2019-05-27 06:55:01 +00:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0-or-later */
|
2009-01-18 05:28:34 +00:00
|
|
|
/*
|
|
|
|
* Implement AES algorithm in Intel AES-NI instructions.
|
|
|
|
*
|
|
|
|
* The white paper of AES-NI instructions can be downloaded from:
|
|
|
|
* http://softwarecommunity.intel.com/isn/downloads/intelavx/AES-Instructions-Set_WP.pdf
|
|
|
|
*
|
|
|
|
* Copyright (C) 2008, Intel Corp.
|
|
|
|
* Author: Huang Ying <ying.huang@intel.com>
|
|
|
|
* Vinodh Gopal <vinodh.gopal@intel.com>
|
|
|
|
* Kahraman Akdemir
|
|
|
|
*
|
2010-11-04 19:00:45 +00:00
|
|
|
* Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
|
|
|
|
* interface for 64-bit kernels.
|
|
|
|
* Authors: Erdinc Ozturk (erdinc.ozturk@intel.com)
|
|
|
|
* Aidan O'Mahony (aidan.o.mahony@intel.com)
|
|
|
|
* Adrian Hoban <adrian.hoban@intel.com>
|
|
|
|
* James Guilford (james.guilford@intel.com)
|
|
|
|
* Gabriele Paoloni <gabriele.paoloni@intel.com>
|
|
|
|
* Tadeusz Struk (tadeusz.struk@intel.com)
|
|
|
|
* Wajdi Feghali (wajdi.k.feghali@intel.com)
|
|
|
|
* Copyright (c) 2010, Intel Corporation.
|
|
|
|
*
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
* Ported x86_64 version to x86:
|
|
|
|
* Author: Mathias Krause <minipli@googlemail.com>
|
2009-01-18 05:28:34 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/linkage.h>
|
2009-11-23 11:54:06 +00:00
|
|
|
#include <asm/inst.h>
|
2016-01-21 22:49:19 +00:00
|
|
|
#include <asm/frame.h>
|
2018-01-11 21:46:27 +00:00
|
|
|
#include <asm/nospec-branch.h>
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
/*
|
|
|
|
* The following macros are used to move an (un)aligned 16 byte value to/from
|
|
|
|
* an XMM register. This can done for either FP or integer values, for FP use
|
|
|
|
* movaps (move aligned packed single) or integer use movdqa (move double quad
|
|
|
|
* aligned). It doesn't make a performance difference which instruction is used
|
|
|
|
* since Nehalem (original Core i7) was released. However, the movaps is a byte
|
|
|
|
* shorter, so that is the one we'll use for now. (same for unaligned).
|
|
|
|
*/
|
|
|
|
#define MOVADQ movaps
|
|
|
|
#define MOVUDQ movups
|
|
|
|
|
2010-11-29 00:35:39 +00:00
|
|
|
#ifdef __x86_64__
|
2015-01-13 18:16:43 +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
|
|
|
# constants in mergeable sections, linker can reorder and merge
|
|
|
|
.section .rodata.cst16.gf128mul_x_ble_mask, "aM", @progbits, 16
|
2013-04-08 18:51:16 +00:00
|
|
|
.align 16
|
|
|
|
.Lgf128mul_x_ble_mask:
|
|
|
|
.octa 0x00000000000000010000000000000087
|
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.POLY, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
POLY: .octa 0xC2000000000000000000000000000001
|
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.TWOONE, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
TWOONE: .octa 0x00000001000000000000000000000001
|
|
|
|
|
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.SHUF_MASK, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
SHUF_MASK: .octa 0x000102030405060708090A0B0C0D0E0F
|
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.MASK1, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
MASK1: .octa 0x0000000000000000ffffffffffffffff
|
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.MASK2, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
MASK2: .octa 0xffffffffffffffff0000000000000000
|
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.ONE, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
ONE: .octa 0x00000000000000000000000000000001
|
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.F_MIN_MASK, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
F_MIN_MASK: .octa 0xf1f2f3f4f5f6f7f8f9fafbfcfdfeff0
|
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.dec, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
dec: .octa 0x1
|
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.enc, "aM", @progbits, 16
|
|
|
|
.align 16
|
2010-11-04 19:00:45 +00:00
|
|
|
enc: .octa 0x2
|
|
|
|
|
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
|
|
|
# order of these constants should not change.
|
|
|
|
# more specifically, ALL_F should follow SHIFT_MASK,
|
|
|
|
# and zero should follow ALL_F
|
|
|
|
.section .rodata, "a", @progbits
|
|
|
|
.align 16
|
|
|
|
SHIFT_MASK: .octa 0x0f0e0d0c0b0a09080706050403020100
|
|
|
|
ALL_F: .octa 0xffffffffffffffffffffffffffffffff
|
|
|
|
.octa 0x00000000000000000000000000000000
|
|
|
|
|
2009-01-18 05:28:34 +00:00
|
|
|
.text
|
|
|
|
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
#define STACK_OFFSET 8*3
|
|
|
|
|
2018-02-14 17:39:23 +00:00
|
|
|
#define AadHash 16*0
|
|
|
|
#define AadLen 16*1
|
|
|
|
#define InLen (16*1)+8
|
|
|
|
#define PBlockEncKey 16*2
|
|
|
|
#define OrigIV 16*3
|
|
|
|
#define CurCount 16*4
|
|
|
|
#define PBlockLen 16*5
|
2018-02-14 17:40:10 +00:00
|
|
|
#define HashKey 16*6 // store HashKey <<1 mod poly here
|
|
|
|
#define HashKey_2 16*7 // store HashKey^2 <<1 mod poly here
|
|
|
|
#define HashKey_3 16*8 // store HashKey^3 <<1 mod poly here
|
|
|
|
#define HashKey_4 16*9 // store HashKey^4 <<1 mod poly here
|
|
|
|
#define HashKey_k 16*10 // store XOR of High 64 bits and Low 64
|
|
|
|
// bits of HashKey <<1 mod poly here
|
|
|
|
//(for Karatsuba purposes)
|
|
|
|
#define HashKey_2_k 16*11 // store XOR of High 64 bits and Low 64
|
|
|
|
// bits of HashKey^2 <<1 mod poly here
|
|
|
|
// (for Karatsuba purposes)
|
|
|
|
#define HashKey_3_k 16*12 // store XOR of High 64 bits and Low 64
|
|
|
|
// bits of HashKey^3 <<1 mod poly here
|
|
|
|
// (for Karatsuba purposes)
|
|
|
|
#define HashKey_4_k 16*13 // store XOR of High 64 bits and Low 64
|
|
|
|
// bits of HashKey^4 <<1 mod poly here
|
|
|
|
// (for Karatsuba purposes)
|
2018-02-14 17:39:23 +00:00
|
|
|
|
2010-11-04 19:00:45 +00:00
|
|
|
#define arg1 rdi
|
|
|
|
#define arg2 rsi
|
|
|
|
#define arg3 rdx
|
|
|
|
#define arg4 rcx
|
|
|
|
#define arg5 r8
|
|
|
|
#define arg6 r9
|
2018-02-14 17:40:10 +00:00
|
|
|
#define arg7 STACK_OFFSET+8(%rsp)
|
|
|
|
#define arg8 STACK_OFFSET+16(%rsp)
|
|
|
|
#define arg9 STACK_OFFSET+24(%rsp)
|
|
|
|
#define arg10 STACK_OFFSET+32(%rsp)
|
|
|
|
#define arg11 STACK_OFFSET+40(%rsp)
|
2015-01-13 18:16:43 +00:00
|
|
|
#define keysize 2*15*16(%arg1)
|
2010-11-29 00:35:39 +00:00
|
|
|
#endif
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
|
2009-01-18 05:28:34 +00:00
|
|
|
#define STATE1 %xmm0
|
|
|
|
#define STATE2 %xmm4
|
|
|
|
#define STATE3 %xmm5
|
|
|
|
#define STATE4 %xmm6
|
|
|
|
#define STATE STATE1
|
|
|
|
#define IN1 %xmm1
|
|
|
|
#define IN2 %xmm7
|
|
|
|
#define IN3 %xmm8
|
|
|
|
#define IN4 %xmm9
|
|
|
|
#define IN IN1
|
|
|
|
#define KEY %xmm2
|
|
|
|
#define IV %xmm3
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
|
2010-03-10 10:28:55 +00:00
|
|
|
#define BSWAP_MASK %xmm10
|
|
|
|
#define CTR %xmm11
|
|
|
|
#define INC %xmm12
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2013-04-08 18:51:16 +00:00
|
|
|
#define GF128MUL_MASK %xmm10
|
|
|
|
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifdef __x86_64__
|
|
|
|
#define AREG %rax
|
2009-01-18 05:28:34 +00:00
|
|
|
#define KEYP %rdi
|
|
|
|
#define OUTP %rsi
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#define UKEYP OUTP
|
2009-01-18 05:28:34 +00:00
|
|
|
#define INP %rdx
|
|
|
|
#define LEN %rcx
|
|
|
|
#define IVP %r8
|
|
|
|
#define KLEN %r9d
|
|
|
|
#define T1 %r10
|
|
|
|
#define TKEYP T1
|
|
|
|
#define T2 %r11
|
2010-03-10 10:28:55 +00:00
|
|
|
#define TCTR_LOW T2
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#else
|
|
|
|
#define AREG %eax
|
|
|
|
#define KEYP %edi
|
|
|
|
#define OUTP AREG
|
|
|
|
#define UKEYP OUTP
|
|
|
|
#define INP %edx
|
|
|
|
#define LEN %esi
|
|
|
|
#define IVP %ebp
|
|
|
|
#define KLEN %ebx
|
|
|
|
#define T1 %ecx
|
|
|
|
#define TKEYP T1
|
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2018-02-14 17:38:35 +00:00
|
|
|
.macro FUNC_SAVE
|
|
|
|
push %r12
|
|
|
|
push %r13
|
|
|
|
push %r14
|
|
|
|
#
|
|
|
|
# states of %xmm registers %xmm6:%xmm15 not saved
|
|
|
|
# all %xmm registers are clobbered
|
|
|
|
#
|
|
|
|
.endm
|
|
|
|
|
|
|
|
|
|
|
|
.macro FUNC_RESTORE
|
|
|
|
pop %r14
|
|
|
|
pop %r13
|
|
|
|
pop %r12
|
|
|
|
.endm
|
2010-11-04 19:00:45 +00:00
|
|
|
|
2018-02-14 17:40:10 +00:00
|
|
|
# Precompute hashkeys.
|
|
|
|
# Input: Hash subkey.
|
|
|
|
# Output: HashKeys stored in gcm_context_data. Only needs to be called
|
|
|
|
# once per key.
|
|
|
|
# clobbers r12, and tmp xmm registers.
|
2018-02-14 17:40:47 +00:00
|
|
|
.macro PRECOMPUTE SUBKEY TMP1 TMP2 TMP3 TMP4 TMP5 TMP6 TMP7
|
|
|
|
mov \SUBKEY, %r12
|
2018-02-14 17:40:10 +00:00
|
|
|
movdqu (%r12), \TMP3
|
|
|
|
movdqa SHUF_MASK(%rip), \TMP2
|
|
|
|
PSHUFB_XMM \TMP2, \TMP3
|
|
|
|
|
|
|
|
# precompute HashKey<<1 mod poly from the HashKey (required for GHASH)
|
|
|
|
|
|
|
|
movdqa \TMP3, \TMP2
|
|
|
|
psllq $1, \TMP3
|
|
|
|
psrlq $63, \TMP2
|
|
|
|
movdqa \TMP2, \TMP1
|
|
|
|
pslldq $8, \TMP2
|
|
|
|
psrldq $8, \TMP1
|
|
|
|
por \TMP2, \TMP3
|
|
|
|
|
|
|
|
# reduce HashKey<<1
|
|
|
|
|
|
|
|
pshufd $0x24, \TMP1, \TMP2
|
|
|
|
pcmpeqd TWOONE(%rip), \TMP2
|
|
|
|
pand POLY(%rip), \TMP2
|
|
|
|
pxor \TMP2, \TMP3
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP3, HashKey(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
|
|
|
|
movdqa \TMP3, \TMP5
|
|
|
|
pshufd $78, \TMP3, \TMP1
|
|
|
|
pxor \TMP3, \TMP1
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP1, HashKey_k(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
|
|
|
|
GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
|
|
|
|
# TMP5 = HashKey^2<<1 (mod poly)
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP5, HashKey_2(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
# HashKey_2 = HashKey^2<<1 (mod poly)
|
|
|
|
pshufd $78, \TMP5, \TMP1
|
|
|
|
pxor \TMP5, \TMP1
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP1, HashKey_2_k(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
|
|
|
|
GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
|
|
|
|
# TMP5 = HashKey^3<<1 (mod poly)
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP5, HashKey_3(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
pshufd $78, \TMP5, \TMP1
|
|
|
|
pxor \TMP5, \TMP1
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP1, HashKey_3_k(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
|
|
|
|
GHASH_MUL \TMP5, \TMP3, \TMP1, \TMP2, \TMP4, \TMP6, \TMP7
|
|
|
|
# TMP5 = HashKey^3<<1 (mod poly)
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP5, HashKey_4(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
pshufd $78, \TMP5, \TMP1
|
|
|
|
pxor \TMP5, \TMP1
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu \TMP1, HashKey_4_k(%arg2)
|
2018-02-14 17:40:10 +00:00
|
|
|
.endm
|
2018-02-14 17:38:45 +00:00
|
|
|
|
|
|
|
# GCM_INIT initializes a gcm_context struct to prepare for encoding/decoding.
|
|
|
|
# Clobbers rax, r10-r13 and xmm0-xmm6, %xmm13
|
2018-02-14 17:40:47 +00:00
|
|
|
.macro GCM_INIT Iv SUBKEY AAD AADLEN
|
|
|
|
mov \AADLEN, %r11
|
2018-02-14 17:39:45 +00:00
|
|
|
mov %r11, AadLen(%arg2) # ctx_data.aad_length = aad_length
|
2018-07-02 10:31:54 +00:00
|
|
|
xor %r11d, %r11d
|
2018-02-14 17:39:45 +00:00
|
|
|
mov %r11, InLen(%arg2) # ctx_data.in_length = 0
|
|
|
|
mov %r11, PBlockLen(%arg2) # ctx_data.partial_block_length = 0
|
|
|
|
mov %r11, PBlockEncKey(%arg2) # ctx_data.partial_block_enc_key = 0
|
2018-02-14 17:40:47 +00:00
|
|
|
mov \Iv, %rax
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu (%rax), %xmm0
|
|
|
|
movdqu %xmm0, OrigIV(%arg2) # ctx_data.orig_IV = iv
|
|
|
|
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm2
|
|
|
|
PSHUFB_XMM %xmm2, %xmm0
|
|
|
|
movdqu %xmm0, CurCount(%arg2) # ctx_data.current_counter = iv
|
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
PRECOMPUTE \SUBKEY, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey(%arg2), %xmm13
|
2018-02-14 17:39:36 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
CALC_AAD_HASH %xmm13, \AAD, \AADLEN, %xmm0, %xmm1, %xmm2, %xmm3, \
|
|
|
|
%xmm4, %xmm5, %xmm6
|
2018-02-14 17:38:45 +00:00
|
|
|
.endm
|
|
|
|
|
2018-02-14 17:39:10 +00:00
|
|
|
# GCM_ENC_DEC Encodes/Decodes given data. Assumes that the passed gcm_context
|
|
|
|
# struct has been initialized by GCM_INIT.
|
|
|
|
# Requires the input data be at least 1 byte long because of READ_PARTIAL_BLOCK
|
|
|
|
# Clobbers rax, r10-r13, and xmm0-xmm15
|
|
|
|
.macro GCM_ENC_DEC operation
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu AadHash(%arg2), %xmm8
|
2018-02-14 17:40:10 +00:00
|
|
|
movdqu HashKey(%arg2), %xmm13
|
2018-02-14 17:39:45 +00:00
|
|
|
add %arg5, InLen(%arg2)
|
2018-02-14 17:40:19 +00:00
|
|
|
|
2018-07-02 10:31:54 +00:00
|
|
|
xor %r11d, %r11d # initialise the data pointer offset as zero
|
2018-02-14 17:40:19 +00:00
|
|
|
PARTIAL_BLOCK %arg3 %arg4 %arg5 %r11 %xmm8 \operation
|
|
|
|
|
|
|
|
sub %r11, %arg5 # sub partial block data used
|
2018-02-14 17:39:45 +00:00
|
|
|
mov %arg5, %r13 # save the number of bytes
|
2018-02-14 17:40:19 +00:00
|
|
|
|
2018-02-14 17:39:45 +00:00
|
|
|
and $-16, %r13 # %r13 = %r13 - (%r13 mod 16)
|
|
|
|
mov %r13, %r12
|
2018-02-14 17:39:10 +00:00
|
|
|
# Encrypt/Decrypt first few blocks
|
|
|
|
|
|
|
|
and $(3<<4), %r12
|
|
|
|
jz _initial_num_blocks_is_0_\@
|
|
|
|
cmp $(2<<4), %r12
|
|
|
|
jb _initial_num_blocks_is_1_\@
|
|
|
|
je _initial_num_blocks_is_2_\@
|
|
|
|
_initial_num_blocks_is_3_\@:
|
|
|
|
INITIAL_BLOCKS_ENC_DEC %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
|
|
|
|
%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 5, 678, \operation
|
|
|
|
sub $48, %r13
|
|
|
|
jmp _initial_blocks_\@
|
|
|
|
_initial_num_blocks_is_2_\@:
|
|
|
|
INITIAL_BLOCKS_ENC_DEC %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
|
|
|
|
%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 6, 78, \operation
|
|
|
|
sub $32, %r13
|
|
|
|
jmp _initial_blocks_\@
|
|
|
|
_initial_num_blocks_is_1_\@:
|
|
|
|
INITIAL_BLOCKS_ENC_DEC %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
|
|
|
|
%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 7, 8, \operation
|
|
|
|
sub $16, %r13
|
|
|
|
jmp _initial_blocks_\@
|
|
|
|
_initial_num_blocks_is_0_\@:
|
|
|
|
INITIAL_BLOCKS_ENC_DEC %xmm9, %xmm10, %xmm13, %xmm11, %xmm12, %xmm0, \
|
|
|
|
%xmm1, %xmm2, %xmm3, %xmm4, %xmm8, %xmm5, %xmm6, 8, 0, \operation
|
|
|
|
_initial_blocks_\@:
|
|
|
|
|
|
|
|
# Main loop - Encrypt/Decrypt remaining blocks
|
|
|
|
|
|
|
|
cmp $0, %r13
|
|
|
|
je _zero_cipher_left_\@
|
|
|
|
sub $64, %r13
|
|
|
|
je _four_cipher_left_\@
|
|
|
|
_crypt_by_4_\@:
|
|
|
|
GHASH_4_ENCRYPT_4_PARALLEL_\operation %xmm9, %xmm10, %xmm11, %xmm12, \
|
|
|
|
%xmm13, %xmm14, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, \
|
|
|
|
%xmm7, %xmm8, enc
|
|
|
|
add $64, %r11
|
|
|
|
sub $64, %r13
|
|
|
|
jne _crypt_by_4_\@
|
|
|
|
_four_cipher_left_\@:
|
|
|
|
GHASH_LAST_4 %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, \
|
|
|
|
%xmm15, %xmm1, %xmm2, %xmm3, %xmm4, %xmm8
|
|
|
|
_zero_cipher_left_\@:
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu %xmm8, AadHash(%arg2)
|
|
|
|
movdqu %xmm0, CurCount(%arg2)
|
|
|
|
|
2018-02-14 17:39:23 +00:00
|
|
|
mov %arg5, %r13
|
|
|
|
and $15, %r13 # %r13 = arg5 (mod 16)
|
2018-02-14 17:39:10 +00:00
|
|
|
je _multiple_of_16_bytes_\@
|
|
|
|
|
2018-02-14 17:39:45 +00:00
|
|
|
mov %r13, PBlockLen(%arg2)
|
|
|
|
|
2018-02-14 17:39:10 +00:00
|
|
|
# Handle the last <16 Byte block separately
|
|
|
|
paddd ONE(%rip), %xmm0 # INCR CNT to get Yn
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu %xmm0, CurCount(%arg2)
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
2018-02-14 17:39:10 +00:00
|
|
|
PSHUFB_XMM %xmm10, %xmm0
|
|
|
|
|
|
|
|
ENCRYPT_SINGLE_BLOCK %xmm0, %xmm1 # Encrypt(K, Yn)
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu %xmm0, PBlockEncKey(%arg2)
|
2018-02-14 17:39:10 +00:00
|
|
|
|
2018-02-14 17:40:31 +00:00
|
|
|
cmp $16, %arg5
|
|
|
|
jge _large_enough_update_\@
|
|
|
|
|
2018-02-14 17:39:23 +00:00
|
|
|
lea (%arg4,%r11,1), %r10
|
2018-02-14 17:39:10 +00:00
|
|
|
mov %r13, %r12
|
|
|
|
READ_PARTIAL_BLOCK %r10 %r12 %xmm2 %xmm1
|
2018-02-14 17:40:31 +00:00
|
|
|
jmp _data_read_\@
|
|
|
|
|
|
|
|
_large_enough_update_\@:
|
|
|
|
sub $16, %r11
|
|
|
|
add %r13, %r11
|
|
|
|
|
|
|
|
# receive the last <16 Byte block
|
|
|
|
movdqu (%arg4, %r11, 1), %xmm1
|
2018-02-14 17:39:10 +00:00
|
|
|
|
2018-02-14 17:40:31 +00:00
|
|
|
sub %r13, %r11
|
|
|
|
add $16, %r11
|
|
|
|
|
|
|
|
lea SHIFT_MASK+16(%rip), %r12
|
|
|
|
# adjust the shuffle mask pointer to be able to shift 16-r13 bytes
|
|
|
|
# (r13 is the number of bytes in plaintext mod 16)
|
|
|
|
sub %r13, %r12
|
|
|
|
# get the appropriate shuffle mask
|
|
|
|
movdqu (%r12), %xmm2
|
|
|
|
# shift right 16-r13 bytes
|
|
|
|
PSHUFB_XMM %xmm2, %xmm1
|
|
|
|
|
|
|
|
_data_read_\@:
|
2018-02-14 17:39:10 +00:00
|
|
|
lea ALL_F+16(%rip), %r12
|
|
|
|
sub %r13, %r12
|
2018-02-14 17:40:31 +00:00
|
|
|
|
2018-02-14 17:39:10 +00:00
|
|
|
.ifc \operation, dec
|
|
|
|
movdqa %xmm1, %xmm2
|
|
|
|
.endif
|
|
|
|
pxor %xmm1, %xmm0 # XOR Encrypt(K, Yn)
|
|
|
|
movdqu (%r12), %xmm1
|
|
|
|
# get the appropriate mask to mask out top 16-r13 bytes of xmm0
|
|
|
|
pand %xmm1, %xmm0 # mask out top 16-r13 bytes of xmm0
|
|
|
|
.ifc \operation, dec
|
|
|
|
pand %xmm1, %xmm2
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
PSHUFB_XMM %xmm10 ,%xmm2
|
|
|
|
|
|
|
|
pxor %xmm2, %xmm8
|
|
|
|
.else
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
PSHUFB_XMM %xmm10,%xmm0
|
|
|
|
|
|
|
|
pxor %xmm0, %xmm8
|
|
|
|
.endif
|
|
|
|
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu %xmm8, AadHash(%arg2)
|
2018-02-14 17:39:10 +00:00
|
|
|
.ifc \operation, enc
|
|
|
|
# GHASH computation for the last <16 byte block
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
# shuffle xmm0 back to output as ciphertext
|
|
|
|
PSHUFB_XMM %xmm10, %xmm0
|
|
|
|
.endif
|
|
|
|
|
|
|
|
# Output %r13 bytes
|
|
|
|
MOVQ_R64_XMM %xmm0, %rax
|
|
|
|
cmp $8, %r13
|
|
|
|
jle _less_than_8_bytes_left_\@
|
2018-02-14 17:39:23 +00:00
|
|
|
mov %rax, (%arg3 , %r11, 1)
|
2018-02-14 17:39:10 +00:00
|
|
|
add $8, %r11
|
|
|
|
psrldq $8, %xmm0
|
|
|
|
MOVQ_R64_XMM %xmm0, %rax
|
|
|
|
sub $8, %r13
|
|
|
|
_less_than_8_bytes_left_\@:
|
2018-02-14 17:39:23 +00:00
|
|
|
mov %al, (%arg3, %r11, 1)
|
2018-02-14 17:39:10 +00:00
|
|
|
add $1, %r11
|
|
|
|
shr $8, %rax
|
|
|
|
sub $1, %r13
|
|
|
|
jne _less_than_8_bytes_left_\@
|
|
|
|
_multiple_of_16_bytes_\@:
|
|
|
|
.endm
|
|
|
|
|
2018-02-14 17:38:57 +00:00
|
|
|
# GCM_COMPLETE Finishes update of tag of last partial block
|
|
|
|
# Output: Authorization Tag (AUTH_TAG)
|
|
|
|
# Clobbers rax, r10-r12, and xmm0, xmm1, xmm5-xmm15
|
2018-02-14 17:40:47 +00:00
|
|
|
.macro GCM_COMPLETE AUTHTAG AUTHTAGLEN
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu AadHash(%arg2), %xmm8
|
2018-02-14 17:40:10 +00:00
|
|
|
movdqu HashKey(%arg2), %xmm13
|
2018-02-14 17:39:55 +00:00
|
|
|
|
|
|
|
mov PBlockLen(%arg2), %r12
|
|
|
|
|
|
|
|
cmp $0, %r12
|
|
|
|
je _partial_done\@
|
|
|
|
|
|
|
|
GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
|
|
|
|
|
|
|
|
_partial_done\@:
|
2018-02-14 17:39:45 +00:00
|
|
|
mov AadLen(%arg2), %r12 # %r13 = aadLen (number of bytes)
|
2018-02-14 17:38:57 +00:00
|
|
|
shl $3, %r12 # convert into number of bits
|
|
|
|
movd %r12d, %xmm15 # len(A) in %xmm15
|
2018-02-14 17:39:45 +00:00
|
|
|
mov InLen(%arg2), %r12
|
|
|
|
shl $3, %r12 # len(C) in bits (*128)
|
|
|
|
MOVQ_R64_XMM %r12, %xmm1
|
|
|
|
|
2018-02-14 17:38:57 +00:00
|
|
|
pslldq $8, %xmm15 # %xmm15 = len(A)||0x0000000000000000
|
|
|
|
pxor %xmm1, %xmm15 # %xmm15 = len(A)||len(C)
|
|
|
|
pxor %xmm15, %xmm8
|
|
|
|
GHASH_MUL %xmm8, %xmm13, %xmm9, %xmm10, %xmm11, %xmm5, %xmm6
|
|
|
|
# final GHASH computation
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
PSHUFB_XMM %xmm10, %xmm8
|
|
|
|
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu OrigIV(%arg2), %xmm0 # %xmm0 = Y0
|
2018-02-14 17:38:57 +00:00
|
|
|
ENCRYPT_SINGLE_BLOCK %xmm0, %xmm1 # E(K, Y0)
|
|
|
|
pxor %xmm8, %xmm0
|
|
|
|
_return_T_\@:
|
2018-02-14 17:40:47 +00:00
|
|
|
mov \AUTHTAG, %r10 # %r10 = authTag
|
|
|
|
mov \AUTHTAGLEN, %r11 # %r11 = auth_tag_len
|
2018-02-14 17:38:57 +00:00
|
|
|
cmp $16, %r11
|
|
|
|
je _T_16_\@
|
|
|
|
cmp $8, %r11
|
|
|
|
jl _T_4_\@
|
|
|
|
_T_8_\@:
|
|
|
|
MOVQ_R64_XMM %xmm0, %rax
|
|
|
|
mov %rax, (%r10)
|
|
|
|
add $8, %r10
|
|
|
|
sub $8, %r11
|
|
|
|
psrldq $8, %xmm0
|
|
|
|
cmp $0, %r11
|
|
|
|
je _return_T_done_\@
|
|
|
|
_T_4_\@:
|
|
|
|
movd %xmm0, %eax
|
|
|
|
mov %eax, (%r10)
|
|
|
|
add $4, %r10
|
|
|
|
sub $4, %r11
|
|
|
|
psrldq $4, %xmm0
|
|
|
|
cmp $0, %r11
|
|
|
|
je _return_T_done_\@
|
|
|
|
_T_123_\@:
|
|
|
|
movd %xmm0, %eax
|
|
|
|
cmp $2, %r11
|
|
|
|
jl _T_1_\@
|
|
|
|
mov %ax, (%r10)
|
|
|
|
cmp $2, %r11
|
|
|
|
je _return_T_done_\@
|
|
|
|
add $2, %r10
|
|
|
|
sar $16, %eax
|
|
|
|
_T_1_\@:
|
|
|
|
mov %al, (%r10)
|
|
|
|
jmp _return_T_done_\@
|
|
|
|
_T_16_\@:
|
|
|
|
movdqu %xmm0, (%r10)
|
|
|
|
_return_T_done_\@:
|
|
|
|
.endm
|
|
|
|
|
2010-11-29 00:35:39 +00:00
|
|
|
#ifdef __x86_64__
|
2010-11-04 19:00:45 +00:00
|
|
|
/* GHASH_MUL MACRO to implement: Data*HashKey mod (128,127,126,121,0)
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* Input: A and B (128-bits each, bit-reflected)
|
|
|
|
* Output: C = A*B*x mod poly, (i.e. >>1 )
|
|
|
|
* To compute GH = GH*HashKey mod poly, give HK = HashKey<<1 mod poly as input
|
|
|
|
* GH = GH * HK * x mod poly which is equivalent to GH*HashKey mod poly.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
.macro GHASH_MUL GH HK TMP1 TMP2 TMP3 TMP4 TMP5
|
|
|
|
movdqa \GH, \TMP1
|
|
|
|
pshufd $78, \GH, \TMP2
|
|
|
|
pshufd $78, \HK, \TMP3
|
|
|
|
pxor \GH, \TMP2 # TMP2 = a1+a0
|
|
|
|
pxor \HK, \TMP3 # TMP3 = b1+b0
|
|
|
|
PCLMULQDQ 0x11, \HK, \TMP1 # TMP1 = a1*b1
|
|
|
|
PCLMULQDQ 0x00, \HK, \GH # GH = a0*b0
|
|
|
|
PCLMULQDQ 0x00, \TMP3, \TMP2 # TMP2 = (a0+a1)*(b1+b0)
|
|
|
|
pxor \GH, \TMP2
|
|
|
|
pxor \TMP1, \TMP2 # TMP2 = (a0*b0)+(a1*b0)
|
|
|
|
movdqa \TMP2, \TMP3
|
|
|
|
pslldq $8, \TMP3 # left shift TMP3 2 DWs
|
|
|
|
psrldq $8, \TMP2 # right shift TMP2 2 DWs
|
|
|
|
pxor \TMP3, \GH
|
|
|
|
pxor \TMP2, \TMP1 # TMP2:GH holds the result of GH*HK
|
|
|
|
|
|
|
|
# first phase of the reduction
|
|
|
|
|
|
|
|
movdqa \GH, \TMP2
|
|
|
|
movdqa \GH, \TMP3
|
|
|
|
movdqa \GH, \TMP4 # copy GH into TMP2,TMP3 and TMP4
|
|
|
|
# in in order to perform
|
|
|
|
# independent shifts
|
|
|
|
pslld $31, \TMP2 # packed right shift <<31
|
|
|
|
pslld $30, \TMP3 # packed right shift <<30
|
|
|
|
pslld $25, \TMP4 # packed right shift <<25
|
|
|
|
pxor \TMP3, \TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4, \TMP2
|
|
|
|
movdqa \TMP2, \TMP5
|
|
|
|
psrldq $4, \TMP5 # right shift TMP5 1 DW
|
|
|
|
pslldq $12, \TMP2 # left shift TMP2 3 DWs
|
|
|
|
pxor \TMP2, \GH
|
|
|
|
|
|
|
|
# second phase of the reduction
|
|
|
|
|
|
|
|
movdqa \GH,\TMP2 # copy GH into TMP2,TMP3 and TMP4
|
|
|
|
# in in order to perform
|
|
|
|
# independent shifts
|
|
|
|
movdqa \GH,\TMP3
|
|
|
|
movdqa \GH,\TMP4
|
|
|
|
psrld $1,\TMP2 # packed left shift >>1
|
|
|
|
psrld $2,\TMP3 # packed left shift >>2
|
|
|
|
psrld $7,\TMP4 # packed left shift >>7
|
|
|
|
pxor \TMP3,\TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4,\TMP2
|
|
|
|
pxor \TMP5, \TMP2
|
|
|
|
pxor \TMP2, \GH
|
|
|
|
pxor \TMP1, \GH # result is in TMP1
|
|
|
|
.endm
|
|
|
|
|
2017-12-21 01:08:37 +00:00
|
|
|
# Reads DLEN bytes starting at DPTR and stores in XMMDst
|
|
|
|
# where 0 < DLEN < 16
|
|
|
|
# Clobbers %rax, DLEN and XMM1
|
|
|
|
.macro READ_PARTIAL_BLOCK DPTR DLEN XMM1 XMMDst
|
|
|
|
cmp $8, \DLEN
|
|
|
|
jl _read_lt8_\@
|
|
|
|
mov (\DPTR), %rax
|
|
|
|
MOVQ_R64_XMM %rax, \XMMDst
|
|
|
|
sub $8, \DLEN
|
|
|
|
jz _done_read_partial_block_\@
|
|
|
|
xor %eax, %eax
|
|
|
|
_read_next_byte_\@:
|
|
|
|
shl $8, %rax
|
|
|
|
mov 7(\DPTR, \DLEN, 1), %al
|
|
|
|
dec \DLEN
|
|
|
|
jnz _read_next_byte_\@
|
|
|
|
MOVQ_R64_XMM %rax, \XMM1
|
|
|
|
pslldq $8, \XMM1
|
|
|
|
por \XMM1, \XMMDst
|
|
|
|
jmp _done_read_partial_block_\@
|
|
|
|
_read_lt8_\@:
|
|
|
|
xor %eax, %eax
|
|
|
|
_read_next_byte_lt8_\@:
|
|
|
|
shl $8, %rax
|
|
|
|
mov -1(\DPTR, \DLEN, 1), %al
|
|
|
|
dec \DLEN
|
|
|
|
jnz _read_next_byte_lt8_\@
|
|
|
|
MOVQ_R64_XMM %rax, \XMMDst
|
|
|
|
_done_read_partial_block_\@:
|
|
|
|
.endm
|
|
|
|
|
2018-02-14 17:39:36 +00:00
|
|
|
# CALC_AAD_HASH: Calculates the hash of the data which will not be encrypted.
|
|
|
|
# clobbers r10-11, xmm14
|
2018-02-14 17:40:47 +00:00
|
|
|
.macro CALC_AAD_HASH HASHKEY AAD AADLEN TMP1 TMP2 TMP3 TMP4 TMP5 \
|
2018-02-14 17:39:36 +00:00
|
|
|
TMP6 TMP7
|
|
|
|
MOVADQ SHUF_MASK(%rip), %xmm14
|
2018-02-14 17:40:47 +00:00
|
|
|
mov \AAD, %r10 # %r10 = AAD
|
|
|
|
mov \AADLEN, %r11 # %r11 = aadLen
|
2018-02-14 17:39:36 +00:00
|
|
|
pxor \TMP7, \TMP7
|
|
|
|
pxor \TMP6, \TMP6
|
2017-04-28 16:11:56 +00:00
|
|
|
|
|
|
|
cmp $16, %r11
|
2018-02-14 17:38:12 +00:00
|
|
|
jl _get_AAD_rest\@
|
|
|
|
_get_AAD_blocks\@:
|
2018-02-14 17:39:36 +00:00
|
|
|
movdqu (%r10), \TMP7
|
|
|
|
PSHUFB_XMM %xmm14, \TMP7 # byte-reflect the AAD data
|
|
|
|
pxor \TMP7, \TMP6
|
|
|
|
GHASH_MUL \TMP6, \HASHKEY, \TMP1, \TMP2, \TMP3, \TMP4, \TMP5
|
2017-04-28 16:11:56 +00:00
|
|
|
add $16, %r10
|
|
|
|
sub $16, %r11
|
|
|
|
cmp $16, %r11
|
2018-02-14 17:38:12 +00:00
|
|
|
jge _get_AAD_blocks\@
|
2017-04-28 16:11:56 +00:00
|
|
|
|
2018-02-14 17:39:36 +00:00
|
|
|
movdqu \TMP6, \TMP7
|
2017-12-21 01:08:38 +00:00
|
|
|
|
|
|
|
/* read the last <16B of AAD */
|
2018-02-14 17:38:12 +00:00
|
|
|
_get_AAD_rest\@:
|
2017-04-28 16:11:56 +00:00
|
|
|
cmp $0, %r11
|
2018-02-14 17:38:12 +00:00
|
|
|
je _get_AAD_done\@
|
2017-04-28 16:11:56 +00:00
|
|
|
|
2018-02-14 17:39:36 +00:00
|
|
|
READ_PARTIAL_BLOCK %r10, %r11, \TMP1, \TMP7
|
|
|
|
PSHUFB_XMM %xmm14, \TMP7 # byte-reflect the AAD data
|
|
|
|
pxor \TMP6, \TMP7
|
|
|
|
GHASH_MUL \TMP7, \HASHKEY, \TMP1, \TMP2, \TMP3, \TMP4, \TMP5
|
|
|
|
movdqu \TMP7, \TMP6
|
2010-12-13 11:51:15 +00:00
|
|
|
|
2018-02-14 17:38:12 +00:00
|
|
|
_get_AAD_done\@:
|
2018-02-14 17:39:36 +00:00
|
|
|
movdqu \TMP6, AadHash(%arg2)
|
|
|
|
.endm
|
|
|
|
|
2018-02-14 17:40:19 +00:00
|
|
|
# PARTIAL_BLOCK: Handles encryption/decryption and the tag partial blocks
|
|
|
|
# between update calls.
|
|
|
|
# Requires the input data be at least 1 byte long due to READ_PARTIAL_BLOCK
|
|
|
|
# Outputs encrypted bytes, and updates hash and partial info in gcm_data_context
|
|
|
|
# Clobbers rax, r10, r12, r13, xmm0-6, xmm9-13
|
|
|
|
.macro PARTIAL_BLOCK CYPH_PLAIN_OUT PLAIN_CYPH_IN PLAIN_CYPH_LEN DATA_OFFSET \
|
|
|
|
AAD_HASH operation
|
|
|
|
mov PBlockLen(%arg2), %r13
|
|
|
|
cmp $0, %r13
|
|
|
|
je _partial_block_done_\@ # Leave Macro if no partial blocks
|
|
|
|
# Read in input data without over reading
|
|
|
|
cmp $16, \PLAIN_CYPH_LEN
|
|
|
|
jl _fewer_than_16_bytes_\@
|
|
|
|
movups (\PLAIN_CYPH_IN), %xmm1 # If more than 16 bytes, just fill xmm
|
|
|
|
jmp _data_read_\@
|
|
|
|
|
|
|
|
_fewer_than_16_bytes_\@:
|
|
|
|
lea (\PLAIN_CYPH_IN, \DATA_OFFSET, 1), %r10
|
|
|
|
mov \PLAIN_CYPH_LEN, %r12
|
|
|
|
READ_PARTIAL_BLOCK %r10 %r12 %xmm0 %xmm1
|
|
|
|
|
|
|
|
mov PBlockLen(%arg2), %r13
|
|
|
|
|
|
|
|
_data_read_\@: # Finished reading in data
|
|
|
|
|
|
|
|
movdqu PBlockEncKey(%arg2), %xmm9
|
|
|
|
movdqu HashKey(%arg2), %xmm13
|
|
|
|
|
|
|
|
lea SHIFT_MASK(%rip), %r12
|
|
|
|
|
|
|
|
# adjust the shuffle mask pointer to be able to shift r13 bytes
|
|
|
|
# r16-r13 is the number of bytes in plaintext mod 16)
|
|
|
|
add %r13, %r12
|
|
|
|
movdqu (%r12), %xmm2 # get the appropriate shuffle mask
|
|
|
|
PSHUFB_XMM %xmm2, %xmm9 # shift right r13 bytes
|
|
|
|
|
|
|
|
.ifc \operation, dec
|
|
|
|
movdqa %xmm1, %xmm3
|
|
|
|
pxor %xmm1, %xmm9 # Cyphertext XOR E(K, Yn)
|
|
|
|
|
|
|
|
mov \PLAIN_CYPH_LEN, %r10
|
|
|
|
add %r13, %r10
|
|
|
|
# Set r10 to be the amount of data left in CYPH_PLAIN_IN after filling
|
|
|
|
sub $16, %r10
|
|
|
|
# Determine if if partial block is not being filled and
|
|
|
|
# shift mask accordingly
|
|
|
|
jge _no_extra_mask_1_\@
|
|
|
|
sub %r10, %r12
|
|
|
|
_no_extra_mask_1_\@:
|
|
|
|
|
|
|
|
movdqu ALL_F-SHIFT_MASK(%r12), %xmm1
|
|
|
|
# get the appropriate mask to mask out bottom r13 bytes of xmm9
|
|
|
|
pand %xmm1, %xmm9 # mask out bottom r13 bytes of xmm9
|
|
|
|
|
|
|
|
pand %xmm1, %xmm3
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
PSHUFB_XMM %xmm10, %xmm3
|
|
|
|
PSHUFB_XMM %xmm2, %xmm3
|
|
|
|
pxor %xmm3, \AAD_HASH
|
|
|
|
|
|
|
|
cmp $0, %r10
|
|
|
|
jl _partial_incomplete_1_\@
|
|
|
|
|
|
|
|
# GHASH computation for the last <16 Byte block
|
|
|
|
GHASH_MUL \AAD_HASH, %xmm13, %xmm0, %xmm10, %xmm11, %xmm5, %xmm6
|
2018-07-02 10:31:54 +00:00
|
|
|
xor %eax, %eax
|
2018-02-14 17:40:19 +00:00
|
|
|
|
|
|
|
mov %rax, PBlockLen(%arg2)
|
|
|
|
jmp _dec_done_\@
|
|
|
|
_partial_incomplete_1_\@:
|
|
|
|
add \PLAIN_CYPH_LEN, PBlockLen(%arg2)
|
|
|
|
_dec_done_\@:
|
|
|
|
movdqu \AAD_HASH, AadHash(%arg2)
|
|
|
|
.else
|
|
|
|
pxor %xmm1, %xmm9 # Plaintext XOR E(K, Yn)
|
|
|
|
|
|
|
|
mov \PLAIN_CYPH_LEN, %r10
|
|
|
|
add %r13, %r10
|
|
|
|
# Set r10 to be the amount of data left in CYPH_PLAIN_IN after filling
|
|
|
|
sub $16, %r10
|
|
|
|
# Determine if if partial block is not being filled and
|
|
|
|
# shift mask accordingly
|
|
|
|
jge _no_extra_mask_2_\@
|
|
|
|
sub %r10, %r12
|
|
|
|
_no_extra_mask_2_\@:
|
|
|
|
|
|
|
|
movdqu ALL_F-SHIFT_MASK(%r12), %xmm1
|
|
|
|
# get the appropriate mask to mask out bottom r13 bytes of xmm9
|
|
|
|
pand %xmm1, %xmm9
|
|
|
|
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm1
|
|
|
|
PSHUFB_XMM %xmm1, %xmm9
|
|
|
|
PSHUFB_XMM %xmm2, %xmm9
|
|
|
|
pxor %xmm9, \AAD_HASH
|
|
|
|
|
|
|
|
cmp $0, %r10
|
|
|
|
jl _partial_incomplete_2_\@
|
|
|
|
|
|
|
|
# GHASH computation for the last <16 Byte block
|
|
|
|
GHASH_MUL \AAD_HASH, %xmm13, %xmm0, %xmm10, %xmm11, %xmm5, %xmm6
|
2018-07-02 10:31:54 +00:00
|
|
|
xor %eax, %eax
|
2018-02-14 17:40:19 +00:00
|
|
|
|
|
|
|
mov %rax, PBlockLen(%arg2)
|
|
|
|
jmp _encode_done_\@
|
|
|
|
_partial_incomplete_2_\@:
|
|
|
|
add \PLAIN_CYPH_LEN, PBlockLen(%arg2)
|
|
|
|
_encode_done_\@:
|
|
|
|
movdqu \AAD_HASH, AadHash(%arg2)
|
|
|
|
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm10
|
|
|
|
# shuffle xmm9 back to output as ciphertext
|
|
|
|
PSHUFB_XMM %xmm10, %xmm9
|
|
|
|
PSHUFB_XMM %xmm2, %xmm9
|
|
|
|
.endif
|
|
|
|
# output encrypted Bytes
|
|
|
|
cmp $0, %r10
|
|
|
|
jl _partial_fill_\@
|
|
|
|
mov %r13, %r12
|
|
|
|
mov $16, %r13
|
|
|
|
# Set r13 to be the number of bytes to write out
|
|
|
|
sub %r12, %r13
|
|
|
|
jmp _count_set_\@
|
|
|
|
_partial_fill_\@:
|
|
|
|
mov \PLAIN_CYPH_LEN, %r13
|
|
|
|
_count_set_\@:
|
|
|
|
movdqa %xmm9, %xmm0
|
|
|
|
MOVQ_R64_XMM %xmm0, %rax
|
|
|
|
cmp $8, %r13
|
|
|
|
jle _less_than_8_bytes_left_\@
|
|
|
|
|
|
|
|
mov %rax, (\CYPH_PLAIN_OUT, \DATA_OFFSET, 1)
|
|
|
|
add $8, \DATA_OFFSET
|
|
|
|
psrldq $8, %xmm0
|
|
|
|
MOVQ_R64_XMM %xmm0, %rax
|
|
|
|
sub $8, %r13
|
|
|
|
_less_than_8_bytes_left_\@:
|
|
|
|
movb %al, (\CYPH_PLAIN_OUT, \DATA_OFFSET, 1)
|
|
|
|
add $1, \DATA_OFFSET
|
|
|
|
shr $8, %rax
|
|
|
|
sub $1, %r13
|
|
|
|
jne _less_than_8_bytes_left_\@
|
|
|
|
_partial_block_done_\@:
|
|
|
|
.endm # PARTIAL_BLOCK
|
|
|
|
|
2018-02-14 17:39:36 +00:00
|
|
|
/*
|
|
|
|
* if a = number of total plaintext bytes
|
|
|
|
* b = floor(a/16)
|
|
|
|
* num_initial_blocks = b mod 4
|
|
|
|
* encrypt the initial num_initial_blocks blocks and apply ghash on
|
|
|
|
* the ciphertext
|
|
|
|
* %r10, %r11, %r12, %rax, %xmm5, %xmm6, %xmm7, %xmm8, %xmm9 registers
|
|
|
|
* are clobbered
|
2018-02-14 17:40:10 +00:00
|
|
|
* arg1, %arg2, %arg3 are used as a pointer only, not modified
|
2018-02-14 17:39:36 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
.macro INITIAL_BLOCKS_ENC_DEC TMP1 TMP2 TMP3 TMP4 TMP5 XMM0 XMM1 \
|
|
|
|
XMM2 XMM3 XMM4 XMMDst TMP6 TMP7 i i_seq operation
|
2018-02-14 17:39:45 +00:00
|
|
|
MOVADQ SHUF_MASK(%rip), %xmm14
|
2018-02-14 17:39:36 +00:00
|
|
|
|
|
|
|
movdqu AadHash(%arg2), %xmm\i # XMM0 = Y0
|
|
|
|
|
2017-04-28 16:11:56 +00:00
|
|
|
# start AES for num_initial_blocks blocks
|
2010-12-13 11:51:15 +00:00
|
|
|
|
2018-02-14 17:39:45 +00:00
|
|
|
movdqu CurCount(%arg2), \XMM0 # XMM0 = Y0
|
2010-12-13 11:51:15 +00:00
|
|
|
|
|
|
|
.if (\i == 5) || (\i == 6) || (\i == 7)
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ ONE(%RIP),\TMP1
|
|
|
|
MOVADQ 0(%arg1),\TMP2
|
2010-12-13 11:51:15 +00:00
|
|
|
.irpc index, \i_seq
|
2015-01-13 18:16:43 +00:00
|
|
|
paddd \TMP1, \XMM0 # INCR Y0
|
2018-02-14 17:38:12 +00:00
|
|
|
.ifc \operation, dec
|
|
|
|
movdqa \XMM0, %xmm\index
|
|
|
|
.else
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ \XMM0, %xmm\index
|
2018-02-14 17:38:12 +00:00
|
|
|
.endif
|
2015-01-13 18:16:43 +00:00
|
|
|
PSHUFB_XMM %xmm14, %xmm\index # perform a 16 byte swap
|
|
|
|
pxor \TMP2, %xmm\index
|
2010-12-13 11:51:15 +00:00
|
|
|
.endr
|
2015-01-13 18:16:43 +00:00
|
|
|
lea 0x10(%arg1),%r10
|
|
|
|
mov keysize,%eax
|
|
|
|
shr $2,%eax # 128->4, 192->6, 256->8
|
|
|
|
add $5,%eax # 128->9, 192->11, 256->13
|
|
|
|
|
2018-02-14 17:38:12 +00:00
|
|
|
aes_loop_initial_\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10),\TMP1
|
|
|
|
.irpc index, \i_seq
|
|
|
|
AESENC \TMP1, %xmm\index
|
2010-12-13 11:51:15 +00:00
|
|
|
.endr
|
2015-01-13 18:16:43 +00:00
|
|
|
add $16,%r10
|
|
|
|
sub $1,%eax
|
2018-02-14 17:38:12 +00:00
|
|
|
jnz aes_loop_initial_\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
|
|
|
MOVADQ (%r10), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
.irpc index, \i_seq
|
2015-01-13 18:16:43 +00:00
|
|
|
AESENCLAST \TMP1, %xmm\index # Last Round
|
2010-12-13 11:51:15 +00:00
|
|
|
.endr
|
|
|
|
.irpc index, \i_seq
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu (%arg4 , %r11, 1), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP1, %xmm\index
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu %xmm\index, (%arg3 , %r11, 1)
|
2010-12-13 11:51:15 +00:00
|
|
|
# write back plaintext/ciphertext for num_initial_blocks
|
|
|
|
add $16, %r11
|
2018-02-14 17:38:12 +00:00
|
|
|
|
|
|
|
.ifc \operation, dec
|
|
|
|
movdqa \TMP1, %xmm\index
|
|
|
|
.endif
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, %xmm\index
|
|
|
|
|
|
|
|
# prepare plaintext/ciphertext for GHASH computation
|
|
|
|
.endr
|
|
|
|
.endif
|
2017-04-28 16:11:56 +00:00
|
|
|
|
2010-12-13 11:51:15 +00:00
|
|
|
# apply GHASH on num_initial_blocks blocks
|
|
|
|
|
|
|
|
.if \i == 5
|
|
|
|
pxor %xmm5, %xmm6
|
|
|
|
GHASH_MUL %xmm6, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
pxor %xmm6, %xmm7
|
|
|
|
GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
pxor %xmm7, %xmm8
|
|
|
|
GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
.elseif \i == 6
|
|
|
|
pxor %xmm6, %xmm7
|
|
|
|
GHASH_MUL %xmm7, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
pxor %xmm7, %xmm8
|
|
|
|
GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
.elseif \i == 7
|
|
|
|
pxor %xmm7, %xmm8
|
|
|
|
GHASH_MUL %xmm8, \TMP3, \TMP1, \TMP2, \TMP4, \TMP5, \XMM1
|
|
|
|
.endif
|
|
|
|
cmp $64, %r13
|
2018-02-14 17:38:12 +00:00
|
|
|
jl _initial_blocks_done\@
|
2010-12-13 11:51:15 +00:00
|
|
|
# no need for precomputed values
|
|
|
|
/*
|
|
|
|
*
|
|
|
|
* Precomputations for HashKey parallel with encryption of first 4 blocks.
|
|
|
|
* Haskey_i_k holds XORed values of the low and high parts of the Haskey_i
|
|
|
|
*/
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ ONE(%RIP),\TMP1
|
|
|
|
paddd \TMP1, \XMM0 # INCR Y0
|
|
|
|
MOVADQ \XMM0, \XMM1
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
paddd \TMP1, \XMM0 # INCR Y0
|
|
|
|
MOVADQ \XMM0, \XMM2
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
paddd \TMP1, \XMM0 # INCR Y0
|
|
|
|
MOVADQ \XMM0, \XMM3
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
paddd \TMP1, \XMM0 # INCR Y0
|
|
|
|
MOVADQ \XMM0, \XMM4
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ 0(%arg1),\TMP1
|
|
|
|
pxor \TMP1, \XMM1
|
|
|
|
pxor \TMP1, \XMM2
|
|
|
|
pxor \TMP1, \XMM3
|
|
|
|
pxor \TMP1, \XMM4
|
2010-12-13 11:51:15 +00:00
|
|
|
.irpc index, 1234 # do 4 rounds
|
|
|
|
movaps 0x10*\index(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
.endr
|
|
|
|
.irpc index, 56789 # do next 5 rounds
|
|
|
|
movaps 0x10*\index(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
.endr
|
2015-01-13 18:16:43 +00:00
|
|
|
lea 0xa0(%arg1),%r10
|
|
|
|
mov keysize,%eax
|
|
|
|
shr $2,%eax # 128->4, 192->6, 256->8
|
|
|
|
sub $4,%eax # 128->0, 192->2, 256->4
|
2018-02-14 17:38:12 +00:00
|
|
|
jz aes_loop_pre_done\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:38:12 +00:00
|
|
|
aes_loop_pre_\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10),\TMP2
|
|
|
|
.irpc index, 1234
|
|
|
|
AESENC \TMP2, %xmm\index
|
|
|
|
.endr
|
|
|
|
add $16,%r10
|
|
|
|
sub $1,%eax
|
2018-02-14 17:38:12 +00:00
|
|
|
jnz aes_loop_pre_\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:38:12 +00:00
|
|
|
aes_loop_pre_done\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10), \TMP2
|
2010-12-13 11:51:15 +00:00
|
|
|
AESENCLAST \TMP2, \XMM1
|
|
|
|
AESENCLAST \TMP2, \XMM2
|
|
|
|
AESENCLAST \TMP2, \XMM3
|
|
|
|
AESENCLAST \TMP2, \XMM4
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16*0(%arg4 , %r11 , 1), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP1, \XMM1
|
2018-02-14 17:38:12 +00:00
|
|
|
.ifc \operation, dec
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM1, 16*0(%arg3 , %r11 , 1)
|
2018-02-14 17:38:12 +00:00
|
|
|
movdqa \TMP1, \XMM1
|
|
|
|
.endif
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16*1(%arg4 , %r11 , 1), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP1, \XMM2
|
2018-02-14 17:38:12 +00:00
|
|
|
.ifc \operation, dec
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM2, 16*1(%arg3 , %r11 , 1)
|
2018-02-14 17:38:12 +00:00
|
|
|
movdqa \TMP1, \XMM2
|
|
|
|
.endif
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16*2(%arg4 , %r11 , 1), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP1, \XMM3
|
2018-02-14 17:38:12 +00:00
|
|
|
.ifc \operation, dec
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM3, 16*2(%arg3 , %r11 , 1)
|
2018-02-14 17:38:12 +00:00
|
|
|
movdqa \TMP1, \XMM3
|
|
|
|
.endif
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16*3(%arg4 , %r11 , 1), \TMP1
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP1, \XMM4
|
2018-02-14 17:38:12 +00:00
|
|
|
.ifc \operation, dec
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM4, 16*3(%arg3 , %r11 , 1)
|
2018-02-14 17:38:12 +00:00
|
|
|
movdqa \TMP1, \XMM4
|
|
|
|
.else
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM1, 16*0(%arg3 , %r11 , 1)
|
|
|
|
movdqu \XMM2, 16*1(%arg3 , %r11 , 1)
|
|
|
|
movdqu \XMM3, 16*2(%arg3 , %r11 , 1)
|
|
|
|
movdqu \XMM4, 16*3(%arg3 , %r11 , 1)
|
2018-02-14 17:38:12 +00:00
|
|
|
.endif
|
2010-12-13 11:51:15 +00:00
|
|
|
|
2010-11-04 19:00:45 +00:00
|
|
|
add $64, %r11
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM1 # perform a 16 byte swap
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor \XMMDst, \XMM1
|
|
|
|
# combine GHASHed value with the corresponding ciphertext
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm14, \XMM2 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm14, \XMM3 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm14, \XMM4 # perform a 16 byte swap
|
|
|
|
|
2018-02-14 17:38:12 +00:00
|
|
|
_initial_blocks_done\@:
|
2010-12-13 11:51:15 +00:00
|
|
|
|
2010-11-04 19:00:45 +00:00
|
|
|
.endm
|
|
|
|
|
|
|
|
/*
|
|
|
|
* encrypt 4 blocks at a time
|
|
|
|
* ghash the 4 previously encrypted ciphertext blocks
|
2018-02-14 17:39:23 +00:00
|
|
|
* arg1, %arg3, %arg4 are used as pointers only, not modified
|
2010-11-04 19:00:45 +00:00
|
|
|
* %r11 is the data offset value
|
|
|
|
*/
|
2010-12-13 11:51:15 +00:00
|
|
|
.macro GHASH_4_ENCRYPT_4_PARALLEL_ENC TMP1 TMP2 TMP3 TMP4 TMP5 \
|
|
|
|
TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
|
|
|
|
|
|
|
|
movdqa \XMM1, \XMM5
|
|
|
|
movdqa \XMM2, \XMM6
|
|
|
|
movdqa \XMM3, \XMM7
|
|
|
|
movdqa \XMM4, \XMM8
|
|
|
|
|
|
|
|
movdqa SHUF_MASK(%rip), %xmm15
|
|
|
|
# multiply TMP5 * HashKey using karatsuba
|
|
|
|
|
|
|
|
movdqa \XMM5, \TMP4
|
|
|
|
pshufd $78, \XMM5, \TMP6
|
|
|
|
pxor \XMM5, \TMP6
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP4 # TMP4 = a1*b1
|
|
|
|
movdqa \XMM0, \XMM1
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM2
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM3
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM4
|
|
|
|
PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM5 # XMM5 = a0*b0
|
|
|
|
PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
|
|
|
|
|
|
|
|
pxor (%arg1), \XMM1
|
|
|
|
pxor (%arg1), \XMM2
|
|
|
|
pxor (%arg1), \XMM3
|
|
|
|
pxor (%arg1), \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4_k(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP6 # TMP6 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x10(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1 # Round 1
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
movaps 0x20(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1 # Round 2
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
movdqa \XMM6, \TMP1
|
|
|
|
pshufd $78, \XMM6, \TMP2
|
|
|
|
pxor \XMM6, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1 * b1
|
|
|
|
movaps 0x30(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 3
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM6 # XMM6 = a0*b0
|
|
|
|
movaps 0x40(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 4
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3_k(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x50(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 5
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
pxor \TMP1, \TMP4
|
|
|
|
# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
|
|
|
|
pxor \XMM6, \XMM5
|
|
|
|
pxor \TMP2, \TMP6
|
|
|
|
movdqa \XMM7, \TMP1
|
|
|
|
pshufd $78, \XMM7, \TMP2
|
|
|
|
pxor \XMM7, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
|
|
|
|
# Multiply TMP5 * HashKey using karatsuba
|
|
|
|
|
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
movaps 0x60(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 6
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM7 # XMM7 = a0*b0
|
|
|
|
movaps 0x70(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 7
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2_k(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x80(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 8
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
pxor \TMP1, \TMP4
|
|
|
|
# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
|
|
|
|
pxor \XMM7, \XMM5
|
|
|
|
pxor \TMP2, \TMP6
|
|
|
|
|
|
|
|
# Multiply XMM8 * HashKey
|
|
|
|
# XMM8 and TMP5 hold the values for the two operands
|
|
|
|
|
|
|
|
movdqa \XMM8, \TMP1
|
|
|
|
pshufd $78, \XMM8, \TMP2
|
|
|
|
pxor \XMM8, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
movaps 0x90(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 9
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM8 # XMM8 = a0*b0
|
2015-01-13 18:16:43 +00:00
|
|
|
lea 0xa0(%arg1),%r10
|
|
|
|
mov keysize,%eax
|
|
|
|
shr $2,%eax # 128->4, 192->6, 256->8
|
|
|
|
sub $4,%eax # 128->0, 192->2, 256->4
|
2018-02-14 17:40:47 +00:00
|
|
|
jz aes_loop_par_enc_done\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
aes_loop_par_enc\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10),\TMP3
|
|
|
|
.irpc index, 1234
|
|
|
|
AESENC \TMP3, %xmm\index
|
|
|
|
.endr
|
|
|
|
add $16,%r10
|
|
|
|
sub $1,%eax
|
2018-02-14 17:40:47 +00:00
|
|
|
jnz aes_loop_par_enc\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
aes_loop_par_enc_done\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10), \TMP3
|
2010-12-13 11:51:15 +00:00
|
|
|
AESENCLAST \TMP3, \XMM1 # Round 10
|
|
|
|
AESENCLAST \TMP3, \XMM2
|
|
|
|
AESENCLAST \TMP3, \XMM3
|
|
|
|
AESENCLAST \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_k(%arg2), \TMP5
|
2010-12-13 11:51:15 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu (%arg4,%r11,1), \TMP3
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP3, \XMM1 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16(%arg4,%r11,1), \TMP3
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP3, \XMM2 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 32(%arg4,%r11,1), \TMP3
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP3, \XMM3 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 48(%arg4,%r11,1), \TMP3
|
2010-12-13 11:51:15 +00:00
|
|
|
pxor \TMP3, \XMM4 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM1, (%arg3,%r11,1) # Write to the ciphertext buffer
|
|
|
|
movdqu \XMM2, 16(%arg3,%r11,1) # Write to the ciphertext buffer
|
|
|
|
movdqu \XMM3, 32(%arg3,%r11,1) # Write to the ciphertext buffer
|
|
|
|
movdqu \XMM4, 48(%arg3,%r11,1) # Write to the ciphertext buffer
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
|
|
|
|
|
|
|
|
pxor \TMP4, \TMP1
|
|
|
|
pxor \XMM8, \XMM5
|
|
|
|
pxor \TMP6, \TMP2
|
|
|
|
pxor \TMP1, \TMP2
|
|
|
|
pxor \XMM5, \TMP2
|
|
|
|
movdqa \TMP2, \TMP3
|
|
|
|
pslldq $8, \TMP3 # left shift TMP3 2 DWs
|
|
|
|
psrldq $8, \TMP2 # right shift TMP2 2 DWs
|
|
|
|
pxor \TMP3, \XMM5
|
|
|
|
pxor \TMP2, \TMP1 # accumulate the results in TMP1:XMM5
|
|
|
|
|
|
|
|
# first phase of reduction
|
|
|
|
|
|
|
|
movdqa \XMM5, \TMP2
|
|
|
|
movdqa \XMM5, \TMP3
|
|
|
|
movdqa \XMM5, \TMP4
|
|
|
|
# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
|
|
|
|
pslld $31, \TMP2 # packed right shift << 31
|
|
|
|
pslld $30, \TMP3 # packed right shift << 30
|
|
|
|
pslld $25, \TMP4 # packed right shift << 25
|
|
|
|
pxor \TMP3, \TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4, \TMP2
|
|
|
|
movdqa \TMP2, \TMP5
|
|
|
|
psrldq $4, \TMP5 # right shift T5 1 DW
|
|
|
|
pslldq $12, \TMP2 # left shift T2 3 DWs
|
|
|
|
pxor \TMP2, \XMM5
|
|
|
|
|
|
|
|
# second phase of reduction
|
|
|
|
|
|
|
|
movdqa \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
|
|
|
|
movdqa \XMM5,\TMP3
|
|
|
|
movdqa \XMM5,\TMP4
|
|
|
|
psrld $1, \TMP2 # packed left shift >>1
|
|
|
|
psrld $2, \TMP3 # packed left shift >>2
|
|
|
|
psrld $7, \TMP4 # packed left shift >>7
|
|
|
|
pxor \TMP3,\TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4,\TMP2
|
|
|
|
pxor \TMP5, \TMP2
|
|
|
|
pxor \TMP2, \XMM5
|
|
|
|
pxor \TMP1, \XMM5 # result is in TMP1
|
|
|
|
|
|
|
|
pxor \XMM5, \XMM1
|
|
|
|
.endm
|
|
|
|
|
|
|
|
/*
|
|
|
|
* decrypt 4 blocks at a time
|
|
|
|
* ghash the 4 previously decrypted ciphertext blocks
|
2018-02-14 17:39:23 +00:00
|
|
|
* arg1, %arg3, %arg4 are used as pointers only, not modified
|
2010-12-13 11:51:15 +00:00
|
|
|
* %r11 is the data offset value
|
|
|
|
*/
|
|
|
|
.macro GHASH_4_ENCRYPT_4_PARALLEL_DEC TMP1 TMP2 TMP3 TMP4 TMP5 \
|
2010-11-04 19:00:45 +00:00
|
|
|
TMP6 XMM0 XMM1 XMM2 XMM3 XMM4 XMM5 XMM6 XMM7 XMM8 operation
|
|
|
|
|
|
|
|
movdqa \XMM1, \XMM5
|
|
|
|
movdqa \XMM2, \XMM6
|
|
|
|
movdqa \XMM3, \XMM7
|
|
|
|
movdqa \XMM4, \XMM8
|
|
|
|
|
2010-12-13 11:51:15 +00:00
|
|
|
movdqa SHUF_MASK(%rip), %xmm15
|
2010-11-04 19:00:45 +00:00
|
|
|
# multiply TMP5 * HashKey using karatsuba
|
|
|
|
|
|
|
|
movdqa \XMM5, \TMP4
|
|
|
|
pshufd $78, \XMM5, \TMP6
|
|
|
|
pxor \XMM5, \TMP6
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP4 # TMP4 = a1*b1
|
|
|
|
movdqa \XMM0, \XMM1
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM2
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM3
|
|
|
|
paddd ONE(%rip), \XMM0 # INCR CNT
|
|
|
|
movdqa \XMM0, \XMM4
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM5 # XMM5 = a0*b0
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
|
|
|
|
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor (%arg1), \XMM1
|
|
|
|
pxor (%arg1), \XMM2
|
|
|
|
pxor (%arg1), \XMM3
|
|
|
|
pxor (%arg1), \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4_k(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP6 # TMP6 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x10(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1 # Round 1
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
movaps 0x20(%arg1), \TMP1
|
|
|
|
AESENC \TMP1, \XMM1 # Round 2
|
|
|
|
AESENC \TMP1, \XMM2
|
|
|
|
AESENC \TMP1, \XMM3
|
|
|
|
AESENC \TMP1, \XMM4
|
|
|
|
movdqa \XMM6, \TMP1
|
|
|
|
pshufd $78, \XMM6, \TMP2
|
|
|
|
pxor \XMM6, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1 * b1
|
|
|
|
movaps 0x30(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 3
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM6 # XMM6 = a0*b0
|
|
|
|
movaps 0x40(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 4
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3_k(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x50(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 5
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
pxor \TMP1, \TMP4
|
|
|
|
# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
|
|
|
|
pxor \XMM6, \XMM5
|
|
|
|
pxor \TMP2, \TMP6
|
|
|
|
movdqa \XMM7, \TMP1
|
|
|
|
pshufd $78, \XMM7, \TMP2
|
|
|
|
pxor \XMM7, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
# Multiply TMP5 * HashKey using karatsuba
|
|
|
|
|
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
movaps 0x60(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 6
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM7 # XMM7 = a0*b0
|
|
|
|
movaps 0x70(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 7
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2_k(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
movaps 0x80(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 8
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
pxor \TMP1, \TMP4
|
|
|
|
# accumulate the results in TMP4:XMM5, TMP6 holds the middle part
|
|
|
|
pxor \XMM7, \XMM5
|
|
|
|
pxor \TMP2, \TMP6
|
|
|
|
|
|
|
|
# Multiply XMM8 * HashKey
|
|
|
|
# XMM8 and TMP5 hold the values for the two operands
|
|
|
|
|
|
|
|
movdqa \XMM8, \TMP1
|
|
|
|
pshufd $78, \XMM8, \TMP2
|
|
|
|
pxor \XMM8, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
movaps 0x90(%arg1), \TMP3
|
|
|
|
AESENC \TMP3, \XMM1 # Round 9
|
|
|
|
AESENC \TMP3, \XMM2
|
|
|
|
AESENC \TMP3, \XMM3
|
|
|
|
AESENC \TMP3, \XMM4
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM8 # XMM8 = a0*b0
|
2015-01-13 18:16:43 +00:00
|
|
|
lea 0xa0(%arg1),%r10
|
|
|
|
mov keysize,%eax
|
|
|
|
shr $2,%eax # 128->4, 192->6, 256->8
|
|
|
|
sub $4,%eax # 128->0, 192->2, 256->4
|
2018-02-14 17:40:47 +00:00
|
|
|
jz aes_loop_par_dec_done\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
aes_loop_par_dec\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10),\TMP3
|
|
|
|
.irpc index, 1234
|
|
|
|
AESENC \TMP3, %xmm\index
|
|
|
|
.endr
|
|
|
|
add $16,%r10
|
|
|
|
sub $1,%eax
|
2018-02-14 17:40:47 +00:00
|
|
|
jnz aes_loop_par_dec\@
|
2015-01-13 18:16:43 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
aes_loop_par_dec_done\@:
|
2015-01-13 18:16:43 +00:00
|
|
|
MOVADQ (%r10), \TMP3
|
|
|
|
AESENCLAST \TMP3, \XMM1 # last round
|
2010-11-04 19:00:45 +00:00
|
|
|
AESENCLAST \TMP3, \XMM2
|
|
|
|
AESENCLAST \TMP3, \XMM3
|
|
|
|
AESENCLAST \TMP3, \XMM4
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_k(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP5, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu (%arg4,%r11,1), \TMP3
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor \TMP3, \XMM1 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM1, (%arg3,%r11,1) # Write to plaintext buffer
|
2010-11-04 19:00:45 +00:00
|
|
|
movdqa \TMP3, \XMM1
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 16(%arg4,%r11,1), \TMP3
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor \TMP3, \XMM2 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM2, 16(%arg3,%r11,1) # Write to plaintext buffer
|
2010-11-04 19:00:45 +00:00
|
|
|
movdqa \TMP3, \XMM2
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 32(%arg4,%r11,1), \TMP3
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor \TMP3, \XMM3 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM3, 32(%arg3,%r11,1) # Write to plaintext buffer
|
2010-11-04 19:00:45 +00:00
|
|
|
movdqa \TMP3, \XMM3
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu 48(%arg4,%r11,1), \TMP3
|
2010-11-04 19:00:45 +00:00
|
|
|
pxor \TMP3, \XMM4 # Ciphertext/Plaintext XOR EK
|
2018-02-14 17:39:23 +00:00
|
|
|
movdqu \XMM4, 48(%arg3,%r11,1) # Write to plaintext buffer
|
2010-11-04 19:00:45 +00:00
|
|
|
movdqa \TMP3, \XMM4
|
2010-12-13 11:51:15 +00:00
|
|
|
PSHUFB_XMM %xmm15, \XMM1 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM2 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM3 # perform a 16 byte swap
|
|
|
|
PSHUFB_XMM %xmm15, \XMM4 # perform a 16 byte swap
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
pxor \TMP4, \TMP1
|
|
|
|
pxor \XMM8, \XMM5
|
|
|
|
pxor \TMP6, \TMP2
|
|
|
|
pxor \TMP1, \TMP2
|
|
|
|
pxor \XMM5, \TMP2
|
|
|
|
movdqa \TMP2, \TMP3
|
|
|
|
pslldq $8, \TMP3 # left shift TMP3 2 DWs
|
|
|
|
psrldq $8, \TMP2 # right shift TMP2 2 DWs
|
|
|
|
pxor \TMP3, \XMM5
|
|
|
|
pxor \TMP2, \TMP1 # accumulate the results in TMP1:XMM5
|
|
|
|
|
|
|
|
# first phase of reduction
|
|
|
|
|
|
|
|
movdqa \XMM5, \TMP2
|
|
|
|
movdqa \XMM5, \TMP3
|
|
|
|
movdqa \XMM5, \TMP4
|
|
|
|
# move XMM5 into TMP2, TMP3, TMP4 in order to perform shifts independently
|
|
|
|
pslld $31, \TMP2 # packed right shift << 31
|
|
|
|
pslld $30, \TMP3 # packed right shift << 30
|
|
|
|
pslld $25, \TMP4 # packed right shift << 25
|
|
|
|
pxor \TMP3, \TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4, \TMP2
|
|
|
|
movdqa \TMP2, \TMP5
|
|
|
|
psrldq $4, \TMP5 # right shift T5 1 DW
|
|
|
|
pslldq $12, \TMP2 # left shift T2 3 DWs
|
|
|
|
pxor \TMP2, \XMM5
|
|
|
|
|
|
|
|
# second phase of reduction
|
|
|
|
|
|
|
|
movdqa \XMM5,\TMP2 # make 3 copies of XMM5 into TMP2, TMP3, TMP4
|
|
|
|
movdqa \XMM5,\TMP3
|
|
|
|
movdqa \XMM5,\TMP4
|
|
|
|
psrld $1, \TMP2 # packed left shift >>1
|
|
|
|
psrld $2, \TMP3 # packed left shift >>2
|
|
|
|
psrld $7, \TMP4 # packed left shift >>7
|
|
|
|
pxor \TMP3,\TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4,\TMP2
|
|
|
|
pxor \TMP5, \TMP2
|
|
|
|
pxor \TMP2, \XMM5
|
|
|
|
pxor \TMP1, \XMM5 # result is in TMP1
|
|
|
|
|
|
|
|
pxor \XMM5, \XMM1
|
|
|
|
.endm
|
|
|
|
|
|
|
|
/* GHASH the last 4 ciphertext blocks. */
|
|
|
|
.macro GHASH_LAST_4 TMP1 TMP2 TMP3 TMP4 TMP5 TMP6 \
|
|
|
|
TMP7 XMM1 XMM2 XMM3 XMM4 XMMDst
|
|
|
|
|
|
|
|
# Multiply TMP6 * HashKey (using Karatsuba)
|
|
|
|
|
|
|
|
movdqa \XMM1, \TMP6
|
|
|
|
pshufd $78, \XMM1, \TMP2
|
|
|
|
pxor \XMM1, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP6 # TMP6 = a1*b1
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM1 # XMM1 = a0*b0
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_4_k(%arg2), \TMP4
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
movdqa \XMM1, \XMMDst
|
|
|
|
movdqa \TMP2, \XMM1 # result in TMP6, XMMDst, XMM1
|
|
|
|
|
|
|
|
# Multiply TMP1 * HashKey (using Karatsuba)
|
|
|
|
|
|
|
|
movdqa \XMM2, \TMP1
|
|
|
|
pshufd $78, \XMM2, \TMP2
|
|
|
|
pxor \XMM2, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM2 # XMM2 = a0*b0
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_3_k(%arg2), \TMP4
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
pxor \TMP1, \TMP6
|
|
|
|
pxor \XMM2, \XMMDst
|
|
|
|
pxor \TMP2, \XMM1
|
|
|
|
# results accumulated in TMP6, XMMDst, XMM1
|
|
|
|
|
|
|
|
# Multiply TMP1 * HashKey (using Karatsuba)
|
|
|
|
|
|
|
|
movdqa \XMM3, \TMP1
|
|
|
|
pshufd $78, \XMM3, \TMP2
|
|
|
|
pxor \XMM3, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM3 # XMM3 = a0*b0
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_2_k(%arg2), \TMP4
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
pxor \TMP1, \TMP6
|
|
|
|
pxor \XMM3, \XMMDst
|
|
|
|
pxor \TMP2, \XMM1 # results accumulated in TMP6, XMMDst, XMM1
|
|
|
|
|
|
|
|
# Multiply TMP1 * HashKey (using Karatsuba)
|
|
|
|
movdqa \XMM4, \TMP1
|
|
|
|
pshufd $78, \XMM4, \TMP2
|
|
|
|
pxor \XMM4, \TMP2
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey(%arg2), \TMP5
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x11, \TMP5, \TMP1 # TMP1 = a1*b1
|
|
|
|
PCLMULQDQ 0x00, \TMP5, \XMM4 # XMM4 = a0*b0
|
2018-08-15 17:29:42 +00:00
|
|
|
movdqu HashKey_k(%arg2), \TMP4
|
2010-11-04 19:00:45 +00:00
|
|
|
PCLMULQDQ 0x00, \TMP4, \TMP2 # TMP2 = (a1+a0)*(b1+b0)
|
|
|
|
pxor \TMP1, \TMP6
|
|
|
|
pxor \XMM4, \XMMDst
|
|
|
|
pxor \XMM1, \TMP2
|
|
|
|
pxor \TMP6, \TMP2
|
|
|
|
pxor \XMMDst, \TMP2
|
|
|
|
# middle section of the temp results combined as in karatsuba algorithm
|
|
|
|
movdqa \TMP2, \TMP4
|
|
|
|
pslldq $8, \TMP4 # left shift TMP4 2 DWs
|
|
|
|
psrldq $8, \TMP2 # right shift TMP2 2 DWs
|
|
|
|
pxor \TMP4, \XMMDst
|
|
|
|
pxor \TMP2, \TMP6
|
|
|
|
# TMP6:XMMDst holds the result of the accumulated carry-less multiplications
|
|
|
|
# first phase of the reduction
|
|
|
|
movdqa \XMMDst, \TMP2
|
|
|
|
movdqa \XMMDst, \TMP3
|
|
|
|
movdqa \XMMDst, \TMP4
|
|
|
|
# move XMMDst into TMP2, TMP3, TMP4 in order to perform 3 shifts independently
|
|
|
|
pslld $31, \TMP2 # packed right shifting << 31
|
|
|
|
pslld $30, \TMP3 # packed right shifting << 30
|
|
|
|
pslld $25, \TMP4 # packed right shifting << 25
|
|
|
|
pxor \TMP3, \TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4, \TMP2
|
|
|
|
movdqa \TMP2, \TMP7
|
|
|
|
psrldq $4, \TMP7 # right shift TMP7 1 DW
|
|
|
|
pslldq $12, \TMP2 # left shift TMP2 3 DWs
|
|
|
|
pxor \TMP2, \XMMDst
|
|
|
|
|
|
|
|
# second phase of the reduction
|
|
|
|
movdqa \XMMDst, \TMP2
|
|
|
|
# make 3 copies of XMMDst for doing 3 shift operations
|
|
|
|
movdqa \XMMDst, \TMP3
|
|
|
|
movdqa \XMMDst, \TMP4
|
|
|
|
psrld $1, \TMP2 # packed left shift >> 1
|
|
|
|
psrld $2, \TMP3 # packed left shift >> 2
|
|
|
|
psrld $7, \TMP4 # packed left shift >> 7
|
|
|
|
pxor \TMP3, \TMP2 # xor the shifted versions
|
|
|
|
pxor \TMP4, \TMP2
|
|
|
|
pxor \TMP7, \TMP2
|
|
|
|
pxor \TMP2, \XMMDst
|
|
|
|
pxor \TMP6, \XMMDst # reduced result is in XMMDst
|
|
|
|
.endm
|
|
|
|
|
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
/* Encryption of a single block
|
|
|
|
* uses eax & r10
|
|
|
|
*/
|
2010-11-04 19:00:45 +00:00
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
.macro ENCRYPT_SINGLE_BLOCK XMM0 TMP1
|
2010-11-04 19:00:45 +00:00
|
|
|
|
2015-01-13 18:16:43 +00:00
|
|
|
pxor (%arg1), \XMM0
|
|
|
|
mov keysize,%eax
|
|
|
|
shr $2,%eax # 128->4, 192->6, 256->8
|
|
|
|
add $5,%eax # 128->9, 192->11, 256->13
|
|
|
|
lea 16(%arg1), %r10 # get first expanded key address
|
|
|
|
|
|
|
|
_esb_loop_\@:
|
|
|
|
MOVADQ (%r10),\TMP1
|
|
|
|
AESENC \TMP1,\XMM0
|
|
|
|
add $16,%r10
|
|
|
|
sub $1,%eax
|
|
|
|
jnz _esb_loop_\@
|
|
|
|
|
|
|
|
MOVADQ (%r10),\TMP1
|
|
|
|
AESENCLAST \TMP1,\XMM0
|
|
|
|
.endm
|
2010-11-04 19:00:45 +00:00
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_dec(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
2018-02-14 17:39:23 +00:00
|
|
|
* struct gcm_context_data *data
|
|
|
|
* // Context data
|
2010-11-04 19:00:45 +00:00
|
|
|
* u8 *out, // Plaintext output. Encrypt in-place is allowed.
|
|
|
|
* const u8 *in, // Ciphertext input
|
|
|
|
* u64 plaintext_len, // Length of data in bytes for decryption.
|
|
|
|
* u8 *iv, // Pre-counter block j0: 4 byte salt (from Security Association)
|
|
|
|
* // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
|
|
|
|
* // concatenated with 0x00000001. 16-byte aligned pointer.
|
|
|
|
* u8 *hash_subkey, // H, the Hash sub key input. Data starts on a 16-byte boundary.
|
|
|
|
* const u8 *aad, // Additional Authentication Data (AAD)
|
|
|
|
* u64 aad_len, // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
|
|
|
|
* u8 *auth_tag, // Authenticated Tag output. The driver will compare this to the
|
|
|
|
* // given authentication tag and only return the plaintext if they match.
|
|
|
|
* u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16
|
|
|
|
* // (most likely), 12 or 8.
|
|
|
|
*
|
|
|
|
* Assumptions:
|
|
|
|
*
|
|
|
|
* keys:
|
|
|
|
* keys are pre-expanded and aligned to 16 bytes. we are using the first
|
|
|
|
* set of 11 keys in the data structure void *aes_ctx
|
|
|
|
*
|
|
|
|
* iv:
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | Salt (From the SA) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | Initialization Vector |
|
|
|
|
* | (This is the sequence number from IPSec header) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x1 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* AAD:
|
|
|
|
* AAD padded to 128 bits with 0
|
|
|
|
* for example, assume AAD is a u32 vector
|
|
|
|
*
|
|
|
|
* if AAD is 8 bytes:
|
|
|
|
* AAD[3] = {A0, A1};
|
|
|
|
* padded AAD in xmm register = {A1 A0 0 0}
|
|
|
|
*
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | SPI (A1) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 32-bit Sequence Number (A0) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x0 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
* AAD Format with 32-bit Sequence Number
|
|
|
|
*
|
|
|
|
* if AAD is 12 bytes:
|
|
|
|
* AAD[3] = {A0, A1, A2};
|
|
|
|
* padded AAD in xmm register = {A2 A1 A0 0}
|
|
|
|
*
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | SPI (A2) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 64-bit Extended Sequence Number {A1,A0} |
|
|
|
|
* | |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x0 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
* AAD Format with 64-bit Extended Sequence Number
|
|
|
|
*
|
|
|
|
* poly = x^128 + x^127 + x^126 + x^121 + 1
|
|
|
|
*
|
|
|
|
*****************************************************************************/
|
|
|
|
ENTRY(aesni_gcm_dec)
|
2018-02-14 17:38:35 +00:00
|
|
|
FUNC_SAVE
|
2010-11-04 19:00:45 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
GCM_INIT %arg6, arg7, arg8, arg9
|
2018-02-14 17:39:10 +00:00
|
|
|
GCM_ENC_DEC dec
|
2018-02-14 17:40:47 +00:00
|
|
|
GCM_COMPLETE arg10, arg11
|
2018-02-14 17:38:35 +00:00
|
|
|
FUNC_RESTORE
|
2010-11-04 19:00:45 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_gcm_dec)
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_enc(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
2018-02-14 17:39:23 +00:00
|
|
|
* struct gcm_context_data *data
|
|
|
|
* // Context data
|
2010-11-04 19:00:45 +00:00
|
|
|
* u8 *out, // Ciphertext output. Encrypt in-place is allowed.
|
|
|
|
* const u8 *in, // Plaintext input
|
|
|
|
* u64 plaintext_len, // Length of data in bytes for encryption.
|
|
|
|
* u8 *iv, // Pre-counter block j0: 4 byte salt (from Security Association)
|
|
|
|
* // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
|
|
|
|
* // concatenated with 0x00000001. 16-byte aligned pointer.
|
|
|
|
* u8 *hash_subkey, // H, the Hash sub key input. Data starts on a 16-byte boundary.
|
|
|
|
* const u8 *aad, // Additional Authentication Data (AAD)
|
|
|
|
* u64 aad_len, // Length of AAD in bytes. With RFC4106 this is going to be 8 or 12 bytes
|
|
|
|
* u8 *auth_tag, // Authenticated Tag output.
|
|
|
|
* u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16 (most likely),
|
|
|
|
* // 12 or 8.
|
|
|
|
*
|
|
|
|
* Assumptions:
|
|
|
|
*
|
|
|
|
* keys:
|
|
|
|
* keys are pre-expanded and aligned to 16 bytes. we are using the
|
|
|
|
* first set of 11 keys in the data structure void *aes_ctx
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* iv:
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | Salt (From the SA) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | Initialization Vector |
|
|
|
|
* | (This is the sequence number from IPSec header) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x1 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* AAD:
|
|
|
|
* AAD padded to 128 bits with 0
|
|
|
|
* for example, assume AAD is a u32 vector
|
|
|
|
*
|
|
|
|
* if AAD is 8 bytes:
|
|
|
|
* AAD[3] = {A0, A1};
|
|
|
|
* padded AAD in xmm register = {A1 A0 0 0}
|
|
|
|
*
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | SPI (A1) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 32-bit Sequence Number (A0) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x0 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
* AAD Format with 32-bit Sequence Number
|
|
|
|
*
|
|
|
|
* if AAD is 12 bytes:
|
|
|
|
* AAD[3] = {A0, A1, A2};
|
|
|
|
* padded AAD in xmm register = {A2 A1 A0 0}
|
|
|
|
*
|
|
|
|
* 0 1 2 3
|
|
|
|
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | SPI (A2) |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 64-bit Extended Sequence Number {A1,A0} |
|
|
|
|
* | |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
* | 0x0 |
|
|
|
|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|
|
|
|
*
|
|
|
|
* AAD Format with 64-bit Extended Sequence Number
|
|
|
|
*
|
|
|
|
* poly = x^128 + x^127 + x^126 + x^121 + 1
|
|
|
|
***************************************************************************/
|
|
|
|
ENTRY(aesni_gcm_enc)
|
2018-02-14 17:38:35 +00:00
|
|
|
FUNC_SAVE
|
2010-11-04 19:00:45 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
GCM_INIT %arg6, arg7, arg8, arg9
|
2018-02-14 17:39:10 +00:00
|
|
|
GCM_ENC_DEC enc
|
2018-02-14 17:40:47 +00:00
|
|
|
|
|
|
|
GCM_COMPLETE arg10, arg11
|
2018-02-14 17:38:35 +00:00
|
|
|
FUNC_RESTORE
|
2010-11-04 19:00:45 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_gcm_enc)
|
2010-12-13 11:51:15 +00:00
|
|
|
|
2018-02-14 17:40:47 +00:00
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_init(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
|
|
|
* struct gcm_context_data *data,
|
|
|
|
* // context data
|
|
|
|
* u8 *iv, // Pre-counter block j0: 4 byte salt (from Security Association)
|
|
|
|
* // concatenated with 8 byte Initialisation Vector (from IPSec ESP Payload)
|
|
|
|
* // concatenated with 0x00000001. 16-byte aligned pointer.
|
|
|
|
* u8 *hash_subkey, // H, the Hash sub key input. Data starts on a 16-byte boundary.
|
|
|
|
* const u8 *aad, // Additional Authentication Data (AAD)
|
|
|
|
* u64 aad_len) // Length of AAD in bytes.
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_gcm_init)
|
|
|
|
FUNC_SAVE
|
|
|
|
GCM_INIT %arg3, %arg4,%arg5, %arg6
|
|
|
|
FUNC_RESTORE
|
|
|
|
ret
|
|
|
|
ENDPROC(aesni_gcm_init)
|
|
|
|
|
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_enc_update(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
|
|
|
* struct gcm_context_data *data,
|
|
|
|
* // context data
|
|
|
|
* u8 *out, // Ciphertext output. Encrypt in-place is allowed.
|
|
|
|
* const u8 *in, // Plaintext input
|
|
|
|
* u64 plaintext_len, // Length of data in bytes for encryption.
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_gcm_enc_update)
|
|
|
|
FUNC_SAVE
|
|
|
|
GCM_ENC_DEC enc
|
|
|
|
FUNC_RESTORE
|
|
|
|
ret
|
|
|
|
ENDPROC(aesni_gcm_enc_update)
|
|
|
|
|
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_dec_update(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
|
|
|
* struct gcm_context_data *data,
|
|
|
|
* // context data
|
|
|
|
* u8 *out, // Ciphertext output. Encrypt in-place is allowed.
|
|
|
|
* const u8 *in, // Plaintext input
|
|
|
|
* u64 plaintext_len, // Length of data in bytes for encryption.
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_gcm_dec_update)
|
|
|
|
FUNC_SAVE
|
|
|
|
GCM_ENC_DEC dec
|
|
|
|
FUNC_RESTORE
|
|
|
|
ret
|
|
|
|
ENDPROC(aesni_gcm_dec_update)
|
|
|
|
|
|
|
|
/*****************************************************************************
|
|
|
|
* void aesni_gcm_finalize(void *aes_ctx, // AES Key schedule. Starts on a 16 byte boundary.
|
|
|
|
* struct gcm_context_data *data,
|
|
|
|
* // context data
|
|
|
|
* u8 *auth_tag, // Authenticated Tag output.
|
|
|
|
* u64 auth_tag_len); // Authenticated Tag Length in bytes. Valid values are 16 (most likely),
|
|
|
|
* // 12 or 8.
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_gcm_finalize)
|
|
|
|
FUNC_SAVE
|
|
|
|
GCM_COMPLETE %arg3 %arg4
|
|
|
|
FUNC_RESTORE
|
|
|
|
ret
|
|
|
|
ENDPROC(aesni_gcm_finalize)
|
|
|
|
|
2010-11-29 00:35:39 +00:00
|
|
|
#endif
|
2010-11-04 19:00:45 +00:00
|
|
|
|
|
|
|
|
2013-01-19 11:38:55 +00:00
|
|
|
.align 4
|
2009-01-18 05:28:34 +00:00
|
|
|
_key_expansion_128:
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_key_expansion_256a)
|
2009-01-18 05:28:34 +00:00
|
|
|
pshufd $0b11111111, %xmm1, %xmm1
|
|
|
|
shufps $0b00010000, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
shufps $0b10001100, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
pxor %xmm1, %xmm0
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm0, (TKEYP)
|
|
|
|
add $0x10, TKEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(_key_expansion_128)
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_key_expansion_256a)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_key_expansion_192a)
|
2009-01-18 05:28:34 +00:00
|
|
|
pshufd $0b01010101, %xmm1, %xmm1
|
|
|
|
shufps $0b00010000, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
shufps $0b10001100, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
pxor %xmm1, %xmm0
|
|
|
|
|
|
|
|
movaps %xmm2, %xmm5
|
|
|
|
movaps %xmm2, %xmm6
|
|
|
|
pslldq $4, %xmm5
|
|
|
|
pshufd $0b11111111, %xmm0, %xmm3
|
|
|
|
pxor %xmm3, %xmm2
|
|
|
|
pxor %xmm5, %xmm2
|
|
|
|
|
|
|
|
movaps %xmm0, %xmm1
|
|
|
|
shufps $0b01000100, %xmm0, %xmm6
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm6, (TKEYP)
|
2009-01-18 05:28:34 +00:00
|
|
|
shufps $0b01001110, %xmm2, %xmm1
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm1, 0x10(TKEYP)
|
|
|
|
add $0x20, TKEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_key_expansion_192a)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_key_expansion_192b)
|
2009-01-18 05:28:34 +00:00
|
|
|
pshufd $0b01010101, %xmm1, %xmm1
|
|
|
|
shufps $0b00010000, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
shufps $0b10001100, %xmm0, %xmm4
|
|
|
|
pxor %xmm4, %xmm0
|
|
|
|
pxor %xmm1, %xmm0
|
|
|
|
|
|
|
|
movaps %xmm2, %xmm5
|
|
|
|
pslldq $4, %xmm5
|
|
|
|
pshufd $0b11111111, %xmm0, %xmm3
|
|
|
|
pxor %xmm3, %xmm2
|
|
|
|
pxor %xmm5, %xmm2
|
|
|
|
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm0, (TKEYP)
|
|
|
|
add $0x10, TKEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_key_expansion_192b)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_key_expansion_256b)
|
2009-01-18 05:28:34 +00:00
|
|
|
pshufd $0b10101010, %xmm1, %xmm1
|
|
|
|
shufps $0b00010000, %xmm2, %xmm4
|
|
|
|
pxor %xmm4, %xmm2
|
|
|
|
shufps $0b10001100, %xmm2, %xmm4
|
|
|
|
pxor %xmm4, %xmm2
|
|
|
|
pxor %xmm1, %xmm2
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm2, (TKEYP)
|
|
|
|
add $0x10, TKEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_key_expansion_256b)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* int aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
|
|
|
|
* unsigned int key_len)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_set_key)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl KEYP
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+8)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+12)(%esp), UKEYP # in_key
|
|
|
|
movl (FRAME_OFFSET+16)(%esp), %edx # key_len
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
|
|
|
movups (UKEYP), %xmm0 # user key (first 16 bytes)
|
|
|
|
movaps %xmm0, (KEYP)
|
|
|
|
lea 0x10(KEYP), TKEYP # key addr
|
|
|
|
movl %edx, 480(KEYP)
|
2009-01-18 05:28:34 +00:00
|
|
|
pxor %xmm4, %xmm4 # xmm4 is assumed 0 in _key_expansion_x
|
|
|
|
cmp $24, %dl
|
|
|
|
jb .Lenc_key128
|
|
|
|
je .Lenc_key192
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movups 0x10(UKEYP), %xmm2 # other user key
|
|
|
|
movaps %xmm2, (TKEYP)
|
|
|
|
add $0x10, TKEYP
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x1 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x2 %xmm2 %xmm1 # round 2
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x2 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x4 %xmm2 %xmm1 # round 3
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x4 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x8 %xmm2 %xmm1 # round 4
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x8 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x10 %xmm2 %xmm1 # round 5
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x10 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x20 %xmm2 %xmm1 # round 6
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x20 %xmm0 %xmm1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x40 %xmm2 %xmm1 # round 7
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_256a
|
|
|
|
jmp .Ldec_key
|
|
|
|
.Lenc_key192:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movq 0x10(UKEYP), %xmm2 # other user key
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x1 %xmm2 %xmm1 # round 1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x2 %xmm2 %xmm1 # round 2
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x4 %xmm2 %xmm1 # round 3
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x8 %xmm2 %xmm1 # round 4
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x10 %xmm2 %xmm1 # round 5
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x20 %xmm2 %xmm1 # round 6
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192b
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x40 %xmm2 %xmm1 # round 7
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192a
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x80 %xmm2 %xmm1 # round 8
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_192b
|
|
|
|
jmp .Ldec_key
|
|
|
|
.Lenc_key128:
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x1 %xmm0 %xmm1 # round 1
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x2 %xmm0 %xmm1 # round 2
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x4 %xmm0 %xmm1 # round 3
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x8 %xmm0 %xmm1 # round 4
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x10 %xmm0 %xmm1 # round 5
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x20 %xmm0 %xmm1 # round 6
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x40 %xmm0 %xmm1 # round 7
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x80 %xmm0 %xmm1 # round 8
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x1b %xmm0 %xmm1 # round 9
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
2009-11-23 11:54:06 +00:00
|
|
|
AESKEYGENASSIST 0x36 %xmm0 %xmm1 # round 10
|
2009-01-18 05:28:34 +00:00
|
|
|
call _key_expansion_128
|
|
|
|
.Ldec_key:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
sub $0x10, TKEYP
|
|
|
|
movaps (KEYP), %xmm0
|
|
|
|
movaps (TKEYP), %xmm1
|
|
|
|
movaps %xmm0, 240(TKEYP)
|
|
|
|
movaps %xmm1, 240(KEYP)
|
|
|
|
add $0x10, KEYP
|
|
|
|
lea 240-16(TKEYP), UKEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.Ldec_key_loop:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps (KEYP), %xmm0
|
2009-11-23 11:54:06 +00:00
|
|
|
AESIMC %xmm0 %xmm1
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
movaps %xmm1, (UKEYP)
|
|
|
|
add $0x10, KEYP
|
|
|
|
sub $0x10, UKEYP
|
|
|
|
cmp TKEYP, KEYP
|
2009-01-18 05:28:34 +00:00
|
|
|
jb .Ldec_key_loop
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
xor AREG, AREG
|
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KEYP
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_set_key)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_enc(struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_enc)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+12)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+16)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+20)(%esp), INP # src
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
movl 480(KEYP), KLEN # key length
|
|
|
|
movups (INP), STATE # input
|
|
|
|
call _aesni_enc1
|
|
|
|
movups STATE, (OUTP) # output
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_enc)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_enc1: internal ABI
|
|
|
|
* input:
|
|
|
|
* KEYP: key struct pointer
|
|
|
|
* KLEN: round count
|
|
|
|
* STATE: initial state (input)
|
|
|
|
* output:
|
|
|
|
* STATE: finial state (output)
|
|
|
|
* changed:
|
|
|
|
* KEY
|
|
|
|
* TKEYP (T1)
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_enc1)
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (KEYP), KEY # key
|
|
|
|
mov KEYP, TKEYP
|
|
|
|
pxor KEY, STATE # round 0
|
|
|
|
add $0x30, TKEYP
|
|
|
|
cmp $24, KLEN
|
|
|
|
jb .Lenc128
|
|
|
|
lea 0x20(TKEYP), TKEYP
|
|
|
|
je .Lenc192
|
|
|
|
add $0x20, TKEYP
|
|
|
|
movaps -0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.Lenc192:
|
|
|
|
movaps -0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.Lenc128:
|
|
|
|
movaps -0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x70(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENCLAST KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_enc1)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_enc4: internal ABI
|
|
|
|
* input:
|
|
|
|
* KEYP: key struct pointer
|
|
|
|
* KLEN: round count
|
|
|
|
* STATE1: initial state (input)
|
|
|
|
* STATE2
|
|
|
|
* STATE3
|
|
|
|
* STATE4
|
|
|
|
* output:
|
|
|
|
* STATE1: finial state (output)
|
|
|
|
* STATE2
|
|
|
|
* STATE3
|
|
|
|
* STATE4
|
|
|
|
* changed:
|
|
|
|
* KEY
|
|
|
|
* TKEYP (T1)
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_enc4)
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (KEYP), KEY # key
|
|
|
|
mov KEYP, TKEYP
|
|
|
|
pxor KEY, STATE1 # round 0
|
|
|
|
pxor KEY, STATE2
|
|
|
|
pxor KEY, STATE3
|
|
|
|
pxor KEY, STATE4
|
|
|
|
add $0x30, TKEYP
|
|
|
|
cmp $24, KLEN
|
|
|
|
jb .L4enc128
|
|
|
|
lea 0x20(TKEYP), TKEYP
|
|
|
|
je .L4enc192
|
|
|
|
add $0x20, TKEYP
|
|
|
|
movaps -0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
#.align 4
|
|
|
|
.L4enc192:
|
|
|
|
movaps -0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
#.align 4
|
|
|
|
.L4enc128:
|
|
|
|
movaps -0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENC KEY STATE1
|
|
|
|
AESENC KEY STATE2
|
|
|
|
AESENC KEY STATE3
|
|
|
|
AESENC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x70(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESENCLAST KEY STATE1 # last round
|
|
|
|
AESENCLAST KEY STATE2
|
|
|
|
AESENCLAST KEY STATE3
|
|
|
|
AESENCLAST KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_enc4)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_dec (struct crypto_aes_ctx *ctx, u8 *dst, const u8 *src)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_dec)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+12)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+16)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+20)(%esp), INP # src
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
mov 480(KEYP), KLEN # key length
|
|
|
|
add $240, KEYP
|
|
|
|
movups (INP), STATE # input
|
|
|
|
call _aesni_dec1
|
|
|
|
movups STATE, (OUTP) #output
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_dec)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_dec1: internal ABI
|
|
|
|
* input:
|
|
|
|
* KEYP: key struct pointer
|
|
|
|
* KLEN: key length
|
|
|
|
* STATE: initial state (input)
|
|
|
|
* output:
|
|
|
|
* STATE: finial state (output)
|
|
|
|
* changed:
|
|
|
|
* KEY
|
|
|
|
* TKEYP (T1)
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_dec1)
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (KEYP), KEY # key
|
|
|
|
mov KEYP, TKEYP
|
|
|
|
pxor KEY, STATE # round 0
|
|
|
|
add $0x30, TKEYP
|
|
|
|
cmp $24, KLEN
|
|
|
|
jb .Ldec128
|
|
|
|
lea 0x20(TKEYP), TKEYP
|
|
|
|
je .Ldec192
|
|
|
|
add $0x20, TKEYP
|
|
|
|
movaps -0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.Ldec192:
|
|
|
|
movaps -0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.Ldec128:
|
|
|
|
movaps -0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x70(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDECLAST KEY STATE
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_dec1)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_dec4: internal ABI
|
|
|
|
* input:
|
|
|
|
* KEYP: key struct pointer
|
|
|
|
* KLEN: key length
|
|
|
|
* STATE1: initial state (input)
|
|
|
|
* STATE2
|
|
|
|
* STATE3
|
|
|
|
* STATE4
|
|
|
|
* output:
|
|
|
|
* STATE1: finial state (output)
|
|
|
|
* STATE2
|
|
|
|
* STATE3
|
|
|
|
* STATE4
|
|
|
|
* changed:
|
|
|
|
* KEY
|
|
|
|
* TKEYP (T1)
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_dec4)
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (KEYP), KEY # key
|
|
|
|
mov KEYP, TKEYP
|
|
|
|
pxor KEY, STATE1 # round 0
|
|
|
|
pxor KEY, STATE2
|
|
|
|
pxor KEY, STATE3
|
|
|
|
pxor KEY, STATE4
|
|
|
|
add $0x30, TKEYP
|
|
|
|
cmp $24, KLEN
|
|
|
|
jb .L4dec128
|
|
|
|
lea 0x20(TKEYP), TKEYP
|
|
|
|
je .L4dec192
|
|
|
|
add $0x20, TKEYP
|
|
|
|
movaps -0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.L4dec192:
|
|
|
|
movaps -0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
.align 4
|
|
|
|
.L4dec128:
|
|
|
|
movaps -0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps -0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps (TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x10(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x20(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x30(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x40(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x50(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x60(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDEC KEY STATE1
|
|
|
|
AESDEC KEY STATE2
|
|
|
|
AESDEC KEY STATE3
|
|
|
|
AESDEC KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
movaps 0x70(TKEYP), KEY
|
2009-11-23 11:54:06 +00:00
|
|
|
AESDECLAST KEY STATE1 # last round
|
|
|
|
AESDECLAST KEY STATE2
|
|
|
|
AESDECLAST KEY STATE3
|
|
|
|
AESDECLAST KEY STATE4
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_dec4)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_ecb_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* size_t len)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_ecb_enc)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl LEN
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+16)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+20)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+24)(%esp), INP # src
|
|
|
|
movl (FRAME_OFFSET+28)(%esp), LEN # len
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
test LEN, LEN # check length
|
|
|
|
jz .Lecb_enc_ret
|
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lecb_enc_ret
|
|
|
|
cmp $64, LEN
|
|
|
|
jb .Lecb_enc_loop1
|
|
|
|
.align 4
|
|
|
|
.Lecb_enc_loop4:
|
|
|
|
movups (INP), STATE1
|
|
|
|
movups 0x10(INP), STATE2
|
|
|
|
movups 0x20(INP), STATE3
|
|
|
|
movups 0x30(INP), STATE4
|
|
|
|
call _aesni_enc4
|
|
|
|
movups STATE1, (OUTP)
|
|
|
|
movups STATE2, 0x10(OUTP)
|
|
|
|
movups STATE3, 0x20(OUTP)
|
|
|
|
movups STATE4, 0x30(OUTP)
|
|
|
|
sub $64, LEN
|
|
|
|
add $64, INP
|
|
|
|
add $64, OUTP
|
|
|
|
cmp $64, LEN
|
|
|
|
jge .Lecb_enc_loop4
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lecb_enc_ret
|
|
|
|
.align 4
|
|
|
|
.Lecb_enc_loop1:
|
|
|
|
movups (INP), STATE1
|
|
|
|
call _aesni_enc1
|
|
|
|
movups STATE1, (OUTP)
|
|
|
|
sub $16, LEN
|
|
|
|
add $16, INP
|
|
|
|
add $16, OUTP
|
|
|
|
cmp $16, LEN
|
|
|
|
jge .Lecb_enc_loop1
|
|
|
|
.Lecb_enc_ret:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
popl LEN
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_ecb_enc)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_ecb_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* size_t len);
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_ecb_dec)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl LEN
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+16)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+20)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+24)(%esp), INP # src
|
|
|
|
movl (FRAME_OFFSET+28)(%esp), LEN # len
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
test LEN, LEN
|
|
|
|
jz .Lecb_dec_ret
|
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
add $240, KEYP
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lecb_dec_ret
|
|
|
|
cmp $64, LEN
|
|
|
|
jb .Lecb_dec_loop1
|
|
|
|
.align 4
|
|
|
|
.Lecb_dec_loop4:
|
|
|
|
movups (INP), STATE1
|
|
|
|
movups 0x10(INP), STATE2
|
|
|
|
movups 0x20(INP), STATE3
|
|
|
|
movups 0x30(INP), STATE4
|
|
|
|
call _aesni_dec4
|
|
|
|
movups STATE1, (OUTP)
|
|
|
|
movups STATE2, 0x10(OUTP)
|
|
|
|
movups STATE3, 0x20(OUTP)
|
|
|
|
movups STATE4, 0x30(OUTP)
|
|
|
|
sub $64, LEN
|
|
|
|
add $64, INP
|
|
|
|
add $64, OUTP
|
|
|
|
cmp $64, LEN
|
|
|
|
jge .Lecb_dec_loop4
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lecb_dec_ret
|
|
|
|
.align 4
|
|
|
|
.Lecb_dec_loop1:
|
|
|
|
movups (INP), STATE1
|
|
|
|
call _aesni_dec1
|
|
|
|
movups STATE1, (OUTP)
|
|
|
|
sub $16, LEN
|
|
|
|
add $16, INP
|
|
|
|
add $16, OUTP
|
|
|
|
cmp $16, LEN
|
|
|
|
jge .Lecb_dec_loop1
|
|
|
|
.Lecb_dec_ret:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
popl LEN
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_ecb_dec)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* size_t len, u8 *iv)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_cbc_enc)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl IVP
|
|
|
|
pushl LEN
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+20)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+24)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+28)(%esp), INP # src
|
|
|
|
movl (FRAME_OFFSET+32)(%esp), LEN # len
|
|
|
|
movl (FRAME_OFFSET+36)(%esp), IVP # iv
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lcbc_enc_ret
|
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
movups (IVP), STATE # load iv as initial state
|
|
|
|
.align 4
|
|
|
|
.Lcbc_enc_loop:
|
|
|
|
movups (INP), IN # load input
|
|
|
|
pxor IN, STATE
|
|
|
|
call _aesni_enc1
|
|
|
|
movups STATE, (OUTP) # store output
|
|
|
|
sub $16, LEN
|
|
|
|
add $16, INP
|
|
|
|
add $16, OUTP
|
|
|
|
cmp $16, LEN
|
|
|
|
jge .Lcbc_enc_loop
|
|
|
|
movups STATE, (IVP)
|
|
|
|
.Lcbc_enc_ret:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
popl LEN
|
|
|
|
popl IVP
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_cbc_enc)
|
2009-01-18 05:28:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* size_t len, u8 *iv)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_cbc_dec)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
pushl IVP
|
|
|
|
pushl LEN
|
|
|
|
pushl KEYP
|
|
|
|
pushl KLEN
|
2016-01-21 22:49:19 +00:00
|
|
|
movl (FRAME_OFFSET+20)(%esp), KEYP # ctx
|
|
|
|
movl (FRAME_OFFSET+24)(%esp), OUTP # dst
|
|
|
|
movl (FRAME_OFFSET+28)(%esp), INP # src
|
|
|
|
movl (FRAME_OFFSET+32)(%esp), LEN # len
|
|
|
|
movl (FRAME_OFFSET+36)(%esp), IVP # iv
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
cmp $16, LEN
|
2009-06-18 11:33:57 +00:00
|
|
|
jb .Lcbc_dec_just_ret
|
2009-01-18 05:28:34 +00:00
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
add $240, KEYP
|
|
|
|
movups (IVP), IV
|
|
|
|
cmp $64, LEN
|
|
|
|
jb .Lcbc_dec_loop1
|
|
|
|
.align 4
|
|
|
|
.Lcbc_dec_loop4:
|
|
|
|
movups (INP), IN1
|
|
|
|
movaps IN1, STATE1
|
|
|
|
movups 0x10(INP), IN2
|
|
|
|
movaps IN2, STATE2
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifdef __x86_64__
|
2009-01-18 05:28:34 +00:00
|
|
|
movups 0x20(INP), IN3
|
|
|
|
movaps IN3, STATE3
|
|
|
|
movups 0x30(INP), IN4
|
|
|
|
movaps IN4, STATE4
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#else
|
|
|
|
movups 0x20(INP), IN1
|
|
|
|
movaps IN1, STATE3
|
|
|
|
movups 0x30(INP), IN2
|
|
|
|
movaps IN2, STATE4
|
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
call _aesni_dec4
|
|
|
|
pxor IV, STATE1
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifdef __x86_64__
|
2009-01-18 05:28:34 +00:00
|
|
|
pxor IN1, STATE2
|
|
|
|
pxor IN2, STATE3
|
|
|
|
pxor IN3, STATE4
|
|
|
|
movaps IN4, IV
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#else
|
|
|
|
pxor IN1, STATE4
|
|
|
|
movaps IN2, IV
|
2012-05-29 23:43:08 +00:00
|
|
|
movups (INP), IN1
|
|
|
|
pxor IN1, STATE2
|
|
|
|
movups 0x10(INP), IN2
|
|
|
|
pxor IN2, STATE3
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|
2009-01-18 05:28:34 +00:00
|
|
|
movups STATE1, (OUTP)
|
|
|
|
movups STATE2, 0x10(OUTP)
|
|
|
|
movups STATE3, 0x20(OUTP)
|
|
|
|
movups STATE4, 0x30(OUTP)
|
|
|
|
sub $64, LEN
|
|
|
|
add $64, INP
|
|
|
|
add $64, OUTP
|
|
|
|
cmp $64, LEN
|
|
|
|
jge .Lcbc_dec_loop4
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lcbc_dec_ret
|
|
|
|
.align 4
|
|
|
|
.Lcbc_dec_loop1:
|
|
|
|
movups (INP), IN
|
|
|
|
movaps IN, STATE
|
|
|
|
call _aesni_dec1
|
|
|
|
pxor IV, STATE
|
|
|
|
movups STATE, (OUTP)
|
|
|
|
movaps IN, IV
|
|
|
|
sub $16, LEN
|
|
|
|
add $16, INP
|
|
|
|
add $16, OUTP
|
|
|
|
cmp $16, LEN
|
|
|
|
jge .Lcbc_dec_loop1
|
|
|
|
.Lcbc_dec_ret:
|
2009-06-18 11:33:57 +00:00
|
|
|
movups IV, (IVP)
|
|
|
|
.Lcbc_dec_just_ret:
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifndef __x86_64__
|
|
|
|
popl KLEN
|
|
|
|
popl KEYP
|
|
|
|
popl LEN
|
|
|
|
popl IVP
|
|
|
|
#endif
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2009-01-18 05:28:34 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_cbc_dec)
|
2010-03-10 10:28:55 +00:00
|
|
|
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#ifdef __x86_64__
|
x86/asm/crypto: Move .Lbswap_mask data to .rodata section
stacktool reports the following warning:
stacktool: arch/x86/crypto/aesni-intel_asm.o: _aesni_inc_init(): can't find starting instruction
stacktool gets confused when it tries to disassemble the following data
in the .text section:
.Lbswap_mask:
.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
Move it to .rodata which is a more appropriate section for read-only
data.
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Borislav Petkov <bp@suse.de>
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/b6a2f3f8bda705143e127c025edb2b53c86e6eb4.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-01-21 22:49:15 +00:00
|
|
|
.pushsection .rodata
|
2010-03-10 10:28:55 +00:00
|
|
|
.align 16
|
|
|
|
.Lbswap_mask:
|
|
|
|
.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
|
x86/asm/crypto: Move .Lbswap_mask data to .rodata section
stacktool reports the following warning:
stacktool: arch/x86/crypto/aesni-intel_asm.o: _aesni_inc_init(): can't find starting instruction
stacktool gets confused when it tries to disassemble the following data
in the .text section:
.Lbswap_mask:
.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
Move it to .rodata which is a more appropriate section for read-only
data.
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Borislav Petkov <bp@suse.de>
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/b6a2f3f8bda705143e127c025edb2b53c86e6eb4.1453405861.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-01-21 22:49:15 +00:00
|
|
|
.popsection
|
2010-03-10 10:28:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_inc_init: internal ABI
|
|
|
|
* setup registers used by _aesni_inc
|
|
|
|
* input:
|
|
|
|
* IV
|
|
|
|
* output:
|
|
|
|
* CTR: == IV, in little endian
|
|
|
|
* TCTR_LOW: == lower qword of CTR
|
|
|
|
* INC: == 1, in little endian
|
|
|
|
* BSWAP_MASK == endian swapping mask
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_inc_init)
|
2010-03-10 10:28:55 +00:00
|
|
|
movaps .Lbswap_mask, BSWAP_MASK
|
|
|
|
movaps IV, CTR
|
|
|
|
PSHUFB_XMM BSWAP_MASK CTR
|
|
|
|
mov $1, TCTR_LOW
|
2010-03-13 08:28:42 +00:00
|
|
|
MOVQ_R64_XMM TCTR_LOW INC
|
|
|
|
MOVQ_R64_XMM CTR TCTR_LOW
|
2010-03-10 10:28:55 +00:00
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_inc_init)
|
2010-03-10 10:28:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_inc: internal ABI
|
|
|
|
* Increase IV by 1, IV is in big endian
|
|
|
|
* input:
|
|
|
|
* IV
|
|
|
|
* CTR: == IV, in little endian
|
|
|
|
* TCTR_LOW: == lower qword of CTR
|
|
|
|
* INC: == 1, in little endian
|
|
|
|
* BSWAP_MASK == endian swapping mask
|
|
|
|
* output:
|
|
|
|
* IV: Increase by 1
|
|
|
|
* changed:
|
|
|
|
* CTR: == output IV, in little endian
|
|
|
|
* TCTR_LOW: == lower qword of CTR
|
|
|
|
*/
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_START_LOCAL(_aesni_inc)
|
2010-03-10 10:28:55 +00:00
|
|
|
paddq INC, CTR
|
|
|
|
add $1, TCTR_LOW
|
|
|
|
jnc .Linc_low
|
|
|
|
pslldq $8, INC
|
|
|
|
paddq INC, CTR
|
|
|
|
psrldq $8, INC
|
|
|
|
.Linc_low:
|
|
|
|
movaps CTR, IV
|
|
|
|
PSHUFB_XMM BSWAP_MASK IV
|
|
|
|
ret
|
2019-10-11 11:50:46 +00:00
|
|
|
SYM_FUNC_END(_aesni_inc)
|
2010-03-10 10:28:55 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_ctr_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* size_t len, u8 *iv)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_ctr_enc)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
2010-03-10 10:28:55 +00:00
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lctr_enc_just_ret
|
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
movups (IVP), IV
|
|
|
|
call _aesni_inc_init
|
|
|
|
cmp $64, LEN
|
|
|
|
jb .Lctr_enc_loop1
|
|
|
|
.align 4
|
|
|
|
.Lctr_enc_loop4:
|
|
|
|
movaps IV, STATE1
|
|
|
|
call _aesni_inc
|
|
|
|
movups (INP), IN1
|
|
|
|
movaps IV, STATE2
|
|
|
|
call _aesni_inc
|
|
|
|
movups 0x10(INP), IN2
|
|
|
|
movaps IV, STATE3
|
|
|
|
call _aesni_inc
|
|
|
|
movups 0x20(INP), IN3
|
|
|
|
movaps IV, STATE4
|
|
|
|
call _aesni_inc
|
|
|
|
movups 0x30(INP), IN4
|
|
|
|
call _aesni_enc4
|
|
|
|
pxor IN1, STATE1
|
|
|
|
movups STATE1, (OUTP)
|
|
|
|
pxor IN2, STATE2
|
|
|
|
movups STATE2, 0x10(OUTP)
|
|
|
|
pxor IN3, STATE3
|
|
|
|
movups STATE3, 0x20(OUTP)
|
|
|
|
pxor IN4, STATE4
|
|
|
|
movups STATE4, 0x30(OUTP)
|
|
|
|
sub $64, LEN
|
|
|
|
add $64, INP
|
|
|
|
add $64, OUTP
|
|
|
|
cmp $64, LEN
|
|
|
|
jge .Lctr_enc_loop4
|
|
|
|
cmp $16, LEN
|
|
|
|
jb .Lctr_enc_ret
|
|
|
|
.align 4
|
|
|
|
.Lctr_enc_loop1:
|
|
|
|
movaps IV, STATE
|
|
|
|
call _aesni_inc
|
|
|
|
movups (INP), IN
|
|
|
|
call _aesni_enc1
|
|
|
|
pxor IN, STATE
|
|
|
|
movups STATE, (OUTP)
|
|
|
|
sub $16, LEN
|
|
|
|
add $16, INP
|
|
|
|
add $16, OUTP
|
|
|
|
cmp $16, LEN
|
|
|
|
jge .Lctr_enc_loop1
|
|
|
|
.Lctr_enc_ret:
|
|
|
|
movups IV, (IVP)
|
|
|
|
.Lctr_enc_just_ret:
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2010-03-10 10:28:55 +00:00
|
|
|
ret
|
2013-01-19 11:38:55 +00:00
|
|
|
ENDPROC(aesni_ctr_enc)
|
2013-04-08 18:51:16 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* _aesni_gf128mul_x_ble: internal ABI
|
|
|
|
* Multiply in GF(2^128) for XTS IVs
|
|
|
|
* input:
|
|
|
|
* IV: current IV
|
|
|
|
* GF128MUL_MASK == mask with 0x87 and 0x01
|
|
|
|
* output:
|
|
|
|
* IV: next IV
|
|
|
|
* changed:
|
|
|
|
* CTR: == temporary value
|
|
|
|
*/
|
|
|
|
#define _aesni_gf128mul_x_ble() \
|
|
|
|
pshufd $0x13, IV, CTR; \
|
|
|
|
paddq IV, IV; \
|
|
|
|
psrad $31, CTR; \
|
|
|
|
pand GF128MUL_MASK, CTR; \
|
|
|
|
pxor CTR, IV;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* void aesni_xts_crypt8(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
|
|
|
|
* bool enc, u8 *iv)
|
|
|
|
*/
|
|
|
|
ENTRY(aesni_xts_crypt8)
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_BEGIN
|
2013-04-08 18:51:16 +00:00
|
|
|
cmpb $0, %cl
|
|
|
|
movl $0, %ecx
|
|
|
|
movl $240, %r10d
|
|
|
|
leaq _aesni_enc4, %r11
|
|
|
|
leaq _aesni_dec4, %rax
|
|
|
|
cmovel %r10d, %ecx
|
|
|
|
cmoveq %rax, %r11
|
|
|
|
|
|
|
|
movdqa .Lgf128mul_x_ble_mask, GF128MUL_MASK
|
|
|
|
movups (IVP), IV
|
|
|
|
|
|
|
|
mov 480(KEYP), KLEN
|
|
|
|
addq %rcx, KEYP
|
|
|
|
|
|
|
|
movdqa IV, STATE1
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x00(INP), INC
|
|
|
|
pxor INC, STATE1
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x00(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE2
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x10(INP), INC
|
|
|
|
pxor INC, STATE2
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x10(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE3
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x20(INP), INC
|
|
|
|
pxor INC, STATE3
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x20(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE4
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x30(INP), INC
|
|
|
|
pxor INC, STATE4
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x30(OUTP)
|
|
|
|
|
2018-01-11 21:46:27 +00:00
|
|
|
CALL_NOSPEC %r11
|
2013-04-08 18:51:16 +00:00
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x00(OUTP), INC
|
|
|
|
pxor INC, STATE1
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE1, 0x00(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE1
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x40(INP), INC
|
|
|
|
pxor INC, STATE1
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x40(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x10(OUTP), INC
|
|
|
|
pxor INC, STATE2
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE2, 0x10(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE2
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x50(INP), INC
|
|
|
|
pxor INC, STATE2
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x50(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x20(OUTP), INC
|
|
|
|
pxor INC, STATE3
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE3, 0x20(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE3
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x60(INP), INC
|
|
|
|
pxor INC, STATE3
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x60(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x30(OUTP), INC
|
|
|
|
pxor INC, STATE4
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE4, 0x30(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movdqa IV, STATE4
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x70(INP), INC
|
|
|
|
pxor INC, STATE4
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu IV, 0x70(OUTP)
|
|
|
|
|
|
|
|
_aesni_gf128mul_x_ble()
|
|
|
|
movups IV, (IVP)
|
|
|
|
|
2018-01-11 21:46:27 +00:00
|
|
|
CALL_NOSPEC %r11
|
2013-04-08 18:51:16 +00:00
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x40(OUTP), INC
|
|
|
|
pxor INC, STATE1
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE1, 0x40(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x50(OUTP), INC
|
|
|
|
pxor INC, STATE2
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE2, 0x50(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x60(OUTP), INC
|
|
|
|
pxor INC, STATE3
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE3, 0x60(OUTP)
|
|
|
|
|
2013-06-11 19:25:22 +00:00
|
|
|
movdqu 0x70(OUTP), INC
|
|
|
|
pxor INC, STATE4
|
2013-04-08 18:51:16 +00:00
|
|
|
movdqu STATE4, 0x70(OUTP)
|
|
|
|
|
2016-01-21 22:49:19 +00:00
|
|
|
FRAME_END
|
2013-04-08 18:51:16 +00:00
|
|
|
ret
|
|
|
|
ENDPROC(aesni_xts_crypt8)
|
|
|
|
|
crypto: aesni-intel - Ported implementation to x86-32
The AES-NI instructions are also available in legacy mode so the 32-bit
architecture may profit from those, too.
To illustrate the performance gain here's a short summary of a dm-crypt
speed test on a Core i7 M620 running at 2.67GHz comparing both assembler
implementations:
x86: i568 aes-ni delta
ECB, 256 bit: 93.8 MB/s 123.3 MB/s +31.4%
CBC, 256 bit: 84.8 MB/s 262.3 MB/s +209.3%
LRW, 256 bit: 108.6 MB/s 222.1 MB/s +104.5%
XTS, 256 bit: 105.0 MB/s 205.5 MB/s +95.7%
Additionally, due to some minor optimizations, the 64-bit version also
got a minor performance gain as seen below:
x86-64: old impl. new impl. delta
ECB, 256 bit: 121.1 MB/s 123.0 MB/s +1.5%
CBC, 256 bit: 285.3 MB/s 290.8 MB/s +1.9%
LRW, 256 bit: 263.7 MB/s 265.3 MB/s +0.6%
XTS, 256 bit: 251.1 MB/s 255.3 MB/s +1.7%
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Reviewed-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2010-11-27 08:34:46 +00:00
|
|
|
#endif
|