AuxXxilium/linux_dsm_epyc7002
1
0
Fork 0
mirror of https://github.com/AuxXxilium/linux_dsm_epyc7002.git synced 2024-12-24 21:57:25 +07:00
Code Issues Actions Packages Projects Releases Wiki Activity
fea79e0b0e
linux_dsm_epyc7002/arch/arm/crypto/aes-cipher-core.S

222 lines
5.5 KiB
ArmAsm
Raw Normal View History

crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
/*
* Scalar AES core transform
*
* Copyright (C) 2017 Linaro Ltd.
* Author: Ard Biesheuvel <ard.biesheuvel@linaro.org>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/linkage.h>
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
#include <asm/cache.h>
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.text
.align 5
rk .req r0
rounds .req r1
in .req r2
out .req r3
crypto: arm/aes - avoid reserved 'tt' mnemonic in asm code The ARMv8-M architecture introduces 'tt' and 'ttt' instructions, which means we can no longer use 'tt' as a register alias on recent versions of binutils for ARM. So replace the alias with 'ttab'. Fixes: 81edb4262975 ("crypto: arm/aes - replace scalar AES cipher") Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-13 15:33:26 +07:00
ttab .req ip
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
t0 .req lr
t1 .req r2
t2 .req r3
.macro __select, out, in, idx
.if __LINUX_ARM_ARCH__ < 7
and \out, \in, #0xff << (8 * \idx)
.else
ubfx \out, \in, #(8 * \idx), #8
.endif
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.macro __load, out, in, idx, sz, op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.if __LINUX_ARM_ARCH__ < 7 && \idx > 0
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
ldr\op \out, [ttab, \in, lsr #(8 * \idx) - \sz]
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.else
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
ldr\op \out, [ttab, \in, lsl #\sz]
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.endif
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.macro __hround, out0, out1, in0, in1, in2, in3, t3, t4, enc, sz, op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
__select \out0, \in0, 0
__select t0, \in1, 1
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
__load \out0, \out0, 0, \sz, \op
__load t0, t0, 1, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.if \enc
__select \out1, \in1, 0
__select t1, \in2, 1
.else
__select \out1, \in3, 0
__select t1, \in0, 1
.endif
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
__load \out1, \out1, 0, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
__select t2, \in2, 2
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
__load t1, t1, 1, \sz, \op
__load t2, t2, 2, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
eor \out0, \out0, t0, ror #24
__select t0, \in3, 3
.if \enc
__select \t3, \in3, 2
__select \t4, \in0, 3
.else
__select \t3, \in1, 2
__select \t4, \in2, 3
.endif
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
__load \t3, \t3, 2, \sz, \op
__load t0, t0, 3, \sz, \op
__load \t4, \t4, 3, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
eor \out1, \out1, t1, ror #24
eor \out0, \out0, t2, ror #16
ldm rk!, {t1, t2}
eor \out1, \out1, \t3, ror #16
eor \out0, \out0, t0, ror #8
eor \out1, \out1, \t4, ror #8
eor \out0, \out0, t1
eor \out1, \out1, t2
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.macro fround, out0, out1, out2, out3, in0, in1, in2, in3, sz=2, op
__hround \out0, \out1, \in0, \in1, \in2, \in3, \out2, \out3, 1, \sz, \op
__hround \out2, \out3, \in2, \in3, \in0, \in1, \in1, \in2, 1, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.macro iround, out0, out1, out2, out3, in0, in1, in2, in3, sz=2, op
__hround \out0, \out1, \in0, \in3, \in2, \in1, \out2, \out3, 0, \sz, \op
__hround \out2, \out3, \in2, \in1, \in0, \in3, \in1, \in0, 0, \sz, \op
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
.endm
.macro __rev, out, in
.if __LINUX_ARM_ARCH__ < 6
lsl t0, \in, #24
and t1, \in, #0xff00
and t2, \in, #0xff0000
orr \out, t0, \in, lsr #24
orr \out, \out, t1, lsl #8
orr \out, \out, t2, lsr #8
.else
rev \out, \in
.endif
.endm
.macro __adrl, out, sym, c
.if __LINUX_ARM_ARCH__ < 7
ldr\c \out, =\sym
.else
movw\c \out, #:lower16:\sym
movt\c \out, #:upper16:\sym
.endif
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.macro do_crypt, round, ttab, ltab, bsz
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
push {r3-r11, lr}
ldr r4, [in]
ldr r5, [in, #4]
ldr r6, [in, #8]
ldr r7, [in, #12]
ldm rk!, {r8-r11}
#ifdef CONFIG_CPU_BIG_ENDIAN
__rev r4, r4
__rev r5, r5
__rev r6, r6
__rev r7, r7
#endif
eor r4, r4, r8
eor r5, r5, r9
eor r6, r6, r10
eor r7, r7, r11
crypto: arm/aes - avoid reserved 'tt' mnemonic in asm code The ARMv8-M architecture introduces 'tt' and 'ttt' instructions, which means we can no longer use 'tt' as a register alias on recent versions of binutils for ARM. So replace the alias with 'ttab'. Fixes: 81edb4262975 ("crypto: arm/aes - replace scalar AES cipher") Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-13 15:33:26 +07:00
__adrl ttab, \ttab
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
tst rounds, #2
bne 1f
0: \round r8, r9, r10, r11, r4, r5, r6, r7
\round r4, r5, r6, r7, r8, r9, r10, r11
1: subs rounds, rounds, #4
\round r8, r9, r10, r11, r4, r5, r6, r7
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
bls 2f
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
\round r4, r5, r6, r7, r8, r9, r10, r11
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
b 0b
2: __adrl ttab, \ltab
\round r4, r5, r6, r7, r8, r9, r10, r11, \bsz, b
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
#ifdef CONFIG_CPU_BIG_ENDIAN
__rev r4, r4
__rev r5, r5
__rev r6, r6
__rev r7, r7
#endif
ldr out, [sp]
str r4, [out]
str r5, [out, #4]
str r6, [out, #8]
str r7, [out, #12]
pop {r3-r11, pc}
.align 3
.ltorg
.endm
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.align L1_CACHE_SHIFT
.type __aes_arm_inverse_sbox, %object
__aes_arm_inverse_sbox:
.byte 0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38
.byte 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb
.byte 0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87
.byte 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb
.byte 0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d
.byte 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e
.byte 0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2
.byte 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25
.byte 0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16
.byte 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92
.byte 0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda
.byte 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84
.byte 0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a
.byte 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06
.byte 0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02
.byte 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b
.byte 0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea
.byte 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73
.byte 0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85
.byte 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e
.byte 0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89
.byte 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b
.byte 0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20
.byte 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4
.byte 0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31
.byte 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f
.byte 0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d
.byte 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef
.byte 0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0
.byte 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61
.byte 0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26
.byte 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d
.size __aes_arm_inverse_sbox, . - __aes_arm_inverse_sbox
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
ENTRY(__aes_arm_encrypt)
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
do_crypt fround, crypto_ft_tab, crypto_ft_tab + 1, 2
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
ENDPROC(__aes_arm_encrypt)
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
.align 5
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
ENTRY(__aes_arm_decrypt)
crypto: arm/aes - avoid expanded lookup tables in the final round For the final round, avoid the expanded and padded lookup tables exported by the generic AES driver. Instead, for encryption, we can perform byte loads from the same table we used for the inner rounds, which will still be hot in the caches. For decryption, use the inverse AES Sbox directly, which is 4x smaller than the inverse lookup table exported by the generic driver. This should significantly reduce the Dcache footprint of our code, which makes the code more robust against timing attacks. It does not introduce any additional module dependencies, given that we already rely on the core AES module for the shared key expansion routines. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-07-24 17:28:19 +07:00
do_crypt iround, crypto_it_tab, __aes_arm_inverse_sbox, 0
crypto: arm/aes - replace scalar AES cipher This replaces the scalar AES cipher that originates in the OpenSSL project with a new implementation that is ~15% (*) faster (on modern cores), and reuses the lookup tables and the key schedule generation routines from the generic C implementation (which is usually compiled in anyway due to networking and other subsystems depending on it). Note that the bit sliced NEON code for AES still depends on the scalar cipher that this patch replaces, so it is not removed entirely yet. * On Cortex-A57, the performance increases from 17.0 to 14.9 cycles per byte for 128-bit keys. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-01-11 23:41:53 +07:00
ENDPROC(__aes_arm_decrypt)
Reference in New Issue Copy Permalink