/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
 * SM4 Cipher Algorithm, AES-NI/AVX optimized.
 * as specified in
 * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
 *
 * Copyright (C) 2018 Markku-Juhani O. Saarinen <mjos@iki.fi>
 * Copyright (C) 2020 Jussi Kivilinna <jussi.kivilinna@iki.fi>
 * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
 */

/* Based on SM4 AES-NI work by libgcrypt and Markku-Juhani O. Saarinen at:
 *  https://github.com/mjosaarinen/sm4ni
 */

#include <linux/linkage.h>
#include <asm/frame.h>

#define rRIP         (%rip)

#define RX0          %xmm0
#define RX1          %xmm1
#define MASK_4BIT    %xmm2
#define RTMP0        %xmm3
#define RTMP1        %xmm4
#define RTMP2        %xmm5
#define RTMP3        %xmm6
#define RTMP4        %xmm7

#define RA0          %xmm8
#define RA1          %xmm9
#define RA2          %xmm10
#define RA3          %xmm11

#define RB0          %xmm12
#define RB1          %xmm13
#define RB2          %xmm14
#define RB3          %xmm15

#define RNOT         %xmm0
#define RBSWAP       %xmm1


/* Transpose four 32-bit words between 128-bit vectors. */
#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
	vpunpckhdq x1, x0, t2;                \
	vpunpckldq x1, x0, x0;                \
	                                      \
	vpunpckldq x3, x2, t1;                \
	vpunpckhdq x3, x2, x2;                \
	                                      \
	vpunpckhqdq t1, x0, x1;               \
	vpunpcklqdq t1, x0, x0;               \
	                                      \
	vpunpckhqdq x2, t2, x3;               \
	vpunpcklqdq x2, t2, x2;

/* pre-SubByte transform. */
#define transform_pre(x, lo_t, hi_t, mask4bit, tmp0) \
	vpand x, mask4bit, tmp0;                     \
	vpandn x, mask4bit, x;                       \
	vpsrld $4, x, x;                             \
	                                             \
	vpshufb tmp0, lo_t, tmp0;                    \
	vpshufb x, hi_t, x;                          \
	vpxor tmp0, x, x;

/* post-SubByte transform. Note: x has been XOR'ed with mask4bit by
 * 'vaeslastenc' instruction.
 */
#define transform_post(x, lo_t, hi_t, mask4bit, tmp0) \
	vpandn mask4bit, x, tmp0;                     \
	vpsrld $4, x, x;                              \
	vpand x, mask4bit, x;                         \
	                                              \
	vpshufb tmp0, lo_t, tmp0;                     \
	vpshufb x, hi_t, x;                           \
	vpxor tmp0, x, x;


.section	.rodata.cst16, "aM", @progbits, 16
.align 16

/*
 * Following four affine transform look-up tables are from work by
 * Markku-Juhani O. Saarinen, at https://github.com/mjosaarinen/sm4ni
 *
 * These allow exposing SM4 S-Box from AES SubByte.
 */

/* pre-SubByte affine transform, from SM4 field to AES field. */
.Lpre_tf_lo_s:
	.quad 0x9197E2E474720701, 0xC7C1B4B222245157
.Lpre_tf_hi_s:
	.quad 0xE240AB09EB49A200, 0xF052B91BF95BB012

/* post-SubByte affine transform, from AES field to SM4 field. */
.Lpost_tf_lo_s:
	.quad 0x5B67F2CEA19D0834, 0xEDD14478172BBE82
.Lpost_tf_hi_s:
	.quad 0xAE7201DD73AFDC00, 0x11CDBE62CC1063BF

/* For isolating SubBytes from AESENCLAST, inverse shift row */
.Linv_shift_row:
	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03

/* Inverse shift row + Rotate left by 8 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_8:
	.byte 0x07, 0x00, 0x0d, 0x0a, 0x0b, 0x04, 0x01, 0x0e
	.byte 0x0f, 0x08, 0x05, 0x02, 0x03, 0x0c, 0x09, 0x06

/* Inverse shift row + Rotate left by 16 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_16:
	.byte 0x0a, 0x07, 0x00, 0x0d, 0x0e, 0x0b, 0x04, 0x01
	.byte 0x02, 0x0f, 0x08, 0x05, 0x06, 0x03, 0x0c, 0x09

/* Inverse shift row + Rotate left by 24 bits on 32-bit words with vpshufb */
.Linv_shift_row_rol_24:
	.byte 0x0d, 0x0a, 0x07, 0x00, 0x01, 0x0e, 0x0b, 0x04
	.byte 0x05, 0x02, 0x0f, 0x08, 0x09, 0x06, 0x03, 0x0c

/* For CTR-mode IV byteswap */
.Lbswap128_mask:
	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0

/* For input word byte-swap */
.Lbswap32_mask:
	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12

.align 4
/* 4-bit mask */
.L0f0f0f0f:
	.long 0x0f0f0f0f

/* 12 bytes, only for padding */
.Lpadding_deadbeef:
	.long 0xdeadbeef, 0xdeadbeef, 0xdeadbeef


.text
.align 16

/*
 * void sm4_aesni_avx_crypt4(const u32 *rk, u8 *dst,
 *                           const u8 *src, int nblocks)
 */
.align 8
SYM_FUNC_START(sm4_aesni_avx_crypt4)
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (1..4 blocks)
	 *	%rdx: src (1..4 blocks)
	 *	%rcx: num blocks (1..4)
	 */
	FRAME_BEGIN

	vmovdqu 0*16(%rdx), RA0;
	vmovdqa RA0, RA1;
	vmovdqa RA0, RA2;
	vmovdqa RA0, RA3;
	cmpq $2, %rcx;
	jb .Lblk4_load_input_done;
	vmovdqu 1*16(%rdx), RA1;
	je .Lblk4_load_input_done;
	vmovdqu 2*16(%rdx), RA2;
	cmpq $3, %rcx;
	je .Lblk4_load_input_done;
	vmovdqu 3*16(%rdx), RA3;

.Lblk4_load_input_done:

	vmovdqa .Lbswap32_mask rRIP, RTMP2;
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;

	vbroadcastss .L0f0f0f0f rRIP, MASK_4BIT;
	vmovdqa .Lpre_tf_lo_s rRIP, RTMP4;
	vmovdqa .Lpre_tf_hi_s rRIP, RB0;
	vmovdqa .Lpost_tf_lo_s rRIP, RB1;
	vmovdqa .Lpost_tf_hi_s rRIP, RB2;
	vmovdqa .Linv_shift_row rRIP, RB3;
	vmovdqa .Linv_shift_row_rol_8 rRIP, RTMP2;
	vmovdqa .Linv_shift_row_rol_16 rRIP, RTMP3;
	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);

#define ROUND(round, s0, s1, s2, s3)                                \
	vbroadcastss (4*(round))(%rdi), RX0;                        \
	vpxor s1, RX0, RX0;                                         \
	vpxor s2, RX0, RX0;                                         \
	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
	                                                            \
	/* sbox, non-linear part */                                 \
	transform_pre(RX0, RTMP4, RB0, MASK_4BIT, RTMP0);           \
	vaesenclast MASK_4BIT, RX0, RX0;                            \
	transform_post(RX0, RB1, RB2, MASK_4BIT, RTMP0);            \
	                                                            \
	/* linear part */                                           \
	vpshufb RB3, RX0, RTMP0;                                    \
	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
	vpshufb RTMP2, RX0, RTMP1;                                  \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
	vpshufb RTMP3, RX0, RTMP1;                                  \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
	vpshufb .Linv_shift_row_rol_24 rRIP, RX0, RTMP1;            \
	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
	vpslld $2, RTMP0, RTMP1;                                    \
	vpsrld $30, RTMP0, RTMP0;                                   \
	vpxor RTMP0, s0, s0;                                        \
	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
	vpxor RTMP1, s0, s0;

	leaq (32*4)(%rdi), %rax;
.align 16
.Lroundloop_blk4:
	ROUND(0, RA0, RA1, RA2, RA3);
	ROUND(1, RA1, RA2, RA3, RA0);
	ROUND(2, RA2, RA3, RA0, RA1);
	ROUND(3, RA3, RA0, RA1, RA2);
	leaq (4*4)(%rdi), %rdi;
	cmpq %rax, %rdi;
	jne .Lroundloop_blk4;

#undef ROUND

	vmovdqa .Lbswap128_mask rRIP, RTMP2;

	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;

	vmovdqu RA0, 0*16(%rsi);
	cmpq $2, %rcx;
	jb .Lblk4_store_output_done;
	vmovdqu RA1, 1*16(%rsi);
	je .Lblk4_store_output_done;
	vmovdqu RA2, 2*16(%rsi);
	cmpq $3, %rcx;
	je .Lblk4_store_output_done;
	vmovdqu RA3, 3*16(%rsi);

.Lblk4_store_output_done:
	vzeroall;
	FRAME_END
	RET;
SYM_FUNC_END(sm4_aesni_avx_crypt4)

.align 8
SYM_FUNC_START_LOCAL(__sm4_crypt_blk8)
	/* input:
	 *	%rdi: round key array, CTX
	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: eight parallel
	 *						plaintext blocks
	 * output:
	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: eight parallel
	 * 						ciphertext blocks
	 */
	FRAME_BEGIN

	vmovdqa .Lbswap32_mask rRIP, RTMP2;
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;
	vpshufb RTMP2, RB0, RB0;
	vpshufb RTMP2, RB1, RB1;
	vpshufb RTMP2, RB2, RB2;
	vpshufb RTMP2, RB3, RB3;

	vbroadcastss .L0f0f0f0f rRIP, MASK_4BIT;
	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);

#define ROUND(round, s0, s1, s2, s3, r0, r1, r2, r3)                \
	vbroadcastss (4*(round))(%rdi), RX0;                        \
	vmovdqa .Lpre_tf_lo_s rRIP, RTMP4;                          \
	vmovdqa .Lpre_tf_hi_s rRIP, RTMP1;                          \
	vmovdqa RX0, RX1;                                           \
	vpxor s1, RX0, RX0;                                         \
	vpxor s2, RX0, RX0;                                         \
	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
	vmovdqa .Lpost_tf_lo_s rRIP, RTMP2;                         \
	vmovdqa .Lpost_tf_hi_s rRIP, RTMP3;                         \
	vpxor r1, RX1, RX1;                                         \
	vpxor r2, RX1, RX1;                                         \
	vpxor r3, RX1, RX1; /* r1 ^ r2 ^ r3 ^ rk */                 \
                                                                    \
	/* sbox, non-linear part */                                 \
	transform_pre(RX0, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
	transform_pre(RX1, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
	vmovdqa .Linv_shift_row rRIP, RTMP4;                        \
	vaesenclast MASK_4BIT, RX0, RX0;                            \
	vaesenclast MASK_4BIT, RX1, RX1;                            \
	transform_post(RX0, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
	transform_post(RX1, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
                                                                    \
	/* linear part */                                           \
	vpshufb RTMP4, RX0, RTMP0;                                  \
	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
	vpshufb RTMP4, RX1, RTMP2;                                  \
	vmovdqa .Linv_shift_row_rol_8 rRIP, RTMP4;                  \
	vpxor RTMP2, r0, r0; /* r0 ^ x */                           \
	vpshufb RTMP4, RX0, RTMP1;                                  \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
	vpshufb RTMP4, RX1, RTMP3;                                  \
	vmovdqa .Linv_shift_row_rol_16 rRIP, RTMP4;                 \
	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) */               \
	vpshufb RTMP4, RX0, RTMP1;                                  \
	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
	vpshufb RTMP4, RX1, RTMP3;                                  \
	vmovdqa .Linv_shift_row_rol_24 rRIP, RTMP4;                 \
	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) ^ rol(x,16) */   \
	vpshufb RTMP4, RX0, RTMP1;                                  \
	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
	vpslld $2, RTMP0, RTMP1;                                    \
	vpsrld $30, RTMP0, RTMP0;                                   \
	vpxor RTMP0, s0, s0;                                        \
	vpxor RTMP1, s0, s0;                                        \
	vpshufb RTMP4, RX1, RTMP3;                                  \
	vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,24) */               \
	/* r0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
	vpslld $2, RTMP2, RTMP3;                                    \
	vpsrld $30, RTMP2, RTMP2;                                   \
	vpxor RTMP2, r0, r0;                                        \
	vpxor RTMP3, r0, r0;

	leaq (32*4)(%rdi), %rax;
.align 16
.Lroundloop_blk8:
	ROUND(0, RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3);
	ROUND(1, RA1, RA2, RA3, RA0, RB1, RB2, RB3, RB0);
	ROUND(2, RA2, RA3, RA0, RA1, RB2, RB3, RB0, RB1);
	ROUND(3, RA3, RA0, RA1, RA2, RB3, RB0, RB1, RB2);
	leaq (4*4)(%rdi), %rdi;
	cmpq %rax, %rdi;
	jne .Lroundloop_blk8;

#undef ROUND

	vmovdqa .Lbswap128_mask rRIP, RTMP2;

	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
	vpshufb RTMP2, RA0, RA0;
	vpshufb RTMP2, RA1, RA1;
	vpshufb RTMP2, RA2, RA2;
	vpshufb RTMP2, RA3, RA3;
	vpshufb RTMP2, RB0, RB0;
	vpshufb RTMP2, RB1, RB1;
	vpshufb RTMP2, RB2, RB2;
	vpshufb RTMP2, RB3, RB3;

	FRAME_END
	RET;
SYM_FUNC_END(__sm4_crypt_blk8)

/*
 * void sm4_aesni_avx_crypt8(const u32 *rk, u8 *dst,
 *                           const u8 *src, int nblocks)
 */
.align 8
SYM_FUNC_START(sm4_aesni_avx_crypt8)
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (1..8 blocks)
	 *	%rdx: src (1..8 blocks)
	 *	%rcx: num blocks (1..8)
	 */
	cmpq $5, %rcx;
	jb sm4_aesni_avx_crypt4;

	FRAME_BEGIN

	vmovdqu (0 * 16)(%rdx), RA0;
	vmovdqu (1 * 16)(%rdx), RA1;
	vmovdqu (2 * 16)(%rdx), RA2;
	vmovdqu (3 * 16)(%rdx), RA3;
	vmovdqu (4 * 16)(%rdx), RB0;
	vmovdqa RB0, RB1;
	vmovdqa RB0, RB2;
	vmovdqa RB0, RB3;
	je .Lblk8_load_input_done;
	vmovdqu (5 * 16)(%rdx), RB1;
	cmpq $7, %rcx;
	jb .Lblk8_load_input_done;
	vmovdqu (6 * 16)(%rdx), RB2;
	je .Lblk8_load_input_done;
	vmovdqu (7 * 16)(%rdx), RB3;

.Lblk8_load_input_done:
	call __sm4_crypt_blk8;

	cmpq $6, %rcx;
	vmovdqu RA0, (0 * 16)(%rsi);
	vmovdqu RA1, (1 * 16)(%rsi);
	vmovdqu RA2, (2 * 16)(%rsi);
	vmovdqu RA3, (3 * 16)(%rsi);
	vmovdqu RB0, (4 * 16)(%rsi);
	jb .Lblk8_store_output_done;
	vmovdqu RB1, (5 * 16)(%rsi);
	je .Lblk8_store_output_done;
	vmovdqu RB2, (6 * 16)(%rsi);
	cmpq $7, %rcx;
	je .Lblk8_store_output_done;
	vmovdqu RB3, (7 * 16)(%rsi);

.Lblk8_store_output_done:
	vzeroall;
	FRAME_END
	RET;
SYM_FUNC_END(sm4_aesni_avx_crypt8)

/*
 * void sm4_aesni_avx_ctr_enc_blk8(const u32 *rk, u8 *dst,
 *                                 const u8 *src, u8 *iv)
 */
.align 8
SYM_FUNC_START(sm4_aesni_avx_ctr_enc_blk8)
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (8 blocks)
	 *	%rdx: src (8 blocks)
	 *	%rcx: iv (big endian, 128bit)
	 */
	FRAME_BEGIN

	/* load IV and byteswap */
	vmovdqu (%rcx), RA0;

	vmovdqa .Lbswap128_mask rRIP, RBSWAP;
	vpshufb RBSWAP, RA0, RTMP0; /* be => le */

	vpcmpeqd RNOT, RNOT, RNOT;
	vpsrldq $8, RNOT, RNOT; /* low: -1, high: 0 */

#define inc_le128(x, minus_one, tmp) \
	vpcmpeqq minus_one, x, tmp;  \
	vpsubq minus_one, x, x;      \
	vpslldq $8, tmp, tmp;        \
	vpsubq tmp, x, x;

	/* construct IVs */
	inc_le128(RTMP0, RNOT, RTMP2); /* +1 */
	vpshufb RBSWAP, RTMP0, RA1;
	inc_le128(RTMP0, RNOT, RTMP2); /* +2 */
	vpshufb RBSWAP, RTMP0, RA2;
	inc_le128(RTMP0, RNOT, RTMP2); /* +3 */
	vpshufb RBSWAP, RTMP0, RA3;
	inc_le128(RTMP0, RNOT, RTMP2); /* +4 */
	vpshufb RBSWAP, RTMP0, RB0;
	inc_le128(RTMP0, RNOT, RTMP2); /* +5 */
	vpshufb RBSWAP, RTMP0, RB1;
	inc_le128(RTMP0, RNOT, RTMP2); /* +6 */
	vpshufb RBSWAP, RTMP0, RB2;
	inc_le128(RTMP0, RNOT, RTMP2); /* +7 */
	vpshufb RBSWAP, RTMP0, RB3;
	inc_le128(RTMP0, RNOT, RTMP2); /* +8 */
	vpshufb RBSWAP, RTMP0, RTMP1;

	/* store new IV */
	vmovdqu RTMP1, (%rcx);

	call __sm4_crypt_blk8;

	vpxor (0 * 16)(%rdx), RA0, RA0;
	vpxor (1 * 16)(%rdx), RA1, RA1;
	vpxor (2 * 16)(%rdx), RA2, RA2;
	vpxor (3 * 16)(%rdx), RA3, RA3;
	vpxor (4 * 16)(%rdx), RB0, RB0;
	vpxor (5 * 16)(%rdx), RB1, RB1;
	vpxor (6 * 16)(%rdx), RB2, RB2;
	vpxor (7 * 16)(%rdx), RB3, RB3;

	vmovdqu RA0, (0 * 16)(%rsi);
	vmovdqu RA1, (1 * 16)(%rsi);
	vmovdqu RA2, (2 * 16)(%rsi);
	vmovdqu RA3, (3 * 16)(%rsi);
	vmovdqu RB0, (4 * 16)(%rsi);
	vmovdqu RB1, (5 * 16)(%rsi);
	vmovdqu RB2, (6 * 16)(%rsi);
	vmovdqu RB3, (7 * 16)(%rsi);

	vzeroall;
	FRAME_END
	RET;
SYM_FUNC_END(sm4_aesni_avx_ctr_enc_blk8)

/*
 * void sm4_aesni_avx_cbc_dec_blk8(const u32 *rk, u8 *dst,
 *                                 const u8 *src, u8 *iv)
 */
.align 8
SYM_FUNC_START(sm4_aesni_avx_cbc_dec_blk8)
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (8 blocks)
	 *	%rdx: src (8 blocks)
	 *	%rcx: iv
	 */
	FRAME_BEGIN

	vmovdqu (0 * 16)(%rdx), RA0;
	vmovdqu (1 * 16)(%rdx), RA1;
	vmovdqu (2 * 16)(%rdx), RA2;
	vmovdqu (3 * 16)(%rdx), RA3;
	vmovdqu (4 * 16)(%rdx), RB0;
	vmovdqu (5 * 16)(%rdx), RB1;
	vmovdqu (6 * 16)(%rdx), RB2;
	vmovdqu (7 * 16)(%rdx), RB3;

	call __sm4_crypt_blk8;

	vmovdqu (7 * 16)(%rdx), RNOT;
	vpxor (%rcx), RA0, RA0;
	vpxor (0 * 16)(%rdx), RA1, RA1;
	vpxor (1 * 16)(%rdx), RA2, RA2;
	vpxor (2 * 16)(%rdx), RA3, RA3;
	vpxor (3 * 16)(%rdx), RB0, RB0;
	vpxor (4 * 16)(%rdx), RB1, RB1;
	vpxor (5 * 16)(%rdx), RB2, RB2;
	vpxor (6 * 16)(%rdx), RB3, RB3;
	vmovdqu RNOT, (%rcx); /* store new IV */

	vmovdqu RA0, (0 * 16)(%rsi);
	vmovdqu RA1, (1 * 16)(%rsi);
	vmovdqu RA2, (2 * 16)(%rsi);
	vmovdqu RA3, (3 * 16)(%rsi);
	vmovdqu RB0, (4 * 16)(%rsi);
	vmovdqu RB1, (5 * 16)(%rsi);
	vmovdqu RB2, (6 * 16)(%rsi);
	vmovdqu RB3, (7 * 16)(%rsi);

	vzeroall;
	FRAME_END
	RET;
SYM_FUNC_END(sm4_aesni_avx_cbc_dec_blk8)

/*
 * void sm4_aesni_avx_cfb_dec_blk8(const u32 *rk, u8 *dst,
 *                                 const u8 *src, u8 *iv)
 */
.align 8
SYM_FUNC_START(sm4_aesni_avx_cfb_dec_blk8)
	/* input:
	 *	%rdi: round key array, CTX
	 *	%rsi: dst (8 blocks)
	 *	%rdx: src (8 blocks)
	 *	%rcx: iv
	 */
	FRAME_BEGIN

	/* Load input */
	vmovdqu (%rcx), RA0;
	vmovdqu 0 * 16(%rdx), RA1;
	vmovdqu 1 * 16(%rdx), RA2;
	vmovdqu 2 * 16(%rdx), RA3;
	vmovdqu 3 * 16(%rdx), RB0;
	vmovdqu 4 * 16(%rdx), RB1;
	vmovdqu 5 * 16(%rdx), RB2;
	vmovdqu 6 * 16(%rdx), RB3;

	/* Update IV */
	vmovdqu 7 * 16(%rdx), RNOT;
	vmovdqu RNOT, (%rcx);

	call __sm4_crypt_blk8;

	vpxor (0 * 16)(%rdx), RA0, RA0;
	vpxor (1 * 16)(%rdx), RA1, RA1;
	vpxor (2 * 16)(%rdx), RA2, RA2;
	vpxor (3 * 16)(%rdx), RA3, RA3;
	vpxor (4 * 16)(%rdx), RB0, RB0;
	vpxor (5 * 16)(%rdx), RB1, RB1;
	vpxor (6 * 16)(%rdx), RB2, RB2;
	vpxor (7 * 16)(%rdx), RB3, RB3;

	vmovdqu RA0, (0 * 16)(%rsi);
	vmovdqu RA1, (1 * 16)(%rsi);
	vmovdqu RA2, (2 * 16)(%rsi);
	vmovdqu RA3, (3 * 16)(%rsi);
	vmovdqu RB0, (4 * 16)(%rsi);
	vmovdqu RB1, (5 * 16)(%rsi);
	vmovdqu RB2, (6 * 16)(%rsi);
	vmovdqu RB3, (7 * 16)(%rsi);

	vzeroall;
	FRAME_END
	RET;
SYM_FUNC_END(sm4_aesni_avx_cfb_dec_blk8)