1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Glue Code for 3-way parallel assembler optimized version of Twofish
4 *
5 * Copyright (c) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
6 */
7
8 #include <crypto/algapi.h>
9 #include <crypto/twofish.h>
10 #include <linux/crypto.h>
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/types.h>
14
15 #include "twofish.h"
16 #include "ecb_cbc_helpers.h"
17
18 EXPORT_SYMBOL_GPL(__twofish_enc_blk_3way);
19 EXPORT_SYMBOL_GPL(twofish_dec_blk_3way);
20
twofish_setkey_skcipher(struct crypto_skcipher * tfm,const u8 * key,unsigned int keylen)21 static int twofish_setkey_skcipher(struct crypto_skcipher *tfm,
22 const u8 *key, unsigned int keylen)
23 {
24 return twofish_setkey(&tfm->base, key, keylen);
25 }
26
twofish_enc_blk_3way(const void * ctx,u8 * dst,const u8 * src)27 static inline void twofish_enc_blk_3way(const void *ctx, u8 *dst, const u8 *src)
28 {
29 __twofish_enc_blk_3way(ctx, dst, src, false);
30 }
31
twofish_dec_blk_cbc_3way(const void * ctx,u8 * dst,const u8 * src)32 void twofish_dec_blk_cbc_3way(const void *ctx, u8 *dst, const u8 *src)
33 {
34 u8 buf[2][TF_BLOCK_SIZE];
35 const u8 *s = src;
36
37 if (dst == src)
38 s = memcpy(buf, src, sizeof(buf));
39 twofish_dec_blk_3way(ctx, dst, src);
40 crypto_xor(dst + TF_BLOCK_SIZE, s, sizeof(buf));
41
42 }
43 EXPORT_SYMBOL_GPL(twofish_dec_blk_cbc_3way);
44
ecb_encrypt(struct skcipher_request * req)45 static int ecb_encrypt(struct skcipher_request *req)
46 {
47 ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
48 ECB_BLOCK(3, twofish_enc_blk_3way);
49 ECB_BLOCK(1, twofish_enc_blk);
50 ECB_WALK_END();
51 }
52
ecb_decrypt(struct skcipher_request * req)53 static int ecb_decrypt(struct skcipher_request *req)
54 {
55 ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
56 ECB_BLOCK(3, twofish_dec_blk_3way);
57 ECB_BLOCK(1, twofish_dec_blk);
58 ECB_WALK_END();
59 }
60
cbc_encrypt(struct skcipher_request * req)61 static int cbc_encrypt(struct skcipher_request *req)
62 {
63 CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
64 CBC_ENC_BLOCK(twofish_enc_blk);
65 CBC_WALK_END();
66 }
67
cbc_decrypt(struct skcipher_request * req)68 static int cbc_decrypt(struct skcipher_request *req)
69 {
70 CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
71 CBC_DEC_BLOCK(3, twofish_dec_blk_cbc_3way);
72 CBC_DEC_BLOCK(1, twofish_dec_blk);
73 CBC_WALK_END();
74 }
75
76 static struct skcipher_alg tf_skciphers[] = {
77 {
78 .base.cra_name = "ecb(twofish)",
79 .base.cra_driver_name = "ecb-twofish-3way",
80 .base.cra_priority = 300,
81 .base.cra_blocksize = TF_BLOCK_SIZE,
82 .base.cra_ctxsize = sizeof(struct twofish_ctx),
83 .base.cra_module = THIS_MODULE,
84 .min_keysize = TF_MIN_KEY_SIZE,
85 .max_keysize = TF_MAX_KEY_SIZE,
86 .setkey = twofish_setkey_skcipher,
87 .encrypt = ecb_encrypt,
88 .decrypt = ecb_decrypt,
89 }, {
90 .base.cra_name = "cbc(twofish)",
91 .base.cra_driver_name = "cbc-twofish-3way",
92 .base.cra_priority = 300,
93 .base.cra_blocksize = TF_BLOCK_SIZE,
94 .base.cra_ctxsize = sizeof(struct twofish_ctx),
95 .base.cra_module = THIS_MODULE,
96 .min_keysize = TF_MIN_KEY_SIZE,
97 .max_keysize = TF_MAX_KEY_SIZE,
98 .ivsize = TF_BLOCK_SIZE,
99 .setkey = twofish_setkey_skcipher,
100 .encrypt = cbc_encrypt,
101 .decrypt = cbc_decrypt,
102 },
103 };
104
is_blacklisted_cpu(void)105 static bool is_blacklisted_cpu(void)
106 {
107 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
108 return false;
109
110 if (boot_cpu_data.x86 == 0x06 &&
111 (boot_cpu_data.x86_model == 0x1c ||
112 boot_cpu_data.x86_model == 0x26 ||
113 boot_cpu_data.x86_model == 0x36)) {
114 /*
115 * On Atom, twofish-3way is slower than original assembler
116 * implementation. Twofish-3way trades off some performance in
117 * storing blocks in 64bit registers to allow three blocks to
118 * be processed parallel. Parallel operation then allows gaining
119 * more performance than was trade off, on out-of-order CPUs.
120 * However Atom does not benefit from this parallelism and
121 * should be blacklisted.
122 */
123 return true;
124 }
125
126 if (boot_cpu_data.x86 == 0x0f) {
127 /*
128 * On Pentium 4, twofish-3way is slower than original assembler
129 * implementation because excessive uses of 64bit rotate and
130 * left-shifts (which are really slow on P4) needed to store and
131 * handle 128bit block in two 64bit registers.
132 */
133 return true;
134 }
135
136 return false;
137 }
138
139 static int force;
140 module_param(force, int, 0);
141 MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
142
init(void)143 static int __init init(void)
144 {
145 if (!force && is_blacklisted_cpu()) {
146 printk(KERN_INFO
147 "twofish-x86_64-3way: performance on this CPU "
148 "would be suboptimal: disabling "
149 "twofish-x86_64-3way.\n");
150 return -ENODEV;
151 }
152
153 return crypto_register_skciphers(tf_skciphers,
154 ARRAY_SIZE(tf_skciphers));
155 }
156
fini(void)157 static void __exit fini(void)
158 {
159 crypto_unregister_skciphers(tf_skciphers, ARRAY_SIZE(tf_skciphers));
160 }
161
162 module_init(init);
163 module_exit(fini);
164
165 MODULE_LICENSE("GPL");
166 MODULE_DESCRIPTION("Twofish Cipher Algorithm, 3-way parallel asm optimized");
167 MODULE_ALIAS_CRYPTO("twofish");
168 MODULE_ALIAS_CRYPTO("twofish-asm");
169