Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
crypto: arm64/sm4 - simplify sm4_ce_expand_key() of CE implementation
Use a 128-bit swap mask and tbl instruction to simplify the implementation
for generating SM4 rkey_dec.
Also fixed the issue of not being wrapped by kernel_neon_begin/end() when
using the sm4_ce_expand_key() function.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
authored by
Tianjia Zhang
and committed by
Herbert Xu
cb9ba02b
ce41fefd
+24
-24
+22
-24
arch/arm64/crypto/sm4-ce-core.S
+22
-24
arch/arm64/crypto/sm4-ce-core.S
···
65
65
sm4ekey v6.4s, v5.4s, v30.4s;
66
66
sm4ekey v7.4s, v6.4s, v31.4s;
67
67
68
+
adr_l x5, .Lbswap128_mask
69
+
ld1 {v24.16b}, [x5]
70
+
68
71
st1 {v0.16b-v3.16b}, [x1], #64;
69
72
st1 {v4.16b-v7.16b}, [x1];
70
-
rev64 v7.4s, v7.4s;
71
-
rev64 v6.4s, v6.4s;
72
-
rev64 v5.4s, v5.4s;
73
-
rev64 v4.4s, v4.4s;
74
-
rev64 v3.4s, v3.4s;
75
-
rev64 v2.4s, v2.4s;
76
-
rev64 v1.4s, v1.4s;
77
-
rev64 v0.4s, v0.4s;
78
-
ext v7.16b, v7.16b, v7.16b, #8;
79
-
ext v6.16b, v6.16b, v6.16b, #8;
80
-
ext v5.16b, v5.16b, v5.16b, #8;
81
-
ext v4.16b, v4.16b, v4.16b, #8;
82
-
ext v3.16b, v3.16b, v3.16b, #8;
83
-
ext v2.16b, v2.16b, v2.16b, #8;
84
-
ext v1.16b, v1.16b, v1.16b, #8;
85
-
ext v0.16b, v0.16b, v0.16b, #8;
86
-
st1 {v7.16b}, [x2], #16;
87
-
st1 {v6.16b}, [x2], #16;
88
-
st1 {v5.16b}, [x2], #16;
89
-
st1 {v4.16b}, [x2], #16;
90
-
st1 {v3.16b}, [x2], #16;
91
-
st1 {v2.16b}, [x2], #16;
92
-
st1 {v1.16b}, [x2], #16;
93
-
st1 {v0.16b}, [x2];
73
+
74
+
tbl v16.16b, {v7.16b}, v24.16b
75
+
tbl v17.16b, {v6.16b}, v24.16b
76
+
tbl v18.16b, {v5.16b}, v24.16b
77
+
tbl v19.16b, {v4.16b}, v24.16b
78
+
tbl v20.16b, {v3.16b}, v24.16b
79
+
tbl v21.16b, {v2.16b}, v24.16b
80
+
tbl v22.16b, {v1.16b}, v24.16b
81
+
tbl v23.16b, {v0.16b}, v24.16b
82
+
83
+
st1 {v16.16b-v19.16b}, [x2], #64
84
+
st1 {v20.16b-v23.16b}, [x2]
94
85
95
86
ret;
96
87
SYM_FUNC_END(sm4_ce_expand_key)
···
569
578
570
579
ret
571
580
SYM_FUNC_END(sm4_ce_ctr_enc)
581
+
582
+
583
+
.section ".rodata", "a"
584
+
.align 4
585
+
.Lbswap128_mask:
586
+
.byte 0x0c, 0x0d, 0x0e, 0x0f, 0x08, 0x09, 0x0a, 0x0b
587
+
.byte 0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02, 0x03