mirror of
https://github.com/guanzhi/GmSSL.git
synced 2026-08-14 01:53:41 +08:00
update
This commit is contained in:
Zhi Guan
@@ -9,8 +9,7 @@
|
||||
|
||||
# I let hardware handle unaligned input(*), except on page boundaries
|
||||
# (see below for details). Otherwise straightforward implementation
|
||||
# with X vector in register bank. The module is big-endian [which is
|
||||
# not big deal as there're no little-endian targets left around].
|
||||
# with X vector in register bank.
|
||||
#
|
||||
# (*) this means that this module is inappropriate for PPC403? Does
|
||||
# anybody know if pre-POWER3 can sustain unaligned load?
|
||||
@@ -24,18 +23,24 @@ $flavour = shift;
|
||||
|
||||
if ($flavour =~ /64/) {
|
||||
$SIZE_T =8;
|
||||
$LRSAVE =2*$SIZE_T;
|
||||
$UCMP ="cmpld";
|
||||
$STU ="stdu";
|
||||
$POP ="ld";
|
||||
$PUSH ="std";
|
||||
} elsif ($flavour =~ /32/) {
|
||||
$SIZE_T =4;
|
||||
$LRSAVE =$SIZE_T;
|
||||
$UCMP ="cmplw";
|
||||
$STU ="stwu";
|
||||
$POP ="lwz";
|
||||
$PUSH ="stw";
|
||||
} else { die "nonsense $flavour"; }
|
||||
|
||||
# Define endianess based on flavour
|
||||
# i.e.: linux64le
|
||||
$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
|
||||
|
||||
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
|
||||
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
|
||||
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
|
||||
@@ -43,7 +48,8 @@ die "can't locate ppc-xlate.pl";
|
||||
|
||||
open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
|
||||
|
||||
$FRAME=24*$SIZE_T;
|
||||
$FRAME=24*$SIZE_T+64;
|
||||
$LOCALS=6*$SIZE_T;
|
||||
|
||||
$K ="r0";
|
||||
$sp ="r1";
|
||||
@@ -65,14 +71,28 @@ $T ="r12";
|
||||
@X=("r16","r17","r18","r19","r20","r21","r22","r23",
|
||||
"r24","r25","r26","r27","r28","r29","r30","r31");
|
||||
|
||||
sub loadbe {
|
||||
my ($dst, $src, $temp_reg) = @_;
|
||||
$code.=<<___ if (!$LITTLE_ENDIAN);
|
||||
lwz $dst,$src
|
||||
___
|
||||
$code.=<<___ if ($LITTLE_ENDIAN);
|
||||
lwz $temp_reg,$src
|
||||
rotlwi $dst,$temp_reg,8
|
||||
rlwimi $dst,$temp_reg,24,0,7
|
||||
rlwimi $dst,$temp_reg,24,16,23
|
||||
___
|
||||
}
|
||||
|
||||
sub BODY_00_19 {
|
||||
my ($i,$a,$b,$c,$d,$e,$f)=@_;
|
||||
my $j=$i+1;
|
||||
$code.=<<___ if ($i==0);
|
||||
lwz @X[$i],`$i*4`($inp)
|
||||
___
|
||||
|
||||
# Since the last value of $f is discarded, we can use
|
||||
# it as a temp reg to swap byte-order when needed.
|
||||
loadbe("@X[$i]","`$i*4`($inp)",$f) if ($i==0);
|
||||
loadbe("@X[$j]","`$j*4`($inp)",$f) if ($i<15);
|
||||
$code.=<<___ if ($i<15);
|
||||
lwz @X[$j],`$j*4`($inp)
|
||||
add $f,$K,$e
|
||||
rotlwi $e,$a,5
|
||||
add $f,$f,@X[$i]
|
||||
@@ -105,31 +125,31 @@ my ($i,$a,$b,$c,$d,$e,$f)=@_;
|
||||
my $j=$i+1;
|
||||
$code.=<<___ if ($i<79);
|
||||
add $f,$K,$e
|
||||
xor $t0,$b,$d
|
||||
rotlwi $e,$a,5
|
||||
xor @X[$j%16],@X[$j%16],@X[($j+2)%16]
|
||||
add $f,$f,@X[$i%16]
|
||||
xor $t0,$b,$c
|
||||
xor $t0,$t0,$c
|
||||
xor @X[$j%16],@X[$j%16],@X[($j+8)%16]
|
||||
add $f,$f,$e
|
||||
rotlwi $b,$b,30
|
||||
xor $t0,$t0,$d
|
||||
xor @X[$j%16],@X[$j%16],@X[($j+13)%16]
|
||||
add $f,$f,$t0
|
||||
rotlwi $b,$b,30
|
||||
xor @X[$j%16],@X[$j%16],@X[($j+13)%16]
|
||||
add $f,$f,$e
|
||||
rotlwi @X[$j%16],@X[$j%16],1
|
||||
___
|
||||
$code.=<<___ if ($i==79);
|
||||
add $f,$K,$e
|
||||
xor $t0,$b,$d
|
||||
rotlwi $e,$a,5
|
||||
lwz r16,0($ctx)
|
||||
add $f,$f,@X[$i%16]
|
||||
xor $t0,$b,$c
|
||||
xor $t0,$t0,$c
|
||||
lwz r17,4($ctx)
|
||||
add $f,$f,$e
|
||||
add $f,$f,$t0
|
||||
rotlwi $b,$b,30
|
||||
lwz r18,8($ctx)
|
||||
xor $t0,$t0,$d
|
||||
lwz r19,12($ctx)
|
||||
add $f,$f,$t0
|
||||
add $f,$f,$e
|
||||
lwz r20,16($ctx)
|
||||
___
|
||||
}
|
||||
@@ -162,9 +182,8 @@ $code=<<___;
|
||||
.globl .sha1_block_data_order
|
||||
.align 4
|
||||
.sha1_block_data_order:
|
||||
$STU $sp,-$FRAME($sp)
|
||||
mflr r0
|
||||
$STU $sp,`-($FRAME+64)`($sp)
|
||||
$PUSH r0,`$FRAME-$SIZE_T*18`($sp)
|
||||
$PUSH r15,`$FRAME-$SIZE_T*17`($sp)
|
||||
$PUSH r16,`$FRAME-$SIZE_T*16`($sp)
|
||||
$PUSH r17,`$FRAME-$SIZE_T*15`($sp)
|
||||
@@ -182,6 +201,7 @@ $code=<<___;
|
||||
$PUSH r29,`$FRAME-$SIZE_T*3`($sp)
|
||||
$PUSH r30,`$FRAME-$SIZE_T*2`($sp)
|
||||
$PUSH r31,`$FRAME-$SIZE_T*1`($sp)
|
||||
$PUSH r0,`$FRAME+$LRSAVE`($sp)
|
||||
lwz $A,0($ctx)
|
||||
lwz $B,4($ctx)
|
||||
lwz $C,8($ctx)
|
||||
@@ -192,8 +212,53 @@ $code=<<___;
|
||||
Laligned:
|
||||
mtctr $num
|
||||
bl Lsha1_block_private
|
||||
b Ldone
|
||||
|
||||
; PowerPC specification allows an implementation to be ill-behaved
|
||||
; upon unaligned access which crosses page boundary. "Better safe
|
||||
; than sorry" principle makes me treat it specially. But I don't
|
||||
; look for particular offending word, but rather for 64-byte input
|
||||
; block which crosses the boundary. Once found that block is aligned
|
||||
; and hashed separately...
|
||||
.align 4
|
||||
Lunaligned:
|
||||
subfic $t1,$inp,4096
|
||||
andi. $t1,$t1,4095 ; distance to closest page boundary
|
||||
srwi. $t1,$t1,6 ; t1/=64
|
||||
beq Lcross_page
|
||||
$UCMP $num,$t1
|
||||
ble- Laligned ; didn't cross the page boundary
|
||||
mtctr $t1
|
||||
subfc $num,$t1,$num
|
||||
bl Lsha1_block_private
|
||||
Lcross_page:
|
||||
li $t1,16
|
||||
mtctr $t1
|
||||
addi r20,$sp,$LOCALS ; spot within the frame
|
||||
Lmemcpy:
|
||||
lbz r16,0($inp)
|
||||
lbz r17,1($inp)
|
||||
lbz r18,2($inp)
|
||||
lbz r19,3($inp)
|
||||
addi $inp,$inp,4
|
||||
stb r16,0(r20)
|
||||
stb r17,1(r20)
|
||||
stb r18,2(r20)
|
||||
stb r19,3(r20)
|
||||
addi r20,r20,4
|
||||
bdnz Lmemcpy
|
||||
|
||||
$PUSH $inp,`$FRAME-$SIZE_T*18`($sp)
|
||||
li $t1,1
|
||||
addi $inp,$sp,$LOCALS
|
||||
mtctr $t1
|
||||
bl Lsha1_block_private
|
||||
$POP $inp,`$FRAME-$SIZE_T*18`($sp)
|
||||
addic. $num,$num,-1
|
||||
bne- Lunaligned
|
||||
|
||||
Ldone:
|
||||
$POP r0,`$FRAME-$SIZE_T*18`($sp)
|
||||
$POP r0,`$FRAME+$LRSAVE`($sp)
|
||||
$POP r15,`$FRAME-$SIZE_T*17`($sp)
|
||||
$POP r16,`$FRAME-$SIZE_T*16`($sp)
|
||||
$POP r17,`$FRAME-$SIZE_T*15`($sp)
|
||||
@@ -212,54 +277,11 @@ Ldone:
|
||||
$POP r30,`$FRAME-$SIZE_T*2`($sp)
|
||||
$POP r31,`$FRAME-$SIZE_T*1`($sp)
|
||||
mtlr r0
|
||||
addi $sp,$sp,`$FRAME+64`
|
||||
addi $sp,$sp,$FRAME
|
||||
blr
|
||||
___
|
||||
|
||||
# PowerPC specification allows an implementation to be ill-behaved
|
||||
# upon unaligned access which crosses page boundary. "Better safe
|
||||
# than sorry" principle makes me treat it specially. But I don't
|
||||
# look for particular offending word, but rather for 64-byte input
|
||||
# block which crosses the boundary. Once found that block is aligned
|
||||
# and hashed separately...
|
||||
$code.=<<___;
|
||||
.align 4
|
||||
Lunaligned:
|
||||
subfic $t1,$inp,4096
|
||||
andi. $t1,$t1,4095 ; distance to closest page boundary
|
||||
srwi. $t1,$t1,6 ; t1/=64
|
||||
beq Lcross_page
|
||||
$UCMP $num,$t1
|
||||
ble- Laligned ; didn't cross the page boundary
|
||||
mtctr $t1
|
||||
subfc $num,$t1,$num
|
||||
bl Lsha1_block_private
|
||||
Lcross_page:
|
||||
li $t1,16
|
||||
mtctr $t1
|
||||
addi r20,$sp,$FRAME ; spot below the frame
|
||||
Lmemcpy:
|
||||
lbz r16,0($inp)
|
||||
lbz r17,1($inp)
|
||||
lbz r18,2($inp)
|
||||
lbz r19,3($inp)
|
||||
addi $inp,$inp,4
|
||||
stb r16,0(r20)
|
||||
stb r17,1(r20)
|
||||
stb r18,2(r20)
|
||||
stb r19,3(r20)
|
||||
addi r20,r20,4
|
||||
bdnz Lmemcpy
|
||||
|
||||
$PUSH $inp,`$FRAME-$SIZE_T*19`($sp)
|
||||
li $t1,1
|
||||
addi $inp,$sp,$FRAME
|
||||
mtctr $t1
|
||||
bl Lsha1_block_private
|
||||
$POP $inp,`$FRAME-$SIZE_T*19`($sp)
|
||||
addic. $num,$num,-1
|
||||
bne- Lunaligned
|
||||
b Ldone
|
||||
.long 0
|
||||
.byte 0,12,4,1,0x80,18,3,0
|
||||
.long 0
|
||||
___
|
||||
|
||||
# This is private block function, which uses tailored calling
|
||||
@@ -309,6 +331,9 @@ $code.=<<___;
|
||||
addi $inp,$inp,`16*4`
|
||||
bdnz- Lsha1_block_private
|
||||
blr
|
||||
.long 0
|
||||
.byte 0,12,0x14,0,0,0,0,0
|
||||
.size .sha1_block_data_order,.-.sha1_block_data_order
|
||||
___
|
||||
$code.=<<___;
|
||||
.asciz "SHA1 block transform for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"
|
||||
|
||||
Reference in New Issue
Block a user