Techwizz/kernel-fxtec-pro1x
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
kernel-fxtec-pro1x/arch/x86/crypto/aes-x86_64-asm_64.S

189 lines
4.7 KiB
ArmAsm
Raw Normal View History

[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
/* AES (Rijndael) implementation (FIPS PUB 197) for x86_64
*
* Copyright (C) 2005 Andreas Steinmetz, <ast@domdv.de>
*
* License:
* This code can be distributed under the terms of the GNU General Public
* License (GPL) Version 2 provided that the above header down to and
* including this sentence is retained in full.
*/
[CRYPTO] aes-x86-64: Remove setkey The setkey() function can be shared with the generic algorithm. Signed-off-by: Sebastian Siewior <sebastian@breakpoint.cc> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2007-11-08 06:25:04 -07:00
.extern crypto_ft_tab
.extern crypto_it_tab
.extern crypto_fl_tab
.extern crypto_il_tab
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
.text
[CRYPTO] all: Pass tfm instead of ctx to algorithms Up until now algorithms have been happy to get a context pointer since they know everything that's in the tfm already (e.g., alignment, block size). However, once we have parameterised algorithms, such information will be specific to each tfm. So the algorithm API needs to be changed to pass the tfm structure instead of the context pointer. This patch is basically a text substitution. The only tricky bit is the assembly routines that need to get the context pointer offset through asm-offsets.h. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-05-16 06:09:29 -06:00
#include <asm/asm-offsets.h>
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
#define R1 %rax
#define R1E %eax
#define R1X %ax
#define R1H %ah
#define R1L %al
#define R2 %rbx
#define R2E %ebx
#define R2X %bx
#define R2H %bh
#define R2L %bl
#define R3 %rcx
#define R3E %ecx
#define R3X %cx
#define R3H %ch
#define R3L %cl
#define R4 %rdx
#define R4E %edx
#define R4X %dx
#define R4H %dh
#define R4L %dl
#define R5 %rsi
#define R5E %esi
#define R6 %rdi
#define R6E %edi
#define R7 %rbp
#define R7E %ebp
#define R8 %r8
#define R9 %r9
#define R10 %r10
#define R11 %r11
[CRYPTO] all: Pass tfm instead of ctx to algorithms Up until now algorithms have been happy to get a context pointer since they know everything that's in the tfm already (e.g., alignment, block size). However, once we have parameterised algorithms, such information will be specific to each tfm. So the algorithm API needs to be changed to pass the tfm structure instead of the context pointer. This patch is basically a text substitution. The only tricky bit is the assembly routines that need to get the context pointer offset through asm-offsets.h. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-05-16 06:09:29 -06:00
#define prologue(FUNC,KEY,B128,B192,r1,r2,r3,r4,r5,r6,r7,r8,r9,r10,r11) \
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
.global FUNC; \
.type FUNC,@function; \
.align 8; \
FUNC: movq r1,r2; \
movq r3,r4; \
crypto: aes - Export x86 AES encrypt/decrypt functions Intel AES-NI AES acceleration instructions touch XMM state, to use that in soft_irq context, general x86 AES implementation is used as fallback. The first parameter is changed from struct crypto_tfm * to struct crypto_aes_ctx * to make it easier to deal with 16 bytes alignment requirement of AES-NI implementation. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-01-08 23:25:50 -07:00
leaq KEY+48(r8),r9; \
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
movq r10,r11; \
movl (r7),r5 ## E; \
movl 4(r7),r1 ## E; \
movl 8(r7),r6 ## E; \
movl 12(r7),r7 ## E; \
crypto: aes - Export x86 AES encrypt/decrypt functions Intel AES-NI AES acceleration instructions touch XMM state, to use that in soft_irq context, general x86 AES implementation is used as fallback. The first parameter is changed from struct crypto_tfm * to struct crypto_aes_ctx * to make it easier to deal with 16 bytes alignment requirement of AES-NI implementation. Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2009-01-08 23:25:50 -07:00
movl 480(r8),r10 ## E; \
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
xorl -48(r9),r5 ## E; \
xorl -44(r9),r1 ## E; \
xorl -40(r9),r6 ## E; \
xorl -36(r9),r7 ## E; \
cmpl $24,r10 ## E; \
jb B128; \
leaq 32(r9),r9; \
je B192; \
leaq 32(r9),r9;
#define epilogue(r1,r2,r3,r4,r5,r6,r7,r8,r9) \
movq r1,r2; \
movq r3,r4; \
movl r5 ## E,(r9); \
movl r6 ## E,4(r9); \
movl r7 ## E,8(r9); \
movl r8 ## E,12(r9); \
ret;
#define round(TAB,OFFSET,r1,r2,r3,r4,r5,r6,r7,r8,ra,rb,rc,rd) \
movzbl r2 ## H,r5 ## E; \
movzbl r2 ## L,r6 ## E; \
movl TAB+1024(,r5,4),r5 ## E;\
movw r4 ## X,r2 ## X; \
movl TAB(,r6,4),r6 ## E; \
roll $16,r2 ## E; \
shrl $16,r4 ## E; \
movzbl r4 ## H,r7 ## E; \
movzbl r4 ## L,r4 ## E; \
xorl OFFSET(r8),ra ## E; \
xorl OFFSET+4(r8),rb ## E; \
xorl TAB+3072(,r7,4),r5 ## E;\
xorl TAB+2048(,r4,4),r6 ## E;\
movzbl r1 ## L,r7 ## E; \
movzbl r1 ## H,r4 ## E; \
movl TAB+1024(,r4,4),r4 ## E;\
movw r3 ## X,r1 ## X; \
roll $16,r1 ## E; \
shrl $16,r3 ## E; \
xorl TAB(,r7,4),r5 ## E; \
movzbl r3 ## H,r7 ## E; \
movzbl r3 ## L,r3 ## E; \
xorl TAB+3072(,r7,4),r4 ## E;\
xorl TAB+2048(,r3,4),r5 ## E;\
movzbl r1 ## H,r7 ## E; \
movzbl r1 ## L,r3 ## E; \
shrl $16,r1 ## E; \
xorl TAB+3072(,r7,4),r6 ## E;\
movl TAB+2048(,r3,4),r3 ## E;\
movzbl r1 ## H,r7 ## E; \
movzbl r1 ## L,r1 ## E; \
xorl TAB+1024(,r7,4),r6 ## E;\
xorl TAB(,r1,4),r3 ## E; \
movzbl r2 ## H,r1 ## E; \
movzbl r2 ## L,r7 ## E; \
shrl $16,r2 ## E; \
xorl TAB+3072(,r1,4),r3 ## E;\
xorl TAB+2048(,r7,4),r4 ## E;\
movzbl r2 ## H,r1 ## E; \
movzbl r2 ## L,r2 ## E; \
xorl OFFSET+8(r8),rc ## E; \
xorl OFFSET+12(r8),rd ## E; \
xorl TAB+1024(,r1,4),r3 ## E;\
xorl TAB(,r2,4),r4 ## E;
#define move_regs(r1,r2,r3,r4) \
movl r3 ## E,r1 ## E; \
movl r4 ## E,r2 ## E;
[CRYPTO] all: Pass tfm instead of ctx to algorithms Up until now algorithms have been happy to get a context pointer since they know everything that's in the tfm already (e.g., alignment, block size). However, once we have parameterised algorithms, such information will be specific to each tfm. So the algorithm API needs to be changed to pass the tfm structure instead of the context pointer. This patch is basically a text substitution. The only tricky bit is the assembly routines that need to get the context pointer offset through asm-offsets.h. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-05-16 06:09:29 -06:00
#define entry(FUNC,KEY,B128,B192) \
prologue(FUNC,KEY,B128,B192,R2,R8,R7,R9,R1,R3,R4,R6,R10,R5,R11)
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
#define return epilogue(R8,R2,R9,R7,R5,R6,R3,R4,R11)
#define encrypt_round(TAB,OFFSET) \
round(TAB,OFFSET,R1,R2,R3,R4,R5,R6,R7,R10,R5,R6,R3,R4) \
move_regs(R1,R2,R5,R6)
#define encrypt_final(TAB,OFFSET) \
round(TAB,OFFSET,R1,R2,R3,R4,R5,R6,R7,R10,R5,R6,R3,R4)
#define decrypt_round(TAB,OFFSET) \
round(TAB,OFFSET,R2,R1,R4,R3,R6,R5,R7,R10,R5,R6,R3,R4) \
move_regs(R1,R2,R5,R6)
#define decrypt_final(TAB,OFFSET) \
round(TAB,OFFSET,R2,R1,R4,R3,R6,R5,R7,R10,R5,R6,R3,R4)
[CRYPTO] aes: Add wrappers for assembly routines The wrapper routines are required when asmlinkage differs from the usual calling convention. So we need to have them. However, by rearranging the parameters, they will get optimised away to a single jump for most people. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-06-01 16:42:25 -06:00
/* void aes_enc_blk(stuct crypto_tfm *tfm, u8 *out, const u8 *in) */
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
[CRYPTO] aes: Add wrappers for assembly routines The wrapper routines are required when asmlinkage differs from the usual calling convention. So we need to have them. However, by rearranging the parameters, they will get optimised away to a single jump for most people. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-06-01 16:42:25 -06:00
entry(aes_enc_blk,0,enc128,enc192)
[CRYPTO] aes-x86-64: Remove setkey The setkey() function can be shared with the generic algorithm. Signed-off-by: Sebastian Siewior <sebastian@breakpoint.cc> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2007-11-08 06:25:04 -07:00
encrypt_round(crypto_ft_tab,-96)
encrypt_round(crypto_ft_tab,-80)
enc192: encrypt_round(crypto_ft_tab,-64)
encrypt_round(crypto_ft_tab,-48)
enc128: encrypt_round(crypto_ft_tab,-32)
encrypt_round(crypto_ft_tab,-16)
encrypt_round(crypto_ft_tab, 0)
encrypt_round(crypto_ft_tab, 16)
encrypt_round(crypto_ft_tab, 32)
encrypt_round(crypto_ft_tab, 48)
encrypt_round(crypto_ft_tab, 64)
encrypt_round(crypto_ft_tab, 80)
encrypt_round(crypto_ft_tab, 96)
encrypt_final(crypto_fl_tab,112)
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
return
[CRYPTO] aes: Add wrappers for assembly routines The wrapper routines are required when asmlinkage differs from the usual calling convention. So we need to have them. However, by rearranging the parameters, they will get optimised away to a single jump for most people. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-06-01 16:42:25 -06:00
/* void aes_dec_blk(struct crypto_tfm *tfm, u8 *out, const u8 *in) */
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
[CRYPTO] aes: Add wrappers for assembly routines The wrapper routines are required when asmlinkage differs from the usual calling convention. So we need to have them. However, by rearranging the parameters, they will get optimised away to a single jump for most people. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2006-06-01 16:42:25 -06:00
entry(aes_dec_blk,240,dec128,dec192)
[CRYPTO] aes-x86-64: Remove setkey The setkey() function can be shared with the generic algorithm. Signed-off-by: Sebastian Siewior <sebastian@breakpoint.cc> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2007-11-08 06:25:04 -07:00
decrypt_round(crypto_it_tab,-96)
decrypt_round(crypto_it_tab,-80)
dec192: decrypt_round(crypto_it_tab,-64)
decrypt_round(crypto_it_tab,-48)
dec128: decrypt_round(crypto_it_tab,-32)
decrypt_round(crypto_it_tab,-16)
decrypt_round(crypto_it_tab, 0)
decrypt_round(crypto_it_tab, 16)
decrypt_round(crypto_it_tab, 32)
decrypt_round(crypto_it_tab, 48)
decrypt_round(crypto_it_tab, 64)
decrypt_round(crypto_it_tab, 80)
decrypt_round(crypto_it_tab, 96)
decrypt_final(crypto_il_tab,112)
[CRYPTO] Add x86_64 asm AES Implementation: =============== The encrypt/decrypt code is based on an x86 implementation I did a while ago which I never published. This unpublished implementation does include an assembler based key schedule and precomputed tables. For simplicity and best acceptance, however, I took Gladman's in-kernel code for table generation and key schedule for the kernel port of my assembler code and modified this code to produce the key schedule as required by my assembler implementation. File locations and Kconfig are kept similar to the i586 AES assembler implementation. It may seem a little bit strange to use 32 bit I/O and registers in the assembler implementation but this gives the best code size. My implementation takes one instruction more per round compared to Gladman's x86 assembler but it doesn't require any stack for local variables or saved registers and it is less serialized than Gladman's code. Note that all comparisons to Gladman's code were done after my code was implemented. I did only use FIPS PUB 197 for the implementation so my implementation is independent work. If anybody has a better assembler solution for x86_64 I'll be pleased to have my code replaced with the better solution. Testing: ======== The implementation passes the in-kernel crypto testing module and I'm running it without any problems on my laptop where it is mainly used for dm-crypt. Microbenchmark: =============== The microbenchmark was done in userspace with similar compile flags as used during kernel compile. Encrypt/decrypt is about 35% faster than the generic C implementation. As the generic C as well as my assembler implementation are both table I don't really expect that there is much room for further improvements though I'll be glad to be corrected here. The key schedule is about 5% slower than the generic C implementation. This is due to the fact that some more work has to be done in the key schedule routine to fit the schedule to the assembler implementation. Code Size: ========== Encrypt and decrypt are together about 2.1 Kbytes smaller than the generic C implementation which is important with regard to L1 cache usage. The key schedule routine is about 100 bytes larger than the generic C implementation. Data Size: ========== There's no difference in data size requirements between the assembler implementation and the generic C implementation. License: ======== Gladmans's code is dual BSD/GPL whereas my assembler code is GPLv2 only (I'm not going to change the license for my code). So I had to change the module license for the x86_64 aes module from 'Dual BSD/GPL' to 'GPL' to reflect the most restrictive license within the module. Signed-off-by: Andreas Steinmetz <ast@domdv.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: David S. Miller <davem@davemloft.net>
2005-07-06 14:55:00 -06:00
return
Reference in a new issue Copy permalink