crypto/modes/asm/aesni-gcm-x86_64.pl

   1 #! /usr/bin/env perl
   2 # Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
   3 #
   4 # Licensed under the OpenSSL license (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 <appro@openssl.org> 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 #
  18 # AES-NI-CTR+GHASH stitch.
  19 #
  20 # February 2013
  21 #
  22 # OpenSSL GCM implementation is organized in such way that its
  23 # performance is rather close to the sum of its streamed components,
  24 # in the context parallelized AES-NI CTR and modulo-scheduled
  25 # PCLMULQDQ-enabled GHASH. Unfortunately, as no stitch implementation
  26 # was observed to perform significantly better than the sum of the
  27 # components on contemporary CPUs, the effort was deemed impossible to
  28 # justify. This module is based on combination of Intel submissions,
  29 # [1] and [2], with MOVBE twist suggested by Ilya Albrekht and Max
  30 # Locktyukhin of Intel Corp. who verified that it reduces shuffles
  31 # pressure with notable relative improvement, achieving 1.0 cycle per
  32 # byte processed with 128-bit key on Haswell processor, 0.74 - on
  33 # Broadwell, 0.63 - on Skylake... [Mentioned results are raw profiled
  34 # measurements for favourable packet size, one divisible by 96.
  35 # Applications using the EVP interface will observe a few percent
  36 # worse performance.]
  37 #
  38 # [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest
  39 # [2] http://www.intel.com/content/dam/www/public/us/en/documents/software-support/enabling-high-performance-gcm.pdf
  40
  41 $flavour = shift;
  42 $output  = shift;
  43 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  44
  45 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  46
  47 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  48 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  49 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
  50 die "can't locate x86_64-xlate.pl";
  51
  52 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
  53                 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
  54         $avx = ($1>=2.20) + ($1>=2.22);
  55 }
  56
  57 if (!$avx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
  58             `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
  59         $avx = ($1>=2.09) + ($1>=2.10);
  60 }
  61
  62 if (!$avx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
  63             `ml64 2>&1` =~ /Version ([0-9]+)\./) {
  64         $avx = ($1>=10) + ($1>=11);
  65 }
  66
  67 if (!$avx && `$ENV{CC} -v 2>&1` =~ /((?:^clang|LLVM) version|.*based on LLVM) ([3-9]\.[0-9]+)/) {
  68         $avx = ($2>=3.0) + ($2>3.0);
  69 }
  70
  71 open OUT,"| \"$^X\" $xlate $flavour $output";
  72 *STDOUT=*OUT;
  73
  74 if ($avx>1) {{{
  75
  76 ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9");
  77
  78 ($Ii,$T1,$T2,$Hkey,
  79  $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8));
  80
  81 ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15));
  82
  83 ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15");
  84
  85 $code=<<___;
  86 .text
  87
  88 .type   _aesni_ctr32_ghash_6x,\@abi-omnipotent
  89 .align  32
  90 _aesni_ctr32_ghash_6x:
  91         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
  92         sub             \$6,$len
  93         vpxor           $Z0,$Z0,$Z0             # $Z0   = 0
  94         vmovdqu         0x00-0x80($key),$rndkey
  95         vpaddb          $T2,$T1,$inout1
  96         vpaddb          $T2,$inout1,$inout2
  97         vpaddb          $T2,$inout2,$inout3
  98         vpaddb          $T2,$inout3,$inout4
  99         vpaddb          $T2,$inout4,$inout5
 100         vpxor           $rndkey,$T1,$inout0
 101         vmovdqu         $Z0,16+8(%rsp)          # "$Z3" = 0
 102         jmp             .Loop6x
 103
 104 .align  32
 105 .Loop6x:
 106         add             \$`6<<24`,$counter
 107         jc              .Lhandle_ctr32          # discard $inout[1-5]?
 108         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 109           vpaddb        $T2,$inout5,$T1         # next counter value
 110           vpxor         $rndkey,$inout1,$inout1
 111           vpxor         $rndkey,$inout2,$inout2
 112
 113 .Lresume_ctr32:
 114         vmovdqu         $T1,($ivp)              # save next counter value
 115         vpclmulqdq      \$0x10,$Hkey,$Z3,$Z1
 116           vpxor         $rndkey,$inout3,$inout3
 117           vmovups       0x10-0x80($key),$T2     # borrow $T2 for $rndkey
 118         vpclmulqdq      \$0x01,$Hkey,$Z3,$Z2
 119         xor             %r12,%r12
 120         cmp             $in0,$end0
 121
 122           vaesenc       $T2,$inout0,$inout0
 123         vmovdqu         0x30+8(%rsp),$Ii        # I[4]
 124           vpxor         $rndkey,$inout4,$inout4
 125         vpclmulqdq      \$0x00,$Hkey,$Z3,$T1
 126           vaesenc       $T2,$inout1,$inout1
 127           vpxor         $rndkey,$inout5,$inout5
 128         setnc           %r12b
 129         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 130           vaesenc       $T2,$inout2,$inout2
 131         vmovdqu         0x10-0x20($Xip),$Hkey   # $Hkey^2
 132         neg             %r12
 133           vaesenc       $T2,$inout3,$inout3
 134          vpxor          $Z1,$Z2,$Z2
 135         vpclmulqdq      \$0x00,$Hkey,$Ii,$Z1
 136          vpxor          $Z0,$Xi,$Xi             # modulo-scheduled
 137           vaesenc       $T2,$inout4,$inout4
 138          vpxor          $Z1,$T1,$Z0
 139         and             \$0x60,%r12
 140           vmovups       0x20-0x80($key),$rndkey
 141         vpclmulqdq      \$0x10,$Hkey,$Ii,$T1
 142           vaesenc       $T2,$inout5,$inout5
 143
 144         vpclmulqdq      \$0x01,$Hkey,$Ii,$T2
 145         lea             ($in0,%r12),$in0
 146           vaesenc       $rndkey,$inout0,$inout0
 147          vpxor          16+8(%rsp),$Xi,$Xi      # modulo-scheduled [vpxor $Z3,$Xi,$Xi]
 148         vpclmulqdq      \$0x11,$Hkey,$Ii,$Hkey
 149          vmovdqu        0x40+8(%rsp),$Ii        # I[3]
 150           vaesenc       $rndkey,$inout1,$inout1
 151         movbe           0x58($in0),%r13
 152           vaesenc       $rndkey,$inout2,$inout2
 153         movbe           0x50($in0),%r12
 154           vaesenc       $rndkey,$inout3,$inout3
 155         mov             %r13,0x20+8(%rsp)
 156           vaesenc       $rndkey,$inout4,$inout4
 157         mov             %r12,0x28+8(%rsp)
 158         vmovdqu         0x30-0x20($Xip),$Z1     # borrow $Z1 for $Hkey^3
 159           vaesenc       $rndkey,$inout5,$inout5
 160
 161           vmovups       0x30-0x80($key),$rndkey
 162          vpxor          $T1,$Z2,$Z2
 163         vpclmulqdq      \$0x00,$Z1,$Ii,$T1
 164           vaesenc       $rndkey,$inout0,$inout0
 165          vpxor          $T2,$Z2,$Z2
 166         vpclmulqdq      \$0x10,$Z1,$Ii,$T2
 167           vaesenc       $rndkey,$inout1,$inout1
 168          vpxor          $Hkey,$Z3,$Z3
 169         vpclmulqdq      \$0x01,$Z1,$Ii,$Hkey
 170           vaesenc       $rndkey,$inout2,$inout2
 171         vpclmulqdq      \$0x11,$Z1,$Ii,$Z1
 172          vmovdqu        0x50+8(%rsp),$Ii        # I[2]
 173           vaesenc       $rndkey,$inout3,$inout3
 174           vaesenc       $rndkey,$inout4,$inout4
 175          vpxor          $T1,$Z0,$Z0
 176         vmovdqu         0x40-0x20($Xip),$T1     # borrow $T1 for $Hkey^4
 177           vaesenc       $rndkey,$inout5,$inout5
 178
 179           vmovups       0x40-0x80($key),$rndkey
 180          vpxor          $T2,$Z2,$Z2
 181         vpclmulqdq      \$0x00,$T1,$Ii,$T2
 182           vaesenc       $rndkey,$inout0,$inout0
 183          vpxor          $Hkey,$Z2,$Z2
 184         vpclmulqdq      \$0x10,$T1,$Ii,$Hkey
 185           vaesenc       $rndkey,$inout1,$inout1
 186         movbe           0x48($in0),%r13
 187          vpxor          $Z1,$Z3,$Z3
 188         vpclmulqdq      \$0x01,$T1,$Ii,$Z1
 189           vaesenc       $rndkey,$inout2,$inout2
 190         movbe           0x40($in0),%r12
 191         vpclmulqdq      \$0x11,$T1,$Ii,$T1
 192          vmovdqu        0x60+8(%rsp),$Ii        # I[1]
 193           vaesenc       $rndkey,$inout3,$inout3
 194         mov             %r13,0x30+8(%rsp)
 195           vaesenc       $rndkey,$inout4,$inout4
 196         mov             %r12,0x38+8(%rsp)
 197          vpxor          $T2,$Z0,$Z0
 198         vmovdqu         0x60-0x20($Xip),$T2     # borrow $T2 for $Hkey^5
 199           vaesenc       $rndkey,$inout5,$inout5
 200
 201           vmovups       0x50-0x80($key),$rndkey
 202          vpxor          $Hkey,$Z2,$Z2
 203         vpclmulqdq      \$0x00,$T2,$Ii,$Hkey
 204           vaesenc       $rndkey,$inout0,$inout0
 205          vpxor          $Z1,$Z2,$Z2
 206         vpclmulqdq      \$0x10,$T2,$Ii,$Z1
 207           vaesenc       $rndkey,$inout1,$inout1
 208         movbe           0x38($in0),%r13
 209          vpxor          $T1,$Z3,$Z3
 210         vpclmulqdq      \$0x01,$T2,$Ii,$T1
 211          vpxor          0x70+8(%rsp),$Xi,$Xi    # accumulate I[0]
 212           vaesenc       $rndkey,$inout2,$inout2
 213         movbe           0x30($in0),%r12
 214         vpclmulqdq      \$0x11,$T2,$Ii,$T2
 215           vaesenc       $rndkey,$inout3,$inout3
 216         mov             %r13,0x40+8(%rsp)
 217           vaesenc       $rndkey,$inout4,$inout4
 218         mov             %r12,0x48+8(%rsp)
 219          vpxor          $Hkey,$Z0,$Z0
 220          vmovdqu        0x70-0x20($Xip),$Hkey   # $Hkey^6
 221           vaesenc       $rndkey,$inout5,$inout5
 222
 223           vmovups       0x60-0x80($key),$rndkey
 224          vpxor          $Z1,$Z2,$Z2
 225         vpclmulqdq      \$0x10,$Hkey,$Xi,$Z1
 226           vaesenc       $rndkey,$inout0,$inout0
 227          vpxor          $T1,$Z2,$Z2
 228         vpclmulqdq      \$0x01,$Hkey,$Xi,$T1
 229           vaesenc       $rndkey,$inout1,$inout1
 230         movbe           0x28($in0),%r13
 231          vpxor          $T2,$Z3,$Z3
 232         vpclmulqdq      \$0x00,$Hkey,$Xi,$T2
 233           vaesenc       $rndkey,$inout2,$inout2
 234         movbe           0x20($in0),%r12
 235         vpclmulqdq      \$0x11,$Hkey,$Xi,$Xi
 236           vaesenc       $rndkey,$inout3,$inout3
 237         mov             %r13,0x50+8(%rsp)
 238           vaesenc       $rndkey,$inout4,$inout4
 239         mov             %r12,0x58+8(%rsp)
 240         vpxor           $Z1,$Z2,$Z2
 241           vaesenc       $rndkey,$inout5,$inout5
 242         vpxor           $T1,$Z2,$Z2
 243
 244           vmovups       0x70-0x80($key),$rndkey
 245         vpslldq         \$8,$Z2,$Z1
 246         vpxor           $T2,$Z0,$Z0
 247         vmovdqu         0x10($const),$Hkey      # .Lpoly
 248
 249           vaesenc       $rndkey,$inout0,$inout0
 250         vpxor           $Xi,$Z3,$Z3
 251           vaesenc       $rndkey,$inout1,$inout1
 252         vpxor           $Z1,$Z0,$Z0
 253         movbe           0x18($in0),%r13
 254           vaesenc       $rndkey,$inout2,$inout2
 255         movbe           0x10($in0),%r12
 256         vpalignr        \$8,$Z0,$Z0,$Ii         # 1st phase
 257         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 258         mov             %r13,0x60+8(%rsp)
 259           vaesenc       $rndkey,$inout3,$inout3
 260         mov             %r12,0x68+8(%rsp)
 261           vaesenc       $rndkey,$inout4,$inout4
 262           vmovups       0x80-0x80($key),$T1     # borrow $T1 for $rndkey
 263           vaesenc       $rndkey,$inout5,$inout5
 264
 265           vaesenc       $T1,$inout0,$inout0
 266           vmovups       0x90-0x80($key),$rndkey
 267           vaesenc       $T1,$inout1,$inout1
 268         vpsrldq         \$8,$Z2,$Z2
 269           vaesenc       $T1,$inout2,$inout2
 270         vpxor           $Z2,$Z3,$Z3
 271           vaesenc       $T1,$inout3,$inout3
 272         vpxor           $Ii,$Z0,$Z0
 273         movbe           0x08($in0),%r13
 274           vaesenc       $T1,$inout4,$inout4
 275         movbe           0x00($in0),%r12
 276           vaesenc       $T1,$inout5,$inout5
 277           vmovups       0xa0-0x80($key),$T1
 278           cmp           \$11,$rounds
 279           jb            .Lenc_tail              # 128-bit key
 280
 281           vaesenc       $rndkey,$inout0,$inout0
 282           vaesenc       $rndkey,$inout1,$inout1
 283           vaesenc       $rndkey,$inout2,$inout2
 284           vaesenc       $rndkey,$inout3,$inout3
 285           vaesenc       $rndkey,$inout4,$inout4
 286           vaesenc       $rndkey,$inout5,$inout5
 287
 288           vaesenc       $T1,$inout0,$inout0
 289           vaesenc       $T1,$inout1,$inout1
 290           vaesenc       $T1,$inout2,$inout2
 291           vaesenc       $T1,$inout3,$inout3
 292           vaesenc       $T1,$inout4,$inout4
 293           vmovups       0xb0-0x80($key),$rndkey
 294           vaesenc       $T1,$inout5,$inout5
 295           vmovups       0xc0-0x80($key),$T1
 296           je            .Lenc_tail              # 192-bit key
 297
 298           vaesenc       $rndkey,$inout0,$inout0
 299           vaesenc       $rndkey,$inout1,$inout1
 300           vaesenc       $rndkey,$inout2,$inout2
 301           vaesenc       $rndkey,$inout3,$inout3
 302           vaesenc       $rndkey,$inout4,$inout4
 303           vaesenc       $rndkey,$inout5,$inout5
 304
 305           vaesenc       $T1,$inout0,$inout0
 306           vaesenc       $T1,$inout1,$inout1
 307           vaesenc       $T1,$inout2,$inout2
 308           vaesenc       $T1,$inout3,$inout3
 309           vaesenc       $T1,$inout4,$inout4
 310           vmovups       0xd0-0x80($key),$rndkey
 311           vaesenc       $T1,$inout5,$inout5
 312           vmovups       0xe0-0x80($key),$T1
 313           jmp           .Lenc_tail              # 256-bit key
 314
 315 .align  32
 316 .Lhandle_ctr32:
 317         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 318           vpshufb       $Ii,$T1,$Z2             # byte-swap counter
 319           vmovdqu       0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 320           vpaddd        0x40($const),$Z2,$inout1        # .Lone_lsb
 321           vpaddd        $Z1,$Z2,$inout2
 322         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 323           vpaddd        $Z1,$inout1,$inout3
 324           vpshufb       $Ii,$inout1,$inout1
 325           vpaddd        $Z1,$inout2,$inout4
 326           vpshufb       $Ii,$inout2,$inout2
 327           vpxor         $rndkey,$inout1,$inout1
 328           vpaddd        $Z1,$inout3,$inout5
 329           vpshufb       $Ii,$inout3,$inout3
 330           vpxor         $rndkey,$inout2,$inout2
 331           vpaddd        $Z1,$inout4,$T1         # byte-swapped next counter value
 332           vpshufb       $Ii,$inout4,$inout4
 333           vpshufb       $Ii,$inout5,$inout5
 334           vpshufb       $Ii,$T1,$T1             # next counter value
 335         jmp             .Lresume_ctr32
 336
 337 .align  32
 338 .Lenc_tail:
 339           vaesenc       $rndkey,$inout0,$inout0
 340         vmovdqu         $Z3,16+8(%rsp)          # postpone vpxor $Z3,$Xi,$Xi
 341         vpalignr        \$8,$Z0,$Z0,$Xi         # 2nd phase
 342           vaesenc       $rndkey,$inout1,$inout1
 343         vpclmulqdq      \$0x10,$Hkey,$Z0,$Z0
 344           vpxor         0x00($inp),$T1,$T2
 345           vaesenc       $rndkey,$inout2,$inout2
 346           vpxor         0x10($inp),$T1,$Ii
 347           vaesenc       $rndkey,$inout3,$inout3
 348           vpxor         0x20($inp),$T1,$Z1
 349           vaesenc       $rndkey,$inout4,$inout4
 350           vpxor         0x30($inp),$T1,$Z2
 351           vaesenc       $rndkey,$inout5,$inout5
 352           vpxor         0x40($inp),$T1,$Z3
 353           vpxor         0x50($inp),$T1,$Hkey
 354           vmovdqu       ($ivp),$T1              # load next counter value
 355
 356           vaesenclast   $T2,$inout0,$inout0
 357           vmovdqu       0x20($const),$T2        # borrow $T2, .Lone_msb
 358           vaesenclast   $Ii,$inout1,$inout1
 359          vpaddb         $T2,$T1,$Ii
 360         mov             %r13,0x70+8(%rsp)
 361         lea             0x60($inp),$inp
 362           vaesenclast   $Z1,$inout2,$inout2
 363          vpaddb         $T2,$Ii,$Z1
 364         mov             %r12,0x78+8(%rsp)
 365         lea             0x60($out),$out
 366           vmovdqu       0x00-0x80($key),$rndkey
 367           vaesenclast   $Z2,$inout3,$inout3
 368          vpaddb         $T2,$Z1,$Z2
 369           vaesenclast   $Z3, $inout4,$inout4
 370          vpaddb         $T2,$Z2,$Z3
 371           vaesenclast   $Hkey,$inout5,$inout5
 372          vpaddb         $T2,$Z3,$Hkey
 373
 374         add             \$0x60,$ret
 375         sub             \$0x6,$len
 376         jc              .L6x_done
 377
 378           vmovups       $inout0,-0x60($out)     # save output
 379          vpxor          $rndkey,$T1,$inout0
 380           vmovups       $inout1,-0x50($out)
 381          vmovdqa        $Ii,$inout1             # 0 latency
 382           vmovups       $inout2,-0x40($out)
 383          vmovdqa        $Z1,$inout2             # 0 latency
 384           vmovups       $inout3,-0x30($out)
 385          vmovdqa        $Z2,$inout3             # 0 latency
 386           vmovups       $inout4,-0x20($out)
 387          vmovdqa        $Z3,$inout4             # 0 latency
 388           vmovups       $inout5,-0x10($out)
 389          vmovdqa        $Hkey,$inout5           # 0 latency
 390         vmovdqu         0x20+8(%rsp),$Z3        # I[5]
 391         jmp             .Loop6x
 392
 393 .L6x_done:
 394         vpxor           16+8(%rsp),$Xi,$Xi      # modulo-scheduled
 395         vpxor           $Z0,$Xi,$Xi             # modulo-scheduled
 396
 397         ret
 398 .size   _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x
 399 ___
 400 ######################################################################
 401 #
 402 # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len,
 403 #               const AES_KEY *key, unsigned char iv[16],
 404 #               struct { u128 Xi,H,Htbl[9]; } *Xip);
 405 $code.=<<___;
 406 .globl  aesni_gcm_decrypt
 407 .type   aesni_gcm_decrypt,\@function,6
 408 .align  32
 409 aesni_gcm_decrypt:
 410         xor     $ret,$ret
 411         cmp     \$0x60,$len                     # minimal accepted length
 412         jb      .Lgcm_dec_abort
 413
 414         lea     (%rsp),%rax                     # save stack pointer
 415         push    %rbx
 416         push    %rbp
 417         push    %r12
 418         push    %r13
 419         push    %r14
 420         push    %r15
 421 ___
 422 $code.=<<___ if ($win64);
 423         lea     -0xa8(%rsp),%rsp
 424         movaps  %xmm6,-0xd8(%rax)
 425         movaps  %xmm7,-0xc8(%rax)
 426         movaps  %xmm8,-0xb8(%rax)
 427         movaps  %xmm9,-0xa8(%rax)
 428         movaps  %xmm10,-0x98(%rax)
 429         movaps  %xmm11,-0x88(%rax)
 430         movaps  %xmm12,-0x78(%rax)
 431         movaps  %xmm13,-0x68(%rax)
 432         movaps  %xmm14,-0x58(%rax)
 433         movaps  %xmm15,-0x48(%rax)
 434 .Lgcm_dec_body:
 435 ___
 436 $code.=<<___;
 437         vzeroupper
 438
 439         vmovdqu         ($ivp),$T1              # input counter value
 440         add             \$-128,%rsp
 441         mov             12($ivp),$counter
 442         lea             .Lbswap_mask(%rip),$const
 443         lea             -0x80($key),$in0        # borrow $in0
 444         mov             \$0xf80,$end0           # borrow $end0
 445         vmovdqu         ($Xip),$Xi              # load Xi
 446         and             \$-128,%rsp             # ensure stack alignment
 447         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 448         lea             0x80($key),$key         # size optimization
 449         lea             0x20+0x20($Xip),$Xip    # size optimization
 450         mov             0xf0-0x80($key),$rounds
 451         vpshufb         $Ii,$Xi,$Xi
 452
 453         and             $end0,$in0
 454         and             %rsp,$end0
 455         sub             $in0,$end0
 456         jc              .Ldec_no_key_aliasing
 457         cmp             \$768,$end0
 458         jnc             .Ldec_no_key_aliasing
 459         sub             $end0,%rsp              # avoid aliasing with key
 460 .Ldec_no_key_aliasing:
 461
 462         vmovdqu         0x50($inp),$Z3          # I[5]
 463         lea             ($inp),$in0
 464         vmovdqu         0x40($inp),$Z0
 465         lea             -0xc0($inp,$len),$end0
 466         vmovdqu         0x30($inp),$Z1
 467         shr             \$4,$len
 468         xor             $ret,$ret
 469         vmovdqu         0x20($inp),$Z2
 470          vpshufb        $Ii,$Z3,$Z3             # passed to _aesni_ctr32_ghash_6x
 471         vmovdqu         0x10($inp),$T2
 472          vpshufb        $Ii,$Z0,$Z0
 473         vmovdqu         ($inp),$Hkey
 474          vpshufb        $Ii,$Z1,$Z1
 475         vmovdqu         $Z0,0x30(%rsp)
 476          vpshufb        $Ii,$Z2,$Z2
 477         vmovdqu         $Z1,0x40(%rsp)
 478          vpshufb        $Ii,$T2,$T2
 479         vmovdqu         $Z2,0x50(%rsp)
 480          vpshufb        $Ii,$Hkey,$Hkey
 481         vmovdqu         $T2,0x60(%rsp)
 482         vmovdqu         $Hkey,0x70(%rsp)
 483
 484         call            _aesni_ctr32_ghash_6x
 485
 486         vmovups         $inout0,-0x60($out)     # save output
 487         vmovups         $inout1,-0x50($out)
 488         vmovups         $inout2,-0x40($out)
 489         vmovups         $inout3,-0x30($out)
 490         vmovups         $inout4,-0x20($out)
 491         vmovups         $inout5,-0x10($out)
 492
 493         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 494         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 495
 496         vzeroupper
 497 ___
 498 $code.=<<___ if ($win64);
 499         movaps  -0xd8(%rax),%xmm6
 500         movaps  -0xc8(%rax),%xmm7
 501         movaps  -0xb8(%rax),%xmm8
 502         movaps  -0xa8(%rax),%xmm9
 503         movaps  -0x98(%rax),%xmm10
 504         movaps  -0x88(%rax),%xmm11
 505         movaps  -0x78(%rax),%xmm12
 506         movaps  -0x68(%rax),%xmm13
 507         movaps  -0x58(%rax),%xmm14
 508         movaps  -0x48(%rax),%xmm15
 509 ___
 510 $code.=<<___;
 511         mov     -48(%rax),%r15
 512         mov     -40(%rax),%r14
 513         mov     -32(%rax),%r13
 514         mov     -24(%rax),%r12
 515         mov     -16(%rax),%rbp
 516         mov     -8(%rax),%rbx
 517         lea     (%rax),%rsp             # restore %rsp
 518 .Lgcm_dec_abort:
 519         mov     $ret,%rax               # return value
 520         ret
 521 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
 522 ___
 523
 524 $code.=<<___;
 525 .type   _aesni_ctr32_6x,\@abi-omnipotent
 526 .align  32
 527 _aesni_ctr32_6x:
 528         vmovdqu         0x00-0x80($key),$Z0     # borrow $Z0 for $rndkey
 529         vmovdqu         0x20($const),$T2        # borrow $T2, .Lone_msb
 530         lea             -1($rounds),%r13
 531         vmovups         0x10-0x80($key),$rndkey
 532         lea             0x20-0x80($key),%r12
 533         vpxor           $Z0,$T1,$inout0
 534         add             \$`6<<24`,$counter
 535         jc              .Lhandle_ctr32_2
 536         vpaddb          $T2,$T1,$inout1
 537         vpaddb          $T2,$inout1,$inout2
 538         vpxor           $Z0,$inout1,$inout1
 539         vpaddb          $T2,$inout2,$inout3
 540         vpxor           $Z0,$inout2,$inout2
 541         vpaddb          $T2,$inout3,$inout4
 542         vpxor           $Z0,$inout3,$inout3
 543         vpaddb          $T2,$inout4,$inout5
 544         vpxor           $Z0,$inout4,$inout4
 545         vpaddb          $T2,$inout5,$T1
 546         vpxor           $Z0,$inout5,$inout5
 547         jmp             .Loop_ctr32
 548
 549 .align  16
 550 .Loop_ctr32:
 551         vaesenc         $rndkey,$inout0,$inout0
 552         vaesenc         $rndkey,$inout1,$inout1
 553         vaesenc         $rndkey,$inout2,$inout2
 554         vaesenc         $rndkey,$inout3,$inout3
 555         vaesenc         $rndkey,$inout4,$inout4
 556         vaesenc         $rndkey,$inout5,$inout5
 557         vmovups         (%r12),$rndkey
 558         lea             0x10(%r12),%r12
 559         dec             %r13d
 560         jnz             .Loop_ctr32
 561
 562         vmovdqu         (%r12),$Hkey            # last round key
 563         vaesenc         $rndkey,$inout0,$inout0
 564         vpxor           0x00($inp),$Hkey,$Z0
 565         vaesenc         $rndkey,$inout1,$inout1
 566         vpxor           0x10($inp),$Hkey,$Z1
 567         vaesenc         $rndkey,$inout2,$inout2
 568         vpxor           0x20($inp),$Hkey,$Z2
 569         vaesenc         $rndkey,$inout3,$inout3
 570         vpxor           0x30($inp),$Hkey,$Xi
 571         vaesenc         $rndkey,$inout4,$inout4
 572         vpxor           0x40($inp),$Hkey,$T2
 573         vaesenc         $rndkey,$inout5,$inout5
 574         vpxor           0x50($inp),$Hkey,$Hkey
 575         lea             0x60($inp),$inp
 576
 577         vaesenclast     $Z0,$inout0,$inout0
 578         vaesenclast     $Z1,$inout1,$inout1
 579         vaesenclast     $Z2,$inout2,$inout2
 580         vaesenclast     $Xi,$inout3,$inout3
 581         vaesenclast     $T2,$inout4,$inout4
 582         vaesenclast     $Hkey,$inout5,$inout5
 583         vmovups         $inout0,0x00($out)
 584         vmovups         $inout1,0x10($out)
 585         vmovups         $inout2,0x20($out)
 586         vmovups         $inout3,0x30($out)
 587         vmovups         $inout4,0x40($out)
 588         vmovups         $inout5,0x50($out)
 589         lea             0x60($out),$out
 590
 591         ret
 592 .align  32
 593 .Lhandle_ctr32_2:
 594         vpshufb         $Ii,$T1,$Z2             # byte-swap counter
 595         vmovdqu         0x30($const),$Z1        # borrow $Z1, .Ltwo_lsb
 596         vpaddd          0x40($const),$Z2,$inout1        # .Lone_lsb
 597         vpaddd          $Z1,$Z2,$inout2
 598         vpaddd          $Z1,$inout1,$inout3
 599         vpshufb         $Ii,$inout1,$inout1
 600         vpaddd          $Z1,$inout2,$inout4
 601         vpshufb         $Ii,$inout2,$inout2
 602         vpxor           $Z0,$inout1,$inout1
 603         vpaddd          $Z1,$inout3,$inout5
 604         vpshufb         $Ii,$inout3,$inout3
 605         vpxor           $Z0,$inout2,$inout2
 606         vpaddd          $Z1,$inout4,$T1         # byte-swapped next counter value
 607         vpshufb         $Ii,$inout4,$inout4
 608         vpxor           $Z0,$inout3,$inout3
 609         vpshufb         $Ii,$inout5,$inout5
 610         vpxor           $Z0,$inout4,$inout4
 611         vpshufb         $Ii,$T1,$T1             # next counter value
 612         vpxor           $Z0,$inout5,$inout5
 613         jmp     .Loop_ctr32
 614 .size   _aesni_ctr32_6x,.-_aesni_ctr32_6x
 615
 616 .globl  aesni_gcm_encrypt
 617 .type   aesni_gcm_encrypt,\@function,6
 618 .align  32
 619 aesni_gcm_encrypt:
 620         xor     $ret,$ret
 621         cmp     \$0x60*3,$len                   # minimal accepted length
 622         jb      .Lgcm_enc_abort
 623
 624         lea     (%rsp),%rax                     # save stack pointer
 625         push    %rbx
 626         push    %rbp
 627         push    %r12
 628         push    %r13
 629         push    %r14
 630         push    %r15
 631 ___
 632 $code.=<<___ if ($win64);
 633         lea     -0xa8(%rsp),%rsp
 634         movaps  %xmm6,-0xd8(%rax)
 635         movaps  %xmm7,-0xc8(%rax)
 636         movaps  %xmm8,-0xb8(%rax)
 637         movaps  %xmm9,-0xa8(%rax)
 638         movaps  %xmm10,-0x98(%rax)
 639         movaps  %xmm11,-0x88(%rax)
 640         movaps  %xmm12,-0x78(%rax)
 641         movaps  %xmm13,-0x68(%rax)
 642         movaps  %xmm14,-0x58(%rax)
 643         movaps  %xmm15,-0x48(%rax)
 644 .Lgcm_enc_body:
 645 ___
 646 $code.=<<___;
 647         vzeroupper
 648
 649         vmovdqu         ($ivp),$T1              # input counter value
 650         add             \$-128,%rsp
 651         mov             12($ivp),$counter
 652         lea             .Lbswap_mask(%rip),$const
 653         lea             -0x80($key),$in0        # borrow $in0
 654         mov             \$0xf80,$end0           # borrow $end0
 655         lea             0x80($key),$key         # size optimization
 656         vmovdqu         ($const),$Ii            # borrow $Ii for .Lbswap_mask
 657         and             \$-128,%rsp             # ensure stack alignment
 658         mov             0xf0-0x80($key),$rounds
 659
 660         and             $end0,$in0
 661         and             %rsp,$end0
 662         sub             $in0,$end0
 663         jc              .Lenc_no_key_aliasing
 664         cmp             \$768,$end0
 665         jnc             .Lenc_no_key_aliasing
 666         sub             $end0,%rsp              # avoid aliasing with key
 667 .Lenc_no_key_aliasing:
 668
 669         lea             ($out),$in0
 670         lea             -0xc0($out,$len),$end0
 671         shr             \$4,$len
 672
 673         call            _aesni_ctr32_6x
 674         vpshufb         $Ii,$inout0,$Xi         # save bswapped output on stack
 675         vpshufb         $Ii,$inout1,$T2
 676         vmovdqu         $Xi,0x70(%rsp)
 677         vpshufb         $Ii,$inout2,$Z0
 678         vmovdqu         $T2,0x60(%rsp)
 679         vpshufb         $Ii,$inout3,$Z1
 680         vmovdqu         $Z0,0x50(%rsp)
 681         vpshufb         $Ii,$inout4,$Z2
 682         vmovdqu         $Z1,0x40(%rsp)
 683         vpshufb         $Ii,$inout5,$Z3         # passed to _aesni_ctr32_ghash_6x
 684         vmovdqu         $Z2,0x30(%rsp)
 685
 686         call            _aesni_ctr32_6x
 687
 688         vmovdqu         ($Xip),$Xi              # load Xi
 689         lea             0x20+0x20($Xip),$Xip    # size optimization
 690         sub             \$12,$len
 691         mov             \$0x60*2,$ret
 692         vpshufb         $Ii,$Xi,$Xi
 693
 694         call            _aesni_ctr32_ghash_6x
 695         vmovdqu         0x20(%rsp),$Z3          # I[5]
 696          vmovdqu        ($const),$Ii            # borrow $Ii for .Lbswap_mask
 697         vmovdqu         0x00-0x20($Xip),$Hkey   # $Hkey^1
 698         vpunpckhqdq     $Z3,$Z3,$T1
 699         vmovdqu         0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK
 700          vmovups        $inout0,-0x60($out)     # save output
 701          vpshufb        $Ii,$inout0,$inout0     # but keep bswapped copy
 702         vpxor           $Z3,$T1,$T1
 703          vmovups        $inout1,-0x50($out)
 704          vpshufb        $Ii,$inout1,$inout1
 705          vmovups        $inout2,-0x40($out)
 706          vpshufb        $Ii,$inout2,$inout2
 707          vmovups        $inout3,-0x30($out)
 708          vpshufb        $Ii,$inout3,$inout3
 709          vmovups        $inout4,-0x20($out)
 710          vpshufb        $Ii,$inout4,$inout4
 711          vmovups        $inout5,-0x10($out)
 712          vpshufb        $Ii,$inout5,$inout5
 713          vmovdqu        $inout0,0x10(%rsp)      # free $inout0
 714 ___
 715 { my ($HK,$T3)=($rndkey,$inout0);
 716
 717 $code.=<<___;
 718          vmovdqu        0x30(%rsp),$Z2          # I[4]
 719          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 720          vpunpckhqdq    $Z2,$Z2,$T2
 721         vpclmulqdq      \$0x00,$Hkey,$Z3,$Z1
 722          vpxor          $Z2,$T2,$T2
 723         vpclmulqdq      \$0x11,$Hkey,$Z3,$Z3
 724         vpclmulqdq      \$0x00,$HK,$T1,$T1
 725
 726          vmovdqu        0x40(%rsp),$T3          # I[3]
 727         vpclmulqdq      \$0x00,$Ii,$Z2,$Z0
 728          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 729         vpxor           $Z1,$Z0,$Z0
 730          vpunpckhqdq    $T3,$T3,$Z1
 731         vpclmulqdq      \$0x11,$Ii,$Z2,$Z2
 732          vpxor          $T3,$Z1,$Z1
 733         vpxor           $Z3,$Z2,$Z2
 734         vpclmulqdq      \$0x10,$HK,$T2,$T2
 735          vmovdqu        0x50-0x20($Xip),$HK
 736         vpxor           $T1,$T2,$T2
 737
 738          vmovdqu        0x50(%rsp),$T1          # I[2]
 739         vpclmulqdq      \$0x00,$Hkey,$T3,$Z3
 740          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 741         vpxor           $Z0,$Z3,$Z3
 742          vpunpckhqdq    $T1,$T1,$Z0
 743         vpclmulqdq      \$0x11,$Hkey,$T3,$T3
 744          vpxor          $T1,$Z0,$Z0
 745         vpxor           $Z2,$T3,$T3
 746         vpclmulqdq      \$0x00,$HK,$Z1,$Z1
 747         vpxor           $T2,$Z1,$Z1
 748
 749          vmovdqu        0x60(%rsp),$T2          # I[1]
 750         vpclmulqdq      \$0x00,$Ii,$T1,$Z2
 751          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 752         vpxor           $Z3,$Z2,$Z2
 753          vpunpckhqdq    $T2,$T2,$Z3
 754         vpclmulqdq      \$0x11,$Ii,$T1,$T1
 755          vpxor          $T2,$Z3,$Z3
 756         vpxor           $T3,$T1,$T1
 757         vpclmulqdq      \$0x10,$HK,$Z0,$Z0
 758          vmovdqu        0x80-0x20($Xip),$HK
 759         vpxor           $Z1,$Z0,$Z0
 760
 761          vpxor          0x70(%rsp),$Xi,$Xi      # accumulate I[0]
 762         vpclmulqdq      \$0x00,$Hkey,$T2,$Z1
 763          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 764          vpunpckhqdq    $Xi,$Xi,$T3
 765         vpxor           $Z2,$Z1,$Z1
 766         vpclmulqdq      \$0x11,$Hkey,$T2,$T2
 767          vpxor          $Xi,$T3,$T3
 768         vpxor           $T1,$T2,$T2
 769         vpclmulqdq      \$0x00,$HK,$Z3,$Z3
 770         vpxor           $Z0,$Z3,$Z0
 771
 772         vpclmulqdq      \$0x00,$Ii,$Xi,$Z2
 773          vmovdqu        0x00-0x20($Xip),$Hkey   # $Hkey^1
 774          vpunpckhqdq    $inout5,$inout5,$T1
 775         vpclmulqdq      \$0x11,$Ii,$Xi,$Xi
 776          vpxor          $inout5,$T1,$T1
 777         vpxor           $Z1,$Z2,$Z1
 778         vpclmulqdq      \$0x10,$HK,$T3,$T3
 779          vmovdqu        0x20-0x20($Xip),$HK
 780         vpxor           $T2,$Xi,$Z3
 781         vpxor           $Z0,$T3,$Z2
 782
 783          vmovdqu        0x10-0x20($Xip),$Ii     # borrow $Ii for $Hkey^2
 784           vpxor         $Z1,$Z3,$T3             # aggregated Karatsuba post-processing
 785         vpclmulqdq      \$0x00,$Hkey,$inout5,$Z0
 786           vpxor         $T3,$Z2,$Z2
 787          vpunpckhqdq    $inout4,$inout4,$T2
 788         vpclmulqdq      \$0x11,$Hkey,$inout5,$inout5
 789          vpxor          $inout4,$T2,$T2
 790           vpslldq       \$8,$Z2,$T3
 791         vpclmulqdq      \$0x00,$HK,$T1,$T1
 792           vpxor         $T3,$Z1,$Xi
 793           vpsrldq       \$8,$Z2,$Z2
 794           vpxor         $Z2,$Z3,$Z3
 795
 796         vpclmulqdq      \$0x00,$Ii,$inout4,$Z1
 797          vmovdqu        0x30-0x20($Xip),$Hkey   # $Hkey^3
 798         vpxor           $Z0,$Z1,$Z1
 799          vpunpckhqdq    $inout3,$inout3,$T3
 800         vpclmulqdq      \$0x11,$Ii,$inout4,$inout4
 801          vpxor          $inout3,$T3,$T3
 802         vpxor           $inout5,$inout4,$inout4
 803           vpalignr      \$8,$Xi,$Xi,$inout5     # 1st phase
 804         vpclmulqdq      \$0x10,$HK,$T2,$T2
 805          vmovdqu        0x50-0x20($Xip),$HK
 806         vpxor           $T1,$T2,$T2
 807
 808         vpclmulqdq      \$0x00,$Hkey,$inout3,$Z0
 809          vmovdqu        0x40-0x20($Xip),$Ii     # borrow $Ii for $Hkey^4
 810         vpxor           $Z1,$Z0,$Z0
 811          vpunpckhqdq    $inout2,$inout2,$T1
 812         vpclmulqdq      \$0x11,$Hkey,$inout3,$inout3
 813          vpxor          $inout2,$T1,$T1
 814         vpxor           $inout4,$inout3,$inout3
 815           vxorps        0x10(%rsp),$Z3,$Z3      # accumulate $inout0
 816         vpclmulqdq      \$0x00,$HK,$T3,$T3
 817         vpxor           $T2,$T3,$T3
 818
 819           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 820           vxorps        $inout5,$Xi,$Xi
 821
 822         vpclmulqdq      \$0x00,$Ii,$inout2,$Z1
 823          vmovdqu        0x60-0x20($Xip),$Hkey   # $Hkey^5
 824         vpxor           $Z0,$Z1,$Z1
 825          vpunpckhqdq    $inout1,$inout1,$T2
 826         vpclmulqdq      \$0x11,$Ii,$inout2,$inout2
 827          vpxor          $inout1,$T2,$T2
 828           vpalignr      \$8,$Xi,$Xi,$inout5     # 2nd phase
 829         vpxor           $inout3,$inout2,$inout2
 830         vpclmulqdq      \$0x10,$HK,$T1,$T1
 831          vmovdqu        0x80-0x20($Xip),$HK
 832         vpxor           $T3,$T1,$T1
 833
 834           vxorps        $Z3,$inout5,$inout5
 835           vpclmulqdq    \$0x10,0x10($const),$Xi,$Xi
 836           vxorps        $inout5,$Xi,$Xi
 837
 838         vpclmulqdq      \$0x00,$Hkey,$inout1,$Z0
 839          vmovdqu        0x70-0x20($Xip),$Ii     # borrow $Ii for $Hkey^6
 840         vpxor           $Z1,$Z0,$Z0
 841          vpunpckhqdq    $Xi,$Xi,$T3
 842         vpclmulqdq      \$0x11,$Hkey,$inout1,$inout1
 843          vpxor          $Xi,$T3,$T3
 844         vpxor           $inout2,$inout1,$inout1
 845         vpclmulqdq      \$0x00,$HK,$T2,$T2
 846         vpxor           $T1,$T2,$T2
 847
 848         vpclmulqdq      \$0x00,$Ii,$Xi,$Z1
 849         vpclmulqdq      \$0x11,$Ii,$Xi,$Z3
 850         vpxor           $Z0,$Z1,$Z1
 851         vpclmulqdq      \$0x10,$HK,$T3,$Z2
 852         vpxor           $inout1,$Z3,$Z3
 853         vpxor           $T2,$Z2,$Z2
 854
 855         vpxor           $Z1,$Z3,$Z0             # aggregated Karatsuba post-processing
 856         vpxor           $Z0,$Z2,$Z2
 857         vpslldq         \$8,$Z2,$T1
 858         vmovdqu         0x10($const),$Hkey      # .Lpoly
 859         vpsrldq         \$8,$Z2,$Z2
 860         vpxor           $T1,$Z1,$Xi
 861         vpxor           $Z2,$Z3,$Z3
 862
 863         vpalignr        \$8,$Xi,$Xi,$T2         # 1st phase
 864         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 865         vpxor           $T2,$Xi,$Xi
 866
 867         vpalignr        \$8,$Xi,$Xi,$T2         # 2nd phase
 868         vpclmulqdq      \$0x10,$Hkey,$Xi,$Xi
 869         vpxor           $Z3,$T2,$T2
 870         vpxor           $T2,$Xi,$Xi
 871 ___
 872 }
 873 $code.=<<___;
 874         vpshufb         ($const),$Xi,$Xi        # .Lbswap_mask
 875         vmovdqu         $Xi,-0x40($Xip)         # output Xi
 876
 877         vzeroupper
 878 ___
 879 $code.=<<___ if ($win64);
 880         movaps  -0xd8(%rax),%xmm6
 881         movaps  -0xc8(%rax),%xmm7
 882         movaps  -0xb8(%rax),%xmm8
 883         movaps  -0xa8(%rax),%xmm9
 884         movaps  -0x98(%rax),%xmm10
 885         movaps  -0x88(%rax),%xmm11
 886         movaps  -0x78(%rax),%xmm12
 887         movaps  -0x68(%rax),%xmm13
 888         movaps  -0x58(%rax),%xmm14
 889         movaps  -0x48(%rax),%xmm15
 890 ___
 891 $code.=<<___;
 892         mov     -48(%rax),%r15
 893         mov     -40(%rax),%r14
 894         mov     -32(%rax),%r13
 895         mov     -24(%rax),%r12
 896         mov     -16(%rax),%rbp
 897         mov     -8(%rax),%rbx
 898         lea     (%rax),%rsp             # restore %rsp
 899 .Lgcm_enc_abort:
 900         mov     $ret,%rax               # return value
 901         ret
 902 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
 903 ___
 904
 905 $code.=<<___;
 906 .align  64
 907 .Lbswap_mask:
 908         .byte   15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0
 909 .Lpoly:
 910         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2
 911 .Lone_msb:
 912         .byte   0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1
 913 .Ltwo_lsb:
 914         .byte   2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 915 .Lone_lsb:
 916         .byte   1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
 917 .asciz  "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 918 .align  64
 919 ___
 920 if ($win64) {
 921 $rec="%rcx";
 922 $frame="%rdx";
 923 $context="%r8";
 924 $disp="%r9";
 925
 926 $code.=<<___
 927 .extern __imp_RtlVirtualUnwind
 928 .type   gcm_se_handler,\@abi-omnipotent
 929 .align  16
 930 gcm_se_handler:
 931         push    %rsi
 932         push    %rdi
 933         push    %rbx
 934         push    %rbp
 935         push    %r12
 936         push    %r13
 937         push    %r14
 938         push    %r15
 939         pushfq
 940         sub     \$64,%rsp
 941
 942         mov     120($context),%rax      # pull context->Rax
 943         mov     248($context),%rbx      # pull context->Rip
 944
 945         mov     8($disp),%rsi           # disp->ImageBase
 946         mov     56($disp),%r11          # disp->HandlerData
 947
 948         mov     0(%r11),%r10d           # HandlerData[0]
 949         lea     (%rsi,%r10),%r10        # prologue label
 950         cmp     %r10,%rbx               # context->Rip<prologue label
 951         jb      .Lcommon_seh_tail
 952
 953         mov     152($context),%rax      # pull context->Rsp
 954
 955         mov     4(%r11),%r10d           # HandlerData[1]
 956         lea     (%rsi,%r10),%r10        # epilogue label
 957         cmp     %r10,%rbx               # context->Rip>=epilogue label
 958         jae     .Lcommon_seh_tail
 959
 960         mov     120($context),%rax      # pull context->Rax
 961
 962         mov     -48(%rax),%r15
 963         mov     -40(%rax),%r14
 964         mov     -32(%rax),%r13
 965         mov     -24(%rax),%r12
 966         mov     -16(%rax),%rbp
 967         mov     -8(%rax),%rbx
 968         mov     %r15,240($context)
 969         mov     %r14,232($context)
 970         mov     %r13,224($context)
 971         mov     %r12,216($context)
 972         mov     %rbp,160($context)
 973         mov     %rbx,144($context)
 974
 975         lea     -0xd8(%rax),%rsi        # %xmm save area
 976         lea     512($context),%rdi      # & context.Xmm6
 977         mov     \$20,%ecx               # 10*sizeof(%xmm0)/sizeof(%rax)
 978         .long   0xa548f3fc              # cld; rep movsq
 979
 980 .Lcommon_seh_tail:
 981         mov     8(%rax),%rdi
 982         mov     16(%rax),%rsi
 983         mov     %rax,152($context)      # restore context->Rsp
 984         mov     %rsi,168($context)      # restore context->Rsi
 985         mov     %rdi,176($context)      # restore context->Rdi
 986
 987         mov     40($disp),%rdi          # disp->ContextRecord
 988         mov     $context,%rsi           # context
 989         mov     \$154,%ecx              # sizeof(CONTEXT)
 990         .long   0xa548f3fc              # cld; rep movsq
 991
 992         mov     $disp,%rsi
 993         xor     %rcx,%rcx               # arg1, UNW_FLAG_NHANDLER
 994         mov     8(%rsi),%rdx            # arg2, disp->ImageBase
 995         mov     0(%rsi),%r8             # arg3, disp->ControlPc
 996         mov     16(%rsi),%r9            # arg4, disp->FunctionEntry
 997         mov     40(%rsi),%r10           # disp->ContextRecord
 998         lea     56(%rsi),%r11           # &disp->HandlerData
 999         lea     24(%rsi),%r12           # &disp->EstablisherFrame
1000         mov     %r10,32(%rsp)           # arg5
1001         mov     %r11,40(%rsp)           # arg6
1002         mov     %r12,48(%rsp)           # arg7
1003         mov     %rcx,56(%rsp)           # arg8, (NULL)
1004         call    *__imp_RtlVirtualUnwind(%rip)
1005
1006         mov     \$1,%eax                # ExceptionContinueSearch
1007         add     \$64,%rsp
1008         popfq
1009         pop     %r15
1010         pop     %r14
1011         pop     %r13
1012         pop     %r12
1013         pop     %rbp
1014         pop     %rbx
1015         pop     %rdi
1016         pop     %rsi
1017         ret
1018 .size   gcm_se_handler,.-gcm_se_handler
1019
1020 .section        .pdata
1021 .align  4
1022         .rva    .LSEH_begin_aesni_gcm_decrypt
1023         .rva    .LSEH_end_aesni_gcm_decrypt
1024         .rva    .LSEH_gcm_dec_info
1025
1026         .rva    .LSEH_begin_aesni_gcm_encrypt
1027         .rva    .LSEH_end_aesni_gcm_encrypt
1028         .rva    .LSEH_gcm_enc_info
1029 .section        .xdata
1030 .align  8
1031 .LSEH_gcm_dec_info:
1032         .byte   9,0,0,0
1033         .rva    gcm_se_handler
1034         .rva    .Lgcm_dec_body,.Lgcm_dec_abort
1035 .LSEH_gcm_enc_info:
1036         .byte   9,0,0,0
1037         .rva    gcm_se_handler
1038         .rva    .Lgcm_enc_body,.Lgcm_enc_abort
1039 ___
1040 }
1041 }}} else {{{
1042 $code=<<___;    # assembler is too old
1043 .text
1044
1045 .globl  aesni_gcm_encrypt
1046 .type   aesni_gcm_encrypt,\@abi-omnipotent
1047 aesni_gcm_encrypt:
1048         xor     %eax,%eax
1049         ret
1050 .size   aesni_gcm_encrypt,.-aesni_gcm_encrypt
1051
1052 .globl  aesni_gcm_decrypt
1053 .type   aesni_gcm_decrypt,\@abi-omnipotent
1054 aesni_gcm_decrypt:
1055         xor     %eax,%eax
1056         ret
1057 .size   aesni_gcm_decrypt,.-aesni_gcm_decrypt
1058 ___
1059 }}}
1060
1061 $code =~ s/\`([^\`]*)\`/eval($1)/gem;
1062
1063 print $code;
1064
1065 close STDOUT;