2019-05-27 14:55:05 +08:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0-or-later */
|
2011-09-26 21:47:25 +08:00
|
|
|
/*
|
|
|
|
|
* Twofish Cipher 3-way parallel algorithm (x86_64)
|
|
|
|
|
*
|
|
|
|
|
* Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
|
|
|
|
|
*/
|
|
|
|
|
|
2013-01-19 19:39:46 +08:00
|
|
|
#include <linux/linkage.h>
|
|
|
|
|
|
2011-09-26 21:47:25 +08:00
|
|
|
.file "twofish-x86_64-asm-3way.S"
|
|
|
|
|
.text
|
|
|
|
|
|
|
|
|
|
/* structure of crypto context */
|
|
|
|
|
#define s0 0
|
|
|
|
|
#define s1 1024
|
|
|
|
|
#define s2 2048
|
|
|
|
|
#define s3 3072
|
|
|
|
|
#define w 4096
|
|
|
|
|
#define k 4128
|
|
|
|
|
|
|
|
|
|
/**********************************************************************
|
|
|
|
|
3-way twofish
|
|
|
|
|
**********************************************************************/
|
|
|
|
|
#define CTX %rdi
|
|
|
|
|
#define RIO %rdx
|
|
|
|
|
|
|
|
|
|
#define RAB0 %rax
|
|
|
|
|
#define RAB1 %rbx
|
|
|
|
|
#define RAB2 %rcx
|
|
|
|
|
|
|
|
|
|
#define RAB0d %eax
|
|
|
|
|
#define RAB1d %ebx
|
|
|
|
|
#define RAB2d %ecx
|
|
|
|
|
|
|
|
|
|
#define RAB0bh %ah
|
|
|
|
|
#define RAB1bh %bh
|
|
|
|
|
#define RAB2bh %ch
|
|
|
|
|
|
|
|
|
|
#define RAB0bl %al
|
|
|
|
|
#define RAB1bl %bl
|
|
|
|
|
#define RAB2bl %cl
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define CD0 0x0(%rsp)
|
|
|
|
|
#define CD1 0x8(%rsp)
|
|
|
|
|
#define CD2 0x10(%rsp)
|
|
|
|
|
|
|
|
|
|
# used only before/after all rounds
|
2011-09-26 21:47:25 +08:00
|
|
|
#define RCD0 %r8
|
|
|
|
|
#define RCD1 %r9
|
|
|
|
|
#define RCD2 %r10
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
# used only during rounds
|
|
|
|
|
#define RX0 %r8
|
|
|
|
|
#define RX1 %r9
|
|
|
|
|
#define RX2 %r10
|
2011-09-26 21:47:25 +08:00
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define RX0d %r8d
|
|
|
|
|
#define RX1d %r9d
|
|
|
|
|
#define RX2d %r10d
|
2011-09-26 21:47:25 +08:00
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define RY0 %r11
|
|
|
|
|
#define RY1 %r12
|
|
|
|
|
#define RY2 %r13
|
2011-09-26 21:47:25 +08:00
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define RY0d %r11d
|
|
|
|
|
#define RY1d %r12d
|
|
|
|
|
#define RY2d %r13d
|
2011-09-26 21:47:25 +08:00
|
|
|
|
|
|
|
|
#define RT0 %rdx
|
|
|
|
|
#define RT1 %rsi
|
|
|
|
|
|
|
|
|
|
#define RT0d %edx
|
|
|
|
|
#define RT1d %esi
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define RT1bl %sil
|
|
|
|
|
|
2011-09-26 21:47:25 +08:00
|
|
|
#define do16bit_ror(rot, op1, op2, T0, T1, tmp1, tmp2, ab, dst) \
|
|
|
|
|
movzbl ab ## bl, tmp2 ## d; \
|
|
|
|
|
movzbl ab ## bh, tmp1 ## d; \
|
|
|
|
|
rorq $(rot), ab; \
|
|
|
|
|
op1##l T0(CTX, tmp2, 4), dst ## d; \
|
|
|
|
|
op2##l T1(CTX, tmp1, 4), dst ## d;
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define swap_ab_with_cd(ab, cd, tmp) \
|
|
|
|
|
movq cd, tmp; \
|
|
|
|
|
movq ab, cd; \
|
|
|
|
|
movq tmp, ab;
|
|
|
|
|
|
2011-09-26 21:47:25 +08:00
|
|
|
/*
|
|
|
|
|
* Combined G1 & G2 function. Reordered with help of rotates to have moves
|
|
|
|
|
* at begining.
|
|
|
|
|
*/
|
|
|
|
|
#define g1g2_3(ab, cd, Tx0, Tx1, Tx2, Tx3, Ty0, Ty1, Ty2, Ty3, x, y) \
|
|
|
|
|
/* G1,1 && G2,1 */ \
|
|
|
|
|
do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 0, ab ## 0, x ## 0); \
|
|
|
|
|
do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 0, ab ## 0, y ## 0); \
|
|
|
|
|
\
|
|
|
|
|
do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 1, ab ## 1, x ## 1); \
|
|
|
|
|
do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 1, ab ## 1, y ## 1); \
|
|
|
|
|
\
|
|
|
|
|
do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 2, ab ## 2, x ## 2); \
|
|
|
|
|
do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 2, ab ## 2, y ## 2); \
|
|
|
|
|
\
|
|
|
|
|
/* G1,2 && G2,2 */ \
|
|
|
|
|
do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 0, x ## 0); \
|
|
|
|
|
do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 0, y ## 0); \
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
swap_ab_with_cd(ab ## 0, cd ## 0, RT0); \
|
2011-09-26 21:47:25 +08:00
|
|
|
\
|
|
|
|
|
do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 1, x ## 1); \
|
|
|
|
|
do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 1, y ## 1); \
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
swap_ab_with_cd(ab ## 1, cd ## 1, RT0); \
|
2011-09-26 21:47:25 +08:00
|
|
|
\
|
|
|
|
|
do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 2, x ## 2); \
|
|
|
|
|
do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 2, y ## 2); \
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
swap_ab_with_cd(ab ## 2, cd ## 2, RT0);
|
2011-09-26 21:47:25 +08:00
|
|
|
|
|
|
|
|
#define enc_round_end(ab, x, y, n) \
|
|
|
|
|
addl y ## d, x ## d; \
|
|
|
|
|
addl x ## d, y ## d; \
|
|
|
|
|
addl k+4*(2*(n))(CTX), x ## d; \
|
|
|
|
|
xorl ab ## d, x ## d; \
|
|
|
|
|
addl k+4*(2*(n)+1)(CTX), y ## d; \
|
|
|
|
|
shrq $32, ab; \
|
|
|
|
|
roll $1, ab ## d; \
|
|
|
|
|
xorl y ## d, ab ## d; \
|
|
|
|
|
shlq $32, ab; \
|
|
|
|
|
rorl $1, x ## d; \
|
|
|
|
|
orq x, ab;
|
|
|
|
|
|
|
|
|
|
#define dec_round_end(ba, x, y, n) \
|
|
|
|
|
addl y ## d, x ## d; \
|
|
|
|
|
addl x ## d, y ## d; \
|
|
|
|
|
addl k+4*(2*(n))(CTX), x ## d; \
|
|
|
|
|
addl k+4*(2*(n)+1)(CTX), y ## d; \
|
|
|
|
|
xorl ba ## d, y ## d; \
|
|
|
|
|
shrq $32, ba; \
|
|
|
|
|
roll $1, ba ## d; \
|
|
|
|
|
xorl x ## d, ba ## d; \
|
|
|
|
|
shlq $32, ba; \
|
|
|
|
|
rorl $1, y ## d; \
|
|
|
|
|
orq y, ba;
|
|
|
|
|
|
|
|
|
|
#define encrypt_round3(ab, cd, n) \
|
|
|
|
|
g1g2_3(ab, cd, s0, s1, s2, s3, s0, s1, s2, s3, RX, RY); \
|
|
|
|
|
\
|
|
|
|
|
enc_round_end(ab ## 0, RX0, RY0, n); \
|
|
|
|
|
enc_round_end(ab ## 1, RX1, RY1, n); \
|
|
|
|
|
enc_round_end(ab ## 2, RX2, RY2, n);
|
|
|
|
|
|
|
|
|
|
#define decrypt_round3(ba, dc, n) \
|
|
|
|
|
g1g2_3(ba, dc, s1, s2, s3, s0, s3, s0, s1, s2, RY, RX); \
|
|
|
|
|
\
|
|
|
|
|
dec_round_end(ba ## 0, RX0, RY0, n); \
|
|
|
|
|
dec_round_end(ba ## 1, RX1, RY1, n); \
|
|
|
|
|
dec_round_end(ba ## 2, RX2, RY2, n);
|
|
|
|
|
|
|
|
|
|
#define encrypt_cycle3(ab, cd, n) \
|
|
|
|
|
encrypt_round3(ab, cd, n*2); \
|
|
|
|
|
encrypt_round3(ab, cd, (n*2)+1);
|
|
|
|
|
|
|
|
|
|
#define decrypt_cycle3(ba, dc, n) \
|
|
|
|
|
decrypt_round3(ba, dc, (n*2)+1); \
|
|
|
|
|
decrypt_round3(ba, dc, (n*2));
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
#define push_cd() \
|
|
|
|
|
pushq RCD2; \
|
|
|
|
|
pushq RCD1; \
|
|
|
|
|
pushq RCD0;
|
|
|
|
|
|
|
|
|
|
#define pop_cd() \
|
|
|
|
|
popq RCD0; \
|
|
|
|
|
popq RCD1; \
|
|
|
|
|
popq RCD2;
|
|
|
|
|
|
2011-09-26 21:47:25 +08:00
|
|
|
#define inpack3(in, n, xy, m) \
|
|
|
|
|
movq 4*(n)(in), xy ## 0; \
|
|
|
|
|
xorq w+4*m(CTX), xy ## 0; \
|
|
|
|
|
\
|
|
|
|
|
movq 4*(4+(n))(in), xy ## 1; \
|
|
|
|
|
xorq w+4*m(CTX), xy ## 1; \
|
|
|
|
|
\
|
|
|
|
|
movq 4*(8+(n))(in), xy ## 2; \
|
|
|
|
|
xorq w+4*m(CTX), xy ## 2;
|
|
|
|
|
|
|
|
|
|
#define outunpack3(op, out, n, xy, m) \
|
|
|
|
|
xorq w+4*m(CTX), xy ## 0; \
|
|
|
|
|
op ## q xy ## 0, 4*(n)(out); \
|
|
|
|
|
\
|
|
|
|
|
xorq w+4*m(CTX), xy ## 1; \
|
|
|
|
|
op ## q xy ## 1, 4*(4+(n))(out); \
|
|
|
|
|
\
|
|
|
|
|
xorq w+4*m(CTX), xy ## 2; \
|
|
|
|
|
op ## q xy ## 2, 4*(8+(n))(out);
|
|
|
|
|
|
|
|
|
|
#define inpack_enc3() \
|
|
|
|
|
inpack3(RIO, 0, RAB, 0); \
|
|
|
|
|
inpack3(RIO, 2, RCD, 2);
|
|
|
|
|
|
|
|
|
|
#define outunpack_enc3(op) \
|
|
|
|
|
outunpack3(op, RIO, 2, RAB, 6); \
|
|
|
|
|
outunpack3(op, RIO, 0, RCD, 4);
|
|
|
|
|
|
|
|
|
|
#define inpack_dec3() \
|
|
|
|
|
inpack3(RIO, 0, RAB, 4); \
|
|
|
|
|
rorq $32, RAB0; \
|
|
|
|
|
rorq $32, RAB1; \
|
|
|
|
|
rorq $32, RAB2; \
|
|
|
|
|
inpack3(RIO, 2, RCD, 6); \
|
|
|
|
|
rorq $32, RCD0; \
|
|
|
|
|
rorq $32, RCD1; \
|
|
|
|
|
rorq $32, RCD2;
|
|
|
|
|
|
|
|
|
|
#define outunpack_dec3() \
|
|
|
|
|
rorq $32, RCD0; \
|
|
|
|
|
rorq $32, RCD1; \
|
|
|
|
|
rorq $32, RCD2; \
|
|
|
|
|
outunpack3(mov, RIO, 0, RCD, 0); \
|
|
|
|
|
rorq $32, RAB0; \
|
|
|
|
|
rorq $32, RAB1; \
|
|
|
|
|
rorq $32, RAB2; \
|
|
|
|
|
outunpack3(mov, RIO, 2, RAB, 2);
|
|
|
|
|
|
2019-10-11 19:51:04 +08:00
|
|
|
SYM_FUNC_START(__twofish_enc_blk_3way)
|
2011-09-26 21:47:25 +08:00
|
|
|
/* input:
|
|
|
|
|
* %rdi: ctx, CTX
|
|
|
|
|
* %rsi: dst
|
|
|
|
|
* %rdx: src, RIO
|
|
|
|
|
* %rcx: bool, if true: xor output
|
|
|
|
|
*/
|
|
|
|
|
pushq %r13;
|
|
|
|
|
pushq %r12;
|
|
|
|
|
pushq %rbx;
|
|
|
|
|
|
|
|
|
|
pushq %rcx; /* bool xor */
|
|
|
|
|
pushq %rsi; /* dst */
|
|
|
|
|
|
|
|
|
|
inpack_enc3();
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
push_cd();
|
|
|
|
|
encrypt_cycle3(RAB, CD, 0);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 1);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 2);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 3);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 4);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 5);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 6);
|
|
|
|
|
encrypt_cycle3(RAB, CD, 7);
|
|
|
|
|
pop_cd();
|
2011-09-26 21:47:25 +08:00
|
|
|
|
|
|
|
|
popq RIO; /* dst */
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
popq RT1; /* bool xor */
|
2011-09-26 21:47:25 +08:00
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
testb RT1bl, RT1bl;
|
2013-01-19 19:39:46 +08:00
|
|
|
jnz .L__enc_xor3;
|
2011-09-26 21:47:25 +08:00
|
|
|
|
|
|
|
|
outunpack_enc3(mov);
|
|
|
|
|
|
|
|
|
|
popq %rbx;
|
|
|
|
|
popq %r12;
|
|
|
|
|
popq %r13;
|
|
|
|
|
ret;
|
|
|
|
|
|
2013-01-19 19:39:46 +08:00
|
|
|
.L__enc_xor3:
|
2011-09-26 21:47:25 +08:00
|
|
|
outunpack_enc3(xor);
|
|
|
|
|
|
|
|
|
|
popq %rbx;
|
|
|
|
|
popq %r12;
|
|
|
|
|
popq %r13;
|
|
|
|
|
ret;
|
2019-10-11 19:51:04 +08:00
|
|
|
SYM_FUNC_END(__twofish_enc_blk_3way)
|
2011-09-26 21:47:25 +08:00
|
|
|
|
2019-10-11 19:51:04 +08:00
|
|
|
SYM_FUNC_START(twofish_dec_blk_3way)
|
2011-09-26 21:47:25 +08:00
|
|
|
/* input:
|
|
|
|
|
* %rdi: ctx, CTX
|
|
|
|
|
* %rsi: dst
|
|
|
|
|
* %rdx: src, RIO
|
|
|
|
|
*/
|
|
|
|
|
pushq %r13;
|
|
|
|
|
pushq %r12;
|
|
|
|
|
pushq %rbx;
|
|
|
|
|
|
|
|
|
|
pushq %rsi; /* dst */
|
|
|
|
|
|
|
|
|
|
inpack_dec3();
|
|
|
|
|
|
crypto: x86/twofish-3way - Fix %rbp usage
Using %rbp as a temporary register breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
In twofish-3way, we can't simply replace %rbp with another register
because there are none available. Instead, we use the stack to hold the
values that %rbp, %r11, and %r12 were holding previously. Each of these
values represents the half of the output from the previous Feistel round
that is being passed on unchanged to the following round. They are only
used once per round, when they are exchanged with %rax, %rbx, and %rcx.
As a result, we free up 3 registers (one per block) and can reassign
them so that %rbp is not used, and additionally %r14 and %r15 are not
used so they do not need to be saved/restored.
There may be a small overhead caused by replacing 'xchg REG, REG' with
the needed sequence 'mov MEM, REG; mov REG, MEM; mov REG, REG' once per
round. But, counterintuitively, when I tested "ctr-twofish-3way" on a
Haswell processor, the new version was actually about 2% faster.
(Perhaps 'xchg' is not as well optimized as plain moves.)
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2017-12-19 08:40:26 +08:00
|
|
|
push_cd();
|
|
|
|
|
decrypt_cycle3(RAB, CD, 7);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 6);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 5);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 4);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 3);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 2);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 1);
|
|
|
|
|
decrypt_cycle3(RAB, CD, 0);
|
|
|
|
|
pop_cd();
|
2011-09-26 21:47:25 +08:00
|
|
|
|
|
|
|
|
popq RIO; /* dst */
|
|
|
|
|
|
|
|
|
|
outunpack_dec3();
|
|
|
|
|
|
|
|
|
|
popq %rbx;
|
|
|
|
|
popq %r12;
|
|
|
|
|
popq %r13;
|
|
|
|
|
ret;
|
2019-10-11 19:51:04 +08:00
|
|
|
SYM_FUNC_END(twofish_dec_blk_3way)
|