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 <asm/cpu_device_id.h>
9 #include <crypto/algapi.h>
10 #include <crypto/twofish.h>
11 #include <linux/crypto.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/types.h>
15
16 #include "twofish.h"
17 #include "ecb_cbc_helpers.h"
18
19 EXPORT_SYMBOL_GPL(__twofish_enc_blk_3way);
20 EXPORT_SYMBOL_GPL(twofish_dec_blk_3way);
21
twofish_setkey_skcipher(struct crypto_skcipher * tfm,const u8 * key,unsigned int keylen)22 static int twofish_setkey_skcipher(struct crypto_skcipher *tfm,
23 const u8 *key, unsigned int keylen)
24 {
25 return twofish_setkey(&tfm->base, key, keylen);
26 }
27
twofish_enc_blk_3way(const void * ctx,u8 * dst,const u8 * src)28 static inline void twofish_enc_blk_3way(const void *ctx, u8 *dst, const u8 *src)
29 {
30 __twofish_enc_blk_3way(ctx, dst, src, false);
31 }
32
twofish_dec_blk_cbc_3way(const void * ctx,u8 * dst,const u8 * src)33 void twofish_dec_blk_cbc_3way(const void *ctx, u8 *dst, const u8 *src)
34 {
35 u8 buf[2][TF_BLOCK_SIZE];
36 const u8 *s = src;
37
38 if (dst == src)
39 s = memcpy(buf, src, sizeof(buf));
40 twofish_dec_blk_3way(ctx, dst, src);
41 crypto_xor(dst + TF_BLOCK_SIZE, s, sizeof(buf));
42
43 }
44 EXPORT_SYMBOL_GPL(twofish_dec_blk_cbc_3way);
45
ecb_encrypt(struct skcipher_request * req)46 static int ecb_encrypt(struct skcipher_request *req)
47 {
48 ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
49 ECB_BLOCK(3, twofish_enc_blk_3way);
50 ECB_BLOCK(1, twofish_enc_blk);
51 ECB_WALK_END();
52 }
53
ecb_decrypt(struct skcipher_request * req)54 static int ecb_decrypt(struct skcipher_request *req)
55 {
56 ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
57 ECB_BLOCK(3, twofish_dec_blk_3way);
58 ECB_BLOCK(1, twofish_dec_blk);
59 ECB_WALK_END();
60 }
61
cbc_encrypt(struct skcipher_request * req)62 static int cbc_encrypt(struct skcipher_request *req)
63 {
64 CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
65 CBC_ENC_BLOCK(twofish_enc_blk);
66 CBC_WALK_END();
67 }
68
cbc_decrypt(struct skcipher_request * req)69 static int cbc_decrypt(struct skcipher_request *req)
70 {
71 CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
72 CBC_DEC_BLOCK(3, twofish_dec_blk_cbc_3way);
73 CBC_DEC_BLOCK(1, twofish_dec_blk);
74 CBC_WALK_END();
75 }
76
77 static struct skcipher_alg tf_skciphers[] = {
78 {
79 .base.cra_name = "ecb(twofish)",
80 .base.cra_driver_name = "ecb-twofish-3way",
81 .base.cra_priority = 300,
82 .base.cra_blocksize = TF_BLOCK_SIZE,
83 .base.cra_ctxsize = sizeof(struct twofish_ctx),
84 .base.cra_module = THIS_MODULE,
85 .min_keysize = TF_MIN_KEY_SIZE,
86 .max_keysize = TF_MAX_KEY_SIZE,
87 .setkey = twofish_setkey_skcipher,
88 .encrypt = ecb_encrypt,
89 .decrypt = ecb_decrypt,
90 }, {
91 .base.cra_name = "cbc(twofish)",
92 .base.cra_driver_name = "cbc-twofish-3way",
93 .base.cra_priority = 300,
94 .base.cra_blocksize = TF_BLOCK_SIZE,
95 .base.cra_ctxsize = sizeof(struct twofish_ctx),
96 .base.cra_module = THIS_MODULE,
97 .min_keysize = TF_MIN_KEY_SIZE,
98 .max_keysize = TF_MAX_KEY_SIZE,
99 .ivsize = TF_BLOCK_SIZE,
100 .setkey = twofish_setkey_skcipher,
101 .encrypt = cbc_encrypt,
102 .decrypt = cbc_decrypt,
103 },
104 };
105
is_blacklisted_cpu(void)106 static bool is_blacklisted_cpu(void)
107 {
108 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
109 return false;
110
111 switch (boot_cpu_data.x86_vfm) {
112 case INTEL_ATOM_BONNELL:
113 case INTEL_ATOM_BONNELL_MID:
114 case INTEL_ATOM_SALTWELL:
115 /*
116 * On Atom, twofish-3way is slower than original assembler
117 * implementation. Twofish-3way trades off some performance in
118 * storing blocks in 64bit registers to allow three blocks to
119 * be processed parallel. Parallel operation then allows gaining
120 * more performance than was trade off, on out-of-order CPUs.
121 * However Atom does not benefit from this parallelism and
122 * should be blacklisted.
123 */
124 return true;
125 }
126
127 if (boot_cpu_data.x86 == 0x0f) {
128 /*
129 * On Pentium 4, twofish-3way is slower than original assembler
130 * implementation because excessive uses of 64bit rotate and
131 * left-shifts (which are really slow on P4) needed to store and
132 * handle 128bit block in two 64bit registers.
133 */
134 return true;
135 }
136
137 return false;
138 }
139
140 static int force;
141 module_param(force, int, 0);
142 MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
143
twofish_3way_init(void)144 static int __init twofish_3way_init(void)
145 {
146 if (!force && is_blacklisted_cpu()) {
147 printk(KERN_INFO
148 "twofish-x86_64-3way: performance on this CPU "
149 "would be suboptimal: disabling "
150 "twofish-x86_64-3way.\n");
151 return -ENODEV;
152 }
153
154 return crypto_register_skciphers(tf_skciphers,
155 ARRAY_SIZE(tf_skciphers));
156 }
157
twofish_3way_fini(void)158 static void __exit twofish_3way_fini(void)
159 {
160 crypto_unregister_skciphers(tf_skciphers, ARRAY_SIZE(tf_skciphers));
161 }
162
163 module_init(twofish_3way_init);
164 module_exit(twofish_3way_fini);
165
166 MODULE_LICENSE("GPL");
167 MODULE_DESCRIPTION("Twofish Cipher Algorithm, 3-way parallel asm optimized");
168 MODULE_ALIAS_CRYPTO("twofish");
169 MODULE_ALIAS_CRYPTO("twofish-asm");
170