1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * sun4i-ss-hash.c - hardware cryptographic accelerator for Allwinner A20 SoC
4  *
5  * Copyright (C) 2013-2015 Corentin LABBE <clabbe.montjoie@gmail.com>
6  *
7  * This file add support for MD5 and SHA1.
8  *
9  * You could find the datasheet in Documentation/arm/sunxi.rst
10  */
11 #include "sun4i-ss.h"
12 #include <asm/unaligned.h>
13 #include <linux/scatterlist.h>
14 
15 /* This is a totally arbitrary value */
16 #define SS_TIMEOUT 100
17 
18 int sun4i_hash_crainit(struct crypto_tfm *tfm)
19 {
20 	struct sun4i_tfm_ctx *op = crypto_tfm_ctx(tfm);
21 	struct ahash_alg *alg = __crypto_ahash_alg(tfm->__crt_alg);
22 	struct sun4i_ss_alg_template *algt;
23 	int err;
24 
25 	memset(op, 0, sizeof(struct sun4i_tfm_ctx));
26 
27 	algt = container_of(alg, struct sun4i_ss_alg_template, alg.hash);
28 	op->ss = algt->ss;
29 
30 	err = pm_runtime_get_sync(op->ss->dev);
31 	if (err < 0)
32 		return err;
33 
34 	crypto_ahash_set_reqsize(__crypto_ahash_cast(tfm),
35 				 sizeof(struct sun4i_req_ctx));
36 	return 0;
37 }
38 
39 void sun4i_hash_craexit(struct crypto_tfm *tfm)
40 {
41 	struct sun4i_tfm_ctx *op = crypto_tfm_ctx(tfm);
42 
43 	pm_runtime_put(op->ss->dev);
44 }
45 
46 /* sun4i_hash_init: initialize request context */
47 int sun4i_hash_init(struct ahash_request *areq)
48 {
49 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
50 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(areq);
51 	struct ahash_alg *alg = __crypto_ahash_alg(tfm->base.__crt_alg);
52 	struct sun4i_ss_alg_template *algt;
53 
54 	memset(op, 0, sizeof(struct sun4i_req_ctx));
55 
56 	algt = container_of(alg, struct sun4i_ss_alg_template, alg.hash);
57 	op->mode = algt->mode;
58 
59 	return 0;
60 }
61 
62 int sun4i_hash_export_md5(struct ahash_request *areq, void *out)
63 {
64 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
65 	struct md5_state *octx = out;
66 	int i;
67 
68 	octx->byte_count = op->byte_count + op->len;
69 
70 	memcpy(octx->block, op->buf, op->len);
71 
72 	if (op->byte_count) {
73 		for (i = 0; i < 4; i++)
74 			octx->hash[i] = op->hash[i];
75 	} else {
76 		octx->hash[0] = SHA1_H0;
77 		octx->hash[1] = SHA1_H1;
78 		octx->hash[2] = SHA1_H2;
79 		octx->hash[3] = SHA1_H3;
80 	}
81 
82 	return 0;
83 }
84 
85 int sun4i_hash_import_md5(struct ahash_request *areq, const void *in)
86 {
87 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
88 	const struct md5_state *ictx = in;
89 	int i;
90 
91 	sun4i_hash_init(areq);
92 
93 	op->byte_count = ictx->byte_count & ~0x3F;
94 	op->len = ictx->byte_count & 0x3F;
95 
96 	memcpy(op->buf, ictx->block, op->len);
97 
98 	for (i = 0; i < 4; i++)
99 		op->hash[i] = ictx->hash[i];
100 
101 	return 0;
102 }
103 
104 int sun4i_hash_export_sha1(struct ahash_request *areq, void *out)
105 {
106 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
107 	struct sha1_state *octx = out;
108 	int i;
109 
110 	octx->count = op->byte_count + op->len;
111 
112 	memcpy(octx->buffer, op->buf, op->len);
113 
114 	if (op->byte_count) {
115 		for (i = 0; i < 5; i++)
116 			octx->state[i] = op->hash[i];
117 	} else {
118 		octx->state[0] = SHA1_H0;
119 		octx->state[1] = SHA1_H1;
120 		octx->state[2] = SHA1_H2;
121 		octx->state[3] = SHA1_H3;
122 		octx->state[4] = SHA1_H4;
123 	}
124 
125 	return 0;
126 }
127 
128 int sun4i_hash_import_sha1(struct ahash_request *areq, const void *in)
129 {
130 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
131 	const struct sha1_state *ictx = in;
132 	int i;
133 
134 	sun4i_hash_init(areq);
135 
136 	op->byte_count = ictx->count & ~0x3F;
137 	op->len = ictx->count & 0x3F;
138 
139 	memcpy(op->buf, ictx->buffer, op->len);
140 
141 	for (i = 0; i < 5; i++)
142 		op->hash[i] = ictx->state[i];
143 
144 	return 0;
145 }
146 
147 #define SS_HASH_UPDATE 1
148 #define SS_HASH_FINAL 2
149 
150 /*
151  * sun4i_hash_update: update hash engine
152  *
153  * Could be used for both SHA1 and MD5
154  * Write data by step of 32bits and put then in the SS.
155  *
156  * Since we cannot leave partial data and hash state in the engine,
157  * we need to get the hash state at the end of this function.
158  * We can get the hash state every 64 bytes
159  *
160  * So the first work is to get the number of bytes to write to SS modulo 64
161  * The extra bytes will go to a temporary buffer op->buf storing op->len bytes
162  *
163  * So at the begin of update()
164  * if op->len + areq->nbytes < 64
165  * => all data will be written to wait buffer (op->buf) and end=0
166  * if not, write all data from op->buf to the device and position end to
167  * complete to 64bytes
168  *
169  * example 1:
170  * update1 60o => op->len=60
171  * update2 60o => need one more word to have 64 bytes
172  * end=4
173  * so write all data from op->buf and one word of SGs
174  * write remaining data in op->buf
175  * final state op->len=56
176  */
177 static int sun4i_hash(struct ahash_request *areq)
178 {
179 	/*
180 	 * i is the total bytes read from SGs, to be compared to areq->nbytes
181 	 * i is important because we cannot rely on SG length since the sum of
182 	 * SG->length could be greater than areq->nbytes
183 	 *
184 	 * end is the position when we need to stop writing to the device,
185 	 * to be compared to i
186 	 *
187 	 * in_i: advancement in the current SG
188 	 */
189 	unsigned int i = 0, end, fill, min_fill, nwait, nbw = 0, j = 0, todo;
190 	unsigned int in_i = 0;
191 	u32 spaces, rx_cnt = SS_RX_DEFAULT, bf[32] = {0}, v, ivmode = 0;
192 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
193 	struct crypto_ahash *tfm = crypto_ahash_reqtfm(areq);
194 	struct sun4i_tfm_ctx *tfmctx = crypto_ahash_ctx(tfm);
195 	struct sun4i_ss_ctx *ss = tfmctx->ss;
196 	struct scatterlist *in_sg = areq->src;
197 	struct sg_mapping_iter mi;
198 	int in_r, err = 0;
199 	size_t copied = 0;
200 	u32 wb = 0;
201 
202 	dev_dbg(ss->dev, "%s %s bc=%llu len=%u mode=%x wl=%u h0=%0x",
203 		__func__, crypto_tfm_alg_name(areq->base.tfm),
204 		op->byte_count, areq->nbytes, op->mode,
205 		op->len, op->hash[0]);
206 
207 	if (unlikely(!areq->nbytes) && !(op->flags & SS_HASH_FINAL))
208 		return 0;
209 
210 	/* protect against overflow */
211 	if (unlikely(areq->nbytes > UINT_MAX - op->len)) {
212 		dev_err(ss->dev, "Cannot process too large request\n");
213 		return -EINVAL;
214 	}
215 
216 	if (op->len + areq->nbytes < 64 && !(op->flags & SS_HASH_FINAL)) {
217 		/* linearize data to op->buf */
218 		copied = sg_pcopy_to_buffer(areq->src, sg_nents(areq->src),
219 					    op->buf + op->len, areq->nbytes, 0);
220 		op->len += copied;
221 		return 0;
222 	}
223 
224 	spin_lock_bh(&ss->slock);
225 
226 	/*
227 	 * if some data have been processed before,
228 	 * we need to restore the partial hash state
229 	 */
230 	if (op->byte_count) {
231 		ivmode = SS_IV_ARBITRARY;
232 		for (i = 0; i < crypto_ahash_digestsize(tfm) / 4; i++)
233 			writel(op->hash[i], ss->base + SS_IV0 + i * 4);
234 	}
235 	/* Enable the device */
236 	writel(op->mode | SS_ENABLED | ivmode, ss->base + SS_CTL);
237 
238 	if (!(op->flags & SS_HASH_UPDATE))
239 		goto hash_final;
240 
241 	/* start of handling data */
242 	if (!(op->flags & SS_HASH_FINAL)) {
243 		end = ((areq->nbytes + op->len) / 64) * 64 - op->len;
244 
245 		if (end > areq->nbytes || areq->nbytes - end > 63) {
246 			dev_err(ss->dev, "ERROR: Bound error %u %u\n",
247 				end, areq->nbytes);
248 			err = -EINVAL;
249 			goto release_ss;
250 		}
251 	} else {
252 		/* Since we have the flag final, we can go up to modulo 4 */
253 		if (areq->nbytes < 4)
254 			end = 0;
255 		else
256 			end = ((areq->nbytes + op->len) / 4) * 4 - op->len;
257 	}
258 
259 	/* TODO if SGlen % 4 and !op->len then DMA */
260 	i = 1;
261 	while (in_sg && i == 1) {
262 		if (in_sg->length % 4)
263 			i = 0;
264 		in_sg = sg_next(in_sg);
265 	}
266 	if (i == 1 && !op->len && areq->nbytes)
267 		dev_dbg(ss->dev, "We can DMA\n");
268 
269 	i = 0;
270 	sg_miter_start(&mi, areq->src, sg_nents(areq->src),
271 		       SG_MITER_FROM_SG | SG_MITER_ATOMIC);
272 	sg_miter_next(&mi);
273 	in_i = 0;
274 
275 	do {
276 		/*
277 		 * we need to linearize in two case:
278 		 * - the buffer is already used
279 		 * - the SG does not have enough byte remaining ( < 4)
280 		 */
281 		if (op->len || (mi.length - in_i) < 4) {
282 			/*
283 			 * if we have entered here we have two reason to stop
284 			 * - the buffer is full
285 			 * - reach the end
286 			 */
287 			while (op->len < 64 && i < end) {
288 				/* how many bytes we can read from current SG */
289 				in_r = min(end - i, 64 - op->len);
290 				in_r = min_t(size_t, mi.length - in_i, in_r);
291 				memcpy(op->buf + op->len, mi.addr + in_i, in_r);
292 				op->len += in_r;
293 				i += in_r;
294 				in_i += in_r;
295 				if (in_i == mi.length) {
296 					sg_miter_next(&mi);
297 					in_i = 0;
298 				}
299 			}
300 			if (op->len > 3 && !(op->len % 4)) {
301 				/* write buf to the device */
302 				writesl(ss->base + SS_RXFIFO, op->buf,
303 					op->len / 4);
304 				op->byte_count += op->len;
305 				op->len = 0;
306 			}
307 		}
308 		if (mi.length - in_i > 3 && i < end) {
309 			/* how many bytes we can read from current SG */
310 			in_r = min_t(size_t, mi.length - in_i, areq->nbytes - i);
311 			in_r = min_t(size_t, ((mi.length - in_i) / 4) * 4, in_r);
312 			/* how many bytes we can write in the device*/
313 			todo = min3((u32)(end - i) / 4, rx_cnt, (u32)in_r / 4);
314 			writesl(ss->base + SS_RXFIFO, mi.addr + in_i, todo);
315 			op->byte_count += todo * 4;
316 			i += todo * 4;
317 			in_i += todo * 4;
318 			rx_cnt -= todo;
319 			if (!rx_cnt) {
320 				spaces = readl(ss->base + SS_FCSR);
321 				rx_cnt = SS_RXFIFO_SPACES(spaces);
322 			}
323 			if (in_i == mi.length) {
324 				sg_miter_next(&mi);
325 				in_i = 0;
326 			}
327 		}
328 	} while (i < end);
329 
330 	/*
331 	 * Now we have written to the device all that we can,
332 	 * store the remaining bytes in op->buf
333 	 */
334 	if ((areq->nbytes - i) < 64) {
335 		while (i < areq->nbytes && in_i < mi.length && op->len < 64) {
336 			/* how many bytes we can read from current SG */
337 			in_r = min(areq->nbytes - i, 64 - op->len);
338 			in_r = min_t(size_t, mi.length - in_i, in_r);
339 			memcpy(op->buf + op->len, mi.addr + in_i, in_r);
340 			op->len += in_r;
341 			i += in_r;
342 			in_i += in_r;
343 			if (in_i == mi.length) {
344 				sg_miter_next(&mi);
345 				in_i = 0;
346 			}
347 		}
348 	}
349 
350 	sg_miter_stop(&mi);
351 
352 	/*
353 	 * End of data process
354 	 * Now if we have the flag final go to finalize part
355 	 * If not, store the partial hash
356 	 */
357 	if (op->flags & SS_HASH_FINAL)
358 		goto hash_final;
359 
360 	writel(op->mode | SS_ENABLED | SS_DATA_END, ss->base + SS_CTL);
361 	i = 0;
362 	do {
363 		v = readl(ss->base + SS_CTL);
364 		i++;
365 	} while (i < SS_TIMEOUT && (v & SS_DATA_END));
366 	if (unlikely(i >= SS_TIMEOUT)) {
367 		dev_err_ratelimited(ss->dev,
368 				    "ERROR: hash end timeout %d>%d ctl=%x len=%u\n",
369 				    i, SS_TIMEOUT, v, areq->nbytes);
370 		err = -EIO;
371 		goto release_ss;
372 	}
373 
374 	/*
375 	 * The datasheet isn't very clear about when to retrieve the digest. The
376 	 * bit SS_DATA_END is cleared when the engine has processed the data and
377 	 * when the digest is computed *but* it doesn't mean the digest is
378 	 * available in the digest registers. Hence the delay to be sure we can
379 	 * read it.
380 	 */
381 	ndelay(1);
382 
383 	for (i = 0; i < crypto_ahash_digestsize(tfm) / 4; i++)
384 		op->hash[i] = readl(ss->base + SS_MD0 + i * 4);
385 
386 	goto release_ss;
387 
388 /*
389  * hash_final: finalize hashing operation
390  *
391  * If we have some remaining bytes, we write them.
392  * Then ask the SS for finalizing the hashing operation
393  *
394  * I do not check RX FIFO size in this function since the size is 32
395  * after each enabling and this function neither write more than 32 words.
396  * If we come from the update part, we cannot have more than
397  * 3 remaining bytes to write and SS is fast enough to not care about it.
398  */
399 
400 hash_final:
401 
402 	/* write the remaining words of the wait buffer */
403 	if (op->len) {
404 		nwait = op->len / 4;
405 		if (nwait) {
406 			writesl(ss->base + SS_RXFIFO, op->buf, nwait);
407 			op->byte_count += 4 * nwait;
408 		}
409 
410 		nbw = op->len - 4 * nwait;
411 		if (nbw) {
412 			wb = le32_to_cpup((__le32 *)(op->buf + nwait * 4));
413 			wb &= GENMASK((nbw * 8) - 1, 0);
414 
415 			op->byte_count += nbw;
416 		}
417 	}
418 
419 	/* write the remaining bytes of the nbw buffer */
420 	wb |= ((1 << 7) << (nbw * 8));
421 	((__le32 *)bf)[j++] = cpu_to_le32(wb);
422 
423 	/*
424 	 * number of space to pad to obtain 64o minus 8(size) minus 4 (final 1)
425 	 * I take the operations from other MD5/SHA1 implementations
426 	 */
427 
428 	/* last block size */
429 	fill = 64 - (op->byte_count % 64);
430 	min_fill = 2 * sizeof(u32) + (nbw ? 0 : sizeof(u32));
431 
432 	/* if we can't fill all data, jump to the next 64 block */
433 	if (fill < min_fill)
434 		fill += 64;
435 
436 	j += (fill - min_fill) / sizeof(u32);
437 
438 	/* write the length of data */
439 	if (op->mode == SS_OP_SHA1) {
440 		__be64 *bits = (__be64 *)&bf[j];
441 		*bits = cpu_to_be64(op->byte_count << 3);
442 		j += 2;
443 	} else {
444 		__le64 *bits = (__le64 *)&bf[j];
445 		*bits = cpu_to_le64(op->byte_count << 3);
446 		j += 2;
447 	}
448 	writesl(ss->base + SS_RXFIFO, bf, j);
449 
450 	/* Tell the SS to stop the hashing */
451 	writel(op->mode | SS_ENABLED | SS_DATA_END, ss->base + SS_CTL);
452 
453 	/*
454 	 * Wait for SS to finish the hash.
455 	 * The timeout could happen only in case of bad overclocking
456 	 * or driver bug.
457 	 */
458 	i = 0;
459 	do {
460 		v = readl(ss->base + SS_CTL);
461 		i++;
462 	} while (i < SS_TIMEOUT && (v & SS_DATA_END));
463 	if (unlikely(i >= SS_TIMEOUT)) {
464 		dev_err_ratelimited(ss->dev,
465 				    "ERROR: hash end timeout %d>%d ctl=%x len=%u\n",
466 				    i, SS_TIMEOUT, v, areq->nbytes);
467 		err = -EIO;
468 		goto release_ss;
469 	}
470 
471 	/*
472 	 * The datasheet isn't very clear about when to retrieve the digest. The
473 	 * bit SS_DATA_END is cleared when the engine has processed the data and
474 	 * when the digest is computed *but* it doesn't mean the digest is
475 	 * available in the digest registers. Hence the delay to be sure we can
476 	 * read it.
477 	 */
478 	ndelay(1);
479 
480 	/* Get the hash from the device */
481 	if (op->mode == SS_OP_SHA1) {
482 		for (i = 0; i < 5; i++) {
483 			v = readl(ss->base + SS_MD0 + i * 4);
484 			if (ss->variant->sha1_in_be)
485 				put_unaligned_le32(v, areq->result + i * 4);
486 			else
487 				put_unaligned_be32(v, areq->result + i * 4);
488 		}
489 	} else {
490 		for (i = 0; i < 4; i++) {
491 			v = readl(ss->base + SS_MD0 + i * 4);
492 			put_unaligned_le32(v, areq->result + i * 4);
493 		}
494 	}
495 
496 release_ss:
497 	writel(0, ss->base + SS_CTL);
498 	spin_unlock_bh(&ss->slock);
499 	return err;
500 }
501 
502 int sun4i_hash_final(struct ahash_request *areq)
503 {
504 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
505 
506 	op->flags = SS_HASH_FINAL;
507 	return sun4i_hash(areq);
508 }
509 
510 int sun4i_hash_update(struct ahash_request *areq)
511 {
512 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
513 
514 	op->flags = SS_HASH_UPDATE;
515 	return sun4i_hash(areq);
516 }
517 
518 /* sun4i_hash_finup: finalize hashing operation after an update */
519 int sun4i_hash_finup(struct ahash_request *areq)
520 {
521 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
522 
523 	op->flags = SS_HASH_UPDATE | SS_HASH_FINAL;
524 	return sun4i_hash(areq);
525 }
526 
527 /* combo of init/update/final functions */
528 int sun4i_hash_digest(struct ahash_request *areq)
529 {
530 	int err;
531 	struct sun4i_req_ctx *op = ahash_request_ctx(areq);
532 
533 	err = sun4i_hash_init(areq);
534 	if (err)
535 		return err;
536 
537 	op->flags = SS_HASH_UPDATE | SS_HASH_FINAL;
538 	return sun4i_hash(areq);
539 }
540