xref: /linux/lib/crypto/arm64/chacha-neon-glue.c (revision 13150742b09e720fdf021de14cd2b98b37415a89)
1 /*
2  * ChaCha and HChaCha functions (ARM64 optimized)
3  *
4  * Copyright (C) 2016 - 2017 Linaro, Ltd. <ard.biesheuvel@linaro.org>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  *
10  * Based on:
11  * ChaCha20 256-bit cipher algorithm, RFC7539, SIMD glue code
12  *
13  * Copyright (C) 2015 Martin Willi
14  *
15  * This program is free software; you can redistribute it and/or modify
16  * it under the terms of the GNU General Public License as published by
17  * the Free Software Foundation; either version 2 of the License, or
18  * (at your option) any later version.
19  */
20 
21 #include <crypto/chacha.h>
22 #include <crypto/internal/simd.h>
23 #include <linux/jump_label.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 
27 #include <asm/hwcap.h>
28 #include <asm/neon.h>
29 #include <asm/simd.h>
30 
31 asmlinkage void chacha_block_xor_neon(const struct chacha_state *state,
32 				      u8 *dst, const u8 *src, int nrounds);
33 asmlinkage void chacha_4block_xor_neon(const struct chacha_state *state,
34 				       u8 *dst, const u8 *src,
35 				       int nrounds, int bytes);
36 asmlinkage void hchacha_block_neon(const struct chacha_state *state,
37 				   u32 out[HCHACHA_OUT_WORDS], int nrounds);
38 
39 static __ro_after_init DEFINE_STATIC_KEY_FALSE(have_neon);
40 
chacha_doneon(struct chacha_state * state,u8 * dst,const u8 * src,int bytes,int nrounds)41 static void chacha_doneon(struct chacha_state *state, u8 *dst, const u8 *src,
42 			  int bytes, int nrounds)
43 {
44 	while (bytes > 0) {
45 		int l = min(bytes, CHACHA_BLOCK_SIZE * 5);
46 
47 		if (l <= CHACHA_BLOCK_SIZE) {
48 			u8 buf[CHACHA_BLOCK_SIZE];
49 
50 			memcpy(buf, src, l);
51 			chacha_block_xor_neon(state, buf, buf, nrounds);
52 			memcpy(dst, buf, l);
53 			state->x[12] += 1;
54 			break;
55 		}
56 		chacha_4block_xor_neon(state, dst, src, nrounds, l);
57 		bytes -= l;
58 		src += l;
59 		dst += l;
60 		state->x[12] += DIV_ROUND_UP(l, CHACHA_BLOCK_SIZE);
61 	}
62 }
63 
hchacha_block_arch(const struct chacha_state * state,u32 out[HCHACHA_OUT_WORDS],int nrounds)64 void hchacha_block_arch(const struct chacha_state *state,
65 			u32 out[HCHACHA_OUT_WORDS], int nrounds)
66 {
67 	if (!static_branch_likely(&have_neon) || !crypto_simd_usable()) {
68 		hchacha_block_generic(state, out, nrounds);
69 	} else {
70 		kernel_neon_begin();
71 		hchacha_block_neon(state, out, nrounds);
72 		kernel_neon_end();
73 	}
74 }
75 EXPORT_SYMBOL(hchacha_block_arch);
76 
chacha_crypt_arch(struct chacha_state * state,u8 * dst,const u8 * src,unsigned int bytes,int nrounds)77 void chacha_crypt_arch(struct chacha_state *state, u8 *dst, const u8 *src,
78 		       unsigned int bytes, int nrounds)
79 {
80 	if (!static_branch_likely(&have_neon) || bytes <= CHACHA_BLOCK_SIZE ||
81 	    !crypto_simd_usable())
82 		return chacha_crypt_generic(state, dst, src, bytes, nrounds);
83 
84 	do {
85 		unsigned int todo = min_t(unsigned int, bytes, SZ_4K);
86 
87 		kernel_neon_begin();
88 		chacha_doneon(state, dst, src, todo, nrounds);
89 		kernel_neon_end();
90 
91 		bytes -= todo;
92 		src += todo;
93 		dst += todo;
94 	} while (bytes);
95 }
96 EXPORT_SYMBOL(chacha_crypt_arch);
97 
chacha_is_arch_optimized(void)98 bool chacha_is_arch_optimized(void)
99 {
100 	return static_key_enabled(&have_neon);
101 }
102 EXPORT_SYMBOL(chacha_is_arch_optimized);
103 
chacha_simd_mod_init(void)104 static int __init chacha_simd_mod_init(void)
105 {
106 	if (cpu_have_named_feature(ASIMD))
107 		static_branch_enable(&have_neon);
108 	return 0;
109 }
110 subsys_initcall(chacha_simd_mod_init);
111 
chacha_simd_mod_exit(void)112 static void __exit chacha_simd_mod_exit(void)
113 {
114 }
115 module_exit(chacha_simd_mod_exit);
116 
117 MODULE_DESCRIPTION("ChaCha and HChaCha functions (ARM64 optimized)");
118 MODULE_AUTHOR("Ard Biesheuvel <ard.biesheuvel@linaro.org>");
119 MODULE_LICENSE("GPL v2");
120