| 1 | #! /usr/bin/env perl
|
|---|
| 2 | # Copyright 2011-2020 The OpenSSL Project Authors. All Rights Reserved.
|
|---|
| 3 | #
|
|---|
| 4 | # Licensed under the Apache License 2.0 (the "License"). You may not use
|
|---|
| 5 | # this file except in compliance with the License. You can obtain a copy
|
|---|
| 6 | # in the file LICENSE in the source distribution or at
|
|---|
| 7 | # https://www.openssl.org/source/license.html
|
|---|
| 8 |
|
|---|
| 9 | #
|
|---|
| 10 | # ====================================================================
|
|---|
| 11 | # Written by Andy Polyakov <[email protected]> for the OpenSSL
|
|---|
| 12 | # project. The module is, however, dual licensed under OpenSSL and
|
|---|
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further
|
|---|
| 14 | # details see http://www.openssl.org/~appro/cryptogams/.
|
|---|
| 15 | # ====================================================================
|
|---|
| 16 |
|
|---|
| 17 | # June 2011
|
|---|
| 18 | #
|
|---|
| 19 | # This is RC4+MD5 "stitch" implementation. The idea, as spelled in
|
|---|
| 20 | # http://download.intel.com/design/intarch/papers/323686.pdf, is that
|
|---|
| 21 | # since both algorithms exhibit instruction-level parallelism, ILP,
|
|---|
| 22 | # below theoretical maximum, interleaving them would allow to utilize
|
|---|
| 23 | # processor resources better and achieve better performance. RC4
|
|---|
| 24 | # instruction sequence is virtually identical to rc4-x86_64.pl, which
|
|---|
| 25 | # is heavily based on submission by Maxim Perminov, Maxim Locktyukhin
|
|---|
| 26 | # and Jim Guilford of Intel. MD5 is fresh implementation aiming to
|
|---|
| 27 | # minimize register usage, which was used as "main thread" with RC4
|
|---|
| 28 | # weaved into it, one RC4 round per one MD5 round. In addition to the
|
|---|
| 29 | # stiched subroutine the script can generate standalone replacement
|
|---|
| 30 | # ossl_md5_block_asm_data_order and RC4. Below are performance numbers in
|
|---|
| 31 | # cycles per processed byte, less is better, for these the standalone
|
|---|
| 32 | # subroutines, sum of them, and stitched one:
|
|---|
| 33 | #
|
|---|
| 34 | # RC4 MD5 RC4+MD5 stitch gain
|
|---|
| 35 | # Opteron 6.5(*) 5.4 11.9 7.0 +70%(*)
|
|---|
| 36 | # Core2 6.5 5.8 12.3 7.7 +60%
|
|---|
| 37 | # Westmere 4.3 5.2 9.5 7.0 +36%
|
|---|
| 38 | # Sandy Bridge 4.2 5.5 9.7 6.8 +43%
|
|---|
| 39 | # Ivy Bridge 4.1 5.2 9.3 6.0 +54%
|
|---|
| 40 | # Haswell 4.0 5.0 9.0 5.7 +60%
|
|---|
| 41 | # Skylake 6.3(**) 5.0 11.3 5.3 +110%
|
|---|
| 42 | # Atom 9.3 6.5 15.8 11.1 +42%
|
|---|
| 43 | # VIA Nano 6.3 5.4 11.7 8.6 +37%
|
|---|
| 44 | # Bulldozer 4.5 5.4 9.9 7.7 +29%
|
|---|
| 45 | #
|
|---|
| 46 | # (*) rc4-x86_64.pl delivers 5.3 on Opteron, so real improvement
|
|---|
| 47 | # is +53%...
|
|---|
| 48 | # (**) unidentified anomaly;
|
|---|
| 49 |
|
|---|
| 50 | my ($rc4,$md5)=(1,1); # what to generate?
|
|---|
| 51 | my $D="#" if (!$md5); # if set to "#", MD5 is stitched into RC4(),
|
|---|
| 52 | # but its result is discarded. Idea here is
|
|---|
| 53 | # to be able to use 'openssl speed rc4' for
|
|---|
| 54 | # benchmarking the stitched subroutine...
|
|---|
| 55 |
|
|---|
| 56 | # $output is the last argument if it looks like a file (it has an extension)
|
|---|
| 57 | # $flavour is the first argument if it doesn't look like a file
|
|---|
| 58 | my $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
|
|---|
| 59 | my $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
|
|---|
| 60 |
|
|---|
| 61 | my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
|
|---|
| 62 |
|
|---|
| 63 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
|
|---|
| 64 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
|
|---|
| 65 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
|
|---|
| 66 | die "can't locate x86_64-xlate.pl";
|
|---|
| 67 |
|
|---|
| 68 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""
|
|---|
| 69 | or die "can't call $xlate: $!";
|
|---|
| 70 | *STDOUT=*OUT;
|
|---|
| 71 |
|
|---|
| 72 | my ($dat,$in0,$out,$ctx,$inp,$len, $func,$nargs);
|
|---|
| 73 |
|
|---|
| 74 | if ($rc4 && !$md5) {
|
|---|
| 75 | ($dat,$len,$in0,$out) = ("%rdi","%rsi","%rdx","%rcx");
|
|---|
| 76 | $func="RC4"; $nargs=4;
|
|---|
| 77 | } elsif ($md5 && !$rc4) {
|
|---|
| 78 | ($ctx,$inp,$len) = ("%rdi","%rsi","%rdx");
|
|---|
| 79 | $func="ossl_md5_block_asm_data_order"; $nargs=3;
|
|---|
| 80 | } else {
|
|---|
| 81 | ($dat,$in0,$out,$ctx,$inp,$len) = ("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
|
|---|
| 82 | $func="rc4_md5_enc"; $nargs=6;
|
|---|
| 83 | # void rc4_md5_enc(
|
|---|
| 84 | # RC4_KEY *key, #
|
|---|
| 85 | # const void *in0, # RC4 input
|
|---|
| 86 | # void *out, # RC4 output
|
|---|
| 87 | # MD5_CTX *ctx, #
|
|---|
| 88 | # const void *inp, # MD5 input
|
|---|
| 89 | # size_t len); # number of 64-byte blocks
|
|---|
| 90 | }
|
|---|
| 91 |
|
|---|
| 92 | my @K=( 0xd76aa478,0xe8c7b756,0x242070db,0xc1bdceee,
|
|---|
| 93 | 0xf57c0faf,0x4787c62a,0xa8304613,0xfd469501,
|
|---|
| 94 | 0x698098d8,0x8b44f7af,0xffff5bb1,0x895cd7be,
|
|---|
| 95 | 0x6b901122,0xfd987193,0xa679438e,0x49b40821,
|
|---|
| 96 |
|
|---|
| 97 | 0xf61e2562,0xc040b340,0x265e5a51,0xe9b6c7aa,
|
|---|
| 98 | 0xd62f105d,0x02441453,0xd8a1e681,0xe7d3fbc8,
|
|---|
| 99 | 0x21e1cde6,0xc33707d6,0xf4d50d87,0x455a14ed,
|
|---|
| 100 | 0xa9e3e905,0xfcefa3f8,0x676f02d9,0x8d2a4c8a,
|
|---|
| 101 |
|
|---|
| 102 | 0xfffa3942,0x8771f681,0x6d9d6122,0xfde5380c,
|
|---|
| 103 | 0xa4beea44,0x4bdecfa9,0xf6bb4b60,0xbebfbc70,
|
|---|
| 104 | 0x289b7ec6,0xeaa127fa,0xd4ef3085,0x04881d05,
|
|---|
| 105 | 0xd9d4d039,0xe6db99e5,0x1fa27cf8,0xc4ac5665,
|
|---|
| 106 |
|
|---|
| 107 | 0xf4292244,0x432aff97,0xab9423a7,0xfc93a039,
|
|---|
| 108 | 0x655b59c3,0x8f0ccc92,0xffeff47d,0x85845dd1,
|
|---|
| 109 | 0x6fa87e4f,0xfe2ce6e0,0xa3014314,0x4e0811a1,
|
|---|
| 110 | 0xf7537e82,0xbd3af235,0x2ad7d2bb,0xeb86d391 );
|
|---|
| 111 |
|
|---|
| 112 | my @V=("%r8d","%r9d","%r10d","%r11d"); # MD5 registers
|
|---|
| 113 | my $tmp="%r12d";
|
|---|
| 114 |
|
|---|
| 115 | my @XX=("%rbp","%rsi"); # RC4 registers
|
|---|
| 116 | my @TX=("%rax","%rbx");
|
|---|
| 117 | my $YY="%rcx";
|
|---|
| 118 | my $TY="%rdx";
|
|---|
| 119 |
|
|---|
| 120 | my $MOD=32; # 16, 32 or 64
|
|---|
| 121 |
|
|---|
| 122 | $code.=<<___;
|
|---|
| 123 | .text
|
|---|
| 124 | .align 16
|
|---|
| 125 |
|
|---|
| 126 | .globl $func
|
|---|
| 127 | .type $func,\@function,$nargs
|
|---|
| 128 | $func:
|
|---|
| 129 | .cfi_startproc
|
|---|
| 130 | cmp \$0,$len
|
|---|
| 131 | je .Labort
|
|---|
| 132 | push %rbx
|
|---|
| 133 | .cfi_push %rbx
|
|---|
| 134 | push %rbp
|
|---|
| 135 | .cfi_push %rbp
|
|---|
| 136 | push %r12
|
|---|
| 137 | .cfi_push %r12
|
|---|
| 138 | push %r13
|
|---|
| 139 | .cfi_push %r13
|
|---|
| 140 | push %r14
|
|---|
| 141 | .cfi_push %r14
|
|---|
| 142 | push %r15
|
|---|
| 143 | .cfi_push %r15
|
|---|
| 144 | sub \$40,%rsp
|
|---|
| 145 | .cfi_adjust_cfa_offset 40
|
|---|
| 146 | .Lbody:
|
|---|
| 147 | ___
|
|---|
| 148 | if ($rc4) {
|
|---|
| 149 | $code.=<<___;
|
|---|
| 150 | $D#md5# mov $ctx,%r11 # reassign arguments
|
|---|
| 151 | mov $len,%r12
|
|---|
| 152 | mov $in0,%r13
|
|---|
| 153 | mov $out,%r14
|
|---|
| 154 | $D#md5# mov $inp,%r15
|
|---|
| 155 | ___
|
|---|
| 156 | $ctx="%r11" if ($md5); # reassign arguments
|
|---|
| 157 | $len="%r12";
|
|---|
| 158 | $in0="%r13";
|
|---|
| 159 | $out="%r14";
|
|---|
| 160 | $inp="%r15" if ($md5);
|
|---|
| 161 | $inp=$in0 if (!$md5);
|
|---|
| 162 | $code.=<<___;
|
|---|
| 163 | xor $XX[0],$XX[0]
|
|---|
| 164 | xor $YY,$YY
|
|---|
| 165 |
|
|---|
| 166 | lea 8($dat),$dat
|
|---|
| 167 | mov -8($dat),$XX[0]#b
|
|---|
| 168 | mov -4($dat),$YY#b
|
|---|
| 169 |
|
|---|
| 170 | inc $XX[0]#b
|
|---|
| 171 | sub $in0,$out
|
|---|
| 172 | movl ($dat,$XX[0],4),$TX[0]#d
|
|---|
| 173 | ___
|
|---|
| 174 | $code.=<<___ if (!$md5);
|
|---|
| 175 | xor $TX[1],$TX[1]
|
|---|
| 176 | test \$-128,$len
|
|---|
| 177 | jz .Loop1
|
|---|
| 178 | sub $XX[0],$TX[1]
|
|---|
| 179 | and \$`$MOD-1`,$TX[1]
|
|---|
| 180 | jz .Loop${MOD}_is_hot
|
|---|
| 181 | sub $TX[1],$len
|
|---|
| 182 | .Loop${MOD}_warmup:
|
|---|
| 183 | add $TX[0]#b,$YY#b
|
|---|
| 184 | movl ($dat,$YY,4),$TY#d
|
|---|
| 185 | movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 186 | movl $TY#d,($dat,$XX[0],4)
|
|---|
| 187 | add $TY#b,$TX[0]#b
|
|---|
| 188 | inc $XX[0]#b
|
|---|
| 189 | movl ($dat,$TX[0],4),$TY#d
|
|---|
| 190 | movl ($dat,$XX[0],4),$TX[0]#d
|
|---|
| 191 | xorb ($in0),$TY#b
|
|---|
| 192 | movb $TY#b,($out,$in0)
|
|---|
| 193 | lea 1($in0),$in0
|
|---|
| 194 | dec $TX[1]
|
|---|
| 195 | jnz .Loop${MOD}_warmup
|
|---|
| 196 |
|
|---|
| 197 | mov $YY,$TX[1]
|
|---|
| 198 | xor $YY,$YY
|
|---|
| 199 | mov $TX[1]#b,$YY#b
|
|---|
| 200 |
|
|---|
| 201 | .Loop${MOD}_is_hot:
|
|---|
| 202 | mov $len,32(%rsp) # save original $len
|
|---|
| 203 | shr \$6,$len # number of 64-byte blocks
|
|---|
| 204 | ___
|
|---|
| 205 | if ($D && !$md5) { # stitch in dummy MD5
|
|---|
| 206 | $md5=1;
|
|---|
| 207 | $ctx="%r11";
|
|---|
| 208 | $inp="%r15";
|
|---|
| 209 | $code.=<<___;
|
|---|
| 210 | mov %rsp,$ctx
|
|---|
| 211 | mov $in0,$inp
|
|---|
| 212 | ___
|
|---|
| 213 | }
|
|---|
| 214 | }
|
|---|
| 215 | $code.=<<___;
|
|---|
| 216 | #rc4# add $TX[0]#b,$YY#b
|
|---|
| 217 | #rc4# lea ($dat,$XX[0],4),$XX[1]
|
|---|
| 218 | shl \$6,$len
|
|---|
| 219 | add $inp,$len # pointer to the end of input
|
|---|
| 220 | mov $len,16(%rsp)
|
|---|
| 221 |
|
|---|
| 222 | #md5# mov $ctx,24(%rsp) # save pointer to MD5_CTX
|
|---|
| 223 | #md5# mov 0*4($ctx),$V[0] # load current hash value from MD5_CTX
|
|---|
| 224 | #md5# mov 1*4($ctx),$V[1]
|
|---|
| 225 | #md5# mov 2*4($ctx),$V[2]
|
|---|
| 226 | #md5# mov 3*4($ctx),$V[3]
|
|---|
| 227 | jmp .Loop
|
|---|
| 228 |
|
|---|
| 229 | .align 16
|
|---|
| 230 | .Loop:
|
|---|
| 231 | #md5# mov $V[0],0*4(%rsp) # put aside current hash value
|
|---|
| 232 | #md5# mov $V[1],1*4(%rsp)
|
|---|
| 233 | #md5# mov $V[2],2*4(%rsp)
|
|---|
| 234 | #md5# mov $V[3],$tmp # forward reference
|
|---|
| 235 | #md5# mov $V[3],3*4(%rsp)
|
|---|
| 236 | ___
|
|---|
| 237 |
|
|---|
| 238 | sub R0 {
|
|---|
| 239 | my ($i,$a,$b,$c,$d)=@_;
|
|---|
| 240 | my @rot0=(7,12,17,22);
|
|---|
| 241 | my $j=$i%16;
|
|---|
| 242 | my $k=$i%$MOD;
|
|---|
| 243 | my $xmm="%xmm".($j&1);
|
|---|
| 244 | $code.=" movdqu ($in0),%xmm2\n" if ($rc4 && $j==15);
|
|---|
| 245 | $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 246 | $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
|
|---|
| 247 | $code.=<<___;
|
|---|
| 248 | #rc4# movl ($dat,$YY,4),$TY#d
|
|---|
| 249 | #md5# xor $c,$tmp
|
|---|
| 250 | #rc4# movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 251 | #md5# and $b,$tmp
|
|---|
| 252 | #md5# add 4*`$j`($inp),$a
|
|---|
| 253 | #rc4# add $TY#b,$TX[0]#b
|
|---|
| 254 | #rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
|
|---|
| 255 | #md5# add \$$K[$i],$a
|
|---|
| 256 | #md5# xor $d,$tmp
|
|---|
| 257 | #rc4# movz $TX[0]#b,$TX[0]#d
|
|---|
| 258 | #rc4# movl $TY#d,4*$k($XX[1])
|
|---|
| 259 | #md5# add $tmp,$a
|
|---|
| 260 | #rc4# add $TX[1]#b,$YY#b
|
|---|
| 261 | #md5# rol \$$rot0[$j%4],$a
|
|---|
| 262 | #md5# mov `$j==15?"$b":"$c"`,$tmp # forward reference
|
|---|
| 263 | #rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
|
|---|
| 264 | #md5# add $b,$a
|
|---|
| 265 | ___
|
|---|
| 266 | $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 267 | mov $YY,$XX[1]
|
|---|
| 268 | xor $YY,$YY # keyword to partial register
|
|---|
| 269 | mov $XX[1]#b,$YY#b
|
|---|
| 270 | lea ($dat,$XX[0],4),$XX[1]
|
|---|
| 271 | ___
|
|---|
| 272 | $code.=<<___ if ($rc4 && $j==15);
|
|---|
| 273 | psllq \$8,%xmm1
|
|---|
| 274 | pxor %xmm0,%xmm2
|
|---|
| 275 | pxor %xmm1,%xmm2
|
|---|
| 276 | ___
|
|---|
| 277 | }
|
|---|
| 278 | sub R1 {
|
|---|
| 279 | my ($i,$a,$b,$c,$d)=@_;
|
|---|
| 280 | my @rot1=(5,9,14,20);
|
|---|
| 281 | my $j=$i%16;
|
|---|
| 282 | my $k=$i%$MOD;
|
|---|
| 283 | my $xmm="%xmm".($j&1);
|
|---|
| 284 | $code.=" movdqu 16($in0),%xmm3\n" if ($rc4 && $j==15);
|
|---|
| 285 | $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 286 | $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
|
|---|
| 287 | $code.=<<___;
|
|---|
| 288 | #rc4# movl ($dat,$YY,4),$TY#d
|
|---|
| 289 | #md5# xor $b,$tmp
|
|---|
| 290 | #rc4# movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 291 | #md5# and $d,$tmp
|
|---|
| 292 | #md5# add 4*`((1+5*$j)%16)`($inp),$a
|
|---|
| 293 | #rc4# add $TY#b,$TX[0]#b
|
|---|
| 294 | #rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
|
|---|
| 295 | #md5# add \$$K[$i],$a
|
|---|
| 296 | #md5# xor $c,$tmp
|
|---|
| 297 | #rc4# movz $TX[0]#b,$TX[0]#d
|
|---|
| 298 | #rc4# movl $TY#d,4*$k($XX[1])
|
|---|
| 299 | #md5# add $tmp,$a
|
|---|
| 300 | #rc4# add $TX[1]#b,$YY#b
|
|---|
| 301 | #md5# rol \$$rot1[$j%4],$a
|
|---|
| 302 | #md5# mov `$j==15?"$c":"$b"`,$tmp # forward reference
|
|---|
| 303 | #rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
|
|---|
| 304 | #md5# add $b,$a
|
|---|
| 305 | ___
|
|---|
| 306 | $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 307 | mov $YY,$XX[1]
|
|---|
| 308 | xor $YY,$YY # keyword to partial register
|
|---|
| 309 | mov $XX[1]#b,$YY#b
|
|---|
| 310 | lea ($dat,$XX[0],4),$XX[1]
|
|---|
| 311 | ___
|
|---|
| 312 | $code.=<<___ if ($rc4 && $j==15);
|
|---|
| 313 | psllq \$8,%xmm1
|
|---|
| 314 | pxor %xmm0,%xmm3
|
|---|
| 315 | pxor %xmm1,%xmm3
|
|---|
| 316 | ___
|
|---|
| 317 | }
|
|---|
| 318 | sub R2 {
|
|---|
| 319 | my ($i,$a,$b,$c,$d)=@_;
|
|---|
| 320 | my @rot2=(4,11,16,23);
|
|---|
| 321 | my $j=$i%16;
|
|---|
| 322 | my $k=$i%$MOD;
|
|---|
| 323 | my $xmm="%xmm".($j&1);
|
|---|
| 324 | $code.=" movdqu 32($in0),%xmm4\n" if ($rc4 && $j==15);
|
|---|
| 325 | $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 326 | $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
|
|---|
| 327 | $code.=<<___;
|
|---|
| 328 | #rc4# movl ($dat,$YY,4),$TY#d
|
|---|
| 329 | #md5# xor $c,$tmp
|
|---|
| 330 | #rc4# movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 331 | #md5# xor $b,$tmp
|
|---|
| 332 | #md5# add 4*`((5+3*$j)%16)`($inp),$a
|
|---|
| 333 | #rc4# add $TY#b,$TX[0]#b
|
|---|
| 334 | #rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
|
|---|
| 335 | #md5# add \$$K[$i],$a
|
|---|
| 336 | #rc4# movz $TX[0]#b,$TX[0]#d
|
|---|
| 337 | #md5# add $tmp,$a
|
|---|
| 338 | #rc4# movl $TY#d,4*$k($XX[1])
|
|---|
| 339 | #rc4# add $TX[1]#b,$YY#b
|
|---|
| 340 | #md5# rol \$$rot2[$j%4],$a
|
|---|
| 341 | #md5# mov `$j==15?"\\\$-1":"$c"`,$tmp # forward reference
|
|---|
| 342 | #rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
|
|---|
| 343 | #md5# add $b,$a
|
|---|
| 344 | ___
|
|---|
| 345 | $code.=<<___ if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 346 | mov $YY,$XX[1]
|
|---|
| 347 | xor $YY,$YY # keyword to partial register
|
|---|
| 348 | mov $XX[1]#b,$YY#b
|
|---|
| 349 | lea ($dat,$XX[0],4),$XX[1]
|
|---|
| 350 | ___
|
|---|
| 351 | $code.=<<___ if ($rc4 && $j==15);
|
|---|
| 352 | psllq \$8,%xmm1
|
|---|
| 353 | pxor %xmm0,%xmm4
|
|---|
| 354 | pxor %xmm1,%xmm4
|
|---|
| 355 | ___
|
|---|
| 356 | }
|
|---|
| 357 | sub R3 {
|
|---|
| 358 | my ($i,$a,$b,$c,$d)=@_;
|
|---|
| 359 | my @rot3=(6,10,15,21);
|
|---|
| 360 | my $j=$i%16;
|
|---|
| 361 | my $k=$i%$MOD;
|
|---|
| 362 | my $xmm="%xmm".($j&1);
|
|---|
| 363 | $code.=" movdqu 48($in0),%xmm5\n" if ($rc4 && $j==15);
|
|---|
| 364 | $code.=" add \$$MOD,$XX[0]#b\n" if ($rc4 && $j==15 && $k==$MOD-1);
|
|---|
| 365 | $code.=" pxor $xmm,$xmm\n" if ($rc4 && $j<=1);
|
|---|
| 366 | $code.=<<___;
|
|---|
| 367 | #rc4# movl ($dat,$YY,4),$TY#d
|
|---|
| 368 | #md5# xor $d,$tmp
|
|---|
| 369 | #rc4# movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 370 | #md5# or $b,$tmp
|
|---|
| 371 | #md5# add 4*`((7*$j)%16)`($inp),$a
|
|---|
| 372 | #rc4# add $TY#b,$TX[0]#b
|
|---|
| 373 | #rc4# movl `4*(($k+1)%$MOD)`(`$k==$MOD-1?"$dat,$XX[0],4":"$XX[1]"`),$TX[1]#d
|
|---|
| 374 | #md5# add \$$K[$i],$a
|
|---|
| 375 | #rc4# movz $TX[0]#b,$TX[0]#d
|
|---|
| 376 | #md5# xor $c,$tmp
|
|---|
| 377 | #rc4# movl $TY#d,4*$k($XX[1])
|
|---|
| 378 | #md5# add $tmp,$a
|
|---|
| 379 | #rc4# add $TX[1]#b,$YY#b
|
|---|
| 380 | #md5# rol \$$rot3[$j%4],$a
|
|---|
| 381 | #md5# mov \$-1,$tmp # forward reference
|
|---|
| 382 | #rc4# pinsrw \$`($j>>1)&7`,($dat,$TX[0],4),$xmm\n
|
|---|
| 383 | #md5# add $b,$a
|
|---|
| 384 | ___
|
|---|
| 385 | $code.=<<___ if ($rc4 && $j==15);
|
|---|
| 386 | mov $XX[0],$XX[1]
|
|---|
| 387 | xor $XX[0],$XX[0] # keyword to partial register
|
|---|
| 388 | mov $XX[1]#b,$XX[0]#b
|
|---|
| 389 | mov $YY,$XX[1]
|
|---|
| 390 | xor $YY,$YY # keyword to partial register
|
|---|
| 391 | mov $XX[1]#b,$YY#b
|
|---|
| 392 | lea ($dat,$XX[0],4),$XX[1]
|
|---|
| 393 | psllq \$8,%xmm1
|
|---|
| 394 | pxor %xmm0,%xmm5
|
|---|
| 395 | pxor %xmm1,%xmm5
|
|---|
| 396 | ___
|
|---|
| 397 | }
|
|---|
| 398 |
|
|---|
| 399 | my $i=0;
|
|---|
| 400 | for(;$i<16;$i++) { R0($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
|
|---|
| 401 | for(;$i<32;$i++) { R1($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
|
|---|
| 402 | for(;$i<48;$i++) { R2($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
|
|---|
| 403 | for(;$i<64;$i++) { R3($i,@V); unshift(@V,pop(@V)); push(@TX,shift(@TX)); }
|
|---|
| 404 |
|
|---|
| 405 | $code.=<<___;
|
|---|
| 406 | #md5# add 0*4(%rsp),$V[0] # accumulate hash value
|
|---|
| 407 | #md5# add 1*4(%rsp),$V[1]
|
|---|
| 408 | #md5# add 2*4(%rsp),$V[2]
|
|---|
| 409 | #md5# add 3*4(%rsp),$V[3]
|
|---|
| 410 |
|
|---|
| 411 | #rc4# movdqu %xmm2,($out,$in0) # write RC4 output
|
|---|
| 412 | #rc4# movdqu %xmm3,16($out,$in0)
|
|---|
| 413 | #rc4# movdqu %xmm4,32($out,$in0)
|
|---|
| 414 | #rc4# movdqu %xmm5,48($out,$in0)
|
|---|
| 415 | #md5# lea 64($inp),$inp
|
|---|
| 416 | #rc4# lea 64($in0),$in0
|
|---|
| 417 | cmp 16(%rsp),$inp # are we done?
|
|---|
| 418 | jb .Loop
|
|---|
| 419 |
|
|---|
| 420 | #md5# mov 24(%rsp),$len # restore pointer to MD5_CTX
|
|---|
| 421 | #rc4# sub $TX[0]#b,$YY#b # correct $YY
|
|---|
| 422 | #md5# mov $V[0],0*4($len) # write MD5_CTX
|
|---|
| 423 | #md5# mov $V[1],1*4($len)
|
|---|
| 424 | #md5# mov $V[2],2*4($len)
|
|---|
| 425 | #md5# mov $V[3],3*4($len)
|
|---|
| 426 | ___
|
|---|
| 427 | $code.=<<___ if ($rc4 && (!$md5 || $D));
|
|---|
| 428 | mov 32(%rsp),$len # restore original $len
|
|---|
| 429 | and \$63,$len # remaining bytes
|
|---|
| 430 | jnz .Loop1
|
|---|
| 431 | jmp .Ldone
|
|---|
| 432 |
|
|---|
| 433 | .align 16
|
|---|
| 434 | .Loop1:
|
|---|
| 435 | add $TX[0]#b,$YY#b
|
|---|
| 436 | movl ($dat,$YY,4),$TY#d
|
|---|
| 437 | movl $TX[0]#d,($dat,$YY,4)
|
|---|
| 438 | movl $TY#d,($dat,$XX[0],4)
|
|---|
| 439 | add $TY#b,$TX[0]#b
|
|---|
| 440 | inc $XX[0]#b
|
|---|
| 441 | movl ($dat,$TX[0],4),$TY#d
|
|---|
| 442 | movl ($dat,$XX[0],4),$TX[0]#d
|
|---|
| 443 | xorb ($in0),$TY#b
|
|---|
| 444 | movb $TY#b,($out,$in0)
|
|---|
| 445 | lea 1($in0),$in0
|
|---|
| 446 | dec $len
|
|---|
| 447 | jnz .Loop1
|
|---|
| 448 |
|
|---|
| 449 | .Ldone:
|
|---|
| 450 | ___
|
|---|
| 451 | $code.=<<___;
|
|---|
| 452 | #rc4# sub \$1,$XX[0]#b
|
|---|
| 453 | #rc4# movl $XX[0]#d,-8($dat)
|
|---|
| 454 | #rc4# movl $YY#d,-4($dat)
|
|---|
| 455 |
|
|---|
| 456 | mov 40(%rsp),%r15
|
|---|
| 457 | .cfi_restore %r15
|
|---|
| 458 | mov 48(%rsp),%r14
|
|---|
| 459 | .cfi_restore %r14
|
|---|
| 460 | mov 56(%rsp),%r13
|
|---|
| 461 | .cfi_restore %r13
|
|---|
| 462 | mov 64(%rsp),%r12
|
|---|
| 463 | .cfi_restore %r12
|
|---|
| 464 | mov 72(%rsp),%rbp
|
|---|
| 465 | .cfi_restore %rbp
|
|---|
| 466 | mov 80(%rsp),%rbx
|
|---|
| 467 | .cfi_restore %rbx
|
|---|
| 468 | lea 88(%rsp),%rsp
|
|---|
| 469 | .cfi_adjust_cfa_offset -88
|
|---|
| 470 | .Lepilogue:
|
|---|
| 471 | .Labort:
|
|---|
| 472 | ret
|
|---|
| 473 | .cfi_endproc
|
|---|
| 474 | .size $func,.-$func
|
|---|
| 475 | ___
|
|---|
| 476 |
|
|---|
| 477 | if ($rc4 && $D) { # sole purpose of this section is to provide
|
|---|
| 478 | # option to use the generated module as drop-in
|
|---|
| 479 | # replacement for rc4-x86_64.pl for debugging
|
|---|
| 480 | # and testing purposes...
|
|---|
| 481 | my ($idx,$ido)=("%r8","%r9");
|
|---|
| 482 | my ($dat,$len,$inp)=("%rdi","%rsi","%rdx");
|
|---|
| 483 |
|
|---|
| 484 | $code.=<<___;
|
|---|
| 485 | .globl RC4_set_key
|
|---|
| 486 | .type RC4_set_key,\@function,3
|
|---|
| 487 | .align 16
|
|---|
| 488 | RC4_set_key:
|
|---|
| 489 | .cfi_startproc
|
|---|
| 490 | lea 8($dat),$dat
|
|---|
| 491 | lea ($inp,$len),$inp
|
|---|
| 492 | neg $len
|
|---|
| 493 | mov $len,%rcx
|
|---|
| 494 | xor %eax,%eax
|
|---|
| 495 | xor $ido,$ido
|
|---|
| 496 | xor %r10,%r10
|
|---|
| 497 | xor %r11,%r11
|
|---|
| 498 | jmp .Lw1stloop
|
|---|
| 499 |
|
|---|
| 500 | .align 16
|
|---|
| 501 | .Lw1stloop:
|
|---|
| 502 | mov %eax,($dat,%rax,4)
|
|---|
| 503 | add \$1,%al
|
|---|
| 504 | jnc .Lw1stloop
|
|---|
| 505 |
|
|---|
| 506 | xor $ido,$ido
|
|---|
| 507 | xor $idx,$idx
|
|---|
| 508 | .align 16
|
|---|
| 509 | .Lw2ndloop:
|
|---|
| 510 | mov ($dat,$ido,4),%r10d
|
|---|
| 511 | add ($inp,$len,1),$idx#b
|
|---|
| 512 | add %r10b,$idx#b
|
|---|
| 513 | add \$1,$len
|
|---|
| 514 | mov ($dat,$idx,4),%r11d
|
|---|
| 515 | cmovz %rcx,$len
|
|---|
| 516 | mov %r10d,($dat,$idx,4)
|
|---|
| 517 | mov %r11d,($dat,$ido,4)
|
|---|
| 518 | add \$1,$ido#b
|
|---|
| 519 | jnc .Lw2ndloop
|
|---|
| 520 |
|
|---|
| 521 | xor %eax,%eax
|
|---|
| 522 | mov %eax,-8($dat)
|
|---|
| 523 | mov %eax,-4($dat)
|
|---|
| 524 | ret
|
|---|
| 525 | .cfi_endproc
|
|---|
| 526 | .size RC4_set_key,.-RC4_set_key
|
|---|
| 527 |
|
|---|
| 528 | .globl RC4_options
|
|---|
| 529 | .type RC4_options,\@abi-omnipotent
|
|---|
| 530 | .align 16
|
|---|
| 531 | RC4_options:
|
|---|
| 532 | lea .Lopts(%rip),%rax
|
|---|
| 533 | ret
|
|---|
| 534 | .align 64
|
|---|
| 535 | .Lopts:
|
|---|
| 536 | .asciz "rc4(64x,int)"
|
|---|
| 537 | .align 64
|
|---|
| 538 | .size RC4_options,.-RC4_options
|
|---|
| 539 | ___
|
|---|
| 540 | }
|
|---|
| 541 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
|
|---|
| 542 | # CONTEXT *context,DISPATCHER_CONTEXT *disp)
|
|---|
| 543 | if ($win64) {
|
|---|
| 544 | my $rec="%rcx";
|
|---|
| 545 | my $frame="%rdx";
|
|---|
| 546 | my $context="%r8";
|
|---|
| 547 | my $disp="%r9";
|
|---|
| 548 |
|
|---|
| 549 | $code.=<<___;
|
|---|
| 550 | .extern __imp_RtlVirtualUnwind
|
|---|
| 551 | .type se_handler,\@abi-omnipotent
|
|---|
| 552 | .align 16
|
|---|
| 553 | se_handler:
|
|---|
| 554 | push %rsi
|
|---|
| 555 | push %rdi
|
|---|
| 556 | push %rbx
|
|---|
| 557 | push %rbp
|
|---|
| 558 | push %r12
|
|---|
| 559 | push %r13
|
|---|
| 560 | push %r14
|
|---|
| 561 | push %r15
|
|---|
| 562 | pushfq
|
|---|
| 563 | sub \$64,%rsp
|
|---|
| 564 |
|
|---|
| 565 | mov 120($context),%rax # pull context->Rax
|
|---|
| 566 | mov 248($context),%rbx # pull context->Rip
|
|---|
| 567 |
|
|---|
| 568 | lea .Lbody(%rip),%r10
|
|---|
| 569 | cmp %r10,%rbx # context->Rip<.Lbody
|
|---|
| 570 | jb .Lin_prologue
|
|---|
| 571 |
|
|---|
| 572 | mov 152($context),%rax # pull context->Rsp
|
|---|
| 573 |
|
|---|
| 574 | lea .Lepilogue(%rip),%r10
|
|---|
| 575 | cmp %r10,%rbx # context->Rip>=.Lepilogue
|
|---|
| 576 | jae .Lin_prologue
|
|---|
| 577 |
|
|---|
| 578 | mov 40(%rax),%r15
|
|---|
| 579 | mov 48(%rax),%r14
|
|---|
| 580 | mov 56(%rax),%r13
|
|---|
| 581 | mov 64(%rax),%r12
|
|---|
| 582 | mov 72(%rax),%rbp
|
|---|
| 583 | mov 80(%rax),%rbx
|
|---|
| 584 | lea 88(%rax),%rax
|
|---|
| 585 |
|
|---|
| 586 | mov %rbx,144($context) # restore context->Rbx
|
|---|
| 587 | mov %rbp,160($context) # restore context->Rbp
|
|---|
| 588 | mov %r12,216($context) # restore context->R12
|
|---|
| 589 | mov %r13,224($context) # restore context->R12
|
|---|
| 590 | mov %r14,232($context) # restore context->R14
|
|---|
| 591 | mov %r15,240($context) # restore context->R15
|
|---|
| 592 |
|
|---|
| 593 | .Lin_prologue:
|
|---|
| 594 | mov 8(%rax),%rdi
|
|---|
| 595 | mov 16(%rax),%rsi
|
|---|
| 596 | mov %rax,152($context) # restore context->Rsp
|
|---|
| 597 | mov %rsi,168($context) # restore context->Rsi
|
|---|
| 598 | mov %rdi,176($context) # restore context->Rdi
|
|---|
| 599 |
|
|---|
| 600 | mov 40($disp),%rdi # disp->ContextRecord
|
|---|
| 601 | mov $context,%rsi # context
|
|---|
| 602 | mov \$154,%ecx # sizeof(CONTEXT)
|
|---|
| 603 | .long 0xa548f3fc # cld; rep movsq
|
|---|
| 604 |
|
|---|
| 605 | mov $disp,%rsi
|
|---|
| 606 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
|
|---|
| 607 | mov 8(%rsi),%rdx # arg2, disp->ImageBase
|
|---|
| 608 | mov 0(%rsi),%r8 # arg3, disp->ControlPc
|
|---|
| 609 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
|
|---|
| 610 | mov 40(%rsi),%r10 # disp->ContextRecord
|
|---|
| 611 | lea 56(%rsi),%r11 # &disp->HandlerData
|
|---|
| 612 | lea 24(%rsi),%r12 # &disp->EstablisherFrame
|
|---|
| 613 | mov %r10,32(%rsp) # arg5
|
|---|
| 614 | mov %r11,40(%rsp) # arg6
|
|---|
| 615 | mov %r12,48(%rsp) # arg7
|
|---|
| 616 | mov %rcx,56(%rsp) # arg8, (NULL)
|
|---|
| 617 | call *__imp_RtlVirtualUnwind(%rip)
|
|---|
| 618 |
|
|---|
| 619 | mov \$1,%eax # ExceptionContinueSearch
|
|---|
| 620 | add \$64,%rsp
|
|---|
| 621 | popfq
|
|---|
| 622 | pop %r15
|
|---|
| 623 | pop %r14
|
|---|
| 624 | pop %r13
|
|---|
| 625 | pop %r12
|
|---|
| 626 | pop %rbp
|
|---|
| 627 | pop %rbx
|
|---|
| 628 | pop %rdi
|
|---|
| 629 | pop %rsi
|
|---|
| 630 | ret
|
|---|
| 631 | .size se_handler,.-se_handler
|
|---|
| 632 |
|
|---|
| 633 | .section .pdata
|
|---|
| 634 | .align 4
|
|---|
| 635 | .rva .LSEH_begin_$func
|
|---|
| 636 | .rva .LSEH_end_$func
|
|---|
| 637 | .rva .LSEH_info_$func
|
|---|
| 638 |
|
|---|
| 639 | .section .xdata
|
|---|
| 640 | .align 8
|
|---|
| 641 | .LSEH_info_$func:
|
|---|
| 642 | .byte 9,0,0,0
|
|---|
| 643 | .rva se_handler
|
|---|
| 644 | ___
|
|---|
| 645 | }
|
|---|
| 646 |
|
|---|
| 647 | sub reg_part {
|
|---|
| 648 | my ($reg,$conv)=@_;
|
|---|
| 649 | if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
|
|---|
| 650 | elsif ($conv eq "b") { $reg =~ s/%[er]([^x]+)x?/%$1l/; }
|
|---|
| 651 | elsif ($conv eq "w") { $reg =~ s/%[er](.+)/%$1/; }
|
|---|
| 652 | elsif ($conv eq "d") { $reg =~ s/%[er](.+)/%e$1/; }
|
|---|
| 653 | return $reg;
|
|---|
| 654 | }
|
|---|
| 655 |
|
|---|
| 656 | $code =~ s/(%[a-z0-9]+)#([bwd])/reg_part($1,$2)/gem;
|
|---|
| 657 | $code =~ s/\`([^\`]*)\`/eval $1/gem;
|
|---|
| 658 | $code =~ s/pinsrw\s+\$0,/movd /gm;
|
|---|
| 659 |
|
|---|
| 660 | $code =~ s/#md5#//gm if ($md5);
|
|---|
| 661 | $code =~ s/#rc4#//gm if ($rc4);
|
|---|
| 662 |
|
|---|
| 663 | print $code;
|
|---|
| 664 |
|
|---|
| 665 | close STDOUT or die "error closing STDOUT: $!";
|
|---|
