1/*
2 * Glue code for the SHA512 Secure Hash Algorithm assembly implementation
3 * using NEON instructions.
4 *
5 * Copyright © 2014 Jussi Kivilinna <jussi.kivilinna@iki.fi>
6 *
7 * This file is based on sha512_ssse3_glue.c:
8 *   Copyright (C) 2013 Intel Corporation
9 *   Author: Tim Chen <tim.c.chen@linux.intel.com>
10 *
11 * This program is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the Free
13 * Software Foundation; either version 2 of the License, or (at your option)
14 * any later version.
15 *
16 */
17
18#include <crypto/internal/hash.h>
19#include <linux/init.h>
20#include <linux/module.h>
21#include <linux/mm.h>
22#include <linux/cryptohash.h>
23#include <linux/types.h>
24#include <linux/string.h>
25#include <crypto/sha.h>
26#include <asm/byteorder.h>
27#include <asm/simd.h>
28#include <asm/neon.h>
29
30
31static const u64 sha512_k[] = {
32	0x428a2f98d728ae22ULL, 0x7137449123ef65cdULL,
33	0xb5c0fbcfec4d3b2fULL, 0xe9b5dba58189dbbcULL,
34	0x3956c25bf348b538ULL, 0x59f111f1b605d019ULL,
35	0x923f82a4af194f9bULL, 0xab1c5ed5da6d8118ULL,
36	0xd807aa98a3030242ULL, 0x12835b0145706fbeULL,
37	0x243185be4ee4b28cULL, 0x550c7dc3d5ffb4e2ULL,
38	0x72be5d74f27b896fULL, 0x80deb1fe3b1696b1ULL,
39	0x9bdc06a725c71235ULL, 0xc19bf174cf692694ULL,
40	0xe49b69c19ef14ad2ULL, 0xefbe4786384f25e3ULL,
41	0x0fc19dc68b8cd5b5ULL, 0x240ca1cc77ac9c65ULL,
42	0x2de92c6f592b0275ULL, 0x4a7484aa6ea6e483ULL,
43	0x5cb0a9dcbd41fbd4ULL, 0x76f988da831153b5ULL,
44	0x983e5152ee66dfabULL, 0xa831c66d2db43210ULL,
45	0xb00327c898fb213fULL, 0xbf597fc7beef0ee4ULL,
46	0xc6e00bf33da88fc2ULL, 0xd5a79147930aa725ULL,
47	0x06ca6351e003826fULL, 0x142929670a0e6e70ULL,
48	0x27b70a8546d22ffcULL, 0x2e1b21385c26c926ULL,
49	0x4d2c6dfc5ac42aedULL, 0x53380d139d95b3dfULL,
50	0x650a73548baf63deULL, 0x766a0abb3c77b2a8ULL,
51	0x81c2c92e47edaee6ULL, 0x92722c851482353bULL,
52	0xa2bfe8a14cf10364ULL, 0xa81a664bbc423001ULL,
53	0xc24b8b70d0f89791ULL, 0xc76c51a30654be30ULL,
54	0xd192e819d6ef5218ULL, 0xd69906245565a910ULL,
55	0xf40e35855771202aULL, 0x106aa07032bbd1b8ULL,
56	0x19a4c116b8d2d0c8ULL, 0x1e376c085141ab53ULL,
57	0x2748774cdf8eeb99ULL, 0x34b0bcb5e19b48a8ULL,
58	0x391c0cb3c5c95a63ULL, 0x4ed8aa4ae3418acbULL,
59	0x5b9cca4f7763e373ULL, 0x682e6ff3d6b2b8a3ULL,
60	0x748f82ee5defb2fcULL, 0x78a5636f43172f60ULL,
61	0x84c87814a1f0ab72ULL, 0x8cc702081a6439ecULL,
62	0x90befffa23631e28ULL, 0xa4506cebde82bde9ULL,
63	0xbef9a3f7b2c67915ULL, 0xc67178f2e372532bULL,
64	0xca273eceea26619cULL, 0xd186b8c721c0c207ULL,
65	0xeada7dd6cde0eb1eULL, 0xf57d4f7fee6ed178ULL,
66	0x06f067aa72176fbaULL, 0x0a637dc5a2c898a6ULL,
67	0x113f9804bef90daeULL, 0x1b710b35131c471bULL,
68	0x28db77f523047d84ULL, 0x32caab7b40c72493ULL,
69	0x3c9ebe0a15c9bebcULL, 0x431d67c49c100d4cULL,
70	0x4cc5d4becb3e42b6ULL, 0x597f299cfc657e2aULL,
71	0x5fcb6fab3ad6faecULL, 0x6c44198c4a475817ULL
72};
73
74
75asmlinkage void sha512_transform_neon(u64 *digest, const void *data,
76				      const u64 k[], unsigned int num_blks);
77
78
79static int sha512_neon_init(struct shash_desc *desc)
80{
81	struct sha512_state *sctx = shash_desc_ctx(desc);
82
83	sctx->state[0] = SHA512_H0;
84	sctx->state[1] = SHA512_H1;
85	sctx->state[2] = SHA512_H2;
86	sctx->state[3] = SHA512_H3;
87	sctx->state[4] = SHA512_H4;
88	sctx->state[5] = SHA512_H5;
89	sctx->state[6] = SHA512_H6;
90	sctx->state[7] = SHA512_H7;
91	sctx->count[0] = sctx->count[1] = 0;
92
93	return 0;
94}
95
96static int __sha512_neon_update(struct shash_desc *desc, const u8 *data,
97				unsigned int len, unsigned int partial)
98{
99	struct sha512_state *sctx = shash_desc_ctx(desc);
100	unsigned int done = 0;
101
102	sctx->count[0] += len;
103	if (sctx->count[0] < len)
104		sctx->count[1]++;
105
106	if (partial) {
107		done = SHA512_BLOCK_SIZE - partial;
108		memcpy(sctx->buf + partial, data, done);
109		sha512_transform_neon(sctx->state, sctx->buf, sha512_k, 1);
110	}
111
112	if (len - done >= SHA512_BLOCK_SIZE) {
113		const unsigned int rounds = (len - done) / SHA512_BLOCK_SIZE;
114
115		sha512_transform_neon(sctx->state, data + done, sha512_k,
116				      rounds);
117
118		done += rounds * SHA512_BLOCK_SIZE;
119	}
120
121	memcpy(sctx->buf, data + done, len - done);
122
123	return 0;
124}
125
126static int sha512_neon_update(struct shash_desc *desc, const u8 *data,
127			     unsigned int len)
128{
129	struct sha512_state *sctx = shash_desc_ctx(desc);
130	unsigned int partial = sctx->count[0] % SHA512_BLOCK_SIZE;
131	int res;
132
133	/* Handle the fast case right here */
134	if (partial + len < SHA512_BLOCK_SIZE) {
135		sctx->count[0] += len;
136		if (sctx->count[0] < len)
137			sctx->count[1]++;
138		memcpy(sctx->buf + partial, data, len);
139
140		return 0;
141	}
142
143	if (!may_use_simd()) {
144		res = crypto_sha512_update(desc, data, len);
145	} else {
146		kernel_neon_begin();
147		res = __sha512_neon_update(desc, data, len, partial);
148		kernel_neon_end();
149	}
150
151	return res;
152}
153
154
155/* Add padding and return the message digest. */
156static int sha512_neon_final(struct shash_desc *desc, u8 *out)
157{
158	struct sha512_state *sctx = shash_desc_ctx(desc);
159	unsigned int i, index, padlen;
160	__be64 *dst = (__be64 *)out;
161	__be64 bits[2];
162	static const u8 padding[SHA512_BLOCK_SIZE] = { 0x80, };
163
164	/* save number of bits */
165	bits[1] = cpu_to_be64(sctx->count[0] << 3);
166	bits[0] = cpu_to_be64(sctx->count[1] << 3 | sctx->count[0] >> 61);
167
168	/* Pad out to 112 mod 128 and append length */
169	index = sctx->count[0] & 0x7f;
170	padlen = (index < 112) ? (112 - index) : ((128+112) - index);
171
172	if (!may_use_simd()) {
173		crypto_sha512_update(desc, padding, padlen);
174		crypto_sha512_update(desc, (const u8 *)&bits, sizeof(bits));
175	} else {
176		kernel_neon_begin();
177		/* We need to fill a whole block for __sha512_neon_update() */
178		if (padlen <= 112) {
179			sctx->count[0] += padlen;
180			if (sctx->count[0] < padlen)
181				sctx->count[1]++;
182			memcpy(sctx->buf + index, padding, padlen);
183		} else {
184			__sha512_neon_update(desc, padding, padlen, index);
185		}
186		__sha512_neon_update(desc, (const u8 *)&bits,
187					sizeof(bits), 112);
188		kernel_neon_end();
189	}
190
191	/* Store state in digest */
192	for (i = 0; i < 8; i++)
193		dst[i] = cpu_to_be64(sctx->state[i]);
194
195	/* Wipe context */
196	memset(sctx, 0, sizeof(*sctx));
197
198	return 0;
199}
200
201static int sha512_neon_export(struct shash_desc *desc, void *out)
202{
203	struct sha512_state *sctx = shash_desc_ctx(desc);
204
205	memcpy(out, sctx, sizeof(*sctx));
206
207	return 0;
208}
209
210static int sha512_neon_import(struct shash_desc *desc, const void *in)
211{
212	struct sha512_state *sctx = shash_desc_ctx(desc);
213
214	memcpy(sctx, in, sizeof(*sctx));
215
216	return 0;
217}
218
219static int sha384_neon_init(struct shash_desc *desc)
220{
221	struct sha512_state *sctx = shash_desc_ctx(desc);
222
223	sctx->state[0] = SHA384_H0;
224	sctx->state[1] = SHA384_H1;
225	sctx->state[2] = SHA384_H2;
226	sctx->state[3] = SHA384_H3;
227	sctx->state[4] = SHA384_H4;
228	sctx->state[5] = SHA384_H5;
229	sctx->state[6] = SHA384_H6;
230	sctx->state[7] = SHA384_H7;
231
232	sctx->count[0] = sctx->count[1] = 0;
233
234	return 0;
235}
236
237static int sha384_neon_final(struct shash_desc *desc, u8 *hash)
238{
239	u8 D[SHA512_DIGEST_SIZE];
240
241	sha512_neon_final(desc, D);
242
243	memcpy(hash, D, SHA384_DIGEST_SIZE);
244	memzero_explicit(D, SHA512_DIGEST_SIZE);
245
246	return 0;
247}
248
249static struct shash_alg algs[] = { {
250	.digestsize	=	SHA512_DIGEST_SIZE,
251	.init		=	sha512_neon_init,
252	.update		=	sha512_neon_update,
253	.final		=	sha512_neon_final,
254	.export		=	sha512_neon_export,
255	.import		=	sha512_neon_import,
256	.descsize	=	sizeof(struct sha512_state),
257	.statesize	=	sizeof(struct sha512_state),
258	.base		=	{
259		.cra_name	=	"sha512",
260		.cra_driver_name =	"sha512-neon",
261		.cra_priority	=	250,
262		.cra_flags	=	CRYPTO_ALG_TYPE_SHASH,
263		.cra_blocksize	=	SHA512_BLOCK_SIZE,
264		.cra_module	=	THIS_MODULE,
265	}
266},  {
267	.digestsize	=	SHA384_DIGEST_SIZE,
268	.init		=	sha384_neon_init,
269	.update		=	sha512_neon_update,
270	.final		=	sha384_neon_final,
271	.export		=	sha512_neon_export,
272	.import		=	sha512_neon_import,
273	.descsize	=	sizeof(struct sha512_state),
274	.statesize	=	sizeof(struct sha512_state),
275	.base		=	{
276		.cra_name	=	"sha384",
277		.cra_driver_name =	"sha384-neon",
278		.cra_priority	=	250,
279		.cra_flags	=	CRYPTO_ALG_TYPE_SHASH,
280		.cra_blocksize	=	SHA384_BLOCK_SIZE,
281		.cra_module	=	THIS_MODULE,
282	}
283} };
284
285static int __init sha512_neon_mod_init(void)
286{
287	if (!cpu_has_neon())
288		return -ENODEV;
289
290	return crypto_register_shashes(algs, ARRAY_SIZE(algs));
291}
292
293static void __exit sha512_neon_mod_fini(void)
294{
295	crypto_unregister_shashes(algs, ARRAY_SIZE(algs));
296}
297
298module_init(sha512_neon_mod_init);
299module_exit(sha512_neon_mod_fini);
300
301MODULE_LICENSE("GPL");
302MODULE_DESCRIPTION("SHA512 Secure Hash Algorithm, NEON accelerated");
303
304MODULE_ALIAS_CRYPTO("sha512");
305MODULE_ALIAS_CRYPTO("sha384");
306