[CRYPTO] twofish: x86-64 assembly version

The patch passed the trycpt tests and automated filesystem tests. This rewrite resulted in some nice perfomance increase over my last patch. Short summary of the tcrypt benchmarks: Twofish Assembler vs. Twofish C (256bit 8kb block CBC) encrypt: -27% Cycles decrypt: -23% Cycles Twofish Assembler vs. AES Assembler (128bit 8kb block CBC) encrypt: +18% Cycles decrypt: +15% Cycles Twofish Assembler vs. AES Assembler (256bit 8kb block CBC) encrypt: -9% Cycles decrypt: -8% Cycles Full Output: http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-c-x86_64.txt http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-asm-x86_64.txt http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-aes-asm-x86_64.txt Here is another bonnie++ benchmark with encrypted filesystems. Most runs maxed out the hd. It should give some idea what the module can do for encrypted filesystem performance even though you can't see the full numbers. http://homepages.tu-darmstadt.de/~fritschi/twofish/output_20060610_130806_x86_64.html Signed-off-by: Joachim Fritschi <jfritschi@freenet.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

[CRYPTO] twofish: x86-64 assembly version
The patch passed the trycpt tests and automated filesystem tests. This rewrite resulted in some nice perfomance increase over my last patch. Short summary of the tcrypt benchmarks: Twofish Assembler vs. Twofish C (256bit 8kb block CBC) encrypt: -27% Cycles decrypt: -23% Cycles Twofish Assembler vs. AES Assembler (128bit 8kb block CBC) encrypt: +18% Cycles decrypt: +15% Cycles Twofish Assembler vs. AES Assembler (256bit 8kb block CBC) encrypt: -9% Cycles decrypt: -8% Cycles Full Output: http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-c-x86_64.txt http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-asm-x86_64.txt http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-aes-asm-x86_64.txt Here is another bonnie++ benchmark with encrypted filesystems. Most runs maxed out the hd. It should give some idea what the module can do for encrypted filesystem performance even though you can't see the full numbers. http://homepages.tu-darmstadt.de/~fritschi/twofish/output_20060610_130806_x86_64.html Signed-off-by: Joachim Fritschi <jfritschi@freenet.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Joachim Fritschi · Herbert Xu
1 parent b9f535ffe3
Showing 4 changed files with 439 additions and 0 deletions Side-by-side Diff
arch/x86_64/crypto/Makefile
arch/x86_64/crypto/twofish-x86_64-asm.S
arch/x86_64/crypto/twofish.c
crypto/Kconfig
@@ -5,6 +5,8 @@
 # 
  
 obj-$(CONFIG_CRYPTO_AES_X86_64) += aes-x86_64.o
+obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o
  
 aes-x86_64-y := aes-x86_64-asm.o aes.o
+twofish-x86_64-y := twofish-x86_64-asm.o twofish.o
+/***************************************************************************
+*   Copyright (C) 2006 by Joachim Fritschi, <jfritschi@freenet.de>        *
+*                                                                         *
+*   This program is free software; you can redistribute it and/or modify  *
+*   it under the terms of the GNU General Public License as published by  *
+*   the Free Software Foundation; either version 2 of the License, or     *
+*   (at your option) any later version.                                   *
+*                                                                         *
+*   This program is distributed in the hope that it will be useful,       *
+*   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
+*   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
+*   GNU General Public License for more details.                          *
+*                                                                         *
+*   You should have received a copy of the GNU General Public License     *
+*   along with this program; if not, write to the                         *
+*   Free Software Foundation, Inc.,                                       *
+*   59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.             *
+***************************************************************************/
+
+.file "twofish-x86_64-asm.S"
+.text
+
+#include <asm/asm-offsets.h>
+
+#define a_offset	0
+#define b_offset	4
+#define c_offset	8
+#define d_offset	12
+
+/* Structure of the crypto context struct*/
+
+#define s0	0	/* S0 Array 256 Words each */
+#define s1	1024	/* S1 Array */
+#define s2	2048	/* S2 Array */
+#define s3	3072	/* S3 Array */
+#define w	4096	/* 8 whitening keys (word) */
+#define k	4128	/* key 1-32 ( word ) */
+
+/* define a few register aliases to allow macro substitution */
+
+#define R0     %rax
+#define R0D    %eax
+#define R0B    %al
+#define R0H    %ah
+
+#define R1     %rbx
+#define R1D    %ebx
+#define R1B    %bl
+#define R1H    %bh
+
+#define R2     %rcx
+#define R2D    %ecx
+#define R2B    %cl
+#define R2H    %ch
+
+#define R3     %rdx
+#define R3D    %edx
+#define R3B    %dl
+#define R3H    %dh
+
+
+/* performs input whitening */
+#define input_whitening(src,context,offset)\
+	xor	w+offset(context),	src;
+
+/* performs input whitening */
+#define output_whitening(src,context,offset)\
+	xor	w+16+offset(context),	src;
+
+
+/*
+ * a input register containing a (rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ */
+#define encrypt_round(a,b,c,d,round)\
+	movzx	b ## B,		%edi;\
+	mov	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$15,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	rol	$15,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;
+
+/*
+ * a input register containing a(rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ * during the round a and b are prepared for the output whitening
+ */
+#define encrypt_last_round(a,b,c,d,round)\
+	mov	b ## D,		%r10d;\
+	shl	$32,		%r10;\
+	movzx	b ## B,		%edi;\
+	mov	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	(%r11,%rdi,4),	%r9d;\
+	xor	a,		%r10;\
+	movzx	b ## H,		%edi;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	ror	$1,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D
+
+/*
+ * a input register containing a
+ * b input register containing b (rotated 16)
+ * c input register containing c (already rol $1)
+ * d input register containing d
+ * operations on a and b are interleaved to increase performance
+ */
+#define decrypt_round(a,b,c,d,round)\
+	movzx	a ## B,		%edi;\
+	mov	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$15,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;\
+	rol	$15,		d ## D;
+
+/*
+ * a input register containing a
+ * b input register containing b
+ * c input register containing c (already rol $1)
+ * d input register containing d
+ * operations on a and b are interleaved to increase performance
+ * during the round a and b are prepared for the output whitening
+ */
+#define decrypt_last_round(a,b,c,d,round)\
+	movzx	a ## B,		%edi;\
+	mov	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%r11,%rdi,4),%r8d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	mov	b ## D,		%r10d;\
+	shl	$32,		%r10;\
+	xor	a,		%r10;\
+	ror	$16,		a ## D;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;\
+	ror	$1,		d ## D;
+
+.align 8
+.global twofish_enc_blk
+.global twofish_dec_blk
+
+twofish_enc_blk:
+	pushq    R1
+
+	/* %rdi contains the crypto tfm adress */
+	/* %rsi contains the output adress */
+	/* %rdx contains the input adress */
+	add	$crypto_tfm_ctx_offset, %rdi	/* set ctx adress */
+	/* ctx adress is moved to free one non-rex register
+	as target for the 8bit high operations */
+	mov	%rdi,		%r11
+
+	movq	(R3),	R1
+	movq	8(R3),	R3
+	input_whitening(R1,%r11,a_offset)
+	input_whitening(R3,%r11,c_offset)
+	mov	R1D,	R0D
+	rol	$16,	R0D
+	shr	$32,	R1
+	mov	R3D,	R2D
+	shr	$32,	R3
+	rol	$1,	R3D
+
+	encrypt_round(R0,R1,R2,R3,0);
+	encrypt_round(R2,R3,R0,R1,8);
+	encrypt_round(R0,R1,R2,R3,2*8);
+	encrypt_round(R2,R3,R0,R1,3*8);
+	encrypt_round(R0,R1,R2,R3,4*8);
+	encrypt_round(R2,R3,R0,R1,5*8);
+	encrypt_round(R0,R1,R2,R3,6*8);
+	encrypt_round(R2,R3,R0,R1,7*8);
+	encrypt_round(R0,R1,R2,R3,8*8);
+	encrypt_round(R2,R3,R0,R1,9*8);
+	encrypt_round(R0,R1,R2,R3,10*8);
+	encrypt_round(R2,R3,R0,R1,11*8);
+	encrypt_round(R0,R1,R2,R3,12*8);
+	encrypt_round(R2,R3,R0,R1,13*8);
+	encrypt_round(R0,R1,R2,R3,14*8);
+	encrypt_last_round(R2,R3,R0,R1,15*8);
+
+
+	output_whitening(%r10,%r11,a_offset)
+	movq	%r10,	(%rsi)
+
+	shl	$32,	R1
+	xor	R0,	R1
+
+	output_whitening(R1,%r11,c_offset)
+	movq	R1,	8(%rsi)
+
+	popq	R1
+	movq	$1,%rax
+	ret
+
+twofish_dec_blk:
+	pushq    R1
+
+	/* %rdi contains the crypto tfm adress */
+	/* %rsi contains the output adress */
+	/* %rdx contains the input adress */
+	add	$crypto_tfm_ctx_offset, %rdi	/* set ctx adress */
+	/* ctx adress is moved to free one non-rex register
+	as target for the 8bit high operations */
+	mov	%rdi,		%r11
+
+	movq	(R3),	R1
+	movq	8(R3),	R3
+	output_whitening(R1,%r11,a_offset)
+	output_whitening(R3,%r11,c_offset)
+	mov	R1D,	R0D
+	shr	$32,	R1
+	rol	$16,	R1D
+	mov	R3D,	R2D
+	shr	$32,	R3
+	rol	$1,	R2D
+
+	decrypt_round(R0,R1,R2,R3,15*8);
+	decrypt_round(R2,R3,R0,R1,14*8);
+	decrypt_round(R0,R1,R2,R3,13*8);
+	decrypt_round(R2,R3,R0,R1,12*8);
+	decrypt_round(R0,R1,R2,R3,11*8);
+	decrypt_round(R2,R3,R0,R1,10*8);
+	decrypt_round(R0,R1,R2,R3,9*8);
+	decrypt_round(R2,R3,R0,R1,8*8);
+	decrypt_round(R0,R1,R2,R3,7*8);
+	decrypt_round(R2,R3,R0,R1,6*8);
+	decrypt_round(R0,R1,R2,R3,5*8);
+	decrypt_round(R2,R3,R0,R1,4*8);
+	decrypt_round(R0,R1,R2,R3,3*8);
+	decrypt_round(R2,R3,R0,R1,2*8);
+	decrypt_round(R0,R1,R2,R3,1*8);
+	decrypt_last_round(R2,R3,R0,R1,0);
+
+	input_whitening(%r10,%r11,a_offset)
+	movq	%r10,	(%rsi)
+
+	shl	$32,	R1
+	xor	R0,	R1
+
+	input_whitening(R1,%r11,c_offset)
+	movq	R1,	8(%rsi)
+
+	popq	R1
+	movq	$1,%rax
+	ret
+/*
+ * Glue Code for optimized x86_64 assembler version of TWOFISH
+ *
+ * Originally Twofish for GPG
+ * By Matthew Skala <mskala@ansuz.sooke.bc.ca>, July 26, 1998
+ * 256-bit key length added March 20, 1999
+ * Some modifications to reduce the text size by Werner Koch, April, 1998
+ * Ported to the kerneli patch by Marc Mutz <Marc@Mutz.com>
+ * Ported to CryptoAPI by Colin Slater <hoho@tacomeat.net>
+ *
+ * The original author has disclaimed all copyright interest in this
+ * code and thus put it in the public domain. The subsequent authors
+ * have put this under the GNU General Public License.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
+ * USA
+ *
+ * This code is a "clean room" implementation, written from the paper
+ * _Twofish: A 128-Bit Block Cipher_ by Bruce Schneier, John Kelsey,
+ * Doug Whiting, David Wagner, Chris Hall, and Niels Ferguson, available
+ * through http://www.counterpane.com/twofish.html
+ *
+ * For background information on multiplication in finite fields, used for
+ * the matrix operations in the key schedule, see the book _Contemporary
+ * Abstract Algebra_ by Joseph A. Gallian, especially chapter 22 in the
+ * Third Edition.
+ */
+
+#include <crypto/twofish.h>
+#include <linux/crypto.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+asmlinkage void twofish_enc_blk(struct crypto_tfm *tfm, u8 *dst, const u8 *src);
+asmlinkage void twofish_dec_blk(struct crypto_tfm *tfm, u8 *dst, const u8 *src);
+
+static void twofish_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	twofish_enc_blk(tfm, dst, src);
+}
+
+static void twofish_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	twofish_dec_blk(tfm, dst, src);
+}
+
+static struct crypto_alg alg = {
+	.cra_name		=	"twofish",
+	.cra_driver_name	=	"twofish-x86_64",
+	.cra_priority		=	200,
+	.cra_flags		=	CRYPTO_ALG_TYPE_CIPHER,
+	.cra_blocksize		=	TF_BLOCK_SIZE,
+	.cra_ctxsize		=	sizeof(struct twofish_ctx),
+	.cra_alignmask		=	3,
+	.cra_module		=	THIS_MODULE,
+	.cra_list		=	LIST_HEAD_INIT(alg.cra_list),
+	.cra_u			=	{
+		.cipher = {
+			.cia_min_keysize	=	TF_MIN_KEY_SIZE,
+			.cia_max_keysize	=	TF_MAX_KEY_SIZE,
+			.cia_setkey		=	twofish_setkey,
+			.cia_encrypt		=	twofish_encrypt,
+			.cia_decrypt		=	twofish_decrypt
+		}
+	}
+};
+
+static int __init init(void)
+{
+	return crypto_register_alg(&alg);
+}
+
+static void __exit fini(void)
+{
+	crypto_unregister_alg(&alg);
+}
+
+module_init(init);
+module_exit(fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION ("Twofish Cipher Algorithm, x86_64 asm optimized");
+MODULE_ALIAS("twofish");
@@ -165,6 +165,21 @@
 	  See also:
 	  <http://www.schneier.com/twofish.html>
  
+config CRYPTO_TWOFISH_X86_64
+	tristate "Twofish cipher algorithm (x86_64)"
+	depends on CRYPTO && ((X86 || UML_X86) && 64BIT)
+	select CRYPTO_TWOFISH_COMMON
+	help
+	  Twofish cipher algorithm (x86_64).
+
+	  Twofish was submitted as an AES (Advanced Encryption Standard)
+	  candidate cipher by researchers at CounterPane Systems.  It is a
+	  16 round block cipher supporting key sizes of 128, 192, and 256
+	  bits.
+
+	  See also:
+	  <http://www.schneier.com/twofish.html>
+
 config CRYPTO_SERPENT
 	tristate "Serpent cipher algorithm"
 	depends on CRYPTO
...	...	@@ -5,6 +5,8 @@
5	5	#
6	6
7	7	obj-$(CONFIG_CRYPTO_AES_X86_64) += aes-x86_64.o
	8	+obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o
8	9
9	10	aes-x86_64-y := aes-x86_64-asm.o aes.o
	11	+twofish-x86_64-y := twofish-x86_64-asm.o twofish.o
	1	+/***************************************************************************
	2	+* Copyright (C) 2006 by Joachim Fritschi, <jfritschi@freenet.de> *
	3	+* *
	4	+* This program is free software; you can redistribute it and/or modify *
	5	+* it under the terms of the GNU General Public License as published by *
	6	+* the Free Software Foundation; either version 2 of the License, or *
	7	+* (at your option) any later version. *
	8	+* *
	9	+* This program is distributed in the hope that it will be useful, *
	10	+* but WITHOUT ANY WARRANTY; without even the implied warranty of *
	11	+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
	12	+* GNU General Public License for more details. *
	13	+* *
	14	+* You should have received a copy of the GNU General Public License *
	15	+* along with this program; if not, write to the *
	16	+* Free Software Foundation, Inc., *
	17	+* 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. *
	18	+***************************************************************************/
	19	+
	20	+.file "twofish-x86_64-asm.S"
	21	+.text
	22	+
	23	+#include <asm/asm-offsets.h>
	24	+
	25	+#define a_offset 0
	26	+#define b_offset 4
	27	+#define c_offset 8
	28	+#define d_offset 12
	29	+
	30	+/* Structure of the crypto context struct*/
	31	+
	32	+#define s0 0 /* S0 Array 256 Words each */
	33	+#define s1 1024 /* S1 Array */
	34	+#define s2 2048 /* S2 Array */
	35	+#define s3 3072 /* S3 Array */
	36	+#define w 4096 /* 8 whitening keys (word) */
	37	+#define k 4128 /* key 1-32 ( word ) */
	38	+
	39	+/* define a few register aliases to allow macro substitution */
	40	+
	41	+#define R0 %rax
	42	+#define R0D %eax
	43	+#define R0B %al
	44	+#define R0H %ah
	45	+
	46	+#define R1 %rbx
	47	+#define R1D %ebx
	48	+#define R1B %bl
	49	+#define R1H %bh
	50	+
	51	+#define R2 %rcx
	52	+#define R2D %ecx
	53	+#define R2B %cl
	54	+#define R2H %ch
	55	+
	56	+#define R3 %rdx
	57	+#define R3D %edx
	58	+#define R3B %dl
	59	+#define R3H %dh
	60	+
	61	+
	62	+/* performs input whitening */
	63	+#define input_whitening(src,context,offset)\
	64	+ xor w+offset(context), src;
	65	+
	66	+/* performs input whitening */
	67	+#define output_whitening(src,context,offset)\
	68	+ xor w+16+offset(context), src;
	69	+
	70	+
	71	+/*
	72	+ * a input register containing a (rotated 16)
	73	+ * b input register containing b
	74	+ * c input register containing c
	75	+ * d input register containing d (already rol $1)
	76	+ * operations on a and b are interleaved to increase performance
	77	+ */
	78	+#define encrypt_round(a,b,c,d,round)\
	79	+ movzx b ## B, %edi;\
	80	+ mov s1(%r11,%rdi,4),%r8d;\
	81	+ movzx a ## B, %edi;\
	82	+ mov s2(%r11,%rdi,4),%r9d;\
	83	+ movzx b ## H, %edi;\
	84	+ ror $16, b ## D;\
	85	+ xor s2(%r11,%rdi,4),%r8d;\
	86	+ movzx a ## H, %edi;\
	87	+ ror $16, a ## D;\
	88	+ xor s3(%r11,%rdi,4),%r9d;\
	89	+ movzx b ## B, %edi;\
	90	+ xor s3(%r11,%rdi,4),%r8d;\
	91	+ movzx a ## B, %edi;\
	92	+ xor (%r11,%rdi,4), %r9d;\
	93	+ movzx b ## H, %edi;\
	94	+ ror $15, b ## D;\
	95	+ xor (%r11,%rdi,4), %r8d;\
	96	+ movzx a ## H, %edi;\
	97	+ xor s1(%r11,%rdi,4),%r9d;\
	98	+ add %r8d, %r9d;\
	99	+ add %r9d, %r8d;\
	100	+ add k+round(%r11), %r9d;\
	101	+ xor %r9d, c ## D;\
	102	+ rol $15, c ## D;\
	103	+ add k+4+round(%r11),%r8d;\
	104	+ xor %r8d, d ## D;
	105	+
	106	+/*
	107	+ * a input register containing a(rotated 16)
	108	+ * b input register containing b
	109	+ * c input register containing c
	110	+ * d input register containing d (already rol $1)
	111	+ * operations on a and b are interleaved to increase performance
	112	+ * during the round a and b are prepared for the output whitening
	113	+ */
	114	+#define encrypt_last_round(a,b,c,d,round)\
	115	+ mov b ## D, %r10d;\
	116	+ shl $32, %r10;\
	117	+ movzx b ## B, %edi;\
	118	+ mov s1(%r11,%rdi,4),%r8d;\
	119	+ movzx a ## B, %edi;\
	120	+ mov s2(%r11,%rdi,4),%r9d;\
	121	+ movzx b ## H, %edi;\
	122	+ ror $16, b ## D;\
	123	+ xor s2(%r11,%rdi,4),%r8d;\
	124	+ movzx a ## H, %edi;\
	125	+ ror $16, a ## D;\
	126	+ xor s3(%r11,%rdi,4),%r9d;\
	127	+ movzx b ## B, %edi;\
	128	+ xor s3(%r11,%rdi,4),%r8d;\
	129	+ movzx a ## B, %edi;\
	130	+ xor (%r11,%rdi,4), %r9d;\
	131	+ xor a, %r10;\
	132	+ movzx b ## H, %edi;\
	133	+ xor (%r11,%rdi,4), %r8d;\
	134	+ movzx a ## H, %edi;\
	135	+ xor s1(%r11,%rdi,4),%r9d;\
	136	+ add %r8d, %r9d;\
	137	+ add %r9d, %r8d;\
	138	+ add k+round(%r11), %r9d;\
	139	+ xor %r9d, c ## D;\
	140	+ ror $1, c ## D;\
	141	+ add k+4+round(%r11),%r8d;\
	142	+ xor %r8d, d ## D
	143	+
	144	+/*
	145	+ * a input register containing a
	146	+ * b input register containing b (rotated 16)
	147	+ * c input register containing c (already rol $1)
	148	+ * d input register containing d
	149	+ * operations on a and b are interleaved to increase performance
	150	+ */
	151	+#define decrypt_round(a,b,c,d,round)\
	152	+ movzx a ## B, %edi;\
	153	+ mov (%r11,%rdi,4), %r9d;\
	154	+ movzx b ## B, %edi;\
	155	+ mov s3(%r11,%rdi,4),%r8d;\
	156	+ movzx a ## H, %edi;\
	157	+ ror $16, a ## D;\
	158	+ xor s1(%r11,%rdi,4),%r9d;\
	159	+ movzx b ## H, %edi;\
	160	+ ror $16, b ## D;\
	161	+ xor (%r11,%rdi,4), %r8d;\
	162	+ movzx a ## B, %edi;\
	163	+ xor s2(%r11,%rdi,4),%r9d;\
	164	+ movzx b ## B, %edi;\
	165	+ xor s1(%r11,%rdi,4),%r8d;\
	166	+ movzx a ## H, %edi;\
	167	+ ror $15, a ## D;\
	168	+ xor s3(%r11,%rdi,4),%r9d;\
	169	+ movzx b ## H, %edi;\
	170	+ xor s2(%r11,%rdi,4),%r8d;\
	171	+ add %r8d, %r9d;\
	172	+ add %r9d, %r8d;\
	173	+ add k+round(%r11), %r9d;\
	174	+ xor %r9d, c ## D;\
	175	+ add k+4+round(%r11),%r8d;\
	176	+ xor %r8d, d ## D;\
	177	+ rol $15, d ## D;
	178	+
	179	+/*
	180	+ * a input register containing a
	181	+ * b input register containing b
	182	+ * c input register containing c (already rol $1)
	183	+ * d input register containing d
	184	+ * operations on a and b are interleaved to increase performance
	185	+ * during the round a and b are prepared for the output whitening
	186	+ */
	187	+#define decrypt_last_round(a,b,c,d,round)\
	188	+ movzx a ## B, %edi;\
	189	+ mov (%r11,%rdi,4), %r9d;\
	190	+ movzx b ## B, %edi;\
	191	+ mov s3(%r11,%rdi,4),%r8d;\
	192	+ movzx b ## H, %edi;\
	193	+ ror $16, b ## D;\
	194	+ xor (%r11,%rdi,4), %r8d;\
	195	+ movzx a ## H, %edi;\
	196	+ mov b ## D, %r10d;\
	197	+ shl $32, %r10;\
	198	+ xor a, %r10;\
	199	+ ror $16, a ## D;\
	200	+ xor s1(%r11,%rdi,4),%r9d;\
	201	+ movzx b ## B, %edi;\
	202	+ xor s1(%r11,%rdi,4),%r8d;\
	203	+ movzx a ## B, %edi;\
	204	+ xor s2(%r11,%rdi,4),%r9d;\
	205	+ movzx b ## H, %edi;\
	206	+ xor s2(%r11,%rdi,4),%r8d;\
	207	+ movzx a ## H, %edi;\
	208	+ xor s3(%r11,%rdi,4),%r9d;\
	209	+ add %r8d, %r9d;\
	210	+ add %r9d, %r8d;\
	211	+ add k+round(%r11), %r9d;\
	212	+ xor %r9d, c ## D;\
	213	+ add k+4+round(%r11),%r8d;\
	214	+ xor %r8d, d ## D;\
	215	+ ror $1, d ## D;
	216	+
	217	+.align 8
	218	+.global twofish_enc_blk
	219	+.global twofish_dec_blk
	220	+
	221	+twofish_enc_blk:
	222	+ pushq R1
	223	+
	224	+ /* %rdi contains the crypto tfm adress */
	225	+ /* %rsi contains the output adress */
	226	+ /* %rdx contains the input adress */
	227	+ add $crypto_tfm_ctx_offset, %rdi /* set ctx adress */
	228	+ /* ctx adress is moved to free one non-rex register
	229	+ as target for the 8bit high operations */
	230	+ mov %rdi, %r11
	231	+
	232	+ movq (R3), R1
	233	+ movq 8(R3), R3
	234	+ input_whitening(R1,%r11,a_offset)
	235	+ input_whitening(R3,%r11,c_offset)
	236	+ mov R1D, R0D
	237	+ rol $16, R0D
	238	+ shr $32, R1
	239	+ mov R3D, R2D
	240	+ shr $32, R3
	241	+ rol $1, R3D
	242	+
	243	+ encrypt_round(R0,R1,R2,R3,0);
	244	+ encrypt_round(R2,R3,R0,R1,8);
	245	+ encrypt_round(R0,R1,R2,R3,2*8);
	246	+ encrypt_round(R2,R3,R0,R1,3*8);
	247	+ encrypt_round(R0,R1,R2,R3,4*8);
	248	+ encrypt_round(R2,R3,R0,R1,5*8);
	249	+ encrypt_round(R0,R1,R2,R3,6*8);
	250	+ encrypt_round(R2,R3,R0,R1,7*8);
	251	+ encrypt_round(R0,R1,R2,R3,8*8);
	252	+ encrypt_round(R2,R3,R0,R1,9*8);
	253	+ encrypt_round(R0,R1,R2,R3,10*8);
	254	+ encrypt_round(R2,R3,R0,R1,11*8);
	255	+ encrypt_round(R0,R1,R2,R3,12*8);
	256	+ encrypt_round(R2,R3,R0,R1,13*8);
	257	+ encrypt_round(R0,R1,R2,R3,14*8);
	258	+ encrypt_last_round(R2,R3,R0,R1,15*8);
	259	+
	260	+
	261	+ output_whitening(%r10,%r11,a_offset)
	262	+ movq %r10, (%rsi)
	263	+
	264	+ shl $32, R1
	265	+ xor R0, R1
	266	+
	267	+ output_whitening(R1,%r11,c_offset)
	268	+ movq R1, 8(%rsi)
	269	+
	270	+ popq R1
	271	+ movq $1,%rax
	272	+ ret
	273	+
	274	+twofish_dec_blk:
	275	+ pushq R1
	276	+
	277	+ /* %rdi contains the crypto tfm adress */
	278	+ /* %rsi contains the output adress */
	279	+ /* %rdx contains the input adress */
	280	+ add $crypto_tfm_ctx_offset, %rdi /* set ctx adress */
	281	+ /* ctx adress is moved to free one non-rex register
	282	+ as target for the 8bit high operations */
	283	+ mov %rdi, %r11
	284	+
	285	+ movq (R3), R1
	286	+ movq 8(R3), R3
	287	+ output_whitening(R1,%r11,a_offset)
	288	+ output_whitening(R3,%r11,c_offset)
	289	+ mov R1D, R0D
	290	+ shr $32, R1
	291	+ rol $16, R1D
	292	+ mov R3D, R2D
	293	+ shr $32, R3
	294	+ rol $1, R2D
	295	+
	296	+ decrypt_round(R0,R1,R2,R3,15*8);
	297	+ decrypt_round(R2,R3,R0,R1,14*8);
	298	+ decrypt_round(R0,R1,R2,R3,13*8);
	299	+ decrypt_round(R2,R3,R0,R1,12*8);
	300	+ decrypt_round(R0,R1,R2,R3,11*8);
	301	+ decrypt_round(R2,R3,R0,R1,10*8);
	302	+ decrypt_round(R0,R1,R2,R3,9*8);
	303	+ decrypt_round(R2,R3,R0,R1,8*8);
	304	+ decrypt_round(R0,R1,R2,R3,7*8);
	305	+ decrypt_round(R2,R3,R0,R1,6*8);
	306	+ decrypt_round(R0,R1,R2,R3,5*8);
	307	+ decrypt_round(R2,R3,R0,R1,4*8);
	308	+ decrypt_round(R0,R1,R2,R3,3*8);
	309	+ decrypt_round(R2,R3,R0,R1,2*8);
	310	+ decrypt_round(R0,R1,R2,R3,1*8);
	311	+ decrypt_last_round(R2,R3,R0,R1,0);
	312	+
	313	+ input_whitening(%r10,%r11,a_offset)
	314	+ movq %r10, (%rsi)
	315	+
	316	+ shl $32, R1
	317	+ xor R0, R1
	318	+
	319	+ input_whitening(R1,%r11,c_offset)
	320	+ movq R1, 8(%rsi)
	321	+
	322	+ popq R1
	323	+ movq $1,%rax
	324	+ ret
	1	+/*
	2	+ * Glue Code for optimized x86_64 assembler version of TWOFISH
	3	+ *
	4	+ * Originally Twofish for GPG
	5	+ * By Matthew Skala <mskala@ansuz.sooke.bc.ca>, July 26, 1998
	6	+ * 256-bit key length added March 20, 1999
	7	+ * Some modifications to reduce the text size by Werner Koch, April, 1998
	8	+ * Ported to the kerneli patch by Marc Mutz <Marc@Mutz.com>
	9	+ * Ported to CryptoAPI by Colin Slater <hoho@tacomeat.net>
	10	+ *
	11	+ * The original author has disclaimed all copyright interest in this
	12	+ * code and thus put it in the public domain. The subsequent authors
	13	+ * have put this under the GNU General Public License.
	14	+ *
	15	+ * This program is free software; you can redistribute it and/or modify
	16	+ * it under the terms of the GNU General Public License as published by
	17	+ * the Free Software Foundation; either version 2 of the License, or
	18	+ * (at your option) any later version.
	19	+ *
	20	+ * This program is distributed in the hope that it will be useful,
	21	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	22	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	23	+ * GNU General Public License for more details.
	24	+ *
	25	+ * You should have received a copy of the GNU General Public License
	26	+ * along with this program; if not, write to the Free Software
	27	+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
	28	+ * USA
	29	+ *
	30	+ * This code is a "clean room" implementation, written from the paper
	31	+ * _Twofish: A 128-Bit Block Cipher_ by Bruce Schneier, John Kelsey,
	32	+ * Doug Whiting, David Wagner, Chris Hall, and Niels Ferguson, available
	33	+ * through http://www.counterpane.com/twofish.html
	34	+ *
	35	+ * For background information on multiplication in finite fields, used for
	36	+ * the matrix operations in the key schedule, see the book _Contemporary
	37	+ * Abstract Algebra_ by Joseph A. Gallian, especially chapter 22 in the
	38	+ * Third Edition.
	39	+ */
	40	+
	41	+#include <crypto/twofish.h>
	42	+#include <linux/crypto.h>
	43	+#include <linux/init.h>
	44	+#include <linux/kernel.h>
	45	+#include <linux/module.h>
	46	+#include <linux/types.h>
	47	+
	48	+asmlinkage void twofish_enc_blk(struct crypto_tfm tfm, u8 dst, const u8 *src);
	49	+asmlinkage void twofish_dec_blk(struct crypto_tfm tfm, u8 dst, const u8 *src);
	50	+
	51	+static void twofish_encrypt(struct crypto_tfm tfm, u8 dst, const u8 *src)
	52	+{
	53	+ twofish_enc_blk(tfm, dst, src);
	54	+}
	55	+
	56	+static void twofish_decrypt(struct crypto_tfm tfm, u8 dst, const u8 *src)
	57	+{
	58	+ twofish_dec_blk(tfm, dst, src);
	59	+}
	60	+
	61	+static struct crypto_alg alg = {
	62	+ .cra_name = "twofish",
	63	+ .cra_driver_name = "twofish-x86_64",
	64	+ .cra_priority = 200,
	65	+ .cra_flags = CRYPTO_ALG_TYPE_CIPHER,
	66	+ .cra_blocksize = TF_BLOCK_SIZE,
	67	+ .cra_ctxsize = sizeof(struct twofish_ctx),
	68	+ .cra_alignmask = 3,
	69	+ .cra_module = THIS_MODULE,
	70	+ .cra_list = LIST_HEAD_INIT(alg.cra_list),
	71	+ .cra_u = {
	72	+ .cipher = {
	73	+ .cia_min_keysize = TF_MIN_KEY_SIZE,
	74	+ .cia_max_keysize = TF_MAX_KEY_SIZE,
	75	+ .cia_setkey = twofish_setkey,
	76	+ .cia_encrypt = twofish_encrypt,
	77	+ .cia_decrypt = twofish_decrypt
	78	+ }
	79	+ }
	80	+};
	81	+
	82	+static int __init init(void)
	83	+{
	84	+ return crypto_register_alg(&alg);
	85	+}
	86	+
	87	+static void __exit fini(void)
	88	+{
	89	+ crypto_unregister_alg(&alg);
	90	+}
	91	+
	92	+module_init(init);
	93	+module_exit(fini);
	94	+
	95	+MODULE_LICENSE("GPL");
	96	+MODULE_DESCRIPTION ("Twofish Cipher Algorithm, x86_64 asm optimized");
	97	+MODULE_ALIAS("twofish");
...	...	@@ -165,6 +165,21 @@
165	165	See also:
166	166	<http://www.schneier.com/twofish.html>
167	167
	168	+config CRYPTO_TWOFISH_X86_64
	169	+ tristate "Twofish cipher algorithm (x86_64)"
	170	+ depends on CRYPTO && ((X86 \|\| UML_X86) && 64BIT)
	171	+ select CRYPTO_TWOFISH_COMMON
	172	+ help
	173	+ Twofish cipher algorithm (x86_64).
	174	+
	175	+ Twofish was submitted as an AES (Advanced Encryption Standard)
	176	+ candidate cipher by researchers at CounterPane Systems. It is a
	177	+ 16 round block cipher supporting key sizes of 128, 192, and 256
	178	+ bits.
	179	+
	180	+ See also:
	181	+ <http://www.schneier.com/twofish.html>
	182	+
168	183	config CRYPTO_SERPENT
169	184	tristate "Serpent cipher algorithm"
170	185	depends on CRYPTO