linux/arch/x86/crypto/cast6-avx-x86_64-asm_64.S

/* SPDX-License-Identifier: GPL-2.0-or-later */
/*
 * Cast6 Cipher 8-way parallel algorithm (AVX/x86_64)
 *
 * Copyright (C) 2012 Johannes Goetzfried
 *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
 *
 * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
 */

#include <linux/linkage.h>
#include <asm/frame.h>
#include "glue_helper-asm-avx.S"

.file "cast6-avx-x86_64-asm_64.S"

.extern cast_s1
.extern cast_s2
.extern cast_s3
.extern cast_s4

/* structure of crypto context */
#define km	0
#define kr	(12*4*4)

/* s-boxes */
#define s1	cast_s1
#define s2	cast_s2
#define s3	cast_s3
#define s4	cast_s4

/**********************************************************************
  8-way AVX cast6
 **********************************************************************/
#define CTX %r15

#define RA1 %xmm0
#define RB1 %xmm1
#define RC1 %xmm2
#define RD1 %xmm3

#define RA2 %xmm4
#define RB2 %xmm5
#define RC2 %xmm6
#define RD2 %xmm7

#define RX  %xmm8

#define RKM  %xmm9
#define RKR  %xmm10
#define RKRF %xmm11
#define RKRR %xmm12
#define R32  %xmm13
#define R1ST %xmm14

#define RTMP %xmm15

#define RID1  %rdi
#define RID1d %edi
#define RID2  %rsi
#define RID2d %esi

#define RGI1   %rdx
#define RGI1bl %dl
#define RGI1bh %dh
#define RGI2   %rcx
#define RGI2bl %cl
#define RGI2bh %ch

#define RGI3   %rax
#define RGI3bl %al
#define RGI3bh %ah
#define RGI4   %rbx
#define RGI4bl %bl
#define RGI4bh %bh

#define RFS1  %r8
#define RFS1d %r8d
#define RFS2  %r9
#define RFS2d %r9d
#define RFS3  %r10
#define RFS3d %r10d


#define lookup_32bit(src, dst, op1, op2, op3, interleave_op, il_reg) \
	movzbl		src ## bh,     RID1d;    \
	movzbl		src ## bl,     RID2d;    \
	shrq $16,	src;                     \
	movl		s1(, RID1, 4), dst ## d; \
	op1		s2(, RID2, 4), dst ## d; \
	movzbl		src ## bh,     RID1d;    \
	movzbl		src ## bl,     RID2d;    \
	interleave_op(il_reg);			 \
	op2		s3(, RID1, 4), dst ## d; \
	op3		s4(, RID2, 4), dst ## d;

#define dummy(d) /* do nothing */

#define shr_next(reg) \
	shrq $16,	reg;

#define F_head(a, x, gi1, gi2, op0) \
	op0	a,	RKM,  x;                 \
	vpslld	RKRF,	x,    RTMP;              \
	vpsrld	RKRR,	x,    x;                 \
	vpor	RTMP,	x,    x;                 \
	\
	vmovq		x,    gi1;               \
	vpextrq $1,	x,    gi2;

#define F_tail(a, x, gi1, gi2, op1, op2, op3) \
	lookup_32bit(##gi1, RFS1, op1, op2, op3, shr_next, ##gi1); \
	lookup_32bit(##gi2, RFS3, op1, op2, op3, shr_next, ##gi2); \
	\
	lookup_32bit(##gi1, RFS2, op1, op2, op3, dummy, none);     \
	shlq $32,	RFS2;                                      \
	orq		RFS1, RFS2;                                \
	lookup_32bit(##gi2, RFS1, op1, op2, op3, dummy, none);     \
	shlq $32,	RFS1;                                      \
	orq		RFS1, RFS3;                                \
	\
	vmovq		RFS2, x;                                   \
	vpinsrq $1,	RFS3, x, x;

#define F_2(a1, b1, a2, b2, op0, op1, op2, op3) \
	F_head(b1, RX, RGI1, RGI2, op0);              \
	F_head(b2, RX, RGI3, RGI4, op0);              \
	\
	F_tail(b1, RX, RGI1, RGI2, op1, op2, op3);    \
	F_tail(b2, RTMP, RGI3, RGI4, op1, op2, op3);  \
	\
	vpxor		a1, RX,   a1;                 \
	vpxor		a2, RTMP, a2;

#define F1_2(a1, b1, a2, b2) \
	F_2(a1, b1, a2, b2, vpaddd, xorl, subl, addl)
#define F2_2(a1, b1, a2, b2) \
	F_2(a1, b1, a2, b2, vpxor, subl, addl, xorl)
#define F3_2(a1, b1, a2, b2) \
	F_2(a1, b1, a2, b2, vpsubd, addl, xorl, subl)

#define qop(in, out, f) \
	F ## f ## _2(out ## 1, in ## 1, out ## 2, in ## 2);

#define get_round_keys(nn) \
	vbroadcastss	(km+(4*(nn)))(CTX), RKM;        \
	vpand		R1ST,               RKR,  RKRF; \
	vpsubq		RKRF,               R32,  RKRR; \
	vpsrldq $1,	RKR,                RKR;

#define Q(n) \
	get_round_keys(4*n+0); \
	qop(RD, RC, 1);        \
	\
	get_round_keys(4*n+1); \
	qop(RC, RB, 2);        \
	\
	get_round_keys(4*n+2); \
	qop(RB, RA, 3);        \
	\
	get_round_keys(4*n+3); \
	qop(RA, RD, 1);

#define QBAR(n) \
	get_round_keys(4*n+3); \
	qop(RA, RD, 1);        \
	\
	get_round_keys(4*n+2); \
	qop(RB, RA, 3);        \
	\
	get_round_keys(4*n+1); \
	qop(RC, RB, 2);        \
	\
	get_round_keys(4*n+0); \
	qop(RD, RC, 1);

#define shuffle(mask) \
	vpshufb		mask,            RKR, RKR;

#define preload_rkr(n, do_mask, mask) \
	vbroadcastss	.L16_mask,                RKR;      \
	/* add 16-bit rotation to key rotations (mod 32) */ \
	vpxor		(kr+n*16)(CTX),           RKR, RKR; \
	do_mask(mask);

#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
	vpunpckldq		x1, x0, t0; \
	vpunpckhdq		x1, x0, t2; \
	vpunpckldq		x3, x2, t1; \
	vpunpckhdq		x3, x2, x3; \
	\
	vpunpcklqdq		t1, t0, x0; \
	vpunpckhqdq		t1, t0, x1; \
	vpunpcklqdq		x3, t2, x2; \
	vpunpckhqdq		x3, t2, x3;

#define inpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
	vpshufb rmask, x0,	x0; \
	vpshufb rmask, x1,	x1; \
	vpshufb rmask, x2,	x2; \
	vpshufb rmask, x3,	x3; \
	\
	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)

#define outunpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
	\
	vpshufb rmask,		x0, x0;       \
	vpshufb rmask,		x1, x1;       \
	vpshufb rmask,		x2, x2;       \
	vpshufb rmask,		x3, x3;

.section	.rodata.cst16, "aM", @progbits, 16
.align 16
.Lxts_gf128mul_and_shl1_mask:
	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
.Lbswap_mask:
	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
.Lbswap128_mask:
	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
.Lrkr_enc_Q_Q_QBAR_QBAR:
	.byte 0, 1, 2, 3, 4, 5, 6, 7, 11, 10, 9, 8, 15, 14, 13, 12
.Lrkr_enc_QBAR_QBAR_QBAR_QBAR:
	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
.Lrkr_dec_Q_Q_Q_Q:
	.byte 12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3
.Lrkr_dec_Q_Q_QBAR_QBAR:
	.byte 12, 13, 14, 15, 8, 9, 10, 11, 7, 6, 5, 4, 3, 2, 1, 0
.Lrkr_dec_QBAR_QBAR_QBAR_QBAR:
	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0

.section	.rodata.cst4.L16_mask, "aM", @progbits, 4
.align 4
.L16_mask:
	.byte 16, 16, 16, 16

.section	.rodata.cst4.L32_mask, "aM", @progbits, 4
.align 4
.L32_mask:
	.byte 32, 0, 0, 0

.section	.rodata.cst4.first_mask, "aM", @progbits, 4
.align 4
.Lfirst_mask:
	.byte 0x1f, 0, 0, 0

.text

.align 8
__cast6_enc_blk8:
	/* input:
	 *	%rdi: ctx
	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
	 * output:
	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
	 */

	pushq %r15;
	pushq %rbx;

	movq %rdi, CTX;

	vmovdqa .Lbswap_mask, RKM;
	vmovd .Lfirst_mask, R1ST;
	vmovd .L32_mask, R32;

	inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
	inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);

	preload_rkr(0, dummy, none);
	Q(0);
	Q(1);
	Q(2);
	Q(3);
	preload_rkr(1, shuffle, .Lrkr_enc_Q_Q_QBAR_QBAR);
	Q(4);
	Q(5);
	QBAR(6);
	QBAR(7);
	preload_rkr(2, shuffle, .Lrkr_enc_QBAR_QBAR_QBAR_QBAR);
	QBAR(8);
	QBAR(9);
	QBAR(10);
	QBAR(11);

	popq %rbx;
	popq %r15;

	vmovdqa .Lbswap_mask, RKM;

	outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
	outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);

	ret;
ENDPROC(__cast6_enc_blk8)

.align 8
__cast6_dec_blk8:
	/* input:
	 *	%rdi: ctx
	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
	 * output:
	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: decrypted blocks
	 */

	pushq %r15;
	pushq %rbx;

	movq %rdi, CTX;

	vmovdqa .Lbswap_mask, RKM;
	vmovd .Lfirst_mask, R1ST;
	vmovd .L32_mask, R32;

	inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
	inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);

	preload_rkr(2, shuffle, .Lrkr_dec_Q_Q_Q_Q);
	Q(11);
	Q(10);
	Q(9);
	Q(8);
	preload_rkr(1, shuffle, .Lrkr_dec_Q_Q_QBAR_QBAR);
	Q(7);
	Q(6);
	QBAR(5);
	QBAR(4);
	preload_rkr(0, shuffle, .Lrkr_dec_QBAR_QBAR_QBAR_QBAR);
	QBAR(3);
	QBAR(2);
	QBAR(1);
	QBAR(0);

	popq %rbx;
	popq %r15;

	vmovdqa .Lbswap_mask, RKM;
	outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
	outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);

	ret;
ENDPROC(__cast6_dec_blk8)

ENTRY(cast6_ecb_enc_8way)
	/* input:
	 *	%rdi: ctx
	 *	%rsi: dst
	 *	%rdx: src
	 */
	FRAME_BEGIN
	pushq %r15;

	movq %rdi, CTX;
	movq %rsi, %r11;

	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	call __cast6_enc_blk8;

	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	FRAME_END
	ret;
ENDPROC(cast6_ecb_enc_8way)

ENTRY(cast6_ecb_dec_8way)
	/* input:
	 *	%rdi: ctx
	 *	%rsi: dst
	 *	%rdx: src
	 */
	FRAME_BEGIN
	pushq %r15;

	movq %rdi, CTX;
	movq %rsi, %r11;

	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	call __cast6_dec_blk8;

	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	FRAME_END
	ret;
ENDPROC(cast6_ecb_dec_8way)

ENTRY(cast6_cbc_dec_8way)
	/* input:
	 *	%rdi: ctx
	 *	%rsi: dst
	 *	%rdx: src
	 */
	FRAME_BEGIN
	pushq %r12;
	pushq %r15;

	movq %rdi, CTX;
	movq %rsi, %r11;
	movq %rdx, %r12;

	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	call __cast6_dec_blk8;

	store_cbc_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	popq %r12;
	FRAME_END
	ret;
ENDPROC(cast6_cbc_dec_8way)

ENTRY(cast6_ctr_8way)
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst
	 *	%rdx: src
	 *	%rcx: iv (little endian, 128bit)
	 */
	FRAME_BEGIN
	pushq %r12;
	pushq %r15

	movq %rdi, CTX;
	movq %rsi, %r11;
	movq %rdx, %r12;

	load_ctr_8way(%rcx, .Lbswap128_mask, RA1, RB1, RC1, RD1, RA2, RB2, RC2,
		      RD2, RX, RKR, RKM);

	call __cast6_enc_blk8;

	store_ctr_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	popq %r12;
	FRAME_END
	ret;
ENDPROC(cast6_ctr_8way)

ENTRY(cast6_xts_enc_8way)
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst
	 *	%rdx: src
	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
	 */
	FRAME_BEGIN
	pushq %r15;

	movq %rdi, CTX
	movq %rsi, %r11;

	/* regs <= src, dst <= IVs, regs <= regs xor IVs */
	load_xts_8way(%rcx, %rdx, %rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2,
		      RX, RKR, RKM, .Lxts_gf128mul_and_shl1_mask);

	call __cast6_enc_blk8;

	/* dst <= regs xor IVs(in dst) */
	store_xts_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	FRAME_END
	ret;
ENDPROC(cast6_xts_enc_8way)

ENTRY(cast6_xts_dec_8way)
	/* input:
	 *	%rdi: ctx, CTX
	 *	%rsi: dst
	 *	%rdx: src
	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
	 */
	FRAME_BEGIN
	pushq %r15;

	movq %rdi, CTX
	movq %rsi, %r11;

	/* regs <= src, dst <= IVs, regs <= regs xor IVs */
	load_xts_8way(%rcx, %rdx, %rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2,
		      RX, RKR, RKM, .Lxts_gf128mul_and_shl1_mask);

	call __cast6_dec_blk8;

	/* dst <= regs xor IVs(in dst) */
	store_xts_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);

	popq %r15;
	FRAME_END
	ret;
ENDPROC(cast6_xts_dec_8way)
-												treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 156

Based on 1 normalized pattern(s):

  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

extracted by the scancode license scanner the SPDX license identifier

  GPL-2.0-or-later

has been chosen to replace the boilerplate/reference in 1334 file(s).

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Reviewed-by: Richard Fontana <rfontana@redhat.com>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190527070033.113240726@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

											
										
										
											2019-05-27 06:55:05 +00:00
+								/* SPDX-License-Identifier: GPL-2.0-or-later */
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								/*
 								 * Cast6 Cipher 8-way parallel algorithm (AVX/x86_64)
 								 *
 								 * Copyright (C) 2012 Johannes Goetzfried
 								 *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
 								 *
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+								 * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								 */
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								#include <linux/linkage.h>
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+								#include <asm/frame.h>
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								#include "glue_helper-asm-avx.S"
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								.file "cast6-avx-x86_64-asm_64.S"
-												crypto: cast5/cast6 - move lookup tables to shared module

CAST5 and CAST6 both use same lookup tables, which can be moved shared module
'cast_common'.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-11-13 09:43:14 +00:00
+								.extern cast_s1
 								.extern cast_s2
 								.extern cast_s3
 								.extern cast_s4
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								/* structure of crypto context */
 								#define km	0
 								#define kr	(12*4*4)
 								/* s-boxes */
-												crypto: cast5/cast6 - move lookup tables to shared module

CAST5 and CAST6 both use same lookup tables, which can be moved shared module
'cast_common'.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-11-13 09:43:14 +00:00
+								#define s1	cast_s1
 								#define s2	cast_s2
 								#define s3	cast_s3
 								#define s4	cast_s4
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								/**********************************************************************
 -way AVX cast6
 								 **********************************************************************/
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+								#define CTX %r15
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define RA1 %xmm0
 								#define RB1 %xmm1
 								#define RC1 %xmm2
 								#define RD1 %xmm3
 								#define RA2 %xmm4
 								#define RB2 %xmm5
 								#define RC2 %xmm6
 								#define RD2 %xmm7
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define RX  %xmm8
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define RKM  %xmm9
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define RKR  %xmm10
 								#define RKRF %xmm11
 								#define RKRR %xmm12
 								#define R32  %xmm13
 								#define R1ST %xmm14
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define RTMP %xmm15
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+								#define RID1  %rdi
 								#define RID1d %edi
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define RID2  %rsi
 								#define RID2d %esi
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define RGI1   %rdx
 								#define RGI1bl %dl
 								#define RGI1bh %dh
 								#define RGI2   %rcx
 								#define RGI2bl %cl
 								#define RGI2bh %ch
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define RGI3   %rax
 								#define RGI3bl %al
 								#define RGI3bh %ah
 								#define RGI4   %rbx
 								#define RGI4bl %bl
 								#define RGI4bh %bh
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								#define RFS1  %r8
 								#define RFS1d %r8d
 								#define RFS2  %r9
 								#define RFS2d %r9d
 								#define RFS3  %r10
 								#define RFS3d %r10d
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define lookup_32bit(src, dst, op1, op2, op3, interleave_op, il_reg) \
 									movzbl		src ## bh,     RID1d;    \
 									movzbl		src ## bl,     RID2d;    \
 									shrq $16,	src;                     \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									movl		s1(, RID1, 4), dst ## d; \
 									op1		s2(, RID2, 4), dst ## d; \
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									movzbl		src ## bh,     RID1d;    \
 									movzbl		src ## bl,     RID2d;    \
 									interleave_op(il_reg);			 \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									op2		s3(, RID1, 4), dst ## d; \
 									op3		s4(, RID2, 4), dst ## d;
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define dummy(d) /* do nothing */
 								#define shr_next(reg) \
 									shrq $16,	reg;
 								#define F_head(a, x, gi1, gi2, op0) \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									op0	a,	RKM,  x;                 \
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vpslld	RKRF,	x,    RTMP;              \
 									vpsrld	RKRR,	x,    x;                 \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									vpor	RTMP,	x,    x;                 \
 									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovq		x,    gi1;               \
 									vpextrq $1,	x,    gi2;
 								#define F_tail(a, x, gi1, gi2, op1, op2, op3) \
 									lookup_32bit(##gi1, RFS1, op1, op2, op3, shr_next, ##gi1); \
 									lookup_32bit(##gi2, RFS3, op1, op2, op3, shr_next, ##gi2); \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									lookup_32bit(##gi1, RFS2, op1, op2, op3, dummy, none);     \
 									shlq $32,	RFS2;                                      \
 									orq		RFS1, RFS2;                                \
 									lookup_32bit(##gi2, RFS1, op1, op2, op3, dummy, none);     \
 									shlq $32,	RFS1;                                      \
 									orq		RFS1, RFS3;                                \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovq		RFS2, x;                                   \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									vpinsrq $1,	RFS3, x, x;
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define F_2(a1, b1, a2, b2, op0, op1, op2, op3) \
 									F_head(b1, RX, RGI1, RGI2, op0);              \
 									F_head(b2, RX, RGI3, RGI4, op0);              \
 									\
 									F_tail(b1, RX, RGI1, RGI2, op1, op2, op3);    \
 									F_tail(b2, RTMP, RGI3, RGI4, op1, op2, op3);  \
 									\
 									vpxor		a1, RX,   a1;                 \
 									vpxor		a2, RTMP, a2;
 								#define F1_2(a1, b1, a2, b2) \
 									F_2(a1, b1, a2, b2, vpaddd, xorl, subl, addl)
 								#define F2_2(a1, b1, a2, b2) \
 									F_2(a1, b1, a2, b2, vpxor, subl, addl, xorl)
 								#define F3_2(a1, b1, a2, b2) \
 									F_2(a1, b1, a2, b2, vpsubd, addl, xorl, subl)
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define qop(in, out, f) \
 									F ## f ## _2(out ## 1, in ## 1, out ## 2, in ## 2);
 								#define get_round_keys(nn) \
 									vbroadcastss	(km+(4*(nn)))(CTX), RKM;        \
 									vpand		R1ST,               RKR,  RKRF; \
 									vpsubq		RKRF,               R32,  RKRR; \
 									vpsrldq $1,	RKR,                RKR;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define Q(n) \
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+0); \
 									qop(RD, RC, 1);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+1); \
 									qop(RC, RB, 2);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+2); \
 									qop(RB, RA, 3);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+3); \
 									qop(RA, RD, 1);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define QBAR(n) \
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+3); \
 									qop(RA, RD, 1);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+2); \
 									qop(RB, RA, 3);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+1); \
 									qop(RC, RB, 2);        \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									get_round_keys(4*n+0); \
 									qop(RD, RC, 1);
 								#define shuffle(mask) \
 									vpshufb		mask,            RKR, RKR;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								#define preload_rkr(n, do_mask, mask) \
 									vbroadcastss	.L16_mask,                RKR;      \
 									/* add 16-bit rotation to key rotations (mod 32) */ \
 									vpxor		(kr+n*16)(CTX),           RKR, RKR; \
 									do_mask(mask);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 								#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
 									vpunpckldq		x1, x0, t0; \
 									vpunpckhdq		x1, x0, t2; \
 									vpunpckldq		x3, x2, t1; \
 									vpunpckhdq		x3, x2, x3; \
 									\
 									vpunpcklqdq		t1, t0, x0; \
 									vpunpckhqdq		t1, t0, x1; \
 									vpunpcklqdq		x3, t2, x2; \
 									vpunpckhqdq		x3, t2, x3;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								#define inpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vpshufb rmask, x0,	x0; \
 									vpshufb rmask, x1,	x1; \
 									vpshufb rmask, x2,	x2; \
 									vpshufb rmask, x3,	x3; \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									\
 									transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								#define outunpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
 									\
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vpshufb rmask,		x0, x0;       \
 									vpshufb rmask,		x1, x1;       \
 									vpshufb rmask,		x2, x2;       \
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									vpshufb rmask,		x3, x3;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: x86 - make constants readonly, allow linker to merge them

A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.

Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.

There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".

GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:

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

It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.

When this is not the case, use non-mergeable section:

	.section .rodata[.VAR_NAME], "a", @progbits

This reduces .data by ~15 kbytes:

    text    data     bss     dec      hex filename
11097415 2705840 2630712 16433967  fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479  fac147 vmlinux.o

Merged objects are visible in System.map:

ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK   <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512

Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-01-19 21:33:04 +00:00
+								.section	.rodata.cst16, "aM", @progbits, 16
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								.align 16
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+								.Lxts_gf128mul_and_shl1_mask:
 									.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								.Lbswap_mask:
 									.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								.Lbswap128_mask:
 									.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								.Lrkr_enc_Q_Q_QBAR_QBAR:
 									.byte 0, 1, 2, 3, 4, 5, 6, 7, 11, 10, 9, 8, 15, 14, 13, 12
 								.Lrkr_enc_QBAR_QBAR_QBAR_QBAR:
 									.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
 								.Lrkr_dec_Q_Q_Q_Q:
 									.byte 12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3
 								.Lrkr_dec_Q_Q_QBAR_QBAR:
 									.byte 12, 13, 14, 15, 8, 9, 10, 11, 7, 6, 5, 4, 3, 2, 1, 0
 								.Lrkr_dec_QBAR_QBAR_QBAR_QBAR:
 									.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
-												crypto: x86 - make constants readonly, allow linker to merge them

A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.

Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.

There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".

GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:

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

It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.

When this is not the case, use non-mergeable section:

	.section .rodata[.VAR_NAME], "a", @progbits

This reduces .data by ~15 kbytes:

    text    data     bss     dec      hex filename
11097415 2705840 2630712 16433967  fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479  fac147 vmlinux.o

Merged objects are visible in System.map:

ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK   <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512

Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-01-19 21:33:04 +00:00
 								.section	.rodata.cst4.L16_mask, "aM", @progbits, 4
 								.align 4
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								.L16_mask:
 									.byte 16, 16, 16, 16
-												crypto: x86 - make constants readonly, allow linker to merge them

A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.

Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.

There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".

GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:

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

It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.

When this is not the case, use non-mergeable section:

	.section .rodata[.VAR_NAME], "a", @progbits

This reduces .data by ~15 kbytes:

    text    data     bss     dec      hex filename
11097415 2705840 2630712 16433967  fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479  fac147 vmlinux.o

Merged objects are visible in System.map:

ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK   <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512

Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-01-19 21:33:04 +00:00
 								.section	.rodata.cst4.L32_mask, "aM", @progbits, 4
 								.align 4
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+								.L32_mask:
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									.byte 32, 0, 0, 0
-												crypto: x86 - make constants readonly, allow linker to merge them

A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.

Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.

There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".

GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:

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

It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.

When this is not the case, use non-mergeable section:

	.section .rodata[.VAR_NAME], "a", @progbits

This reduces .data by ~15 kbytes:

    text    data     bss     dec      hex filename
11097415 2705840 2630712 16433967  fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479  fac147 vmlinux.o

Merged objects are visible in System.map:

ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK   <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512

Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Josh Poimboeuf <jpoimboe@redhat.com>
CC: Xiaodong Liu <xiaodong.liu@intel.com>
CC: Megha Dey <megha.dey@intel.com>
CC: linux-crypto@vger.kernel.org
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-01-19 21:33:04 +00:00
 								.section	.rodata.cst4.first_mask, "aM", @progbits, 4
 								.align 4
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+								.Lfirst_mask:
 									.byte 0x1f, 0, 0, 0
 								.text
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								.align 8
 								__cast6_enc_blk8:
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									/* input:
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									 *	%rdi: ctx
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
 									 * output:
 									 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									 */
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									pushq %rbx;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovdqa .Lbswap_mask, RKM;
 									vmovd .Lfirst_mask, R1ST;
 									vmovd .L32_mask, R32;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
 									inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(0, dummy, none);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									Q(0);
 									Q(1);
 									Q(2);
 									Q(3);
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(1, shuffle, .Lrkr_enc_Q_Q_QBAR_QBAR);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									Q(4);
 									Q(5);
 									QBAR(6);
 									QBAR(7);
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(2, shuffle, .Lrkr_enc_QBAR_QBAR_QBAR_QBAR);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									QBAR(8);
 									QBAR(9);
 									QBAR(10);
 									QBAR(11);
 									popq %rbx;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovdqa .Lbswap_mask, RKM;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
 									outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
 									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(__cast6_enc_blk8)
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+								.align 8
 								__cast6_dec_blk8:
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									/* input:
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									 *	%rdi: ctx
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
 									 * output:
 									 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: decrypted blocks
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									 */
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									pushq %rbx;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovdqa .Lbswap_mask, RKM;
 									vmovd .Lfirst_mask, R1ST;
 									vmovd .L32_mask, R32;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
 									inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(2, shuffle, .Lrkr_dec_Q_Q_Q_Q);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									Q(11);
 									Q(10);
 									Q(9);
 									Q(8);
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(1, shuffle, .Lrkr_dec_Q_Q_QBAR_QBAR);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									Q(7);
 									Q(6);
 									QBAR(5);
 									QBAR(4);
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									preload_rkr(0, shuffle, .Lrkr_dec_QBAR_QBAR_QBAR_QBAR);
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									QBAR(3);
 									QBAR(2);
 									QBAR(1);
 									QBAR(0);
 									popq %rbx;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
-												crypto: cast6-avx - tune assembler code for more performance

Patch replaces 'movb' instructions with 'movzbl' to break false register
dependencies, interleaves instructions better for out-of-order scheduling
and merges constant 16-bit rotation with round-key variable rotation.

tcrypt ECB results:

Intel Core i5-2450M:

size    old-vs-new      new-vs-generic  old-vs-generic
        enc     dec     enc     dec     enc     dec
256     1.13x   1.19x   2.05x   2.17x   1.82x   1.82x
1k      1.18x   1.21x   2.26x   2.33x   1.93x   1.93x
8k      1.19x   1.19x   2.32x   2.33x   1.95x   1.95x

[v2]
 - Do instruction interleaving another way to avoid adding new FPU<=>CPU
   register moves as these cause performance drop on Bulldozer.
 - Improvements to round-key variable rotation handling.
 - Further interleaving improvements for better out-of-order scheduling.

Cc: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-08-28 11:24:54 +00:00
+									vmovdqa .Lbswap_mask, RKM;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
 									outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
 									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(__cast6_dec_blk8)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENTRY(cast6_ecb_enc_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									/* input:
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									 *	%rdi: ctx
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									 *	%rsi: dst
 									 *	%rdx: src
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									movq %rsi, %r11;
 									load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
 									call __cast6_enc_blk8;
 									store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(cast6_ecb_enc_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENTRY(cast6_ecb_dec_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									/* input:
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									 *	%rdi: ctx
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									 *	%rsi: dst
 									 *	%rdx: src
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									movq %rsi, %r11;
 									load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
 									call __cast6_dec_blk8;
 									store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(cast6_ecb_dec_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENTRY(cast6_cbc_dec_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									/* input:
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									 *	%rdi: ctx
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									 *	%rsi: dst
 									 *	%rdx: src
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									pushq %r12;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									movq %rsi, %r11;
 									movq %rdx, %r12;
 									load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
 									call __cast6_dec_blk8;
 									store_cbc_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									popq %r12;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(cast6_cbc_dec_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENTRY(cast6_ctr_8way)
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									/* input:
 									 *	%rdi: ctx, CTX
 									 *	%rsi: dst
 									 *	%rdx: src
 									 *	%rcx: iv (little endian, 128bit)
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									pushq %r12;
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									movq %rsi, %r11;
 									movq %rdx, %r12;
 									load_ctr_8way(%rcx, .Lbswap128_mask, RA1, RB1, RC1, RD1, RA2, RB2, RC2,
 										      RD2, RX, RKR, RKM);
 									call __cast6_enc_blk8;
 									store_ctr_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												crypto: cast6/avx - avoid using temporary stack buffers

Introduce new assembler functions to avoid use temporary stack buffers in
glue code. This also allows use of vector instructions for xoring output
in CTR and CBC modes and construction of IVs for CTR mode.

ECB mode sees ~0.5% decrease in speed because added one extra function
call. CBC mode decryption and CTR mode benefit from vector operations
and gain ~2%.

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

											
										
										
											2012-10-20 12:06:41 +00:00
+									popq %r12;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6 - add x86_64/avx assembler implementation

This patch adds a x86_64/avx assembler implementation of the Cast6 block
cipher. The implementation processes eight blocks in parallel (two 4 block
chunk AVX operations). The table-lookups are done in general-purpose registers.
For small blocksizes the functions from the generic module are called. A good
performance increase is provided for blocksizes greater or equal to 128B.

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmark results:

Intel Core i5-2500 CPU (fam:6, model:42, step:7)

cast6-avx-x86_64 vs. cast6-generic
128bit key:                                             (lrw:256bit)    (xts:256bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   1.00x   1.01x   1.01x   0.99x   0.97x   0.98x   1.01x   0.96x   0.98x
64B     0.98x   0.99x   1.02x   1.01x   0.99x   1.00x   1.01x   0.99x   1.00x   0.99x
256B    1.77x   1.84x   0.99x   1.85x   1.77x   1.77x   1.70x   1.74x   1.69x   1.72x
1024B   1.93x   1.95x   0.99x   1.96x   1.93x   1.93x   1.84x   1.85x   1.89x   1.87x
8192B   1.91x   1.95x   0.99x   1.97x   1.95x   1.91x   1.86x   1.87x   1.93x   1.90x

256bit key:                                             (lrw:384bit)    (xts:512bit)
size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec lrw-enc lrw-dec xts-enc xts-dec
16B     0.97x   0.99x   1.02x   1.01x   0.98x   0.99x   1.00x   1.00x   0.98x   0.98x
64B     0.98x   0.99x   1.01x   1.00x   1.00x   1.00x   1.01x   1.01x   0.97x   1.00x
256B    1.77x   1.83x   1.00x   1.86x   1.79x   1.78x   1.70x   1.76x   1.71x   1.69x
1024B   1.92x   1.95x   0.99x   1.96x   1.93x   1.93x   1.83x   1.86x   1.89x   1.87x
8192B   1.94x   1.95x   0.99x   1.97x   1.95x   1.95x   1.87x   1.87x   1.93x   1.91x

Signed-off-by: Johannes Goetzfried <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2012-07-11 17:38:57 +00:00
+									ret;
-												crypto: cast6-avx: use ENTRY()/ENDPROC() for assembler functions

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2013-01-19 11:39:16 +00:00
+								ENDPROC(cast6_ctr_8way)
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
 								ENTRY(cast6_xts_enc_8way)
 									/* input:
 									 *	%rdi: ctx, CTX
 									 *	%rsi: dst
 									 *	%rdx: src
 									 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+									movq %rsi, %r11;
 									/* regs <= src, dst <= IVs, regs <= regs xor IVs */
 									load_xts_8way(%rcx, %rdx, %rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2,
 										      RX, RKR, RKM, .Lxts_gf128mul_and_shl1_mask);
 									call __cast6_enc_blk8;
 									/* dst <= regs xor IVs(in dst) */
 									store_xts_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+									ret;
 								ENDPROC(cast6_xts_enc_8way)
 								ENTRY(cast6_xts_dec_8way)
 									/* input:
 									 *	%rdi: ctx, CTX
 									 *	%rsi: dst
 									 *	%rdx: src
 									 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
 									 */
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_BEGIN
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									pushq %r15;
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									movq %rdi, CTX
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+									movq %rsi, %r11;
 									/* regs <= src, dst <= IVs, regs <= regs xor IVs */
 									load_xts_8way(%rcx, %rdx, %rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2,
 										      RX, RKR, RKM, .Lxts_gf128mul_and_shl1_mask);
 									call __cast6_dec_blk8;
 									/* dst <= regs xor IVs(in dst) */
 									store_xts_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
-												crypto: x86/cast6 - Fix RBP usage

Using RBP as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.

Use R15 instead of RBP.  R15 can't be used as the RID1 register because
of x86 instruction encoding limitations.  So use R15 for CTX and RDI for
CTX.  This means that CTX is no longer an implicit function argument.
Instead it needs to be explicitly copied from RDI.

Reported-by: Eric Biggers <ebiggers@google.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric Biggers <ebiggers@google.com>
Acked-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

											
										
										
											2017-09-18 19:42:03 +00:00
+									popq %r15;
-												x86/asm/crypto: Create stack frames in crypto functions

The crypto code has several callable non-leaf functions which don't
honor CONFIG_FRAME_POINTER, which can result in bad stack traces.

Create stack frames for them when CONFIG_FRAME_POINTER is enabled.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Chris J Arges <chris.j.arges@canonical.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: live-patching@vger.kernel.org
Link: http://lkml.kernel.org/r/6c20192bcf1102ae18ae5a242cabf30ce9b29895.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

											
										
										
											2016-01-21 22:49:19 +00:00
+									FRAME_END
-												crypto: cast6-avx: use new optimized XTS code

Change cast6-avx to use the new XTS code, for smaller stack usage and small
boost to performance.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.01x   1.01x
64B     1.01x   1.00x
256B    1.09x   1.02x
1024B   1.08x   1.06x
8192B   1.08x   1.07x

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

											
										
										
											2013-04-08 18:51:05 +00:00
+									ret;
 								ENDPROC(cast6_xts_dec_8way)