1
2
3
4
5
6
7
8
9
10
11
12#include <crypto/algapi.h>
13#include <crypto/chacha20.h>
14#include <linux/crypto.h>
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <asm/fpu/api.h>
18#include <asm/simd.h>
19
20#define CHACHA20_STATE_ALIGN 16
21
22asmlinkage void chacha20_block_xor_ssse3(u32 *state, u8 *dst, const u8 *src);
23asmlinkage void chacha20_4block_xor_ssse3(u32 *state, u8 *dst, const u8 *src);
24#ifdef CONFIG_AS_AVX2
25asmlinkage void chacha20_8block_xor_avx2(u32 *state, u8 *dst, const u8 *src);
26static bool chacha20_use_avx2;
27#endif
28
29static void chacha20_dosimd(u32 *state, u8 *dst, const u8 *src,
30 unsigned int bytes)
31{
32 u8 buf[CHACHA20_BLOCK_SIZE];
33
34#ifdef CONFIG_AS_AVX2
35 if (chacha20_use_avx2) {
36 while (bytes >= CHACHA20_BLOCK_SIZE * 8) {
37 chacha20_8block_xor_avx2(state, dst, src);
38 bytes -= CHACHA20_BLOCK_SIZE * 8;
39 src += CHACHA20_BLOCK_SIZE * 8;
40 dst += CHACHA20_BLOCK_SIZE * 8;
41 state[12] += 8;
42 }
43 }
44#endif
45 while (bytes >= CHACHA20_BLOCK_SIZE * 4) {
46 chacha20_4block_xor_ssse3(state, dst, src);
47 bytes -= CHACHA20_BLOCK_SIZE * 4;
48 src += CHACHA20_BLOCK_SIZE * 4;
49 dst += CHACHA20_BLOCK_SIZE * 4;
50 state[12] += 4;
51 }
52 while (bytes >= CHACHA20_BLOCK_SIZE) {
53 chacha20_block_xor_ssse3(state, dst, src);
54 bytes -= CHACHA20_BLOCK_SIZE;
55 src += CHACHA20_BLOCK_SIZE;
56 dst += CHACHA20_BLOCK_SIZE;
57 state[12]++;
58 }
59 if (bytes) {
60 memcpy(buf, src, bytes);
61 chacha20_block_xor_ssse3(state, buf, buf);
62 memcpy(dst, buf, bytes);
63 }
64}
65
66static int chacha20_simd(struct blkcipher_desc *desc, struct scatterlist *dst,
67 struct scatterlist *src, unsigned int nbytes)
68{
69 u32 *state, state_buf[16 + (CHACHA20_STATE_ALIGN / sizeof(u32)) - 1];
70 struct blkcipher_walk walk;
71 int err;
72
73 if (!may_use_simd())
74 return crypto_chacha20_crypt(desc, dst, src, nbytes);
75
76 state = (u32 *)roundup((uintptr_t)state_buf, CHACHA20_STATE_ALIGN);
77
78 blkcipher_walk_init(&walk, dst, src, nbytes);
79 err = blkcipher_walk_virt_block(desc, &walk, CHACHA20_BLOCK_SIZE);
80
81 crypto_chacha20_init(state, crypto_blkcipher_ctx(desc->tfm), walk.iv);
82
83 kernel_fpu_begin();
84
85 while (walk.nbytes >= CHACHA20_BLOCK_SIZE) {
86 chacha20_dosimd(state, walk.dst.virt.addr, walk.src.virt.addr,
87 rounddown(walk.nbytes, CHACHA20_BLOCK_SIZE));
88 err = blkcipher_walk_done(desc, &walk,
89 walk.nbytes % CHACHA20_BLOCK_SIZE);
90 }
91
92 if (walk.nbytes) {
93 chacha20_dosimd(state, walk.dst.virt.addr, walk.src.virt.addr,
94 walk.nbytes);
95 err = blkcipher_walk_done(desc, &walk, 0);
96 }
97
98 kernel_fpu_end();
99
100 return err;
101}
102
103static struct crypto_alg alg = {
104 .cra_name = "chacha20",
105 .cra_driver_name = "chacha20-simd",
106 .cra_priority = 300,
107 .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
108 .cra_blocksize = 1,
109 .cra_type = &crypto_blkcipher_type,
110 .cra_ctxsize = sizeof(struct chacha20_ctx),
111 .cra_alignmask = sizeof(u32) - 1,
112 .cra_module = THIS_MODULE,
113 .cra_u = {
114 .blkcipher = {
115 .min_keysize = CHACHA20_KEY_SIZE,
116 .max_keysize = CHACHA20_KEY_SIZE,
117 .ivsize = CHACHA20_IV_SIZE,
118 .geniv = "seqiv",
119 .setkey = crypto_chacha20_setkey,
120 .encrypt = chacha20_simd,
121 .decrypt = chacha20_simd,
122 },
123 },
124};
125
126static int __init chacha20_simd_mod_init(void)
127{
128 if (!boot_cpu_has(X86_FEATURE_SSSE3))
129 return -ENODEV;
130
131#ifdef CONFIG_AS_AVX2
132 chacha20_use_avx2 = cpu_has_avx && cpu_has_avx2 &&
133 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL);
134#endif
135 return crypto_register_alg(&alg);
136}
137
138static void __exit chacha20_simd_mod_fini(void)
139{
140 crypto_unregister_alg(&alg);
141}
142
143module_init(chacha20_simd_mod_init);
144module_exit(chacha20_simd_mod_fini);
145
146MODULE_LICENSE("GPL");
147MODULE_AUTHOR("Martin Willi <martin@strongswan.org>");
148MODULE_DESCRIPTION("chacha20 cipher algorithm, SIMD accelerated");
149MODULE_ALIAS_CRYPTO("chacha20");
150MODULE_ALIAS_CRYPTO("chacha20-simd");
151
