Optimize SM2 on aarch64

Signed-off-by: Xu Yizhou <xuyizhou1@huawei.com>

Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20754)
This commit is contained in:
Xu Yizhou 2023-08-23 17:30:09 +08:00 committed by Tomas Mraz
parent ce7a9e23fb
commit 6399d7856c
10 changed files with 18045 additions and 3 deletions

View File

@ -102,6 +102,7 @@ jobs:
no-siphash,
no-siv,
no-sm2,
no-sm2-precomp,
no-sm3,
no-sm4,
no-sock,

View File

@ -25,6 +25,13 @@ OpenSSL 3.2
### Changes between 3.1 and 3.2 [xx XXX xxxx]
* Added optimization for SM2 algorithm on aarch64. It uses a huge precomputed
table for point multiplication of the base point, which increases the size of
libcrypto from 4.4 MB to 4.9 MB. A new configure option `no-sm2-precomp` has
been added to disable the precomputed table.
*Xu Yizhou*
* Added client side support for QUIC
*Hugo Landau, Matt Caswell, Paul Dale, Tomáš Mráz, Richard Levitte*

View File

@ -497,6 +497,7 @@ my @disablables = (
"siphash",
"siv",
"sm2",
"sm2-precomp",
"sm3",
"sm4",
"sock",

View File

@ -914,6 +914,10 @@ Do not create shared libraries, only static ones.
See [Notes on shared libraries](#notes-on-shared-libraries) below.
### no-sm2-precomp
Disable using the SM2 precomputed table on aarch64 to make the library smaller.
### no-sock
Don't build support for socket BIOs.

View File

@ -0,0 +1,820 @@
#! /usr/bin/env perl
# Copyright 2023 The OpenSSL Project Authors. All Rights Reserved.
#
# Licensed under the OpenSSL license (the "License"). You may not use
# this file except in compliance with the License. You can obtain a copy
# in the file LICENSE in the source distribution or at
# https://www.openssl.org/source/license.html
$flavour = shift;
while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
die "can't locate arm-xlate.pl";
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;
my ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("x$_",(7..14));
my ($a8,$a10,$a12,$a14,$a9,$a11,$a13,$a15)=map("x$_",(7..14));
my ($t0,$t1,$t2,$t3)=map("x$_",(3..6));
my ($t4,$t5,$t6,$t7,$t8)=map("x$_",(15..19));
sub bn_mod_add() {
my $mod = shift;
$code.=<<___;
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Addition
adds $s0,$s0,$s4
adcs $s1,$s1,$s5
adcs $s2,$s2,$s6
adcs $s3,$s3,$s7
adc $t4,xzr,xzr
# Load polynomial
adr x2,$mod
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Backup Addition
mov $t0,$s0
mov $t1,$s1
mov $t2,$s2
mov $t3,$s3
# Sub polynomial
subs $t0,$t0,$s4
sbcs $t1,$t1,$s5
sbcs $t2,$t2,$s6
sbcs $t3,$t3,$s7
sbcs $t4,$t4,xzr
# Select based on carry
csel $s0,$s0,$t0,cc
csel $s1,$s1,$t1,cc
csel $s2,$s2,$t2,cc
csel $s3,$s3,$t3,cc
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
___
}
sub bn_mod_sub() {
my $mod = shift;
$code.=<<___;
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Subtraction
subs $s0,$s0,$s4
sbcs $s1,$s1,$s5
sbcs $s2,$s2,$s6
sbcs $s3,$s3,$s7
sbc $t4,xzr,xzr
# Load polynomial
adr x2,$mod
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Backup subtraction
mov $t0,$s0
mov $t1,$s1
mov $t2,$s2
mov $t3,$s3
# Add polynomial
adds $t0,$t0,$s4
adcs $t1,$t1,$s5
adcs $t2,$t2,$s6
adcs $t3,$t3,$s7
tst $t4,$t4
# Select based on carry
csel $s0,$s0,$t0,eq
csel $s1,$s1,$t1,eq
csel $s2,$s2,$t2,eq
csel $s3,$s3,$t3,eq
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
___
}
sub bn_mod_div_by_2() {
my $mod = shift;
$code.=<<___;
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
# Save the least significant bit
mov $t0,$s0
# Right shift 1
extr $s0,$s1,$s0,#1
extr $s1,$s2,$s1,#1
extr $s2,$s3,$s2,#1
lsr $s3,$s3,#1
# Load mod
adr x2,$mod
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Parity check
tst $t0,#1
csel $s4,xzr,$s4,eq
csel $s5,xzr,$s5,eq
csel $s6,xzr,$s6,eq
csel $s7,xzr,$s7,eq
# Add
adds $s0,$s0,$s4
adcs $s1,$s1,$s5
adcs $s2,$s2,$s6
adc $s3,$s3,$s7
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
___
}
{
$code.=<<___;
#include "arm_arch.h"
.arch armv8-a
.text
.align 5
// The polynomial p
.Lpoly:
.quad 0xffffffffffffffff,0xffffffff00000000,0xffffffffffffffff,0xfffffffeffffffff
// The order of polynomial n
.Lord:
.quad 0x53bbf40939d54123,0x7203df6b21c6052b,0xffffffffffffffff,0xfffffffeffffffff
// (p + 1) / 2
.Lpoly_div_2:
.quad 0x8000000000000000,0xffffffff80000000,0xffffffffffffffff,0x7fffffff7fffffff
// (n + 1) / 2
.Lord_div_2:
.quad 0xa9ddfa049ceaa092,0xb901efb590e30295,0xffffffffffffffff,0x7fffffff7fffffff
// void bn_rshift1(BN_ULONG *a);
.globl bn_rshift1
.type bn_rshift1,%function
.align 5
bn_rshift1:
AARCH64_VALID_CALL_TARGET
# Load inputs
ldp $s0,$s1,[x0]
ldp $s2,$s3,[x0,#16]
# Right shift
extr $s0,$s1,$s0,#1
extr $s1,$s2,$s1,#1
extr $s2,$s3,$s2,#1
lsr $s3,$s3,#1
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
ret
.size bn_rshift1,.-bn_rshift1
// void bn_sub(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
.globl bn_sub
.type bn_sub,%function
.align 5
bn_sub:
AARCH64_VALID_CALL_TARGET
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
# Subtraction
subs $s0,$s0,$s4
sbcs $s1,$s1,$s5
sbcs $s2,$s2,$s6
sbc $s3,$s3,$s7
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
ret
.size bn_sub,.-bn_sub
// void ecp_sm2p256_div_by_2(BN_ULONG *r,const BN_ULONG *a);
.globl ecp_sm2p256_div_by_2
.type ecp_sm2p256_div_by_2,%function
.align 5
ecp_sm2p256_div_by_2:
AARCH64_VALID_CALL_TARGET
___
&bn_mod_div_by_2(".Lpoly_div_2");
$code.=<<___;
ret
.size ecp_sm2p256_div_by_2,.-ecp_sm2p256_div_by_2
// void ecp_sm2p256_div_by_2_mod_ord(BN_ULONG *r,const BN_ULONG *a);
.globl ecp_sm2p256_div_by_2_mod_ord
.type ecp_sm2p256_div_by_2_mod_ord,%function
.align 5
ecp_sm2p256_div_by_2_mod_ord:
AARCH64_VALID_CALL_TARGET
___
&bn_mod_div_by_2(".Lord_div_2");
$code.=<<___;
ret
.size ecp_sm2p256_div_by_2_mod_ord,.-ecp_sm2p256_div_by_2_mod_ord
// void ecp_sm2p256_mul_by_3(BN_ULONG *r,const BN_ULONG *a);
.globl ecp_sm2p256_mul_by_3
.type ecp_sm2p256_mul_by_3,%function
.align 5
ecp_sm2p256_mul_by_3:
AARCH64_VALID_CALL_TARGET
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
# 2*a
adds $s0,$s0,$s0
adcs $s1,$s1,$s1
adcs $s2,$s2,$s2
adcs $s3,$s3,$s3
adcs $t4,xzr,xzr
mov $t0,$s0
mov $t1,$s1
mov $t2,$s2
mov $t3,$s3
# Sub polynomial
adr x2,.Lpoly
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
subs $s0,$s0,$s4
sbcs $s1,$s1,$s5
sbcs $s2,$s2,$s6
sbcs $s3,$s3,$s7
sbcs $t4,$t4,xzr
csel $s0,$s0,$t0,cs
csel $s1,$s1,$t1,cs
csel $s2,$s2,$t2,cs
csel $s3,$s3,$t3,cs
eor $t4,$t4,$t4
# 3*a
ldp $s4,$s5,[x1]
ldp $s6,$s7,[x1,#16]
adds $s0,$s0,$s4
adcs $s1,$s1,$s5
adcs $s2,$s2,$s6
adcs $s3,$s3,$s7
adcs $t4,xzr,xzr
mov $t0,$s0
mov $t1,$s1
mov $t2,$s2
mov $t3,$s3
# Sub polynomial
adr x2,.Lpoly
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
subs $s0,$s0,$s4
sbcs $s1,$s1,$s5
sbcs $s2,$s2,$s6
sbcs $s3,$s3,$s7
sbcs $t4,$t4,xzr
csel $s0,$s0,$t0,cs
csel $s1,$s1,$t1,cs
csel $s2,$s2,$t2,cs
csel $s3,$s3,$t3,cs
# Store results
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
ret
.size ecp_sm2p256_mul_by_3,.-ecp_sm2p256_mul_by_3
// void ecp_sm2p256_add(BN_ULONG *r,const BN_ULONG *a,const BN_ULONG *b);
.globl ecp_sm2p256_add
.type ecp_sm2p256_add,%function
.align 5
ecp_sm2p256_add:
AARCH64_VALID_CALL_TARGET
___
&bn_mod_add(".Lpoly");
$code.=<<___;
ret
.size ecp_sm2p256_add,.-ecp_sm2p256_add
// void ecp_sm2p256_sub(BN_ULONG *r,const BN_ULONG *a,const BN_ULONG *b);
.globl ecp_sm2p256_sub
.type ecp_sm2p256_sub,%function
.align 5
ecp_sm2p256_sub:
AARCH64_VALID_CALL_TARGET
___
&bn_mod_sub(".Lpoly");
$code.=<<___;
ret
.size ecp_sm2p256_sub,.-ecp_sm2p256_sub
// void ecp_sm2p256_sub_mod_ord(BN_ULONG *r,const BN_ULONG *a,const BN_ULONG *b);
.globl ecp_sm2p256_sub_mod_ord
.type ecp_sm2p256_sub_mod_ord,%function
.align 5
ecp_sm2p256_sub_mod_ord:
AARCH64_VALID_CALL_TARGET
___
&bn_mod_sub(".Lord");
$code.=<<___;
ret
.size ecp_sm2p256_sub_mod_ord,.-ecp_sm2p256_sub_mod_ord
.macro RDC
# a = | s7 | ... | s0 |, where si are 64-bit quantities
# = |a15|a14| ... |a1|a0|, where ai are 32-bit quantities
# | s7 | s6 | s5 | s4 |
# | a15 | a14 | a13 | a12 | a11 | a10 | a9 | a8 |
# | s3 | s2 | s1 | s0 |
# | a7 | a6 | a5 | a4 | a3 | a2 | a1 | a0 |
# =================================================
# | a8 | a11 | a10 | a9 | a8 | 0 | s4 | (+)
# | a9 | a15 | s6 | a11 | 0 | a10 | a9 | (+)
# | a10 | 0 | a14 | a13 | a12 | 0 | s5 | (+)
# | a11 | 0 | s7 | a13 | 0 | a12 | a11 | (+)
# | a12 | 0 | s7 | a13 | 0 | s6 | (+)
# | a12 | 0 | 0 | a15 | a14 | 0 | a14 | a13 | (+)
# | a13 | 0 | 0 | 0 | a15 | 0 | a14 | a13 | (+)
# | a13 | 0 | 0 | 0 | 0 | 0 | s7 | (+)
# | a14 | 0 | 0 | 0 | 0 | 0 | s7 | (+)
# | a14 | 0 | 0 | 0 | 0 | 0 | 0 | a15 | (+)
# | a15 | 0 | 0 | 0 | 0 | 0 | 0 | a15 | (+)
# | a15 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | (+)
# | s7 | 0 | 0 | 0 | 0 | 0 | 0 | (+)
# | 0 | 0 | 0 | 0 | 0 | a8 | 0 | 0 | (-)
# | 0 | 0 | 0 | 0 | 0 | a9 | 0 | 0 | (-)
# | 0 | 0 | 0 | 0 | 0 | a13 | 0 | 0 | (-)
# | 0 | 0 | 0 | 0 | 0 | a14 | 0 | 0 | (-)
# | U[7]| U[6]| U[5]| U[4]| U[3]| U[2]| U[1]| U[0]|
# | V[3] | V[2] | V[1] | V[0] |
# 1. 64-bit addition
# t2=s6+s7+s7
adds $t2,$s6,$s7
adcs $t1,xzr,xzr
adds $t2,$t2,$s7
adcs $t1,$t1,xzr
# t3=s4+s5+t2
adds $t3,$s4,$t2
adcs $t4,$t1,xzr
adds $t3,$t3,$s5
adcs $t4,$t4,xzr
# sum
adds $s0,$s0,$t3
adcs $s1,$s1,$t4
adcs $s2,$s2,$t2
adcs $s3,$s3,$s7
adcs $t0,xzr,xzr
adds $s3,$s3,$t1
adcs $t0,$t0,xzr
stp $s0,$s1,[sp,#32]
stp $s2,$s3,[sp,#48]
# 2. 64-bit to 32-bit spread
mov $t1,#0xffffffff
mov $s0,$s4
mov $s1,$s5
mov $s2,$s6
mov $s3,$s7
and $s0,$s0,$t1 // a8
and $s1,$s1,$t1 // a10
and $s2,$s2,$t1 // a12
and $s3,$s3,$t1 // a14
lsr $s4,$s4,#32 // a9
lsr $s5,$s5,#32 // a11
lsr $s6,$s6,#32 // a13
lsr $s7,$s7,#32 // a15
# 3. 32-bit addition
add $t1,$a14,$a12 // t1 <- a12 + a14
add $t2,$a15,$a13 // t2 <- a13 + a15
add $t3,$a8,$a9 // t3 <- a8 + a9
add $t4,$a14,$a10 // t4 <- a10 + a14
add $a15,$a15,$a11 // a15 <- a11 + a15
add $a12,$t2,$t1 // a12 <- a12 + a13 + a14 + a15
add $a10,$a10,$a12 // a10 <- a10 + a12 + a13 + a14 + a15
add $a10,$a10,$a12 // a10 <- a10 + 2*(a12 + a13 + a14 + a15)
add $a10,$a10,$t3 // a10 <- a8 + a9 + a10 + 2*(a12 + a13 + a14 + a15)
add $a10,$a10,$a11 // a10 <- a8 + a9 + a10 + a11 + 2*(a12 + a13 + a14 + a15)
add $a12,$a12,$a13 // a12 <- a12 + 2*a13 + a14 + a15
add $a12,$a12,$a11 // a12 <- a11 + a12 + 2*a13 + a14 + a15
add $a12,$a12,$a8 // a12 <- a8 + a11 + a12 + 2*a13 + a14 + a15
add $t3,$t3,$a14 // t3 <- a8 + a9 + a14
add $t3,$t3,$a13 // t3 <- a8 + a9 + a13 + a14
add $a9,$a9,$t2 // a9 <- a9 + a13 + a15
add $a11,$a11,$a9 // a11 <- a9 + a11 + a13 + a15
add $a11,$a11,$t2 // a11 <- a9 + a11 + 2*(a13 + a15)
add $t1,$t1,$t4 // t1 <- a10 + a12 + 2*a14
# U[0] s5 a9 + a11 + 2*(a13 + a15)
# U[1] t1 a10 + a12 + 2*a14
# U[2] -t3 a8 + a9 + a13 + a14
# U[3] s2 a8 + a11 + a12 + 2*a13 + a14 + a15
# U[4] s4 a9 + a13 + a15
# U[5] t4 a10 + a14
# U[6] s7 a11 + a15
# U[7] s1 a8 + a9 + a10 + a11 + 2*(a12 + a13 + a14 + a15)
# 4. 32-bit to 64-bit
lsl $s0,$t1,#32
extr $t1,$s2,$t1,#32
extr $s2,$t4,$s2,#32
extr $t4,$s1,$t4,#32
lsr $s1,$s1,#32
# 5. 64-bit addition
adds $s5,$s5,$s0
adcs $t1,$t1,xzr
adcs $s4,$s4,$s2
adcs $s7,$s7,$t4
adcs $t0,$t0,$s1
# V[0] s5
# V[1] t1
# V[2] s4
# V[3] s7
# carry t0
# sub t3
# 5. Process s0-s3
ldp $s0,$s1,[sp,#32]
ldp $s2,$s3,[sp,#48]
# add with V0-V3
adds $s0,$s0,$s5
adcs $s1,$s1,$t1
adcs $s2,$s2,$s4
adcs $s3,$s3,$s7
adcs $t0,$t0,xzr
# sub with t3
subs $s1,$s1,$t3
sbcs $s2,$s2,xzr
sbcs $s3,$s3,xzr
sbcs $t0,$t0,xzr
# 6. MOD
# First Mod
lsl $t1,$t0,#32
subs $t2,$t1,$t0
adds $s0,$s0,$t0
adcs $s1,$s1,$t2
adcs $s2,$s2,xzr
adcs $s3,$s3,$t1
# Last Mod
# return y - p if y > p else y
mov $s4,$s0
mov $s5,$s1
mov $s6,$s2
mov $s7,$s3
adr $t0,.Lpoly
ldp $t1,$t2,[$t0]
ldp $t3,$t4,[$t0,#16]
adcs $t5,xzr,xzr
subs $s0,$s0,$t1
sbcs $s1,$s1,$t2
sbcs $s2,$s2,$t3
sbcs $s3,$s3,$t4
sbcs $t5,$t5,xzr
csel $s0,$s0,$s4,cs
csel $s1,$s1,$s5,cs
csel $s2,$s2,$s6,cs
csel $s3,$s3,$s7,cs
.endm
// void ecp_sm2p256_mul(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
.globl ecp_sm2p256_mul
.type ecp_sm2p256_mul,%function
.align 5
ecp_sm2p256_mul:
AARCH64_SIGN_LINK_REGISTER
# Store scalar registers
stp x29,x30,[sp,#-80]!
add x29,sp,#0
stp x16,x17,[sp,#16]
stp x18,x19,[sp,#64]
# Load inputs
ldp $s0,$s1,[x1]
ldp $s2,$s3,[x1,#16]
ldp $s4,$s5,[x2]
ldp $s6,$s7,[x2,#16]
### multiplication ###
# ========================
# s3 s2 s1 s0
# * s7 s6 s5 s4
# ------------------------
# + s0 s0 s0 s0
# * * * *
# s7 s6 s5 s4
# s1 s1 s1 s1
# * * * *
# s7 s6 s5 s4
# s2 s2 s2 s2
# * * * *
# s7 s6 s5 s4
# s3 s3 s3 s3
# * * * *
# s7 s6 s5 s4
# ------------------------
# s7 s6 s5 s4 s3 s2 s1 s0
# ========================
### s0*s4 ###
mul $t5,$s0,$s4
umulh $t2,$s0,$s4
### s1*s4 + s0*s5 ###
mul $t0,$s1,$s4
umulh $t1,$s1,$s4
adds $t2,$t2,$t0
adcs $t3,$t1,xzr
mul $t0,$s0,$s5
umulh $t1,$s0,$s5
adds $t2,$t2,$t0
adcs $t3,$t3,$t1
adcs $t4,xzr,xzr
### s2*s4 + s1*s5 + s0*s6 ###
mul $t0,$s2,$s4
umulh $t1,$s2,$s4
adds $t3,$t3,$t0
adcs $t4,$t4,$t1
mul $t0,$s1,$s5
umulh $t1,$s1,$s5
adds $t3,$t3,$t0
adcs $t4,$t4,$t1
adcs $t6,xzr,xzr
mul $t0,$s0,$s6
umulh $t1,$s0,$s6
adds $t3,$t3,$t0
adcs $t4,$t4,$t1
adcs $t6,$t6,xzr
### s3*s4 + s2*s5 + s1*s6 + s0*s7 ###
mul $t0,$s3,$s4
umulh $t1,$s3,$s4
adds $t4,$t4,$t0
adcs $t6,$t6,$t1
adcs $t7,xzr,xzr
mul $t0,$s2,$s5
umulh $t1,$s2,$s5
adds $t4,$t4,$t0
adcs $t6,$t6,$t1
adcs $t7,$t7,xzr
mul $t0,$s1,$s6
umulh $t1,$s1,$s6
adds $t4,$t4,$t0
adcs $t6,$t6,$t1
adcs $t7,$t7,xzr
mul $t0,$s0,$s7
umulh $t1,$s0,$s7
adds $t4,$t4,$t0
adcs $t6,$t6,$t1
adcs $t7,$t7,xzr
### s3*s5 + s2*s6 + s1*s7 ###
mul $t0,$s3,$s5
umulh $t1,$s3,$s5
adds $t6,$t6,$t0
adcs $t7,$t7,$t1
adcs $t8,xzr,xzr
mul $t0,$s2,$s6
umulh $t1,$s2,$s6
adds $t6,$t6,$t0
adcs $t7,$t7,$t1
adcs $t8,$t8,xzr
mul $t0,$s1,$s7
umulh $t1,$s1,$s7
adds $s4,$t6,$t0
adcs $t7,$t7,$t1
adcs $t8,$t8,xzr
### s3*s6 + s2*s7 ###
mul $t0,$s3,$s6
umulh $t1,$s3,$s6
adds $t7,$t7,$t0
adcs $t8,$t8,$t1
adcs $t6,xzr,xzr
mul $t0,$s2,$s7
umulh $t1,$s2,$s7
adds $s5,$t7,$t0
adcs $t8,$t8,$t1
adcs $t6,$t6,xzr
### s3*s7 ###
mul $t0,$s3,$s7
umulh $t1,$s3,$s7
adds $s6,$t8,$t0
adcs $s7,$t6,$t1
mov $s0,$t5
mov $s1,$t2
mov $s2,$t3
mov $s3,$t4
# result of mul: s7 s6 s5 s4 s3 s2 s1 s0
### Reduction ###
RDC
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
# Restore scalar registers
ldp x16,x17,[sp,#16]
ldp x18,x19,[sp,#64]
ldp x29,x30,[sp],#80
AARCH64_VALIDATE_LINK_REGISTER
ret
.size ecp_sm2p256_mul,.-ecp_sm2p256_mul
// void ecp_sm2p256_sqr(BN_ULONG *r, const BN_ULONG *a);
.globl ecp_sm2p256_sqr
.type ecp_sm2p256_sqr,%function
.align 5
ecp_sm2p256_sqr:
AARCH64_SIGN_LINK_REGISTER
# Store scalar registers
stp x29,x30,[sp,#-80]!
add x29,sp,#0
stp x16,x17,[sp,#16]
stp x18,x19,[sp,#64]
# Load inputs
ldp $s4,$s5,[x1]
ldp $s6,$s7,[x1,#16]
### square ###
# ========================
# s7 s6 s5 s4
# * s7 s6 s5 s4
# ------------------------
# + s4 s4 s4 s4
# * * * *
# s7 s6 s5 s4
# s5 s5 s5 s5
# * * * *
# s7 s6 s5 s4
# s6 s6 s6 s6
# * * * *
# s7 s6 s5 s4
# s7 s7 s7 s7
# * * * *
# s7 s6 s5 s4
# ------------------------
# s7 s6 s5 s4 s3 s2 s1 s0
# ========================
### s4*s5 ###
mul $s1,$s4,$s5
umulh $s2,$s4,$s5
### s4*s6 ###
mul $t0,$s6,$s4
umulh $s3,$s6,$s4
adds $s2,$s2,$t0
adcs $s3,$s3,xzr
### s4*s7 + s5*s6 ###
mul $t0,$s7,$s4
umulh $t1,$s7,$s4
adds $s3,$s3,$t0
adcs $s0,$t1,xzr
mul $t0,$s6,$s5
umulh $t1,$s6,$s5
adds $s3,$s3,$t0
adcs $s0,$s0,$t1
adcs $t2,xzr,xzr
### s5*s7 ###
mul $t0,$s7,$s5
umulh $t1,$s7,$s5
adds $s0,$s0,$t0
adcs $t2,$t2,$t1
### s6*s7 ###
mul $t0,$s7,$s6
umulh $t1,$s7,$s6
adds $t2,$t2,$t0
adcs $t3,$t1,xzr
### 2*(t3,t2,s0,s3,s2,s1) ###
adds $s1,$s1,$s1
adcs $s2,$s2,$s2
adcs $s3,$s3,$s3
adcs $s0,$s0,$s0
adcs $t2,$t2,$t2
adcs $t3,$t3,$t3
adcs $t4,xzr,xzr
### s4*s4 ###
mul $t5,$s4,$s4
umulh $t6,$s4,$s4
### s5*s5 ###
mul $s4,$s5,$s5
umulh $s5,$s5,$s5
### s6*s6 ###
mul $t0,$s6,$s6
umulh $t1,$s6,$s6
### s7*s7 ###
mul $t7,$s7,$s7
umulh $t8,$s7,$s7
adds $s1,$s1,$t6
adcs $s2,$s2,$s4
adcs $s3,$s3,$s5
adcs $s0,$s0,$t0
adcs $t2,$t2,$t1
adcs $t3,$t3,$t7
adcs $t4,$t4,$t8
mov $s4,$s0
mov $s0,$t5
mov $s5,$t2
mov $s6,$t3
mov $s7,$t4
# result of mul: s7 s6 s5 s4 s3 s2 s1 s0
### Reduction ###
RDC
stp $s0,$s1,[x0]
stp $s2,$s3,[x0,#16]
# Restore scalar registers
ldp x16,x17,[sp,#16]
ldp x18,x19,[sp,#64]
ldp x29,x30,[sp],#80
AARCH64_VALIDATE_LINK_REGISTER
ret
.size ecp_sm2p256_sqr,.-ecp_sm2p256_sqr
___
}
foreach (split("\n",$code)) {
s/\`([^\`]*)\`/eval $1/ge;
print $_,"\n";
}
close STDOUT or die "error closing STDOUT: $!"; # enforce flush

View File

@ -31,6 +31,13 @@ IF[{- !$disabled{asm} -}]
$ECDEF_armv4=ECP_NISTZ256_ASM
$ECASM_aarch64=ecp_nistz256.c ecp_nistz256-armv8.S
$ECDEF_aarch64=ECP_NISTZ256_ASM
IF[{- !$disabled{'sm2'} -}]
$ECASM_aarch64=$ECASM_aarch64 ecp_sm2p256.c ecp_sm2p256-armv8.S
IF[{- !$disabled{'sm2-precomp'} -}]
$ECASM_aarch64=$ECASM_aarch64 ecp_sm2p256_table.c
ENDIF
$ECDEF_aarch64=$ECDEF_aarch64 ECP_SM2P256_ASM
ENDIF
$ECASM_parisc11=
$ECASM_parisc20_64=
@ -127,3 +134,8 @@ IF[{- !$disabled{'ecx'} -}]
GENERATE[x25519-x86_64.s]=asm/x25519-x86_64.pl
GENERATE[x25519-ppc64.s]=asm/x25519-ppc64.pl
ENDIF
IF[{- !$disabled{'sm2'} -}]
GENERATE[ecp_sm2p256-armv8.S]=asm/ecp_sm2p256-armv8.pl
INCLUDE[ecp_sm2p256-armv8.o]=..
ENDIF

View File

@ -1,5 +1,5 @@
/*
* Copyright 2002-2021 The OpenSSL Project Authors. All Rights Reserved.
* Copyright 2002-2023 The OpenSSL Project Authors. All Rights Reserved.
* Copyright (c) 2002, Oracle and/or its affiliates. All rights reserved
*
* Licensed under the Apache License 2.0 (the "License"). You may not use
@ -3111,8 +3111,13 @@ static const ec_list_element curve_list[] = {
"RFC 5639 curve over a 512 bit prime field"},
{NID_brainpoolP512t1, &_EC_brainpoolP512t1.h, 0,
"RFC 5639 curve over a 512 bit prime field"},
# ifndef OPENSSL_NO_SM2
{NID_sm2, &_EC_sm2p256v1.h, 0,
#ifndef OPENSSL_NO_SM2
{NID_sm2, &_EC_sm2p256v1.h,
# ifdef ECP_SM2P256_ASM
EC_GFp_sm2p256_method,
# else
0,
# endif
"SM2 curve over a 256 bit prime field"},
# endif
};

View File

@ -653,6 +653,11 @@ int ossl_ec_key_simple_generate_key(EC_KEY *eckey);
int ossl_ec_key_simple_generate_public_key(EC_KEY *eckey);
int ossl_ec_key_simple_check_key(const EC_KEY *eckey);
#ifdef ECP_SM2P256_ASM
/* Returns optimized methods for SM2 */
const EC_METHOD *EC_GFp_sm2p256_method(void);
#endif
int ossl_ec_curve_nid_from_params(const EC_GROUP *group, BN_CTX *ctx);
/* EC_METHOD definitions */

800
crypto/ec/ecp_sm2p256.c Normal file
View File

@ -0,0 +1,800 @@
/*
* Copyright 2023 The OpenSSL Project Authors. All Rights Reserved.
*
* Licensed under the Apache License 2.0 (the "License"). You may not use
* this file except in compliance with the License. You can obtain a copy
* in the file LICENSE in the source distribution or at
* https://www.openssl.org/source/license.html
*
*/
/*
* SM2 low level APIs are deprecated for public use, but still ok for
* internal use.
*/
#include "internal/deprecated.h"
#include <string.h>
#include <openssl/err.h>
#include "crypto/bn.h"
#include "ec_local.h"
#include "internal/constant_time.h"
#if defined(__GNUC__)
# define ALIGN32 __attribute((aligned(32)))
# define ALIGN64 __attribute((aligned(64)))
#elif defined(_MSC_VER)
# define ALIGN32 __declspec(align(32))
# define ALIGN64 __declspec(align(64))
#else
# define ALIGN32
# define ALIGN64
#endif
#define P256_LIMBS (256 / BN_BITS2)
#if !defined(OPENSSL_NO_SM2_PRECOMP)
extern const BN_ULONG ecp_sm2p256_precomputed[8 * 32 * 256];
#endif
typedef struct {
BN_ULONG X[P256_LIMBS];
BN_ULONG Y[P256_LIMBS];
BN_ULONG Z[P256_LIMBS];
} P256_POINT;
typedef struct {
BN_ULONG X[P256_LIMBS];
BN_ULONG Y[P256_LIMBS];
} P256_POINT_AFFINE;
#if !defined(OPENSSL_NO_SM2_PRECOMP)
/* Coordinates of G, for which we have precomputed tables */
static const BN_ULONG def_xG[P256_LIMBS] ALIGN32 = {
0x715a4589334c74c7, 0x8fe30bbff2660be1,
0x5f9904466a39c994, 0x32c4ae2c1f198119
};
static const BN_ULONG def_yG[P256_LIMBS] ALIGN32 = {
0x02df32e52139f0a0, 0xd0a9877cc62a4740,
0x59bdcee36b692153, 0xbc3736a2f4f6779c,
};
#endif
/* p and order for SM2 according to GB/T 32918.5-2017 */
static const BN_ULONG def_p[P256_LIMBS] ALIGN32 = {
0xffffffffffffffff, 0xffffffff00000000,
0xffffffffffffffff, 0xfffffffeffffffff
};
static const BN_ULONG def_ord[P256_LIMBS] ALIGN32 = {
0x53bbf40939d54123, 0x7203df6b21c6052b,
0xffffffffffffffff, 0xfffffffeffffffff
};
static const BN_ULONG ONE[P256_LIMBS] ALIGN32 = {1, 0, 0, 0};
/* Functions implemented in assembly */
/*
* Most of below mentioned functions *preserve* the property of inputs
* being fully reduced, i.e. being in [0, modulus) range. Simply put if
* inputs are fully reduced, then output is too.
*/
/* Right shift: a >> 1 */
void bn_rshift1(BN_ULONG *a);
/* Sub: r = a - b */
void bn_sub(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
/* Modular div by 2: r = a / 2 mod p */
void ecp_sm2p256_div_by_2(BN_ULONG *r, const BN_ULONG *a);
/* Modular div by 2: r = a / 2 mod n, where n = ord(p) */
void ecp_sm2p256_div_by_2_mod_ord(BN_ULONG *r, const BN_ULONG *a);
/* Modular add: r = a + b mod p */
void ecp_sm2p256_add(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
/* Modular sub: r = a - b mod p */
void ecp_sm2p256_sub(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
/* Modular sub: r = a - b mod n, where n = ord(p) */
void ecp_sm2p256_sub_mod_ord(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
/* Modular mul by 3: out = 3 * a mod p */
void ecp_sm2p256_mul_by_3(BN_ULONG *r, const BN_ULONG *a);
/* Modular mul: r = a * b mod p */
void ecp_sm2p256_mul(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
/* Modular sqr: r = a ^ 2 mod p */
void ecp_sm2p256_sqr(BN_ULONG *r, const BN_ULONG *a);
static ossl_inline BN_ULONG is_zeros(const BN_ULONG *a)
{
BN_ULONG res;
res = a[0] | a[1] | a[2] | a[3];
return constant_time_is_zero_64(res);
}
static ossl_inline int is_equal(const BN_ULONG *a, const BN_ULONG *b)
{
BN_ULONG res;
res = a[0] ^ b[0];
res |= a[1] ^ b[1];
res |= a[2] ^ b[2];
res |= a[3] ^ b[3];
return constant_time_is_zero_64(res);
}
static ossl_inline int is_greater(const BN_ULONG *a, const BN_ULONG *b)
{
int i;
for (i = P256_LIMBS - 1; i >= 0; --i) {
if (a[i] > b[i])
return 1;
if (a[i] < b[i])
return -1;
}
return 0;
}
#define is_one(a) is_equal(a, ONE)
#define is_even(a) !(a[0] & 1)
#define is_point_equal(a, b) \
is_equal(a->X, b->X) && \
is_equal(a->Y, b->Y) && \
is_equal(a->Z, b->Z)
/* Bignum and field elements conversion */
#define ecp_sm2p256_bignum_field_elem(out, in) \
bn_copy_words(out, in, P256_LIMBS)
/* Binary algorithm for inversion in Fp */
#define BN_MOD_INV(out, in, mod_div, mod_sub, mod) \
do { \
BN_ULONG u[4] ALIGN32; \
BN_ULONG v[4] ALIGN32; \
BN_ULONG x1[4] ALIGN32 = {1, 0, 0, 0}; \
BN_ULONG x2[4] ALIGN32 = {0}; \
\
if (is_zeros(in)) \
return; \
memcpy(u, in, 32); \
memcpy(v, mod, 32); \
while (!is_one(u) && !is_one(v)) { \
while (is_even(u)) { \
bn_rshift1(u); \
mod_div(x1, x1); \
} \
while (is_even(v)) { \
bn_rshift1(v); \
mod_div(x2, x2); \
} \
if (is_greater(u, v) == 1) { \
bn_sub(u, u, v); \
mod_sub(x1, x1, x2); \
} else { \
bn_sub(v, v, u); \
mod_sub(x2, x2, x1); \
} \
} \
if (is_one(u)) \
memcpy(out, x1, 32); \
else \
memcpy(out, x2, 32); \
} while (0)
/* Modular inverse |out| = |in|^(-1) mod |p|. */
static ossl_inline void ecp_sm2p256_mod_inverse(BN_ULONG* out,
const BN_ULONG* in) {
BN_MOD_INV(out, in, ecp_sm2p256_div_by_2, ecp_sm2p256_sub, def_p);
}
/* Modular inverse mod order |out| = |in|^(-1) % |ord|. */
static ossl_inline void ecp_sm2p256_mod_ord_inverse(BN_ULONG* out,
const BN_ULONG* in) {
BN_MOD_INV(out, in, ecp_sm2p256_div_by_2_mod_ord, ecp_sm2p256_sub_mod_ord,
def_ord);
}
/* Point double: R <- P + P */
static void ecp_sm2p256_point_double(P256_POINT *R, const P256_POINT *P)
{
unsigned int i;
BN_ULONG tmp0[P256_LIMBS] ALIGN32;
BN_ULONG tmp1[P256_LIMBS] ALIGN32;
BN_ULONG tmp2[P256_LIMBS] ALIGN32;
/* zero-check P->Z */
if (is_zeros(P->Z)) {
for (i = 0; i < P256_LIMBS; ++i)
R->Z[i] = 0;
return;
}
ecp_sm2p256_sqr(tmp0, P->Z);
ecp_sm2p256_sub(tmp1, P->X, tmp0);
ecp_sm2p256_add(tmp0, P->X, tmp0);
ecp_sm2p256_mul(tmp1, tmp1, tmp0);
ecp_sm2p256_mul_by_3(tmp1, tmp1);
ecp_sm2p256_add(R->Y, P->Y, P->Y);
ecp_sm2p256_mul(R->Z, R->Y, P->Z);
ecp_sm2p256_sqr(R->Y, R->Y);
ecp_sm2p256_mul(tmp2, R->Y, P->X);
ecp_sm2p256_sqr(R->Y, R->Y);
ecp_sm2p256_div_by_2(R->Y, R->Y);
ecp_sm2p256_sqr(R->X, tmp1);
ecp_sm2p256_add(tmp0, tmp2, tmp2);
ecp_sm2p256_sub(R->X, R->X, tmp0);
ecp_sm2p256_sub(tmp0, tmp2, R->X);
ecp_sm2p256_mul(tmp0, tmp0, tmp1);
ecp_sm2p256_sub(tmp1, tmp0, R->Y);
memcpy(R->Y, tmp1, 32);
}
/* Point add affine: R <- P + Q */
static void ecp_sm2p256_point_add_affine(P256_POINT *R, const P256_POINT *P,
const P256_POINT_AFFINE *Q)
{
unsigned int i;
BN_ULONG tmp0[P256_LIMBS] ALIGN32 = {0};
BN_ULONG tmp1[P256_LIMBS] ALIGN32 = {0};
BN_ULONG tmp2[P256_LIMBS] ALIGN32 = {0};
BN_ULONG tmp3[P256_LIMBS] ALIGN32 = {0};
/* zero-check P->Z */
if (is_zeros(P->Z)) {
for (i = 0; i < P256_LIMBS; ++i) {
R->X[i] = Q->X[i];
R->Y[i] = Q->Y[i];
R->Z[i] = 0;
}
R->Z[0] = 1;
return;
}
ecp_sm2p256_sqr(tmp0, P->Z);
ecp_sm2p256_mul(tmp1, tmp0, P->Z);
ecp_sm2p256_mul(tmp0, tmp0, Q->X);
ecp_sm2p256_mul(tmp1, tmp1, Q->Y);
ecp_sm2p256_sub(tmp0, tmp0, P->X);
ecp_sm2p256_sub(tmp1, tmp1, P->Y);
/* zero-check tmp0, tmp1 */
if (is_zeros(tmp0)) {
if (is_zeros(tmp1)) {
P256_POINT K;
for (i = 0; i < P256_LIMBS; ++i) {
K.X[i] = Q->X[i];
K.Y[i] = Q->Y[i];
K.Z[i] = 0;
}
K.Z[0] = 1;
ecp_sm2p256_point_double(R, &K);
} else {
for (i = 0; i < P256_LIMBS; ++i)
R->Z[i] = 0;
}
return;
}
ecp_sm2p256_mul(R->Z, P->Z, tmp0);
ecp_sm2p256_sqr(tmp2, tmp0);
ecp_sm2p256_mul(tmp3, tmp2, tmp0);
ecp_sm2p256_mul(tmp2, tmp2, P->X);
ecp_sm2p256_add(tmp0, tmp2, tmp2);
ecp_sm2p256_sqr(R->X, tmp1);
ecp_sm2p256_sub(R->X, R->X, tmp0);
ecp_sm2p256_sub(R->X, R->X, tmp3);
ecp_sm2p256_sub(tmp2, tmp2, R->X);
ecp_sm2p256_mul(tmp2, tmp2, tmp1);
ecp_sm2p256_mul(tmp3, tmp3, P->Y);
ecp_sm2p256_sub(R->Y, tmp2, tmp3);
}
/* Point add: R <- P + Q */
static void ecp_sm2p256_point_add(P256_POINT *R, const P256_POINT *P,
const P256_POINT *Q)
{
unsigned int i;
BN_ULONG tmp0[P256_LIMBS] ALIGN32 = {0};
BN_ULONG tmp1[P256_LIMBS] ALIGN32 = {0};
BN_ULONG tmp2[P256_LIMBS] ALIGN32 = {0};
/* zero-check P | Q ->Z */
if (is_zeros(P->Z)) {
for (i = 0; i < P256_LIMBS; ++i) {
R->X[i] = Q->X[i];
R->Y[i] = Q->Y[i];
R->Z[i] = Q->Z[i];
}
return;
} else if (is_zeros(Q->Z)) {
for (i = 0; i < P256_LIMBS; ++i) {
R->X[i] = P->X[i];
R->Y[i] = P->Y[i];
R->Z[i] = P->Z[i];
}
return;
} else if (is_point_equal(P, Q)) {
ecp_sm2p256_point_double(R, Q);
return;
}
ecp_sm2p256_sqr(tmp0, P->Z);
ecp_sm2p256_mul(tmp1, tmp0, P->Z);
ecp_sm2p256_mul(tmp0, tmp0, Q->X);
ecp_sm2p256_mul(tmp1, tmp1, Q->Y);
ecp_sm2p256_mul(R->Y, P->Y, Q->Z);
ecp_sm2p256_mul(R->Z, Q->Z, P->Z);
ecp_sm2p256_sqr(tmp2, Q->Z);
ecp_sm2p256_mul(R->Y, tmp2, R->Y);
ecp_sm2p256_mul(R->X, tmp2, P->X);
ecp_sm2p256_sub(tmp0, tmp0, R->X);
ecp_sm2p256_mul(R->Z, tmp0, R->Z);
ecp_sm2p256_sub(tmp1, tmp1, R->Y);
ecp_sm2p256_sqr(tmp2, tmp0);
ecp_sm2p256_mul(tmp0, tmp0, tmp2);
ecp_sm2p256_mul(tmp2, tmp2, R->X);
ecp_sm2p256_sqr(R->X, tmp1);
ecp_sm2p256_sub(R->X, R->X, tmp2);
ecp_sm2p256_sub(R->X, R->X, tmp2);
ecp_sm2p256_sub(R->X, R->X, tmp0);
ecp_sm2p256_sub(tmp2, tmp2, R->X);
ecp_sm2p256_mul(tmp2, tmp1, tmp2);
ecp_sm2p256_mul(tmp0, tmp0, R->Y);
ecp_sm2p256_sub(R->Y, tmp2, tmp0);
}
#if !defined(OPENSSL_NO_SM2_PRECOMP)
/* Base point mul by scalar: k - scalar, G - base point */
static void ecp_sm2p256_point_G_mul_by_scalar(P256_POINT *R, const BN_ULONG *k)
{
unsigned int i, index, mask = 0xff;
P256_POINT_AFFINE Q;
memset(R, 0, sizeof(P256_POINT));
if (is_zeros(k))
return;
index = k[0] & mask;
if (index) {
index = index * 8;
memcpy(R->X, ecp_sm2p256_precomputed + index, 32);
memcpy(R->Y, ecp_sm2p256_precomputed + index + P256_LIMBS, 32);
R->Z[0] = 1;
}
for (i = 1; i < 32; ++i) {
index = (k[i / 8] >> (8 * (i % 8))) & mask;
if (index) {
index = index + i * 256;
index = index * 8;
memcpy(Q.X, ecp_sm2p256_precomputed + index, 32);
memcpy(Q.Y, ecp_sm2p256_precomputed + index + P256_LIMBS, 32);
ecp_sm2p256_point_add_affine(R, R, &Q);
}
}
}
#endif
/*
* Affine point mul by scalar: k - scalar, P - affine point
*/
static void ecp_sm2p256_point_P_mul_by_scalar(P256_POINT *R, const BN_ULONG *k,
P256_POINT_AFFINE P)
{
int i, init = 0;
unsigned int index, mask = 0x0f;
P256_POINT precomputed[16] ALIGN64;
memset(R, 0, sizeof(P256_POINT));
if (is_zeros(k))
return;
/* The first value of the precomputed table is P. */
memcpy(precomputed[1].X, P.X, 32);
memcpy(precomputed[1].Y, P.Y, 32);
precomputed[1].Z[0] = 1;
precomputed[1].Z[1] = 0;
precomputed[1].Z[2] = 0;
precomputed[1].Z[3] = 0;
/* The second value of the precomputed table is 2P. */
ecp_sm2p256_point_double(&precomputed[2], &precomputed[1]);
/* The subsequent elements are 3P, 4P, and so on. */
for (i = 3; i < 16; ++i)
ecp_sm2p256_point_add_affine(&precomputed[i], &precomputed[i - 1], &P);
for (i = 64 - 1; i >= 0; --i) {
index = (k[i / 16] >> (4 * (i % 16))) & mask;
if (init == 0) {
if (index) {
memcpy(R, &precomputed[index], sizeof(P256_POINT));
init = 1;
}
} else {
ecp_sm2p256_point_double(R, R);
ecp_sm2p256_point_double(R, R);
ecp_sm2p256_point_double(R, R);
ecp_sm2p256_point_double(R, R);
if (index)
ecp_sm2p256_point_add(R, R, &precomputed[index]);
}
}
}
/* Get affine point */
static void ecp_sm2p256_point_get_affine(P256_POINT_AFFINE *R,
const P256_POINT *P)
{
BN_ULONG z_inv3[P256_LIMBS] ALIGN32 = {0};
BN_ULONG z_inv2[P256_LIMBS] ALIGN32 = {0};
if (is_one(P->Z)) {
memcpy(R->X, P->X, 32);
memcpy(R->Y, P->Y, 32);
return;
}
ecp_sm2p256_mod_inverse(z_inv3, P->Z);
ecp_sm2p256_sqr(z_inv2, z_inv3);
ecp_sm2p256_mul(R->X, P->X, z_inv2);
ecp_sm2p256_mul(z_inv3, z_inv3, z_inv2);
ecp_sm2p256_mul(R->Y, P->Y, z_inv3);
}
#if !defined(OPENSSL_NO_SM2_PRECOMP)
static int ecp_sm2p256_is_affine_G(const EC_POINT *generator)
{
return (bn_get_top(generator->X) == P256_LIMBS)
&& (bn_get_top(generator->Y) == P256_LIMBS)
&& is_equal(bn_get_words(generator->X), def_xG)
&& is_equal(bn_get_words(generator->Y), def_yG)
&& (generator->Z_is_one == 1);
}
#endif
/*
* Convert Jacobian coordinate point into affine coordinate (x,y)
*/
static int ecp_sm2p256_get_affine(const EC_GROUP *group,
const EC_POINT *point,
BIGNUM *x, BIGNUM *y, BN_CTX *ctx)
{
BN_ULONG z_inv2[P256_LIMBS] ALIGN32 = {0};
BN_ULONG z_inv3[P256_LIMBS] ALIGN32 = {0};
BN_ULONG x_aff[P256_LIMBS] ALIGN32 = {0};
BN_ULONG y_aff[P256_LIMBS] ALIGN32 = {0};
BN_ULONG point_x[P256_LIMBS] ALIGN32 = {0};
BN_ULONG point_y[P256_LIMBS] ALIGN32 = {0};
BN_ULONG point_z[P256_LIMBS] ALIGN32 = {0};
if (EC_POINT_is_at_infinity(group, point)) {
ECerr(ERR_LIB_EC, EC_R_POINT_AT_INFINITY);
return 0;
}
if (ecp_sm2p256_bignum_field_elem(point_x, point->X) <= 0
|| ecp_sm2p256_bignum_field_elem(point_y, point->Y) <= 0
|| ecp_sm2p256_bignum_field_elem(point_z, point->Z) <= 0) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
return 0;
}
ecp_sm2p256_mod_inverse(z_inv3, point_z);
ecp_sm2p256_sqr(z_inv2, z_inv3);
if (x != NULL) {
ecp_sm2p256_mul(x_aff, point_x, z_inv2);
if (!bn_set_words(x, x_aff, P256_LIMBS))
return 0;
}
if (y != NULL) {
ecp_sm2p256_mul(z_inv3, z_inv3, z_inv2);
ecp_sm2p256_mul(y_aff, point_y, z_inv3);
if (!bn_set_words(y, y_aff, P256_LIMBS))
return 0;
}
return 1;
}
/* r = sum(scalar[i]*point[i]) */
static int ecp_sm2p256_windowed_mul(const EC_GROUP *group,
P256_POINT *r,
const BIGNUM **scalar,
const EC_POINT **point,
size_t num, BN_CTX *ctx)
{
unsigned int i;
int ret = 0;
const BIGNUM **scalars = NULL;
BN_ULONG k[P256_LIMBS] ALIGN32 = {0};
P256_POINT kP;
ALIGN32 union {
P256_POINT p;
P256_POINT_AFFINE a;
} t, p;
if (num > OPENSSL_MALLOC_MAX_NELEMS(P256_POINT)
|| (scalars = OPENSSL_malloc(num * sizeof(BIGNUM *))) == NULL) {
ECerr(ERR_LIB_EC, ERR_R_MALLOC_FAILURE);
goto err;
}
memset(r, 0, sizeof(P256_POINT));
for (i = 0; i < num; i++) {
if (EC_POINT_is_at_infinity(group, point[i]))
continue;
if ((BN_num_bits(scalar[i]) > 256) || BN_is_negative(scalar[i])) {
BIGNUM *tmp;
if ((tmp = BN_CTX_get(ctx)) == NULL)
goto err;
if (!BN_nnmod(tmp, scalar[i], group->order, ctx)) {
ECerr(ERR_LIB_EC, ERR_R_BN_LIB);
goto err;
}
scalars[i] = tmp;
} else {
scalars[i] = scalar[i];
}
if (ecp_sm2p256_bignum_field_elem(k, scalars[i]) <= 0
|| ecp_sm2p256_bignum_field_elem(p.p.X, point[i]->X) <= 0
|| ecp_sm2p256_bignum_field_elem(p.p.Y, point[i]->Y) <= 0
|| ecp_sm2p256_bignum_field_elem(p.p.Z, point[i]->Z) <= 0) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
goto err;
}
ecp_sm2p256_point_get_affine(&t.a, &p.p);
ecp_sm2p256_point_P_mul_by_scalar(&kP, k, t.a);
ecp_sm2p256_point_add(r, r, &kP);
}
ret = 1;
err:
OPENSSL_free(scalars);
return ret;
}
/* r = scalar*G + sum(scalars[i]*points[i]) */
static int ecp_sm2p256_points_mul(const EC_GROUP *group,
EC_POINT *r,
const BIGNUM *scalar,
size_t num,
const EC_POINT *points[],
const BIGNUM *scalars[], BN_CTX *ctx)
{
int ret = 0, p_is_infinity = 0;
const EC_POINT *generator = NULL;
BN_ULONG k[P256_LIMBS] ALIGN32 = {0};
ALIGN32 union {
P256_POINT p;
P256_POINT_AFFINE a;
} t, p;
if ((num + 1) == 0 || (num + 1) > OPENSSL_MALLOC_MAX_NELEMS(void *)) {
ECerr(ERR_LIB_EC, ERR_R_MALLOC_FAILURE);
goto err;
}
BN_CTX_start(ctx);
if (scalar) {
generator = EC_GROUP_get0_generator(group);
if (generator == NULL) {
ECerr(ERR_LIB_EC, EC_R_UNDEFINED_GENERATOR);
goto err;
}
if (!ecp_sm2p256_bignum_field_elem(k, scalar)) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
goto err;
}
#if !defined(OPENSSL_NO_SM2_PRECOMP)
if (ecp_sm2p256_is_affine_G(generator)) {
ecp_sm2p256_point_G_mul_by_scalar(&p.p, k);
} else
#endif
{
/* if no precomputed table */
const EC_POINT *new_generator[1];
const BIGNUM *g_scalars[1];
new_generator[0] = generator;
g_scalars[0] = scalar;
if (!ecp_sm2p256_windowed_mul(group, &p.p, g_scalars, new_generator,
(new_generator[0] != NULL
&& g_scalars[0] != NULL), ctx))
goto err;
}
} else {
p_is_infinity = 1;
}
if (num) {
P256_POINT *out = &t.p;
if (p_is_infinity)
out = &p.p;
if (!ecp_sm2p256_windowed_mul(group, out, scalars, points, num, ctx))
goto err;
if (!p_is_infinity)
ecp_sm2p256_point_add(&p.p, &p.p, out);
}
/* Not constant-time, but we're only operating on the public output. */
if (!bn_set_words(r->X, p.p.X, P256_LIMBS)
|| !bn_set_words(r->Y, p.p.Y, P256_LIMBS)
|| !bn_set_words(r->Z, p.p.Z, P256_LIMBS))
goto err;
r->Z_is_one = is_equal(bn_get_words(r->Z), ONE) & 1;
ret = 1;
err:
BN_CTX_end(ctx);
return ret;
}
static int ecp_sm2p256_field_mul(const EC_GROUP *group, BIGNUM *r,
const BIGNUM *a, const BIGNUM *b, BN_CTX *ctx)
{
BN_ULONG a_fe[P256_LIMBS] ALIGN32 = {0};
BN_ULONG b_fe[P256_LIMBS] ALIGN32 = {0};
BN_ULONG r_fe[P256_LIMBS] ALIGN32 = {0};
if (a == NULL || b == NULL || r == NULL)
return 0;
if (!ecp_sm2p256_bignum_field_elem(a_fe, a)
|| !ecp_sm2p256_bignum_field_elem(b_fe, b)) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
return 0;
}
ecp_sm2p256_mul(r_fe, a_fe, b_fe);
if (!bn_set_words(r, r_fe, P256_LIMBS))
return 0;
return 1;
}
static int ecp_sm2p256_field_sqr(const EC_GROUP *group, BIGNUM *r,
const BIGNUM *a, BN_CTX *ctx)
{
BN_ULONG a_fe[P256_LIMBS] ALIGN32 = {0};
BN_ULONG r_fe[P256_LIMBS] ALIGN32 = {0};
if (a == NULL || r == NULL)
return 0;
if (!ecp_sm2p256_bignum_field_elem(a_fe, a)) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
return 0;
}
ecp_sm2p256_sqr(r_fe, a_fe);
if (!bn_set_words(r, r_fe, P256_LIMBS))
return 0;
return 1;
}
static int ecp_sm2p256_inv_mod_ord(const EC_GROUP *group, BIGNUM *r,
const BIGNUM *x, BN_CTX *ctx)
{
int ret = 0;
BN_ULONG t[P256_LIMBS] ALIGN32 = {0};
BN_ULONG out[P256_LIMBS] ALIGN32 = {0};
if (bn_wexpand(r, P256_LIMBS) == NULL) {
ECerr(ERR_LIB_EC, ERR_R_BN_LIB);
goto err;
}
if ((BN_num_bits(x) > 256) || BN_is_negative(x)) {
BIGNUM *tmp;
if ((tmp = BN_CTX_get(ctx)) == NULL
|| !BN_nnmod(tmp, x, group->order, ctx)) {
ECerr(ERR_LIB_EC, ERR_R_BN_LIB);
goto err;
}
x = tmp;
}
if (!ecp_sm2p256_bignum_field_elem(t, x)) {
ECerr(ERR_LIB_EC, EC_R_COORDINATES_OUT_OF_RANGE);
goto err;
}
ecp_sm2p256_mod_ord_inverse(out, t);
if (!bn_set_words(r, out, P256_LIMBS))
goto err;
ret = 1;
err:
return ret;
}
const EC_METHOD *EC_GFp_sm2p256_method(void)
{
static const EC_METHOD ret = {
EC_FLAGS_DEFAULT_OCT,
NID_X9_62_prime_field,
ossl_ec_GFp_simple_group_init,
ossl_ec_GFp_simple_group_finish,
ossl_ec_GFp_simple_group_clear_finish,
ossl_ec_GFp_simple_group_copy,
ossl_ec_GFp_simple_group_set_curve,
ossl_ec_GFp_simple_group_get_curve,
ossl_ec_GFp_simple_group_get_degree,
ossl_ec_group_simple_order_bits,
ossl_ec_GFp_simple_group_check_discriminant,
ossl_ec_GFp_simple_point_init,
ossl_ec_GFp_simple_point_finish,
ossl_ec_GFp_simple_point_clear_finish,
ossl_ec_GFp_simple_point_copy,
ossl_ec_GFp_simple_point_set_to_infinity,
ossl_ec_GFp_simple_point_set_affine_coordinates,
ecp_sm2p256_get_affine,
0, 0, 0,
ossl_ec_GFp_simple_add,
ossl_ec_GFp_simple_dbl,
ossl_ec_GFp_simple_invert,
ossl_ec_GFp_simple_is_at_infinity,
ossl_ec_GFp_simple_is_on_curve,
ossl_ec_GFp_simple_cmp,
ossl_ec_GFp_simple_make_affine,
ossl_ec_GFp_simple_points_make_affine,
ecp_sm2p256_points_mul, /* mul */
0 /* precompute_mult */,
0 /* have_precompute_mult */,
ecp_sm2p256_field_mul,
ecp_sm2p256_field_sqr,
0 /* field_div */,
0 /* field_inv */,
0 /* field_encode */,
0 /* field_decode */,
0 /* field_set_to_one */,
ossl_ec_key_simple_priv2oct,
ossl_ec_key_simple_oct2priv,
0, /* set private */
ossl_ec_key_simple_generate_key,
ossl_ec_key_simple_check_key,
ossl_ec_key_simple_generate_public_key,
0, /* keycopy */
0, /* keyfinish */
ossl_ecdh_simple_compute_key,
ossl_ecdsa_simple_sign_setup,
ossl_ecdsa_simple_sign_sig,
ossl_ecdsa_simple_verify_sig,
ecp_sm2p256_inv_mod_ord,
0, /* blind_coordinates */
0, /* ladder_pre */
0, /* ladder_step */
0 /* ladder_post */
};
return &ret;
}

16387
crypto/ec/ecp_sm2p256_table.c Normal file

File diff suppressed because it is too large Load Diff