tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * Cryptographic API.
3 *
4 * Glue code for the SHA1 Secure Hash Algorithm assembler implementation using
5 * Supplemental SSE3 instructions.
6 *
7 * This file is based on sha1_generic.c
8 *
9 * Copyright (c) Alan Smithee.
10 * Copyright (c) Andrew McDonald <andrew@mcdonald.org.uk>
11 * Copyright (c) Jean-Francois Dive <jef@linuxbe.org>
12 * Copyright (c) Mathias Krause <minipli@googlemail.com>
13 * Copyright (c) Chandramouli Narayanan <mouli@linux.intel.com>
14 *
15 * This program is free software; you can redistribute it and/or modify it
16 * under the terms of the GNU General Public License as published by the Free
17 * Software Foundation; either version 2 of the License, or (at your option)
18 * any later version.
19 *
20 */
21
22#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23
24#include <crypto/internal/hash.h>
25#include <linux/init.h>
26#include <linux/module.h>
27#include <linux/mm.h>
28#include <linux/cryptohash.h>
29#include <linux/types.h>
30#include <crypto/sha.h>
31#include <crypto/sha1_base.h>
32#include <asm/fpu/api.h>
33
34
35asmlinkage void sha1_transform_ssse3(u32 *digest, const char *data,
36 unsigned int rounds);
37#ifdef CONFIG_AS_AVX
38asmlinkage void sha1_transform_avx(u32 *digest, const char *data,
39 unsigned int rounds);
40#endif
41#ifdef CONFIG_AS_AVX2
42#define SHA1_AVX2_BLOCK_OPTSIZE 4 /* optimal 4*64 bytes of SHA1 blocks */
43
44asmlinkage void sha1_transform_avx2(u32 *digest, const char *data,
45 unsigned int rounds);
46#endif
47
48static void (*sha1_transform_asm)(u32 *, const char *, unsigned int);
49
50static int sha1_ssse3_update(struct shash_desc *desc, const u8 *data,
51 unsigned int len)
52{
53 struct sha1_state *sctx = shash_desc_ctx(desc);
54
55 if (!irq_fpu_usable() ||
56 (sctx->count % SHA1_BLOCK_SIZE) + len < SHA1_BLOCK_SIZE)
57 return crypto_sha1_update(desc, data, len);
58
59 /* make sure casting to sha1_block_fn() is safe */
60 BUILD_BUG_ON(offsetof(struct sha1_state, state) != 0);
61
62 kernel_fpu_begin();
63 sha1_base_do_update(desc, data, len,
64 (sha1_block_fn *)sha1_transform_asm);
65 kernel_fpu_end();
66
67 return 0;
68}
69
70static int sha1_ssse3_finup(struct shash_desc *desc, const u8 *data,
71 unsigned int len, u8 *out)
72{
73 if (!irq_fpu_usable())
74 return crypto_sha1_finup(desc, data, len, out);
75
76 kernel_fpu_begin();
77 if (len)
78 sha1_base_do_update(desc, data, len,
79 (sha1_block_fn *)sha1_transform_asm);
80 sha1_base_do_finalize(desc, (sha1_block_fn *)sha1_transform_asm);
81 kernel_fpu_end();
82
83 return sha1_base_finish(desc, out);
84}
85
86/* Add padding and return the message digest. */
87static int sha1_ssse3_final(struct shash_desc *desc, u8 *out)
88{
89 return sha1_ssse3_finup(desc, NULL, 0, out);
90}
91
92#ifdef CONFIG_AS_AVX2
93static void sha1_apply_transform_avx2(u32 *digest, const char *data,
94 unsigned int rounds)
95{
96 /* Select the optimal transform based on data block size */
97 if (rounds >= SHA1_AVX2_BLOCK_OPTSIZE)
98 sha1_transform_avx2(digest, data, rounds);
99 else
100 sha1_transform_avx(digest, data, rounds);
101}
102#endif
103
104static struct shash_alg alg = {
105 .digestsize = SHA1_DIGEST_SIZE,
106 .init = sha1_base_init,
107 .update = sha1_ssse3_update,
108 .final = sha1_ssse3_final,
109 .finup = sha1_ssse3_finup,
110 .descsize = sizeof(struct sha1_state),
111 .base = {
112 .cra_name = "sha1",
113 .cra_driver_name= "sha1-ssse3",
114 .cra_priority = 150,
115 .cra_flags = CRYPTO_ALG_TYPE_SHASH,
116 .cra_blocksize = SHA1_BLOCK_SIZE,
117 .cra_module = THIS_MODULE,
118 }
119};
120
121#ifdef CONFIG_AS_AVX
122static bool __init avx_usable(void)
123{
124 if (!cpu_has_xfeatures(XSTATE_SSE | XSTATE_YMM, NULL)) {
125 if (cpu_has_avx)
126 pr_info("AVX detected but unusable.\n");
127 return false;
128 }
129
130 return true;
131}
132
133#ifdef CONFIG_AS_AVX2
134static bool __init avx2_usable(void)
135{
136 if (avx_usable() && cpu_has_avx2 && boot_cpu_has(X86_FEATURE_BMI1) &&
137 boot_cpu_has(X86_FEATURE_BMI2))
138 return true;
139
140 return false;
141}
142#endif
143#endif
144
145static int __init sha1_ssse3_mod_init(void)
146{
147 char *algo_name;
148
149 /* test for SSSE3 first */
150 if (cpu_has_ssse3) {
151 sha1_transform_asm = sha1_transform_ssse3;
152 algo_name = "SSSE3";
153 }
154
155#ifdef CONFIG_AS_AVX
156 /* allow AVX to override SSSE3, it's a little faster */
157 if (avx_usable()) {
158 sha1_transform_asm = sha1_transform_avx;
159 algo_name = "AVX";
160#ifdef CONFIG_AS_AVX2
161 /* allow AVX2 to override AVX, it's a little faster */
162 if (avx2_usable()) {
163 sha1_transform_asm = sha1_apply_transform_avx2;
164 algo_name = "AVX2";
165 }
166#endif
167 }
168#endif
169
170 if (sha1_transform_asm) {
171 pr_info("Using %s optimized SHA-1 implementation\n", algo_name);
172 return crypto_register_shash(&alg);
173 }
174 pr_info("Neither AVX nor AVX2 nor SSSE3 is available/usable.\n");
175
176 return -ENODEV;
177}
178
179static void __exit sha1_ssse3_mod_fini(void)
180{
181 crypto_unregister_shash(&alg);
182}
183
184module_init(sha1_ssse3_mod_init);
185module_exit(sha1_ssse3_mod_fini);
186
187MODULE_LICENSE("GPL");
188MODULE_DESCRIPTION("SHA1 Secure Hash Algorithm, Supplemental SSE3 accelerated");
189
190MODULE_ALIAS_CRYPTO("sha1");