xref: /linux/arch/arm/crypto/sha1_neon_glue.c (revision add4b1b02da7e7ec35c34dd04d351ac53f3f0dd8)
1 /*
2  * Glue code for the SHA1 Secure Hash Algorithm assembler implementation using
3  * ARM NEON instructions.
4  *
5  * Copyright © 2014 Jussi Kivilinna <jussi.kivilinna@iki.fi>
6  *
7  * This file is based on sha1_generic.c and sha1_ssse3_glue.c:
8  *  Copyright (c) Alan Smithee.
9  *  Copyright (c) Andrew McDonald <andrew@mcdonald.org.uk>
10  *  Copyright (c) Jean-Francois Dive <jef@linuxbe.org>
11  *  Copyright (c) Mathias Krause <minipli@googlemail.com>
12  *  Copyright (c) Chandramouli Narayanan <mouli@linux.intel.com>
13  *
14  * This program is free software; you can redistribute it and/or modify it
15  * under the terms of the GNU General Public License as published by the Free
16  * Software Foundation; either version 2 of the License, or (at your option)
17  * any later version.
18  *
19  */
20 
21 #include <crypto/internal/hash.h>
22 #include <linux/init.h>
23 #include <linux/module.h>
24 #include <linux/mm.h>
25 #include <linux/cryptohash.h>
26 #include <linux/types.h>
27 #include <crypto/sha.h>
28 #include <asm/byteorder.h>
29 #include <asm/neon.h>
30 #include <asm/simd.h>
31 #include <asm/crypto/sha1.h>
32 
33 
34 asmlinkage void sha1_transform_neon(void *state_h, const char *data,
35 				    unsigned int rounds);
36 
37 
38 static int sha1_neon_init(struct shash_desc *desc)
39 {
40 	struct sha1_state *sctx = shash_desc_ctx(desc);
41 
42 	*sctx = (struct sha1_state){
43 		.state = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 },
44 	};
45 
46 	return 0;
47 }
48 
49 static int __sha1_neon_update(struct shash_desc *desc, const u8 *data,
50 			       unsigned int len, unsigned int partial)
51 {
52 	struct sha1_state *sctx = shash_desc_ctx(desc);
53 	unsigned int done = 0;
54 
55 	sctx->count += len;
56 
57 	if (partial) {
58 		done = SHA1_BLOCK_SIZE - partial;
59 		memcpy(sctx->buffer + partial, data, done);
60 		sha1_transform_neon(sctx->state, sctx->buffer, 1);
61 	}
62 
63 	if (len - done >= SHA1_BLOCK_SIZE) {
64 		const unsigned int rounds = (len - done) / SHA1_BLOCK_SIZE;
65 
66 		sha1_transform_neon(sctx->state, data + done, rounds);
67 		done += rounds * SHA1_BLOCK_SIZE;
68 	}
69 
70 	memcpy(sctx->buffer, data + done, len - done);
71 
72 	return 0;
73 }
74 
75 static int sha1_neon_update(struct shash_desc *desc, const u8 *data,
76 			     unsigned int len)
77 {
78 	struct sha1_state *sctx = shash_desc_ctx(desc);
79 	unsigned int partial = sctx->count % SHA1_BLOCK_SIZE;
80 	int res;
81 
82 	/* Handle the fast case right here */
83 	if (partial + len < SHA1_BLOCK_SIZE) {
84 		sctx->count += len;
85 		memcpy(sctx->buffer + partial, data, len);
86 
87 		return 0;
88 	}
89 
90 	if (!may_use_simd()) {
91 		res = sha1_update_arm(desc, data, len);
92 	} else {
93 		kernel_neon_begin();
94 		res = __sha1_neon_update(desc, data, len, partial);
95 		kernel_neon_end();
96 	}
97 
98 	return res;
99 }
100 
101 
102 /* Add padding and return the message digest. */
103 static int sha1_neon_final(struct shash_desc *desc, u8 *out)
104 {
105 	struct sha1_state *sctx = shash_desc_ctx(desc);
106 	unsigned int i, index, padlen;
107 	__be32 *dst = (__be32 *)out;
108 	__be64 bits;
109 	static const u8 padding[SHA1_BLOCK_SIZE] = { 0x80, };
110 
111 	bits = cpu_to_be64(sctx->count << 3);
112 
113 	/* Pad out to 56 mod 64 and append length */
114 	index = sctx->count % SHA1_BLOCK_SIZE;
115 	padlen = (index < 56) ? (56 - index) : ((SHA1_BLOCK_SIZE+56) - index);
116 	if (!may_use_simd()) {
117 		sha1_update_arm(desc, padding, padlen);
118 		sha1_update_arm(desc, (const u8 *)&bits, sizeof(bits));
119 	} else {
120 		kernel_neon_begin();
121 		/* We need to fill a whole block for __sha1_neon_update() */
122 		if (padlen <= 56) {
123 			sctx->count += padlen;
124 			memcpy(sctx->buffer + index, padding, padlen);
125 		} else {
126 			__sha1_neon_update(desc, padding, padlen, index);
127 		}
128 		__sha1_neon_update(desc, (const u8 *)&bits, sizeof(bits), 56);
129 		kernel_neon_end();
130 	}
131 
132 	/* Store state in digest */
133 	for (i = 0; i < 5; i++)
134 		dst[i] = cpu_to_be32(sctx->state[i]);
135 
136 	/* Wipe context */
137 	memset(sctx, 0, sizeof(*sctx));
138 
139 	return 0;
140 }
141 
142 static int sha1_neon_export(struct shash_desc *desc, void *out)
143 {
144 	struct sha1_state *sctx = shash_desc_ctx(desc);
145 
146 	memcpy(out, sctx, sizeof(*sctx));
147 
148 	return 0;
149 }
150 
151 static int sha1_neon_import(struct shash_desc *desc, const void *in)
152 {
153 	struct sha1_state *sctx = shash_desc_ctx(desc);
154 
155 	memcpy(sctx, in, sizeof(*sctx));
156 
157 	return 0;
158 }
159 
160 static struct shash_alg alg = {
161 	.digestsize	=	SHA1_DIGEST_SIZE,
162 	.init		=	sha1_neon_init,
163 	.update		=	sha1_neon_update,
164 	.final		=	sha1_neon_final,
165 	.export		=	sha1_neon_export,
166 	.import		=	sha1_neon_import,
167 	.descsize	=	sizeof(struct sha1_state),
168 	.statesize	=	sizeof(struct sha1_state),
169 	.base		=	{
170 		.cra_name		= "sha1",
171 		.cra_driver_name	= "sha1-neon",
172 		.cra_priority		= 250,
173 		.cra_flags		= CRYPTO_ALG_TYPE_SHASH,
174 		.cra_blocksize		= SHA1_BLOCK_SIZE,
175 		.cra_module		= THIS_MODULE,
176 	}
177 };
178 
179 static int __init sha1_neon_mod_init(void)
180 {
181 	if (!cpu_has_neon())
182 		return -ENODEV;
183 
184 	return crypto_register_shash(&alg);
185 }
186 
187 static void __exit sha1_neon_mod_fini(void)
188 {
189 	crypto_unregister_shash(&alg);
190 }
191 
192 module_init(sha1_neon_mod_init);
193 module_exit(sha1_neon_mod_fini);
194 
195 MODULE_LICENSE("GPL");
196 MODULE_DESCRIPTION("SHA1 Secure Hash Algorithm, NEON accelerated");
197 MODULE_ALIAS_CRYPTO("sha1");
198