1/*
2 * ARM NEON and scalar accelerated ChaCha and XChaCha stream ciphers,
3 * including ChaCha20 (RFC7539)
4 *
5 * Copyright (C) 2016 - 2017 Linaro, Ltd. <ard.biesheuvel@linaro.org>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * Based on:
12 * ChaCha20 256-bit cipher algorithm, RFC7539, SIMD glue code
13 *
14 * Copyright (C) 2015 Martin Willi
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License as published by
18 * the Free Software Foundation; either version 2 of the License, or
19 * (at your option) any later version.
20 */
21
22#include <crypto/algapi.h>
23#include <crypto/internal/chacha.h>
24#include <crypto/internal/simd.h>
25#include <crypto/internal/skcipher.h>
26#include <linux/jump_label.h>
27#include <linux/kernel.h>
28#include <linux/module.h>
29
30#include <asm/hwcap.h>
31#include <asm/neon.h>
32#include <asm/simd.h>
33
34asmlinkage void chacha_block_xor_neon(u32 *state, u8 *dst, const u8 *src,
35				      int nrounds);
36asmlinkage void chacha_4block_xor_neon(u32 *state, u8 *dst, const u8 *src,
37				       int nrounds, int bytes);
38asmlinkage void hchacha_block_neon(const u32 *state, u32 *out, int nrounds);
39
40static __ro_after_init DEFINE_STATIC_KEY_FALSE(have_neon);
41
42static void chacha_doneon(u32 *state, u8 *dst, const u8 *src,
43			  int bytes, int nrounds)
44{
45	while (bytes > 0) {
46		int l = min(bytes, CHACHA_BLOCK_SIZE * 5);
47
48		if (l <= CHACHA_BLOCK_SIZE) {
49			u8 buf[CHACHA_BLOCK_SIZE];
50
51			memcpy(buf, src, l);
52			chacha_block_xor_neon(state, buf, buf, nrounds);
53			memcpy(dst, buf, l);
54			state[12] += 1;
55			break;
56		}
57		chacha_4block_xor_neon(state, dst, src, nrounds, l);
58		bytes -= l;
59		src += l;
60		dst += l;
61		state[12] += DIV_ROUND_UP(l, CHACHA_BLOCK_SIZE);
62	}
63}
64
65void hchacha_block_arch(const u32 *state, u32 *stream, int nrounds)
66{
67	if (!static_branch_likely(&have_neon) || !crypto_simd_usable()) {
68		hchacha_block_generic(state, stream, nrounds);
69	} else {
70		kernel_neon_begin();
71		hchacha_block_neon(state, stream, nrounds);
72		kernel_neon_end();
73	}
74}
75EXPORT_SYMBOL(hchacha_block_arch);
76
77void chacha_init_arch(u32 *state, const u32 *key, const u8 *iv)
78{
79	chacha_init_generic(state, key, iv);
80}
81EXPORT_SYMBOL(chacha_init_arch);
82
83void chacha_crypt_arch(u32 *state, u8 *dst, const u8 *src, unsigned int bytes,
84		       int nrounds)
85{
86	if (!static_branch_likely(&have_neon) || bytes <= CHACHA_BLOCK_SIZE ||
87	    !crypto_simd_usable())
88		return chacha_crypt_generic(state, dst, src, bytes, nrounds);
89
90	do {
91		unsigned int todo = min_t(unsigned int, bytes, SZ_4K);
92
93		kernel_neon_begin();
94		chacha_doneon(state, dst, src, todo, nrounds);
95		kernel_neon_end();
96
97		bytes -= todo;
98		src += todo;
99		dst += todo;
100	} while (bytes);
101}
102EXPORT_SYMBOL(chacha_crypt_arch);
103
104static int chacha_neon_stream_xor(struct skcipher_request *req,
105				  const struct chacha_ctx *ctx, const u8 *iv)
106{
107	struct skcipher_walk walk;
108	u32 state[16];
109	int err;
110
111	err = skcipher_walk_virt(&walk, req, false);
112
113	chacha_init_generic(state, ctx->key, iv);
114
115	while (walk.nbytes > 0) {
116		unsigned int nbytes = walk.nbytes;
117
118		if (nbytes < walk.total)
119			nbytes = rounddown(nbytes, walk.stride);
120
121		if (!static_branch_likely(&have_neon) ||
122		    !crypto_simd_usable()) {
123			chacha_crypt_generic(state, walk.dst.virt.addr,
124					     walk.src.virt.addr, nbytes,
125					     ctx->nrounds);
126		} else {
127			kernel_neon_begin();
128			chacha_doneon(state, walk.dst.virt.addr,
129				      walk.src.virt.addr, nbytes, ctx->nrounds);
130			kernel_neon_end();
131		}
132		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
133	}
134
135	return err;
136}
137
138static int chacha_neon(struct skcipher_request *req)
139{
140	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
141	struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
142
143	return chacha_neon_stream_xor(req, ctx, req->iv);
144}
145
146static int xchacha_neon(struct skcipher_request *req)
147{
148	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
149	struct chacha_ctx *ctx = crypto_skcipher_ctx(tfm);
150	struct chacha_ctx subctx;
151	u32 state[16];
152	u8 real_iv[16];
153
154	chacha_init_generic(state, ctx->key, req->iv);
155	hchacha_block_arch(state, subctx.key, ctx->nrounds);
156	subctx.nrounds = ctx->nrounds;
157
158	memcpy(&real_iv[0], req->iv + 24, 8);
159	memcpy(&real_iv[8], req->iv + 16, 8);
160	return chacha_neon_stream_xor(req, &subctx, real_iv);
161}
162
163static struct skcipher_alg algs[] = {
164	{
165		.base.cra_name		= "chacha20",
166		.base.cra_driver_name	= "chacha20-neon",
167		.base.cra_priority	= 300,
168		.base.cra_blocksize	= 1,
169		.base.cra_ctxsize	= sizeof(struct chacha_ctx),
170		.base.cra_module	= THIS_MODULE,
171
172		.min_keysize		= CHACHA_KEY_SIZE,
173		.max_keysize		= CHACHA_KEY_SIZE,
174		.ivsize			= CHACHA_IV_SIZE,
175		.chunksize		= CHACHA_BLOCK_SIZE,
176		.walksize		= 5 * CHACHA_BLOCK_SIZE,
177		.setkey			= chacha20_setkey,
178		.encrypt		= chacha_neon,
179		.decrypt		= chacha_neon,
180	}, {
181		.base.cra_name		= "xchacha20",
182		.base.cra_driver_name	= "xchacha20-neon",
183		.base.cra_priority	= 300,
184		.base.cra_blocksize	= 1,
185		.base.cra_ctxsize	= sizeof(struct chacha_ctx),
186		.base.cra_module	= THIS_MODULE,
187
188		.min_keysize		= CHACHA_KEY_SIZE,
189		.max_keysize		= CHACHA_KEY_SIZE,
190		.ivsize			= XCHACHA_IV_SIZE,
191		.chunksize		= CHACHA_BLOCK_SIZE,
192		.walksize		= 5 * CHACHA_BLOCK_SIZE,
193		.setkey			= chacha20_setkey,
194		.encrypt		= xchacha_neon,
195		.decrypt		= xchacha_neon,
196	}, {
197		.base.cra_name		= "xchacha12",
198		.base.cra_driver_name	= "xchacha12-neon",
199		.base.cra_priority	= 300,
200		.base.cra_blocksize	= 1,
201		.base.cra_ctxsize	= sizeof(struct chacha_ctx),
202		.base.cra_module	= THIS_MODULE,
203
204		.min_keysize		= CHACHA_KEY_SIZE,
205		.max_keysize		= CHACHA_KEY_SIZE,
206		.ivsize			= XCHACHA_IV_SIZE,
207		.chunksize		= CHACHA_BLOCK_SIZE,
208		.walksize		= 5 * CHACHA_BLOCK_SIZE,
209		.setkey			= chacha12_setkey,
210		.encrypt		= xchacha_neon,
211		.decrypt		= xchacha_neon,
212	}
213};
214
215static int __init chacha_simd_mod_init(void)
216{
217	if (!cpu_have_named_feature(ASIMD))
218		return 0;
219
220	static_branch_enable(&have_neon);
221
222	return IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) ?
223		crypto_register_skciphers(algs, ARRAY_SIZE(algs)) : 0;
224}
225
226static void __exit chacha_simd_mod_fini(void)
227{
228	if (IS_REACHABLE(CONFIG_CRYPTO_SKCIPHER) && cpu_have_named_feature(ASIMD))
229		crypto_unregister_skciphers(algs, ARRAY_SIZE(algs));
230}
231
232module_init(chacha_simd_mod_init);
233module_exit(chacha_simd_mod_fini);
234
235MODULE_DESCRIPTION("ChaCha and XChaCha stream ciphers (NEON accelerated)");
236MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
237MODULE_LICENSE("GPL v2");
238MODULE_ALIAS_CRYPTO("chacha20");
239MODULE_ALIAS_CRYPTO("chacha20-neon");
240MODULE_ALIAS_CRYPTO("xchacha20");
241MODULE_ALIAS_CRYPTO("xchacha20-neon");
242MODULE_ALIAS_CRYPTO("xchacha12");
243MODULE_ALIAS_CRYPTO("xchacha12-neon");
244