xref: /linux/drivers/crypto/intel/keembay/ocs-hcu.c (revision 5c61f59824b5e46516ea5d0543ad7a8871567416)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Intel Keem Bay OCS HCU Crypto Driver.
4  *
5  * Copyright (C) 2018-2020 Intel Corporation
6  */
7 
8 #include <linux/delay.h>
9 #include <linux/device.h>
10 #include <linux/iopoll.h>
11 #include <linux/irq.h>
12 #include <linux/module.h>
13 
14 #include <crypto/sha2.h>
15 
16 #include "ocs-hcu.h"
17 
18 /* Registers. */
19 #define OCS_HCU_MODE			0x00
20 #define OCS_HCU_CHAIN			0x04
21 #define OCS_HCU_OPERATION		0x08
22 #define OCS_HCU_KEY_0			0x0C
23 #define OCS_HCU_ISR			0x50
24 #define OCS_HCU_IER			0x54
25 #define OCS_HCU_STATUS			0x58
26 #define OCS_HCU_MSG_LEN_LO		0x60
27 #define OCS_HCU_MSG_LEN_HI		0x64
28 #define OCS_HCU_KEY_BYTE_ORDER_CFG	0x80
29 #define OCS_HCU_DMA_SRC_ADDR		0x400
30 #define OCS_HCU_DMA_SRC_SIZE		0x408
31 #define OCS_HCU_DMA_DST_SIZE		0x40C
32 #define OCS_HCU_DMA_DMA_MODE		0x410
33 #define OCS_HCU_DMA_NEXT_SRC_DESCR	0x418
34 #define OCS_HCU_DMA_MSI_ISR		0x480
35 #define OCS_HCU_DMA_MSI_IER		0x484
36 #define OCS_HCU_DMA_MSI_MASK		0x488
37 
38 /* Register bit definitions. */
39 #define HCU_MODE_ALGO_SHIFT		16
40 #define HCU_MODE_HMAC_SHIFT		22
41 
42 #define HCU_STATUS_BUSY			BIT(0)
43 
44 #define HCU_BYTE_ORDER_SWAP		BIT(0)
45 
46 #define HCU_IRQ_HASH_DONE		BIT(2)
47 #define HCU_IRQ_HASH_ERR_MASK		(BIT(3) | BIT(1) | BIT(0))
48 
49 #define HCU_DMA_IRQ_SRC_DONE		BIT(0)
50 #define HCU_DMA_IRQ_SAI_ERR		BIT(2)
51 #define HCU_DMA_IRQ_BAD_COMP_ERR	BIT(3)
52 #define HCU_DMA_IRQ_INBUF_RD_ERR	BIT(4)
53 #define HCU_DMA_IRQ_INBUF_WD_ERR	BIT(5)
54 #define HCU_DMA_IRQ_OUTBUF_WR_ERR	BIT(6)
55 #define HCU_DMA_IRQ_OUTBUF_RD_ERR	BIT(7)
56 #define HCU_DMA_IRQ_CRD_ERR		BIT(8)
57 #define HCU_DMA_IRQ_ERR_MASK		(HCU_DMA_IRQ_SAI_ERR | \
58 					 HCU_DMA_IRQ_BAD_COMP_ERR | \
59 					 HCU_DMA_IRQ_INBUF_RD_ERR | \
60 					 HCU_DMA_IRQ_INBUF_WD_ERR | \
61 					 HCU_DMA_IRQ_OUTBUF_WR_ERR | \
62 					 HCU_DMA_IRQ_OUTBUF_RD_ERR | \
63 					 HCU_DMA_IRQ_CRD_ERR)
64 
65 #define HCU_DMA_SNOOP_MASK		(0x7 << 28)
66 #define HCU_DMA_SRC_LL_EN		BIT(25)
67 #define HCU_DMA_EN			BIT(31)
68 
69 #define OCS_HCU_ENDIANNESS_VALUE	0x2A
70 
71 #define HCU_DMA_MSI_UNMASK		BIT(0)
72 #define HCU_DMA_MSI_DISABLE		0
73 #define HCU_IRQ_DISABLE			0
74 
75 #define OCS_HCU_START			BIT(0)
76 #define OCS_HCU_TERMINATE		BIT(1)
77 
78 #define OCS_LL_DMA_FLAG_TERMINATE	BIT(31)
79 
80 #define OCS_HCU_HW_KEY_LEN_U32		(OCS_HCU_HW_KEY_LEN / sizeof(u32))
81 
82 #define HCU_DATA_WRITE_ENDIANNESS_OFFSET	26
83 
84 #define OCS_HCU_NUM_CHAINS_SHA256_224_SM3	(SHA256_DIGEST_SIZE / sizeof(u32))
85 #define OCS_HCU_NUM_CHAINS_SHA384_512		(SHA512_DIGEST_SIZE / sizeof(u32))
86 
87 /*
88  * While polling on a busy HCU, wait maximum 200us between one check and the
89  * other.
90  */
91 #define OCS_HCU_WAIT_BUSY_RETRY_DELAY_US	200
92 /* Wait on a busy HCU for maximum 1 second. */
93 #define OCS_HCU_WAIT_BUSY_TIMEOUT_US		1000000
94 
95 /**
96  * struct ocs_hcu_dma_entry - An entry in an OCS DMA linked list.
97  * @src_addr:  Source address of the data.
98  * @src_len:   Length of data to be fetched.
99  * @nxt_desc:  Next descriptor to fetch.
100  * @ll_flags:  Flags (Freeze @ terminate) for the DMA engine.
101  */
102 struct ocs_hcu_dma_entry {
103 	u32 src_addr;
104 	u32 src_len;
105 	u32 nxt_desc;
106 	u32 ll_flags;
107 };
108 
109 /**
110  * struct ocs_hcu_dma_list - OCS-specific DMA linked list.
111  * @head:	The head of the list (points to the array backing the list).
112  * @tail:	The current tail of the list; NULL if the list is empty.
113  * @dma_addr:	The DMA address of @head (i.e., the DMA address of the backing
114  *		array).
115  * @max_nents:	Maximum number of entries in the list (i.e., number of elements
116  *		in the backing array).
117  *
118  * The OCS DMA list is an array-backed list of OCS DMA descriptors. The array
119  * backing the list is allocated with dma_alloc_coherent() and pointed by
120  * @head.
121  */
122 struct ocs_hcu_dma_list {
123 	struct ocs_hcu_dma_entry	*head;
124 	struct ocs_hcu_dma_entry	*tail;
125 	dma_addr_t			dma_addr;
126 	size_t				max_nents;
127 };
128 
129 static inline u32 ocs_hcu_num_chains(enum ocs_hcu_algo algo)
130 {
131 	switch (algo) {
132 	case OCS_HCU_ALGO_SHA224:
133 	case OCS_HCU_ALGO_SHA256:
134 	case OCS_HCU_ALGO_SM3:
135 		return OCS_HCU_NUM_CHAINS_SHA256_224_SM3;
136 	case OCS_HCU_ALGO_SHA384:
137 	case OCS_HCU_ALGO_SHA512:
138 		return OCS_HCU_NUM_CHAINS_SHA384_512;
139 	default:
140 		return 0;
141 	};
142 }
143 
144 static inline u32 ocs_hcu_digest_size(enum ocs_hcu_algo algo)
145 {
146 	switch (algo) {
147 	case OCS_HCU_ALGO_SHA224:
148 		return SHA224_DIGEST_SIZE;
149 	case OCS_HCU_ALGO_SHA256:
150 	case OCS_HCU_ALGO_SM3:
151 		/* SM3 shares the same block size. */
152 		return SHA256_DIGEST_SIZE;
153 	case OCS_HCU_ALGO_SHA384:
154 		return SHA384_DIGEST_SIZE;
155 	case OCS_HCU_ALGO_SHA512:
156 		return SHA512_DIGEST_SIZE;
157 	default:
158 		return 0;
159 	}
160 }
161 
162 /**
163  * ocs_hcu_wait_busy() - Wait for HCU OCS hardware to became usable.
164  * @hcu_dev:	OCS HCU device to wait for.
165  *
166  * Return: 0 if device free, -ETIMEOUT if device busy and internal timeout has
167  *	   expired.
168  */
169 static int ocs_hcu_wait_busy(struct ocs_hcu_dev *hcu_dev)
170 {
171 	long val;
172 
173 	return readl_poll_timeout(hcu_dev->io_base + OCS_HCU_STATUS, val,
174 				  !(val & HCU_STATUS_BUSY),
175 				  OCS_HCU_WAIT_BUSY_RETRY_DELAY_US,
176 				  OCS_HCU_WAIT_BUSY_TIMEOUT_US);
177 }
178 
179 static void ocs_hcu_done_irq_en(struct ocs_hcu_dev *hcu_dev)
180 {
181 	/* Clear any pending interrupts. */
182 	writel(0xFFFFFFFF, hcu_dev->io_base + OCS_HCU_ISR);
183 	hcu_dev->irq_err = false;
184 	/* Enable error and HCU done interrupts. */
185 	writel(HCU_IRQ_HASH_DONE | HCU_IRQ_HASH_ERR_MASK,
186 	       hcu_dev->io_base + OCS_HCU_IER);
187 }
188 
189 static void ocs_hcu_dma_irq_en(struct ocs_hcu_dev *hcu_dev)
190 {
191 	/* Clear any pending interrupts. */
192 	writel(0xFFFFFFFF, hcu_dev->io_base + OCS_HCU_DMA_MSI_ISR);
193 	hcu_dev->irq_err = false;
194 	/* Only operating on DMA source completion and error interrupts. */
195 	writel(HCU_DMA_IRQ_ERR_MASK | HCU_DMA_IRQ_SRC_DONE,
196 	       hcu_dev->io_base + OCS_HCU_DMA_MSI_IER);
197 	/* Unmask */
198 	writel(HCU_DMA_MSI_UNMASK, hcu_dev->io_base + OCS_HCU_DMA_MSI_MASK);
199 }
200 
201 static void ocs_hcu_irq_dis(struct ocs_hcu_dev *hcu_dev)
202 {
203 	writel(HCU_IRQ_DISABLE, hcu_dev->io_base + OCS_HCU_IER);
204 	writel(HCU_DMA_MSI_DISABLE, hcu_dev->io_base + OCS_HCU_DMA_MSI_IER);
205 }
206 
207 static int ocs_hcu_wait_and_disable_irq(struct ocs_hcu_dev *hcu_dev)
208 {
209 	int rc;
210 
211 	rc = wait_for_completion_interruptible(&hcu_dev->irq_done);
212 	if (rc)
213 		goto exit;
214 
215 	if (hcu_dev->irq_err) {
216 		/* Unset flag and return error. */
217 		hcu_dev->irq_err = false;
218 		rc = -EIO;
219 		goto exit;
220 	}
221 
222 exit:
223 	ocs_hcu_irq_dis(hcu_dev);
224 
225 	return rc;
226 }
227 
228 /**
229  * ocs_hcu_get_intermediate_data() - Get intermediate data.
230  * @hcu_dev:	The target HCU device.
231  * @data:	Where to store the intermediate.
232  * @algo:	The algorithm being used.
233  *
234  * This function is used to save the current hashing process state in order to
235  * continue it in the future.
236  *
237  * Note: once all data has been processed, the intermediate data actually
238  * contains the hashing result. So this function is also used to retrieve the
239  * final result of a hashing process.
240  *
241  * Return: 0 on success, negative error code otherwise.
242  */
243 static int ocs_hcu_get_intermediate_data(struct ocs_hcu_dev *hcu_dev,
244 					 struct ocs_hcu_idata *data,
245 					 enum ocs_hcu_algo algo)
246 {
247 	const int n = ocs_hcu_num_chains(algo);
248 	u32 *chain;
249 	int rc;
250 	int i;
251 
252 	/* Data not requested. */
253 	if (!data)
254 		return -EINVAL;
255 
256 	chain = (u32 *)data->digest;
257 
258 	/* Ensure that the OCS is no longer busy before reading the chains. */
259 	rc = ocs_hcu_wait_busy(hcu_dev);
260 	if (rc)
261 		return rc;
262 
263 	/*
264 	 * This loops is safe because data->digest is an array of
265 	 * SHA512_DIGEST_SIZE bytes and the maximum value returned by
266 	 * ocs_hcu_num_chains() is OCS_HCU_NUM_CHAINS_SHA384_512 which is equal
267 	 * to SHA512_DIGEST_SIZE / sizeof(u32).
268 	 */
269 	for (i = 0; i < n; i++)
270 		chain[i] = readl(hcu_dev->io_base + OCS_HCU_CHAIN);
271 
272 	data->msg_len_lo = readl(hcu_dev->io_base + OCS_HCU_MSG_LEN_LO);
273 	data->msg_len_hi = readl(hcu_dev->io_base + OCS_HCU_MSG_LEN_HI);
274 
275 	return 0;
276 }
277 
278 /**
279  * ocs_hcu_set_intermediate_data() - Set intermediate data.
280  * @hcu_dev:	The target HCU device.
281  * @data:	The intermediate data to be set.
282  * @algo:	The algorithm being used.
283  *
284  * This function is used to continue a previous hashing process.
285  */
286 static void ocs_hcu_set_intermediate_data(struct ocs_hcu_dev *hcu_dev,
287 					  const struct ocs_hcu_idata *data,
288 					  enum ocs_hcu_algo algo)
289 {
290 	const int n = ocs_hcu_num_chains(algo);
291 	u32 *chain = (u32 *)data->digest;
292 	int i;
293 
294 	/*
295 	 * This loops is safe because data->digest is an array of
296 	 * SHA512_DIGEST_SIZE bytes and the maximum value returned by
297 	 * ocs_hcu_num_chains() is OCS_HCU_NUM_CHAINS_SHA384_512 which is equal
298 	 * to SHA512_DIGEST_SIZE / sizeof(u32).
299 	 */
300 	for (i = 0; i < n; i++)
301 		writel(chain[i], hcu_dev->io_base + OCS_HCU_CHAIN);
302 
303 	writel(data->msg_len_lo, hcu_dev->io_base + OCS_HCU_MSG_LEN_LO);
304 	writel(data->msg_len_hi, hcu_dev->io_base + OCS_HCU_MSG_LEN_HI);
305 }
306 
307 static int ocs_hcu_get_digest(struct ocs_hcu_dev *hcu_dev,
308 			      enum ocs_hcu_algo algo, u8 *dgst, size_t dgst_len)
309 {
310 	u32 *chain;
311 	int rc;
312 	int i;
313 
314 	if (!dgst)
315 		return -EINVAL;
316 
317 	/* Length of the output buffer must match the algo digest size. */
318 	if (dgst_len != ocs_hcu_digest_size(algo))
319 		return -EINVAL;
320 
321 	/* Ensure that the OCS is no longer busy before reading the chains. */
322 	rc = ocs_hcu_wait_busy(hcu_dev);
323 	if (rc)
324 		return rc;
325 
326 	chain = (u32 *)dgst;
327 	for (i = 0; i < dgst_len / sizeof(u32); i++)
328 		chain[i] = readl(hcu_dev->io_base + OCS_HCU_CHAIN);
329 
330 	return 0;
331 }
332 
333 /**
334  * ocs_hcu_hw_cfg() - Configure the HCU hardware.
335  * @hcu_dev:	The HCU device to configure.
336  * @algo:	The algorithm to be used by the HCU device.
337  * @use_hmac:	Whether or not HW HMAC should be used.
338  *
339  * Return: 0 on success, negative error code otherwise.
340  */
341 static int ocs_hcu_hw_cfg(struct ocs_hcu_dev *hcu_dev, enum ocs_hcu_algo algo,
342 			  bool use_hmac)
343 {
344 	u32 cfg;
345 	int rc;
346 
347 	if (algo != OCS_HCU_ALGO_SHA256 && algo != OCS_HCU_ALGO_SHA224 &&
348 	    algo != OCS_HCU_ALGO_SHA384 && algo != OCS_HCU_ALGO_SHA512 &&
349 	    algo != OCS_HCU_ALGO_SM3)
350 		return -EINVAL;
351 
352 	rc = ocs_hcu_wait_busy(hcu_dev);
353 	if (rc)
354 		return rc;
355 
356 	/* Ensure interrupts are disabled. */
357 	ocs_hcu_irq_dis(hcu_dev);
358 
359 	/* Configure endianness, hashing algorithm and HW HMAC (if needed) */
360 	cfg = OCS_HCU_ENDIANNESS_VALUE << HCU_DATA_WRITE_ENDIANNESS_OFFSET;
361 	cfg |= algo << HCU_MODE_ALGO_SHIFT;
362 	if (use_hmac)
363 		cfg |= BIT(HCU_MODE_HMAC_SHIFT);
364 
365 	writel(cfg, hcu_dev->io_base + OCS_HCU_MODE);
366 
367 	return 0;
368 }
369 
370 /**
371  * ocs_hcu_clear_key() - Clear key stored in OCS HMAC KEY registers.
372  * @hcu_dev:	The OCS HCU device whose key registers should be cleared.
373  */
374 static void ocs_hcu_clear_key(struct ocs_hcu_dev *hcu_dev)
375 {
376 	int reg_off;
377 
378 	/* Clear OCS_HCU_KEY_[0..15] */
379 	for (reg_off = 0; reg_off < OCS_HCU_HW_KEY_LEN; reg_off += sizeof(u32))
380 		writel(0, hcu_dev->io_base + OCS_HCU_KEY_0 + reg_off);
381 }
382 
383 /**
384  * ocs_hcu_write_key() - Write key to OCS HMAC KEY registers.
385  * @hcu_dev:	The OCS HCU device the key should be written to.
386  * @key:	The key to be written.
387  * @len:	The size of the key to write. It must be OCS_HCU_HW_KEY_LEN.
388  *
389  * Return:	0 on success, negative error code otherwise.
390  */
391 static int ocs_hcu_write_key(struct ocs_hcu_dev *hcu_dev, const u8 *key, size_t len)
392 {
393 	u32 key_u32[OCS_HCU_HW_KEY_LEN_U32];
394 	int i;
395 
396 	if (len > OCS_HCU_HW_KEY_LEN)
397 		return -EINVAL;
398 
399 	/* Copy key into temporary u32 array. */
400 	memcpy(key_u32, key, len);
401 
402 	/*
403 	 * Hardware requires all the bytes of the HW Key vector to be
404 	 * written. So pad with zero until we reach OCS_HCU_HW_KEY_LEN.
405 	 */
406 	memzero_explicit((u8 *)key_u32 + len, OCS_HCU_HW_KEY_LEN - len);
407 
408 	/*
409 	 * OCS hardware expects the MSB of the key to be written at the highest
410 	 * address of the HCU Key vector; in other word, the key must be
411 	 * written in reverse order.
412 	 *
413 	 * Therefore, we first enable byte swapping for the HCU key vector;
414 	 * so that bytes of 32-bit word written to OCS_HCU_KEY_[0..15] will be
415 	 * swapped:
416 	 * 3 <---> 0, 2 <---> 1.
417 	 */
418 	writel(HCU_BYTE_ORDER_SWAP,
419 	       hcu_dev->io_base + OCS_HCU_KEY_BYTE_ORDER_CFG);
420 	/*
421 	 * And then we write the 32-bit words composing the key starting from
422 	 * the end of the key.
423 	 */
424 	for (i = 0; i < OCS_HCU_HW_KEY_LEN_U32; i++)
425 		writel(key_u32[OCS_HCU_HW_KEY_LEN_U32 - 1 - i],
426 		       hcu_dev->io_base + OCS_HCU_KEY_0 + (sizeof(u32) * i));
427 
428 	memzero_explicit(key_u32, OCS_HCU_HW_KEY_LEN);
429 
430 	return 0;
431 }
432 
433 /**
434  * ocs_hcu_ll_dma_start() - Start OCS HCU hashing via DMA
435  * @hcu_dev:	The OCS HCU device to use.
436  * @dma_list:	The OCS DMA list mapping the data to hash.
437  * @finalize:	Whether or not this is the last hashing operation and therefore
438  *		the final hash should be compute even if data is not
439  *		block-aligned.
440  *
441  * Return: 0 on success, negative error code otherwise.
442  */
443 static int ocs_hcu_ll_dma_start(struct ocs_hcu_dev *hcu_dev,
444 				const struct ocs_hcu_dma_list *dma_list,
445 				bool finalize)
446 {
447 	u32 cfg = HCU_DMA_SNOOP_MASK | HCU_DMA_SRC_LL_EN | HCU_DMA_EN;
448 	int rc;
449 
450 	if (!dma_list)
451 		return -EINVAL;
452 
453 	/*
454 	 * For final requests we use HCU_DONE IRQ to be notified when all input
455 	 * data has been processed by the HCU; however, we cannot do so for
456 	 * non-final requests, because we don't get a HCU_DONE IRQ when we
457 	 * don't terminate the operation.
458 	 *
459 	 * Therefore, for non-final requests, we use the DMA IRQ, which
460 	 * triggers when DMA has finishing feeding all the input data to the
461 	 * HCU, but the HCU may still be processing it. This is fine, since we
462 	 * will wait for the HCU processing to be completed when we try to read
463 	 * intermediate results, in ocs_hcu_get_intermediate_data().
464 	 */
465 	if (finalize)
466 		ocs_hcu_done_irq_en(hcu_dev);
467 	else
468 		ocs_hcu_dma_irq_en(hcu_dev);
469 
470 	reinit_completion(&hcu_dev->irq_done);
471 	writel(dma_list->dma_addr, hcu_dev->io_base + OCS_HCU_DMA_NEXT_SRC_DESCR);
472 	writel(0, hcu_dev->io_base + OCS_HCU_DMA_SRC_SIZE);
473 	writel(0, hcu_dev->io_base + OCS_HCU_DMA_DST_SIZE);
474 
475 	writel(OCS_HCU_START, hcu_dev->io_base + OCS_HCU_OPERATION);
476 
477 	writel(cfg, hcu_dev->io_base + OCS_HCU_DMA_DMA_MODE);
478 
479 	if (finalize)
480 		writel(OCS_HCU_TERMINATE, hcu_dev->io_base + OCS_HCU_OPERATION);
481 
482 	rc = ocs_hcu_wait_and_disable_irq(hcu_dev);
483 	if (rc)
484 		return rc;
485 
486 	return 0;
487 }
488 
489 struct ocs_hcu_dma_list *ocs_hcu_dma_list_alloc(struct ocs_hcu_dev *hcu_dev,
490 						int max_nents)
491 {
492 	struct ocs_hcu_dma_list *dma_list;
493 
494 	dma_list = kmalloc(sizeof(*dma_list), GFP_KERNEL);
495 	if (!dma_list)
496 		return NULL;
497 
498 	/* Total size of the DMA list to allocate. */
499 	dma_list->head = dma_alloc_coherent(hcu_dev->dev,
500 					    sizeof(*dma_list->head) * max_nents,
501 					    &dma_list->dma_addr, GFP_KERNEL);
502 	if (!dma_list->head) {
503 		kfree(dma_list);
504 		return NULL;
505 	}
506 	dma_list->max_nents = max_nents;
507 	dma_list->tail = NULL;
508 
509 	return dma_list;
510 }
511 
512 void ocs_hcu_dma_list_free(struct ocs_hcu_dev *hcu_dev,
513 			   struct ocs_hcu_dma_list *dma_list)
514 {
515 	if (!dma_list)
516 		return;
517 
518 	dma_free_coherent(hcu_dev->dev,
519 			  sizeof(*dma_list->head) * dma_list->max_nents,
520 			  dma_list->head, dma_list->dma_addr);
521 
522 	kfree(dma_list);
523 }
524 
525 /* Add a new DMA entry at the end of the OCS DMA list. */
526 int ocs_hcu_dma_list_add_tail(struct ocs_hcu_dev *hcu_dev,
527 			      struct ocs_hcu_dma_list *dma_list,
528 			      dma_addr_t addr, u32 len)
529 {
530 	struct device *dev = hcu_dev->dev;
531 	struct ocs_hcu_dma_entry *old_tail;
532 	struct ocs_hcu_dma_entry *new_tail;
533 
534 	if (!len)
535 		return 0;
536 
537 	if (!dma_list)
538 		return -EINVAL;
539 
540 	if (addr & ~OCS_HCU_DMA_BIT_MASK) {
541 		dev_err(dev,
542 			"Unexpected error: Invalid DMA address for OCS HCU\n");
543 		return -EINVAL;
544 	}
545 
546 	old_tail = dma_list->tail;
547 	new_tail = old_tail ? old_tail + 1 : dma_list->head;
548 
549 	/* Check if list is full. */
550 	if (new_tail - dma_list->head >= dma_list->max_nents)
551 		return -ENOMEM;
552 
553 	/*
554 	 * If there was an old tail (i.e., this is not the first element we are
555 	 * adding), un-terminate the old tail and make it point to the new one.
556 	 */
557 	if (old_tail) {
558 		old_tail->ll_flags &= ~OCS_LL_DMA_FLAG_TERMINATE;
559 		/*
560 		 * The old tail 'nxt_desc' must point to the DMA address of the
561 		 * new tail.
562 		 */
563 		old_tail->nxt_desc = dma_list->dma_addr +
564 				     sizeof(*dma_list->tail) * (new_tail -
565 								dma_list->head);
566 	}
567 
568 	new_tail->src_addr = (u32)addr;
569 	new_tail->src_len = (u32)len;
570 	new_tail->ll_flags = OCS_LL_DMA_FLAG_TERMINATE;
571 	new_tail->nxt_desc = 0;
572 
573 	/* Update list tail with new tail. */
574 	dma_list->tail = new_tail;
575 
576 	return 0;
577 }
578 
579 /**
580  * ocs_hcu_hash_init() - Initialize hash operation context.
581  * @ctx:	The context to initialize.
582  * @algo:	The hashing algorithm to use.
583  *
584  * Return:	0 on success, negative error code otherwise.
585  */
586 int ocs_hcu_hash_init(struct ocs_hcu_hash_ctx *ctx, enum ocs_hcu_algo algo)
587 {
588 	if (!ctx)
589 		return -EINVAL;
590 
591 	ctx->algo = algo;
592 	ctx->idata.msg_len_lo = 0;
593 	ctx->idata.msg_len_hi = 0;
594 	/* No need to set idata.digest to 0. */
595 
596 	return 0;
597 }
598 
599 /**
600  * ocs_hcu_hash_update() - Perform a hashing iteration.
601  * @hcu_dev:	The OCS HCU device to use.
602  * @ctx:	The OCS HCU hashing context.
603  * @dma_list:	The OCS DMA list mapping the input data to process.
604  *
605  * Return: 0 on success; negative error code otherwise.
606  */
607 int ocs_hcu_hash_update(struct ocs_hcu_dev *hcu_dev,
608 			struct ocs_hcu_hash_ctx *ctx,
609 			const struct ocs_hcu_dma_list *dma_list)
610 {
611 	int rc;
612 
613 	if (!hcu_dev || !ctx)
614 		return -EINVAL;
615 
616 	/* Configure the hardware for the current request. */
617 	rc = ocs_hcu_hw_cfg(hcu_dev, ctx->algo, false);
618 	if (rc)
619 		return rc;
620 
621 	/* If we already processed some data, idata needs to be set. */
622 	if (ctx->idata.msg_len_lo || ctx->idata.msg_len_hi)
623 		ocs_hcu_set_intermediate_data(hcu_dev, &ctx->idata, ctx->algo);
624 
625 	/* Start linked-list DMA hashing. */
626 	rc = ocs_hcu_ll_dma_start(hcu_dev, dma_list, false);
627 	if (rc)
628 		return rc;
629 
630 	/* Update idata and return. */
631 	return ocs_hcu_get_intermediate_data(hcu_dev, &ctx->idata, ctx->algo);
632 }
633 
634 /**
635  * ocs_hcu_hash_finup() - Update and finalize hash computation.
636  * @hcu_dev:	The OCS HCU device to use.
637  * @ctx:	The OCS HCU hashing context.
638  * @dma_list:	The OCS DMA list mapping the input data to process.
639  * @dgst:	The buffer where to save the computed digest.
640  * @dgst_len:	The length of @dgst.
641  *
642  * Return: 0 on success; negative error code otherwise.
643  */
644 int ocs_hcu_hash_finup(struct ocs_hcu_dev *hcu_dev,
645 		       const struct ocs_hcu_hash_ctx *ctx,
646 		       const struct ocs_hcu_dma_list *dma_list,
647 		       u8 *dgst, size_t dgst_len)
648 {
649 	int rc;
650 
651 	if (!hcu_dev || !ctx)
652 		return -EINVAL;
653 
654 	/* Configure the hardware for the current request. */
655 	rc = ocs_hcu_hw_cfg(hcu_dev, ctx->algo, false);
656 	if (rc)
657 		return rc;
658 
659 	/* If we already processed some data, idata needs to be set. */
660 	if (ctx->idata.msg_len_lo || ctx->idata.msg_len_hi)
661 		ocs_hcu_set_intermediate_data(hcu_dev, &ctx->idata, ctx->algo);
662 
663 	/* Start linked-list DMA hashing. */
664 	rc = ocs_hcu_ll_dma_start(hcu_dev, dma_list, true);
665 	if (rc)
666 		return rc;
667 
668 	/* Get digest and return. */
669 	return ocs_hcu_get_digest(hcu_dev, ctx->algo, dgst, dgst_len);
670 }
671 
672 /**
673  * ocs_hcu_hash_final() - Finalize hash computation.
674  * @hcu_dev:		The OCS HCU device to use.
675  * @ctx:		The OCS HCU hashing context.
676  * @dgst:		The buffer where to save the computed digest.
677  * @dgst_len:		The length of @dgst.
678  *
679  * Return: 0 on success; negative error code otherwise.
680  */
681 int ocs_hcu_hash_final(struct ocs_hcu_dev *hcu_dev,
682 		       const struct ocs_hcu_hash_ctx *ctx, u8 *dgst,
683 		       size_t dgst_len)
684 {
685 	int rc;
686 
687 	if (!hcu_dev || !ctx)
688 		return -EINVAL;
689 
690 	/* Configure the hardware for the current request. */
691 	rc = ocs_hcu_hw_cfg(hcu_dev, ctx->algo, false);
692 	if (rc)
693 		return rc;
694 
695 	/* If we already processed some data, idata needs to be set. */
696 	if (ctx->idata.msg_len_lo || ctx->idata.msg_len_hi)
697 		ocs_hcu_set_intermediate_data(hcu_dev, &ctx->idata, ctx->algo);
698 
699 	/*
700 	 * Enable HCU interrupts, so that HCU_DONE will be triggered once the
701 	 * final hash is computed.
702 	 */
703 	ocs_hcu_done_irq_en(hcu_dev);
704 	reinit_completion(&hcu_dev->irq_done);
705 	writel(OCS_HCU_TERMINATE, hcu_dev->io_base + OCS_HCU_OPERATION);
706 
707 	rc = ocs_hcu_wait_and_disable_irq(hcu_dev);
708 	if (rc)
709 		return rc;
710 
711 	/* Get digest and return. */
712 	return ocs_hcu_get_digest(hcu_dev, ctx->algo, dgst, dgst_len);
713 }
714 
715 /**
716  * ocs_hcu_digest() - Compute hash digest.
717  * @hcu_dev:		The OCS HCU device to use.
718  * @algo:		The hash algorithm to use.
719  * @data:		The input data to process.
720  * @data_len:		The length of @data.
721  * @dgst:		The buffer where to save the computed digest.
722  * @dgst_len:		The length of @dgst.
723  *
724  * Return: 0 on success; negative error code otherwise.
725  */
726 int ocs_hcu_digest(struct ocs_hcu_dev *hcu_dev, enum ocs_hcu_algo algo,
727 		   void *data, size_t data_len, u8 *dgst, size_t dgst_len)
728 {
729 	struct device *dev = hcu_dev->dev;
730 	dma_addr_t dma_handle;
731 	u32 reg;
732 	int rc;
733 
734 	/* Configure the hardware for the current request. */
735 	rc = ocs_hcu_hw_cfg(hcu_dev, algo, false);
736 	if (rc)
737 		return rc;
738 
739 	dma_handle = dma_map_single(dev, data, data_len, DMA_TO_DEVICE);
740 	if (dma_mapping_error(dev, dma_handle))
741 		return -EIO;
742 
743 	reg = HCU_DMA_SNOOP_MASK | HCU_DMA_EN;
744 
745 	ocs_hcu_done_irq_en(hcu_dev);
746 
747 	reinit_completion(&hcu_dev->irq_done);
748 
749 	writel(dma_handle, hcu_dev->io_base + OCS_HCU_DMA_SRC_ADDR);
750 	writel(data_len, hcu_dev->io_base + OCS_HCU_DMA_SRC_SIZE);
751 	writel(OCS_HCU_START, hcu_dev->io_base + OCS_HCU_OPERATION);
752 	writel(reg, hcu_dev->io_base + OCS_HCU_DMA_DMA_MODE);
753 
754 	writel(OCS_HCU_TERMINATE, hcu_dev->io_base + OCS_HCU_OPERATION);
755 
756 	rc = ocs_hcu_wait_and_disable_irq(hcu_dev);
757 	if (rc)
758 		return rc;
759 
760 	dma_unmap_single(dev, dma_handle, data_len, DMA_TO_DEVICE);
761 
762 	return ocs_hcu_get_digest(hcu_dev, algo, dgst, dgst_len);
763 }
764 
765 /**
766  * ocs_hcu_hmac() - Compute HMAC.
767  * @hcu_dev:		The OCS HCU device to use.
768  * @algo:		The hash algorithm to use with HMAC.
769  * @key:		The key to use.
770  * @dma_list:	The OCS DMA list mapping the input data to process.
771  * @key_len:		The length of @key.
772  * @dgst:		The buffer where to save the computed HMAC.
773  * @dgst_len:		The length of @dgst.
774  *
775  * Return: 0 on success; negative error code otherwise.
776  */
777 int ocs_hcu_hmac(struct ocs_hcu_dev *hcu_dev, enum ocs_hcu_algo algo,
778 		 const u8 *key, size_t key_len,
779 		 const struct ocs_hcu_dma_list *dma_list,
780 		 u8 *dgst, size_t dgst_len)
781 {
782 	int rc;
783 
784 	/* Ensure 'key' is not NULL. */
785 	if (!key || key_len == 0)
786 		return -EINVAL;
787 
788 	/* Configure the hardware for the current request. */
789 	rc = ocs_hcu_hw_cfg(hcu_dev, algo, true);
790 	if (rc)
791 		return rc;
792 
793 	rc = ocs_hcu_write_key(hcu_dev, key, key_len);
794 	if (rc)
795 		return rc;
796 
797 	rc = ocs_hcu_ll_dma_start(hcu_dev, dma_list, true);
798 
799 	/* Clear HW key before processing return code. */
800 	ocs_hcu_clear_key(hcu_dev);
801 
802 	if (rc)
803 		return rc;
804 
805 	return ocs_hcu_get_digest(hcu_dev, algo, dgst, dgst_len);
806 }
807 
808 irqreturn_t ocs_hcu_irq_handler(int irq, void *dev_id)
809 {
810 	struct ocs_hcu_dev *hcu_dev = dev_id;
811 	u32 hcu_irq;
812 	u32 dma_irq;
813 
814 	/* Read and clear the HCU interrupt. */
815 	hcu_irq = readl(hcu_dev->io_base + OCS_HCU_ISR);
816 	writel(hcu_irq, hcu_dev->io_base + OCS_HCU_ISR);
817 
818 	/* Read and clear the HCU DMA interrupt. */
819 	dma_irq = readl(hcu_dev->io_base + OCS_HCU_DMA_MSI_ISR);
820 	writel(dma_irq, hcu_dev->io_base + OCS_HCU_DMA_MSI_ISR);
821 
822 	/* Check for errors. */
823 	if (hcu_irq & HCU_IRQ_HASH_ERR_MASK || dma_irq & HCU_DMA_IRQ_ERR_MASK) {
824 		hcu_dev->irq_err = true;
825 		goto complete;
826 	}
827 
828 	/* Check for DONE IRQs. */
829 	if (hcu_irq & HCU_IRQ_HASH_DONE || dma_irq & HCU_DMA_IRQ_SRC_DONE)
830 		goto complete;
831 
832 	return IRQ_NONE;
833 
834 complete:
835 	complete(&hcu_dev->irq_done);
836 
837 	return IRQ_HANDLED;
838 }
839 
840 MODULE_DESCRIPTION("Intel Keem Bay OCS HCU Crypto Driver");
841 MODULE_LICENSE("GPL");
842