc(block_size + digest_size, GFP_KERNEL);
1739 		if (!hmac_buf) {
1740 			ret = -ENOMEM;
1741 			goto e_data;
1742 		}
1743 		sg_init_one(&sg, hmac_buf, block_size + digest_size);
1744 
1745 		scatterwalk_map_and_copy(hmac_buf, sha->opad, 0, block_size, 0);
1746 		switch (sha->type) {
1747 		case CCP_SHA_TYPE_1:
1748 		case CCP_SHA_TYPE_224:
1749 		case CCP_SHA_TYPE_256:
1750 			memcpy(hmac_buf + block_size,
1751 			       ctx.address + ooffset,
1752 			       digest_size);
1753 			break;
1754 		case CCP_SHA_TYPE_384:
1755 		case CCP_SHA_TYPE_512:
1756 			memcpy(hmac_buf + block_size,
1757 			       ctx.address + LSB_ITEM_SIZE + ooffset,
1758 			       LSB_ITEM_SIZE);
1759 			memcpy(hmac_buf + block_size +
1760 			       (LSB_ITEM_SIZE - ooffset),
1761 			       ctx.address,
1762 			       LSB_ITEM_SIZE);
1763 			break;
1764 		default:
1765 			ret = -EINVAL;
1766 			goto e_ctx;
1767 		}
1768 
1769 		memset(&hmac_cmd, 0, sizeof(hmac_cmd));
1770 		hmac_cmd.engine = CCP_ENGINE_SHA;
1771 		hmac_cmd.u.sha.type = sha->type;
1772 		hmac_cmd.u.sha.ctx = sha->ctx;
1773 		hmac_cmd.u.sha.ctx_len = sha->ctx_len;
1774 		hmac_cmd.u.sha.src = &sg;
1775 		hmac_cmd.u.sha.src_len = block_size + digest_size;
1776 		hmac_cmd.u.sha.opad = NULL;
1777 		hmac_cmd.u.sha.opad_len = 0;
1778 		hmac_cmd.u.sha.first = 1;
1779 		hmac_cmd.u.sha.final = 1;
1780 		hmac_cmd.u.sha.msg_bits = (block_size + digest_size) << 3;
1781 
1782 		ret = ccp_run_sha_cmd(cmd_q, &hmac_cmd);
1783 		if (ret)
1784 			cmd->engine_error = hmac_cmd.engine_error;
1785 
1786 		kfree(hmac_buf);
1787 	}
1788 
1789 e_data:
1790 	if (sha->src)
1791 		ccp_free_data(&src, cmd_q);
1792 
1793 e_ctx:
1794 	ccp_dm_free(&ctx);
1795 
1796 	return ret;
1797 }
1798 
1799 static noinline_for_stack int
1800 ccp_run_rsa_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
1801 {
1802 	struct ccp_rsa_engine *rsa = &cmd->u.rsa;
1803 	struct ccp_dm_workarea exp, src, dst;
1804 	struct ccp_op op;
1805 	unsigned int sb_count, i_len, o_len;
1806 	int ret;
1807 
1808 	/* Check against the maximum allowable size, in bits */
1809 	if (rsa->key_size > cmd_q->ccp->vdata->rsamax)
1810 		return -EINVAL;
1811 
1812 	if (!rsa->exp || !rsa->mod || !rsa->src || !rsa->dst)
1813 		return -EINVAL;
1814 
1815 	memset(&op, 0, sizeof(op));
1816 	op.cmd_q = cmd_q;
1817 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
1818 
1819 	/* The RSA modulus must precede the message being acted upon, so
1820 	 * it must be copied to a DMA area where the message and the
1821 	 * modulus can be concatenated.  Therefore the input buffer
1822 	 * length required is twice the output buffer length (which
1823 	 * must be a multiple of 256-bits).  Compute o_len, i_len in bytes.
1824 	 * Buffer sizes must be a multiple of 32 bytes; rounding up may be
1825 	 * required.
1826 	 */
1827 	o_len = 32 * ((rsa->key_size + 255) / 256);
1828 	i_len = o_len * 2;
1829 
1830 	sb_count = 0;
1831 	if (cmd_q->ccp->vdata->version < CCP_VERSION(5, 0)) {
1832 		/* sb_count is the number of storage block slots required
1833 		 * for the modulus.
1834 		 */
1835 		sb_count = o_len / CCP_SB_BYTES;
1836 		op.sb_key = cmd_q->ccp->vdata->perform->sballoc(cmd_q,
1837 								sb_count);
1838 		if (!op.sb_key)
1839 			return -EIO;
1840 	} else {
1841 		/* A version 5 device allows a modulus size that will not fit
1842 		 * in the LSB, so the command will transfer it from memory.
1843 		 * Set the sb key to the default, even though it's not used.
1844 		 */
1845 		op.sb_key = cmd_q->sb_key;
1846 	}
1847 
1848 	/* The RSA exponent must be in little endian format. Reverse its
1849 	 * byte order.
1850 	 */
1851 	ret = ccp_init_dm_workarea(&exp, cmd_q, o_len, DMA_TO_DEVICE);
1852 	if (ret)
1853 		goto e_sb;
1854 
1855 	ret = ccp_reverse_set_dm_area(&exp, 0, rsa->exp, 0, rsa->exp_len);
1856 	if (ret)
1857 		goto e_exp;
1858 
1859 	if (cmd_q->ccp->vdata->version < CCP_VERSION(5, 0)) {
1860 		/* Copy the exponent to the local storage block, using
1861 		 * as many 32-byte blocks as were allocated above. It's
1862 		 * already little endian, so no further change is required.
1863 		 */
1864 		ret = ccp_copy_to_sb(cmd_q, &exp, op.jobid, op.sb_key,
1865 				     CCP_PASSTHRU_BYTESWAP_NOOP);
1866 		if (ret) {
1867 			cmd->engine_error = cmd_q->cmd_error;
1868 			goto e_exp;
1869 		}
1870 	} else {
1871 		/* The exponent can be retrieved from memory via DMA. */
1872 		op.exp.u.dma.address = exp.dma.address;
1873 		op.exp.u.dma.offset = 0;
1874 	}
1875 
1876 	/* Concatenate the modulus and the message. Both the modulus and
1877 	 * the operands must be in little endian format.  Since the input
1878 	 * is in big endian format it must be converted.
1879 	 */
1880 	ret = ccp_init_dm_workarea(&src, cmd_q, i_len, DMA_TO_DEVICE);
1881 	if (ret)
1882 		goto e_exp;
1883 
1884 	ret = ccp_reverse_set_dm_area(&src, 0, rsa->mod, 0, rsa->mod_len);
1885 	if (ret)
1886 		goto e_src;
1887 	ret = ccp_reverse_set_dm_area(&src, o_len, rsa->src, 0, rsa->src_len);
1888 	if (ret)
1889 		goto e_src;
1890 
1891 	/* Prepare the output area for the operation */
1892 	ret = ccp_init_dm_workarea(&dst, cmd_q, o_len, DMA_FROM_DEVICE);
1893 	if (ret)
1894 		goto e_src;
1895 
1896 	op.soc = 1;
1897 	op.src.u.dma.address = src.dma.address;
1898 	op.src.u.dma.offset = 0;
1899 	op.src.u.dma.length = i_len;
1900 	op.dst.u.dma.address = dst.dma.address;
1901 	op.dst.u.dma.offset = 0;
1902 	op.dst.u.dma.length = o_len;
1903 
1904 	op.u.rsa.mod_size = rsa->key_size;
1905 	op.u.rsa.input_len = i_len;
1906 
1907 	ret = cmd_q->ccp->vdata->perform->rsa(&op);
1908 	if (ret) {
1909 		cmd->engine_error = cmd_q->cmd_error;
1910 		goto e_dst;
1911 	}
1912 
1913 	ccp_reverse_get_dm_area(&dst, 0, rsa->dst, 0, rsa->mod_len);
1914 
1915 e_dst:
1916 	ccp_dm_free(&dst);
1917 
1918 e_src:
1919 	ccp_dm_free(&src);
1920 
1921 e_exp:
1922 	ccp_dm_free(&exp);
1923 
1924 e_sb:
1925 	if (sb_count)
1926 		cmd_q->ccp->vdata->perform->sbfree(cmd_q, op.sb_key, sb_count);
1927 
1928 	return ret;
1929 }
1930 
1931 static noinline_for_stack int
1932 ccp_run_passthru_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
1933 {
1934 	struct ccp_passthru_engine *pt = &cmd->u.passthru;
1935 	struct ccp_dm_workarea mask;
1936 	struct ccp_data src, dst;
1937 	struct ccp_op op;
1938 	bool in_place = false;
1939 	unsigned int i;
1940 	int ret = 0;
1941 
1942 	if (!pt->final && (pt->src_len & (CCP_PASSTHRU_BLOCKSIZE - 1)))
1943 		return -EINVAL;
1944 
1945 	if (!pt->src || !pt->dst)
1946 		return -EINVAL;
1947 
1948 	if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) {
1949 		if (pt->mask_len != CCP_PASSTHRU_MASKSIZE)
1950 			return -EINVAL;
1951 		if (!pt->mask)
1952 			return -EINVAL;
1953 	}
1954 
1955 	BUILD_BUG_ON(CCP_PASSTHRU_SB_COUNT != 1);
1956 
1957 	memset(&op, 0, sizeof(op));
1958 	op.cmd_q = cmd_q;
1959 	op.jobid = CCP_NEW_JOBID(cmd_q->ccp);
1960 
1961 	if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP) {
1962 		/* Load the mask */
1963 		op.sb_key = cmd_q->sb_key;
1964 
1965 		ret = ccp_init_dm_workarea(&mask, cmd_q,
1966 					   CCP_PASSTHRU_SB_COUNT *
1967 					   CCP_SB_BYTES,
1968 					   DMA_TO_DEVICE);
1969 		if (ret)
1970 			return ret;
1971 
1972 		ret = ccp_set_dm_area(&mask, 0, pt->mask, 0, pt->mask_len);
1973 		if (ret)
1974 			goto e_mask;
1975 		ret = ccp_copy_to_sb(cmd_q, &mask, op.jobid, op.sb_key,
1976 				     CCP_PASSTHRU_BYTESWAP_NOOP);
1977 		if (ret) {
1978 			cmd->engine_error = cmd_q->cmd_error;
1979 			goto e_mask;
1980 		}
1981 	}
1982 
1983 	/* Prepare the input and output data workareas. For in-place
1984 	 * operations we need to set the dma direction to BIDIRECTIONAL
1985 	 * and copy the src workarea to the dst workarea.
1986 	 */
1987 	if (sg_virt(pt->src) == sg_virt(pt->dst))
1988 		in_place = true;
1989 
1990 	ret = ccp_init_data(&src, cmd_q, pt->src, pt->src_len,
1991 			    CCP_PASSTHRU_MASKSIZE,
1992 			    in_place ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE);
1993 	if (ret)
1994 		goto e_mask;
1995 
1996 	if (in_place) {
1997 		dst = src;
1998 	} else {
1999 		ret = ccp_init_data(&dst, cmd_q, pt->dst, pt->src_len,
2000 				    CCP_PASSTHRU_MASKSIZE, DMA_FROM_DEVICE);
2001 		if (ret)
2002 			goto e_src;
2003 	}
2004 
2005 	/* Send data to the CCP Passthru engine
2006 	 *   Because the CCP engine works on a single source and destination
2007 	 *   dma address at a time, each entry in the source scatterlist
2008 	 *   (after the dma_map_sg call) must be less than or equal to the
2009 	 *   (remaining) length in the destination scatterlist entry and the
2010 	 *   length must be a multiple of CCP_PASSTHRU_BLOCKSIZE
2011 	 */
2012 	dst.sg_wa.sg_used = 0;
2013 	for (i = 1; i <= src.sg_wa.dma_count; i++) {
2014 		if (!dst.sg_wa.sg ||
2015 		    (dst.sg_wa.sg->length < src.sg_wa.sg->length)) {
2016 			ret = -EINVAL;
2017 			goto e_dst;
2018 		}
2019 
2020 		if (i == src.sg_wa.dma_count) {
2021 			op.eom = 1;
2022 			op.soc = 1;
2023 		}
2024 
2025 		op.src.type = CCP_MEMTYPE_SYSTEM;
2026 		op.src.u.dma.address = sg_dma_address(src.sg_wa.sg);
2027 		op.src.u.dma.offset = 0;
2028 		op.src.u.dma.length = sg_dma_len(src.sg_wa.sg);
2029 
2030 		op.dst.type = CCP_MEMTYPE_SYSTEM;
2031 		op.dst.u.dma.address = sg_dma_address(dst.sg_wa.sg);
2032 		op.dst.u.dma.offset = dst.sg_wa.sg_used;
2033 		op.dst.u.dma.length = op.src.u.dma.length;
2034 
2035 		ret = cmd_q->ccp->vdata->perform->passthru(&op);
2036 		if (ret) {
2037 			cmd->engine_error = cmd_q->cmd_error;
2038 			goto e_dst;
2039 		}
2040 
2041 		dst.sg_wa.sg_used += src.sg_wa.sg->length;
2042 		if (dst.sg_wa.sg_used == dst.sg_wa.sg->length) {
2043 			dst.sg_wa.sg = sg_next(dst.sg_wa.sg);
2044 			dst.sg_wa.sg_used = 0;
2045 		}
2046 		src.sg_wa.sg = sg_next(src.sg_wa.sg);
2047 	}
2048 
2049 e_dst:
2050 	if (!in_place)
2051 		ccp_free_data(&dst, cmd_q);
2052 
2053 e_src:
2054 	ccp_free_data(&src, cmd_q);
2055 
2056 e_mask:
2057 	if (pt->bit_mod != CCP_PASSTHRU_BITWISE_NOOP)
2058 		ccp_dm_free(&mask);
2059 
2060 	return ret;
2061 }
2062 
2063 static noinline_for_stack int
2064 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 noinline_for_stack int
2406 ccp_run_ecc_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
2407 {
2408 	struct ccp_ecc_engine *ecc = &cmd->u.ecc;
2409 
2410 	ecc->ecc_result = 0;
2411 
2412 	if (!ecc->mod ||
2413 	    (ecc->mod_len > CCP_ECC_MODULUS_BYTES))
2414 		return -EINVAL;
2415 
2416 	switch (ecc->function) {
2417 	case CCP_ECC_FUNCTION_MMUL_384BIT:
2418 	case CCP_ECC_FUNCTION_MADD_384BIT:
2419 	case CCP_ECC_FUNCTION_MINV_384BIT:
2420 		return ccp_run_ecc_mm_cmd(cmd_q, cmd);
2421 
2422 	case CCP_ECC_FUNCTION_PADD_384BIT:
2423 	case CCP_ECC_FUNCTION_PMUL_384BIT:
2424 	case CCP_ECC_FUNCTION_PDBL_384BIT:
2425 		return ccp_run_ecc_pm_cmd(cmd_q, cmd);
2426 
2427 	default:
2428 		return -EINVAL;
2429 	}
2430 }
2431 
2432 int ccp_run_cmd(struct ccp_cmd_queue *cmd_q, struct ccp_cmd *cmd)
2433 {
2434 	int ret;
2435 
2436 	cmd->engine_error = 0;
2437 	cmd_q->cmd_error = 0;
2438 	cmd_q->int_rcvd = 0;
2439 	cmd_q->free_slots = cmd_q->ccp->vdata->perform->get_free_slots(cmd_q);
2440 
2441 	switch (cmd->engine) {
2442 	case CCP_ENGINE_AES:
2443 		switch (cmd->u.aes.mode) {
2444 		case CCP_AES_MODE_CMAC:
2445 			ret = ccp_run_aes_cmac_cmd(cmd_q, cmd);
2446 			break;
2447 		case CCP_AES_MODE_GCM:
2448 			ret = ccp_run_aes_gcm_cmd(cmd_q, cmd);
2449 			break;
2450 		default:
2451 			ret = ccp_run_aes_cmd(cmd_q, cmd);
2452 			break;
2453 		}
2454 		break;
2455 	case CCP_ENGINE_XTS_AES_128:
2456 		ret = ccp_run_xts_aes_cmd(cmd_q, cmd);
2457 		break;
2458 	case CCP_ENGINE_DES3:
2459 		ret = ccp_run_des3_cmd(cmd_q, cmd);
2460 		break;
2461 	case CCP_ENGINE_SHA:
2462 		ret = ccp_run_sha_cmd(cmd_q, cmd);
2463 		break;
2464 	case CCP_ENGINE_RSA:
2465 		ret = ccp_run_rsa_cmd(cmd_q, cmd);
2466 		break;
2467 	case CCP_ENGINE_PASSTHRU:
2468 		if (cmd->flags & CCP_CMD_PASSTHRU_NO_DMA_MAP)
2469 			ret = ccp_run_passthru_nomap_cmd(cmd_q, cmd);
2470 		else
2471 			ret = ccp_run_passthru_cmd(cmd_q, cmd);
2472 		break;
2473 	case CCP_ENGINE_ECC:
2474 		ret = ccp_run_ecc_cmd(cmd_q, cmd);
2475 		break;
2476 	default:
2477 		ret = -EINVAL;
2478 	}
2479 
2480 	return ret;
2481 }
2482