crypto: serpent - add 4-way parallel i586/SSE2 assembler implementation

Patch adds i586/SSE2 assembler implementation of serpent cipher. Assembler
functions crypt data in four block chunks.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmarks results (serpent-sse2/serpent_generic speed ratios):

Intel Atom N270:

size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec
16      0.95x   1.12x   1.02x   1.07x   0.97x   0.98x
64      1.73x   1.82x   1.08x   1.82x   1.72x   1.73x
256     2.08x   2.00x   1.04x   2.07x   1.99x   2.01x
1024    2.28x   2.18x   1.05x   2.23x   2.17x   2.20x
8192    2.28x   2.13x   1.05x   2.23x   2.18x   2.20x

Full output:
 http://koti.mbnet.fi/axh/kernel/crypto/atom-n270/serpent-generic.txt
 http://koti.mbnet.fi/axh/kernel/crypto/atom-n270/serpent-sse2.txt

Userspace test results:

Encryption/decryption of sse2-i586 vs generic on Intel Atom N270:
 encrypt: 2.35x
 decrypt: 2.54x

Encryption/decryption of sse2-i586 vs generic on AMD Phenom II:
 encrypt: 1.82x
 decrypt: 2.51x

Encryption/decryption of sse2-i586 vs generic on Intel Xeon E7330:
 encrypt: 2.99x
 decrypt: 3.48x

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

4 files changed, 688 insertions, 0 deletions

diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
index 12ebdbd..2b0b963 100644
--- a/arch/x86/crypto/Makefile
+++ b/arch/x86/crypto/Makefile
@@ -5,6 +5,7 @@
 obj-$(CONFIG_CRYPTO_AES_586) += aes-i586.o
 obj-$(CONFIG_CRYPTO_TWOFISH_586) += twofish-i586.o
 obj-$(CONFIG_CRYPTO_SALSA20_586) += salsa20-i586.o
+obj-$(CONFIG_CRYPTO_SERPENT_SSE2_586) += serpent-sse2-i586.o
 
 obj-$(CONFIG_CRYPTO_AES_X86_64) += aes-x86_64.o
 obj-$(CONFIG_CRYPTO_BLOWFISH_X86_64) += blowfish-x86_64.o
@@ -21,6 +22,7 @@ obj-$(CONFIG_CRYPTO_SHA1_SSSE3) += sha1-ssse3.o
 aes-i586-y := aes-i586-asm_32.o aes_glue.o
 twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
 salsa20-i586-y := salsa20-i586-asm_32.o salsa20_glue.o
+serpent-sse2-i586-y := serpent-sse2-i586-asm_32.o serpent_sse2_glue.o
 
 aes-x86_64-y := aes-x86_64-asm_64.o aes_glue.o
 blowfish-x86_64-y := blowfish-x86_64-asm_64.o blowfish_glue.o
diff --git a/arch/x86/crypto/serpent-sse2-i586-asm_32.S b/arch/x86/crypto/serpent-sse2-i586-asm_32.S
new file mode 100644
index 0000000..4e37677
--- /dev/null
+++ b/arch/x86/crypto/serpent-sse2-i586-asm_32.S
@@ -0,0 +1,638 @@
+/*
+ * Serpent Cipher 4-way parallel algorithm (i586/SSE2)
+ *
+ * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Based on crypto/serpent.c by
+ *  Copyright (C) 2002 Dag Arne Osvik <osvik@ii.uib.no>
+ *                2003 Herbert Valerio Riedel <hvr@gnu.org>
+ *
+ * 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 "serpent-sse2-i586-asm_32.S"
+.text
+
+#define arg_ctx 4
+#define arg_dst 8
+#define arg_src 12
+#define arg_xor 16
+
+/**********************************************************************
+  4-way SSE2 serpent
+ **********************************************************************/
+#define CTX %edx
+
+#define RA %xmm0
+#define RB %xmm1
+#define RC %xmm2
+#define RD %xmm3
+#define RE %xmm4
+
+#define RT0 %xmm5
+#define RT1 %xmm6
+
+#define RNOT %xmm7
+
+#define get_key(i, j, t) \
+	movd (4*(i)+(j))*4(CTX), t; \
+	pshufd $0, t, t;
+
+#define K(x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, x4); \
+	get_key(i, 1, RT0); \
+	get_key(i, 2, RT1); \
+	pxor x4,		x0; \
+	pxor RT0,		x1; \
+	pxor RT1,		x2; \
+	get_key(i, 3, x4); \
+	pxor x4,		x3;
+
+#define LK(x0, x1, x2, x3, x4, i) \
+	movdqa x0,		x4; \
+	pslld $13,		x0; \
+	psrld $(32 - 13),	x4; \
+	por x4,			x0; \
+	pxor x0,		x1; \
+	movdqa x2,		x4; \
+	pslld $3,		x2; \
+	psrld $(32 - 3),	x4; \
+	por x4,			x2; \
+	pxor x2,		x1; \
+	movdqa x1,		x4; \
+	pslld $1,		x1; \
+	psrld $(32 - 1),	x4; \
+	por x4,			x1; \
+	movdqa x0,		x4; \
+	pslld $3,		x4; \
+	pxor x2,		x3; \
+	pxor x4,		x3; \
+	movdqa x3,		x4; \
+	pslld $7,		x3; \
+	psrld $(32 - 7),	x4; \
+	por x4,			x3; \
+	movdqa x1,		x4; \
+	pslld $7,		x4; \
+	pxor x1,		x0; \
+	pxor x3,		x0; \
+	pxor x3,		x2; \
+	pxor x4,		x2; \
+	movdqa x0,		x4; \
+	get_key(i, 1, RT0); \
+	pxor RT0,		x1; \
+	get_key(i, 3, RT0); \
+	pxor RT0,		x3; \
+	pslld $5,		x0; \
+	psrld $(32 - 5),	x4; \
+	por x4,			x0; \
+	movdqa x2,		x4; \
+	pslld $22,		x2; \
+	psrld $(32 - 22),	x4; \
+	por x4,			x2; \
+	get_key(i, 0, RT0); \
+	pxor RT0,		x0; \
+	get_key(i, 2, RT0); \
+	pxor RT0,		x2;
+
+#define KL(x0, x1, x2, x3, x4, i) \
+	K(x0, x1, x2, x3, x4, i); \
+	movdqa x0,		x4; \
+	psrld $5,		x0; \
+	pslld $(32 - 5),	x4; \
+	por x4,			x0; \
+	movdqa x2,		x4; \
+	psrld $22,		x2; \
+	pslld $(32 - 22),	x4; \
+	por x4,			x2; \
+	pxor x3,		x2; \
+	pxor x3,		x0; \
+	movdqa x1,		x4; \
+	pslld $7,		x4; \
+	pxor x1,		x0; \
+	pxor x4,		x2; \
+	movdqa x1,		x4; \
+	psrld $1,		x1; \
+	pslld $(32 - 1),	x4; \
+	por x4,			x1; \
+	movdqa x3,		x4; \
+	psrld $7,		x3; \
+	pslld $(32 - 7),	x4; \
+	por x4,			x3; \
+	pxor x0,		x1; \
+	movdqa x0,		x4; \
+	pslld $3,		x4; \
+	pxor x4,		x3; \
+	movdqa x0,		x4; \
+	psrld $13,		x0; \
+	pslld $(32 - 13),	x4; \
+	por x4,			x0; \
+	pxor x2,		x1; \
+	pxor x2,		x3; \
+	movdqa x2,		x4; \
+	psrld $3,		x2; \
+	pslld $(32 - 3),	x4; \
+	por x4,			x2;
+
+#define S0(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	por x0,			x3; \
+	pxor x4,		x0; \
+	pxor x2,		x4; \
+	pxor RNOT,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pxor x0,		x2; \
+	pxor x3,		x0; \
+	por x0,			x4; \
+	pxor x2,		x0; \
+	pand x1,		x2; \
+	pxor x2,		x3; \
+	pxor RNOT,		x1; \
+	pxor x4,		x2; \
+	pxor x2,		x1;
+
+#define S1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x1; \
+	pxor x3,		x0; \
+	pxor RNOT,		x3; \
+	pand x1,		x4; \
+	por x1,			x0; \
+	pxor x2,		x3; \
+	pxor x3,		x0; \
+	pxor x3,		x1; \
+	pxor x4,		x3; \
+	por x4,			x1; \
+	pxor x2,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x2; \
+	por x0,			x1; \
+	pxor RNOT,		x0; \
+	pxor x2,		x0; \
+	pxor x1,		x4;
+
+#define S2(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x3; \
+	pxor x0,		x1; \
+	movdqa x0,		x4; \
+	pand x2,		x0; \
+	pxor x3,		x0; \
+	por x4,			x3; \
+	pxor x1,		x2; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x2,		x0; \
+	pand x3,		x2; \
+	por x1,			x3; \
+	pxor RNOT,		x0; \
+	pxor x0,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	por x2,			x1;
+
+#define S3(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x3,		x1; \
+	por x0,			x3; \
+	pand x0,		x4; \
+	pxor x2,		x0; \
+	pxor x1,		x2; \
+	pand x3,		x1; \
+	pxor x3,		x2; \
+	por x4,			x0; \
+	pxor x3,		x4; \
+	pxor x0,		x1; \
+	pand x3,		x0; \
+	pand x4,		x3; \
+	pxor x2,		x3; \
+	por x1,			x4; \
+	pand x1,		x2; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	pxor x2,		x3;
+
+#define S4(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x4,		x0; \
+	pxor x2,		x3; \
+	por x4,			x2; \
+	pxor x1,		x0; \
+	pxor x3,		x4; \
+	por x0,			x2; \
+	pxor x1,		x2; \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pand x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x4; \
+	por x1,			x3; \
+	pxor RNOT,		x1; \
+	pxor x0,		x3;
+
+#define S5(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x0,			x1; \
+	pxor x1,		x2; \
+	pxor RNOT,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	pand x4,		x1; \
+	por x3,			x4; \
+	pxor x0,		x4; \
+	pand x3,		x0; \
+	pxor x3,		x1; \
+	pxor x2,		x3; \
+	pxor x1,		x0; \
+	pand x4,		x2; \
+	pxor x2,		x1; \
+	pand x0,		x2; \
+	pxor x2,		x3;
+
+#define S6(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x3; \
+	pxor x2,		x1; \
+	pxor x0,		x2; \
+	pand x3,		x0; \
+	por x3,			x1; \
+	pxor RNOT,		x4; \
+	pxor x1,		x0; \
+	pxor x2,		x1; \
+	pxor x4,		x3; \
+	pxor x0,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x4; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x3; \
+	pxor x2,		x1;
+
+#define S7(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x1; \
+	movdqa x1,		x4; \
+	pxor RNOT,		x0; \
+	pand x2,		x1; \
+	pxor x3,		x1; \
+	por x4,			x3; \
+	pxor x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	por x1,			x0; \
+	pand x0,		x2; \
+	pxor x4,		x0; \
+	pxor x3,		x4; \
+	pand x0,		x3; \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pxor x1,		x3; \
+	por x0,			x4; \
+	pxor x1,		x4;
+
+#define SI0(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pxor x0,		x1; \
+	por x1,			x3; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	pand x1,		x0; \
+	pxor x2,		x0; \
+	pand x3,		x2; \
+	pxor x4,		x3; \
+	pxor x3,		x2; \
+	pxor x3,		x1; \
+	pand x0,		x3; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x4;
+
+#define SI1(x0, x1, x2, x3, x4) \
+	pxor x3,		x1; \
+	movdqa x0,		x4; \
+	pxor x2,		x0; \
+	pxor RNOT,		x2; \
+	por x1,			x4; \
+	pxor x3,		x4; \
+	pand x1,		x3; \
+	pxor x2,		x1; \
+	pand x4,		x2; \
+	pxor x1,		x4; \
+	por x3,			x1; \
+	pxor x0,		x3; \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x4,		x2; \
+	pxor x0,		x1; \
+	pxor x1,		x4;
+
+#define SI2(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x3,		x4; \
+	pxor RNOT,		x3; \
+	por x2,			x3; \
+	pxor x4,		x2; \
+	pxor x0,		x4; \
+	pxor x1,		x3; \
+	por x2,			x1; \
+	pxor x0,		x2; \
+	pxor x4,		x1; \
+	por x3,			x4; \
+	pxor x3,		x2; \
+	pxor x2,		x4; \
+	pand x1,		x2; \
+	pxor x3,		x2; \
+	pxor x4,		x3; \
+	pxor x0,		x4;
+
+#define SI3(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x1,		x4; \
+	pand x2,		x1; \
+	pxor x0,		x1; \
+	por x4,			x0; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	por x1,			x3; \
+	pxor x2,		x1; \
+	pxor x3,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x4; \
+	pxor x0,		x3; \
+	pxor x1,		x0;
+
+#define SI4(x0, x1, x2, x3, x4) \
+	pxor x3,		x2; \
+	movdqa x0,		x4; \
+	pand x1,		x0; \
+	pxor x2,		x0; \
+	por x3,			x2; \
+	pxor RNOT,		x4; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pand x4,		x2; \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x3,		x0; \
+	pand x2,		x3; \
+	pxor x3,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x1,		x4; \
+	pxor x3,		x0;
+
+#define SI5(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x2,			x1; \
+	pxor x4,		x2; \
+	pxor x3,		x1; \
+	pand x4,		x3; \
+	pxor x3,		x2; \
+	por x0,			x3; \
+	pxor RNOT,		x0; \
+	pxor x2,		x3; \
+	por x0,			x2; \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pand x0,		x4; \
+	pxor x1,		x0; \
+	pxor x3,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x4,		x2; \
+	pxor x3,		x4;
+
+#define SI6(x0, x1, x2, x3, x4) \
+	pxor x2,		x0; \
+	movdqa x0,		x4; \
+	pand x3,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x1,		x3; \
+	por x4,			x2; \
+	pxor x3,		x2; \
+	pand x0,		x3; \
+	pxor RNOT,		x0; \
+	pxor x1,		x3; \
+	pand x2,		x1; \
+	pxor x0,		x4; \
+	pxor x4,		x3; \
+	pxor x2,		x4; \
+	pxor x1,		x0; \
+	pxor x0,		x2;
+
+#define SI7(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x2,		x0; \
+	por x4,			x2; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	por x3,			x1; \
+	pxor x0,		x4; \
+	pand x2,		x0; \
+	pxor x1,		x0; \
+	pand x2,		x1; \
+	pxor x2,		x3; \
+	pxor x3,		x4; \
+	pand x3,		x2; \
+	por x0,			x3; \
+	pxor x4,		x1; \
+	pxor x4,		x3; \
+	pand x0,		x4; \
+	pxor x2,		x4;
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2, t3) \
+	movdqa x2,		t3; \
+	movdqa x0,		t1; \
+	unpcklps x3,		t3; \
+	movdqa x0,		t2; \
+	unpcklps x1,		t1; \
+	unpckhps x1,		t2; \
+	movdqa t3,		x1; \
+	unpckhps x3,		x2; \
+	movdqa t1,		x0; \
+	movhlps t1,		x1; \
+	movdqa t2,		t1; \
+	movlhps t3,		x0; \
+	movlhps x2,		t1; \
+	movhlps t2,		x2; \
+	movdqa x2,		x3; \
+	movdqa t1,		x2;
+
+#define read_blocks(in, x0, x1, x2, x3, t0, t1, t2) \
+	movdqu (0*4*4)(in),	x0; \
+	movdqu (1*4*4)(in),	x1; \
+	movdqu (2*4*4)(in),	x2; \
+	movdqu (3*4*4)(in),	x3; \
+	\
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define write_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu x0, (0*4*4)(out); \
+	movdqu x1, (1*4*4)(out); \
+	movdqu x2, (2*4*4)(out); \
+	movdqu x3, (3*4*4)(out);
+
+#define xor_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu (0*4*4)(out),	t0; \
+	pxor t0,		x0; \
+	movdqu x0,		(0*4*4)(out); \
+	movdqu (1*4*4)(out),	t0; \
+	pxor t0,		x1; \
+	movdqu x1,		(1*4*4)(out); \
+	movdqu (2*4*4)(out),	t0; \
+	pxor t0,		x2; \
+	movdqu x2,		(2*4*4)(out); \
+	movdqu (3*4*4)(out),	t0; \
+	pxor t0,		x3; \
+	movdqu x3,		(3*4*4)(out);
+
+.align 8
+.global __serpent_enc_blk_4way
+.type   __serpent_enc_blk_4way,@function;
+
+__serpent_enc_blk_4way:
+	/* input:
+	 *	arg_ctx(%esp): ctx, CTX
+	 *	arg_dst(%esp): dst
+	 *	arg_src(%esp): src
+	 *	arg_xor(%esp): bool, if true: xor output
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	movl arg_ctx(%esp), CTX;
+
+	movl arg_src(%esp), %eax;
+	read_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+					 K(RA, RB, RC, RD, RE, 0);
+	S0(RA, RB, RC, RD, RE);		LK(RC, RB, RD, RA, RE, 1);
+	S1(RC, RB, RD, RA, RE);		LK(RE, RD, RA, RC, RB, 2);
+	S2(RE, RD, RA, RC, RB);		LK(RB, RD, RE, RC, RA, 3);
+	S3(RB, RD, RE, RC, RA);		LK(RC, RA, RD, RB, RE, 4);
+	S4(RC, RA, RD, RB, RE);		LK(RA, RD, RB, RE, RC, 5);
+	S5(RA, RD, RB, RE, RC);		LK(RC, RA, RD, RE, RB, 6);
+	S6(RC, RA, RD, RE, RB);		LK(RD, RB, RA, RE, RC, 7);
+	S7(RD, RB, RA, RE, RC);		LK(RC, RA, RE, RD, RB, 8);
+	S0(RC, RA, RE, RD, RB);		LK(RE, RA, RD, RC, RB, 9);
+	S1(RE, RA, RD, RC, RB);		LK(RB, RD, RC, RE, RA, 10);
+	S2(RB, RD, RC, RE, RA);		LK(RA, RD, RB, RE, RC, 11);
+	S3(RA, RD, RB, RE, RC);		LK(RE, RC, RD, RA, RB, 12);
+	S4(RE, RC, RD, RA, RB);		LK(RC, RD, RA, RB, RE, 13);
+	S5(RC, RD, RA, RB, RE);		LK(RE, RC, RD, RB, RA, 14);
+	S6(RE, RC, RD, RB, RA);		LK(RD, RA, RC, RB, RE, 15);
+	S7(RD, RA, RC, RB, RE);		LK(RE, RC, RB, RD, RA, 16);
+	S0(RE, RC, RB, RD, RA);		LK(RB, RC, RD, RE, RA, 17);
+	S1(RB, RC, RD, RE, RA);		LK(RA, RD, RE, RB, RC, 18);
+	S2(RA, RD, RE, RB, RC);		LK(RC, RD, RA, RB, RE, 19);
+	S3(RC, RD, RA, RB, RE);		LK(RB, RE, RD, RC, RA, 20);
+	S4(RB, RE, RD, RC, RA);		LK(RE, RD, RC, RA, RB, 21);
+	S5(RE, RD, RC, RA, RB);		LK(RB, RE, RD, RA, RC, 22);
+	S6(RB, RE, RD, RA, RC);		LK(RD, RC, RE, RA, RB, 23);
+	S7(RD, RC, RE, RA, RB);		LK(RB, RE, RA, RD, RC, 24);
+	S0(RB, RE, RA, RD, RC);		LK(RA, RE, RD, RB, RC, 25);
+	S1(RA, RE, RD, RB, RC);		LK(RC, RD, RB, RA, RE, 26);
+	S2(RC, RD, RB, RA, RE);		LK(RE, RD, RC, RA, RB, 27);
+	S3(RE, RD, RC, RA, RB);		LK(RA, RB, RD, RE, RC, 28);
+	S4(RA, RB, RD, RE, RC);		LK(RB, RD, RE, RC, RA, 29);
+	S5(RB, RD, RE, RC, RA);		LK(RA, RB, RD, RC, RE, 30);
+	S6(RA, RB, RD, RC, RE);		LK(RD, RE, RB, RC, RA, 31);
+	S7(RD, RE, RB, RC, RA);		 K(RA, RB, RC, RD, RE, 32);
+
+	movl arg_dst(%esp), %eax;
+
+	cmpb $0, arg_xor(%esp);
+	jnz __enc_xor4;
+
+	write_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+	ret;
+
+__enc_xor4:
+	xor_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+	ret;
+
+.align 8
+.global serpent_dec_blk_4way
+.type   serpent_dec_blk_4way,@function;
+
+serpent_dec_blk_4way:
+	/* input:
+	 *	arg_ctx(%esp): ctx, CTX
+	 *	arg_dst(%esp): dst
+	 *	arg_src(%esp): src
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	movl arg_ctx(%esp), CTX;
+
+	movl arg_src(%esp), %eax;
+	read_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+					 K(RA, RB, RC, RD, RE, 32);
+	SI7(RA, RB, RC, RD, RE);	KL(RB, RD, RA, RE, RC, 31);
+	SI6(RB, RD, RA, RE, RC);	KL(RA, RC, RE, RB, RD, 30);
+	SI5(RA, RC, RE, RB, RD);	KL(RC, RD, RA, RE, RB, 29);
+	SI4(RC, RD, RA, RE, RB);	KL(RC, RA, RB, RE, RD, 28);
+	SI3(RC, RA, RB, RE, RD);	KL(RB, RC, RD, RE, RA, 27);
+	SI2(RB, RC, RD, RE, RA);	KL(RC, RA, RE, RD, RB, 26);
+	SI1(RC, RA, RE, RD, RB);	KL(RB, RA, RE, RD, RC, 25);
+	SI0(RB, RA, RE, RD, RC);	KL(RE, RC, RA, RB, RD, 24);
+	SI7(RE, RC, RA, RB, RD);	KL(RC, RB, RE, RD, RA, 23);
+	SI6(RC, RB, RE, RD, RA);	KL(RE, RA, RD, RC, RB, 22);
+	SI5(RE, RA, RD, RC, RB);	KL(RA, RB, RE, RD, RC, 21);
+	SI4(RA, RB, RE, RD, RC);	KL(RA, RE, RC, RD, RB, 20);
+	SI3(RA, RE, RC, RD, RB);	KL(RC, RA, RB, RD, RE, 19);
+	SI2(RC, RA, RB, RD, RE);	KL(RA, RE, RD, RB, RC, 18);
+	SI1(RA, RE, RD, RB, RC);	KL(RC, RE, RD, RB, RA, 17);
+	SI0(RC, RE, RD, RB, RA);	KL(RD, RA, RE, RC, RB, 16);
+	SI7(RD, RA, RE, RC, RB);	KL(RA, RC, RD, RB, RE, 15);
+	SI6(RA, RC, RD, RB, RE);	KL(RD, RE, RB, RA, RC, 14);
+	SI5(RD, RE, RB, RA, RC);	KL(RE, RC, RD, RB, RA, 13);
+	SI4(RE, RC, RD, RB, RA);	KL(RE, RD, RA, RB, RC, 12);
+	SI3(RE, RD, RA, RB, RC);	KL(RA, RE, RC, RB, RD, 11);
+	SI2(RA, RE, RC, RB, RD);	KL(RE, RD, RB, RC, RA, 10);
+	SI1(RE, RD, RB, RC, RA);	KL(RA, RD, RB, RC, RE, 9);
+	SI0(RA, RD, RB, RC, RE);	KL(RB, RE, RD, RA, RC, 8);
+	SI7(RB, RE, RD, RA, RC);	KL(RE, RA, RB, RC, RD, 7);
+	SI6(RE, RA, RB, RC, RD);	KL(RB, RD, RC, RE, RA, 6);
+	SI5(RB, RD, RC, RE, RA);	KL(RD, RA, RB, RC, RE, 5);
+	SI4(RD, RA, RB, RC, RE);	KL(RD, RB, RE, RC, RA, 4);
+	SI3(RD, RB, RE, RC, RA);	KL(RE, RD, RA, RC, RB, 3);
+	SI2(RE, RD, RA, RC, RB);	KL(RD, RB, RC, RA, RE, 2);
+	SI1(RD, RB, RC, RA, RE);	KL(RE, RB, RC, RA, RD, 1);
+	SI0(RE, RB, RC, RA, RD);	 K(RC, RD, RB, RE, RA, 0);
+
+	movl arg_dst(%esp), %eax;
+	write_blocks(%eax, RC, RD, RB, RE, RT0, RT1, RA);
+
+	ret;
diff --git a/arch/x86/include/asm/serpent.h b/arch/x86/include/asm/serpent.h
index b7fd3b5..d3ef63f 100644
--- a/arch/x86/include/asm/serpent.h
+++ b/arch/x86/include/asm/serpent.h
@@ -4,6 +4,35 @@
 #include <linux/crypto.h>
 #include <crypto/serpent.h>
 
+#ifdef CONFIG_X86_32
+
+#define SERPENT_PARALLEL_BLOCKS 4
+
+asmlinkage void __serpent_enc_blk_4way(struct serpent_ctx *ctx, u8 *dst,
+				       const u8 *src, bool xor);
+asmlinkage void serpent_dec_blk_4way(struct serpent_ctx *ctx, u8 *dst,
+				     const u8 *src);
+
+static inline void serpent_enc_blk_xway(struct serpent_ctx *ctx, u8 *dst,
+					const u8 *src)
+{
+	__serpent_enc_blk_4way(ctx, dst, src, false);
+}
+
+static inline void serpent_enc_blk_xway_xor(struct serpent_ctx *ctx, u8 *dst,
+					    const u8 *src)
+{
+	__serpent_enc_blk_4way(ctx, dst, src, true);
+}
+
+static inline void serpent_dec_blk_xway(struct serpent_ctx *ctx, u8 *dst,
+					const u8 *src)
+{
+	serpent_dec_blk_4way(ctx, dst, src);
+}
+
+#else
+
 #define SERPENT_PARALLEL_BLOCKS 8
 
 asmlinkage void __serpent_enc_blk_8way(struct serpent_ctx *ctx, u8 *dst,
@@ -30,3 +59,5 @@ static inline void serpent_dec_blk_xway(struct serpent_ctx *ctx, u8 *dst,
 }
 
 #endif
+
+#endif
diff --git a/crypto/Kconfig b/crypto/Kconfig
index 2df61e4..16e0c13 100644
--- a/crypto/Kconfig
+++ b/crypto/Kconfig
@@ -783,6 +783,23 @@ config CRYPTO_SERPENT_SSE2_X86_64
 	  See also:
 	  <http://www.cl.cam.ac.uk/~rja14/serpent.html>
 
+config CRYPTO_SERPENT_SSE2_586
+	tristate "Serpent cipher algorithm (i586/SSE2)"
+	depends on X86 && !64BIT
+	select CRYPTO_ALGAPI
+	select CRYPTO_SERPENT
+	help
+	  Serpent cipher algorithm, by Anderson, Biham & Knudsen.
+
+	  Keys are allowed to be from 0 to 256 bits in length, in steps
+	  of 8 bits.
+
+	  This module provides Serpent cipher algorithm that processes four
+	  blocks parallel using SSE2 instruction set.
+
+	  See also:
+	  <http://www.cl.cam.ac.uk/~rja14/serpent.html>
+
 config CRYPTO_TEA
 	tristate "TEA, XTEA and XETA cipher algorithms"
 	select CRYPTO_ALGAPI