lib/accelerated/x86/macosx/padlock-x86-macosx.s

   1 # Copyright (c) 2011-2012, Andy Polyakov <appro@openssl.org>
   2 # All rights reserved.
   3 #
   4 # Redistribution and use in source and binary forms, with or without
   5 # modification, are permitted provided that the following conditions
   6 # are met:
   7 #
   8 #     * Redistributions of source code must retain copyright notices,
   9 #      this list of conditions and the following disclaimer.
  10 #
  11 #     * Redistributions in binary form must reproduce the above
  12 #      copyright notice, this list of conditions and the following
  13 #      disclaimer in the documentation and/or other materials
  14 #      provided with the distribution.
  15 #
  16 #     * Neither the name of the Andy Polyakov nor the names of its
  17 #      copyright holder and contributors may be used to endorse or
  18 #      promote products derived from this software without specific
  19 #      prior written permission.
  20 #
  21 # ALTERNATIVELY, provided that this notice is retained in full, this
  22 # product may be distributed under the terms of the GNU General Public
  23 # License (GPL), in which case the provisions of the GPL apply INSTEAD OF
  24 # those given above.
  25 #
  26 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS
  27 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  28 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  29 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  30 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  31 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  32 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  33 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  34 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  35 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  36 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  37 #
  38 # *** This file is auto-generated ***
  39 #
  40 .file   "devel/perlasm/e_padlock-x86.s"
  41 .text
  42 .globl  _padlock_capability
  43 .align  4
  44 _padlock_capability:
  45 L_padlock_capability_begin:
  46         pushl   %ebx
  47         pushfl
  48         popl    %eax
  49         movl    %eax,%ecx
  50         xorl    $2097152,%eax
  51         pushl   %eax
  52         popfl
  53         pushfl
  54         popl    %eax
  55         xorl    %eax,%ecx
  56         xorl    %eax,%eax
  57         btl     $21,%ecx
  58         jnc     L000noluck
  59         .byte   0x0f,0xa2
  60         xorl    %eax,%eax
  61         cmpl    $0x746e6543,%ebx
  62         jne     L000noluck
  63         cmpl    $0x48727561,%edx
  64         jne     L000noluck
  65         cmpl    $0x736c7561,%ecx
  66         jne     L000noluck
  67         movl    $3221225472,%eax
  68         .byte   0x0f,0xa2
  69         movl    %eax,%edx
  70         xorl    %eax,%eax
  71         cmpl    $3221225473,%edx
  72         jb      L000noluck
  73         movl    $1,%eax
  74         .byte   0x0f,0xa2
  75         orl     $15,%eax
  76         xorl    %ebx,%ebx
  77         andl    $4095,%eax
  78         cmpl    $1791,%eax
  79         sete    %bl
  80         movl    $3221225473,%eax
  81         pushl   %ebx
  82         .byte   0x0f,0xa2
  83         popl    %ebx
  84         movl    %edx,%eax
  85         shll    $4,%ebx
  86         andl    $4294967279,%eax
  87         orl     %ebx,%eax
  88 L000noluck:
  89         popl    %ebx
  90         ret
  91 .globl  _padlock_key_bswap
  92 .align  4
  93 _padlock_key_bswap:
  94 L_padlock_key_bswap_begin:
  95         movl    4(%esp),%edx
  96         movl    240(%edx),%ecx
  97 L001bswap_loop:
  98         movl    (%edx),%eax
  99         bswap   %eax
 100         movl    %eax,(%edx)
 101         leal    4(%edx),%edx
 102         subl    $1,%ecx
 103         jnz     L001bswap_loop
 104         ret
 105 .globl  _padlock_verify_context
 106 .align  4
 107 _padlock_verify_context:
 108 L_padlock_verify_context_begin:
 109         movl    4(%esp),%edx
 110         leal    Lpadlock_saved_context-L002verify_pic_point,%eax
 111         pushfl
 112         call    __padlock_verify_ctx
 113 L002verify_pic_point:
 114         leal    4(%esp),%esp
 115         ret
 116 .align  4
 117 __padlock_verify_ctx:
 118         addl    (%esp),%eax
 119         btl     $30,4(%esp)
 120         jnc     L003verified
 121         cmpl    (%eax),%edx
 122         je      L003verified
 123         pushfl
 124         popfl
 125 L003verified:
 126         movl    %edx,(%eax)
 127         ret
 128 .globl  _padlock_reload_key
 129 .align  4
 130 _padlock_reload_key:
 131 L_padlock_reload_key_begin:
 132         pushfl
 133         popfl
 134         ret
 135 .globl  _padlock_aes_block
 136 .align  4
 137 _padlock_aes_block:
 138 L_padlock_aes_block_begin:
 139         pushl   %edi
 140         pushl   %esi
 141         pushl   %ebx
 142         movl    16(%esp),%edi
 143         movl    20(%esp),%esi
 144         movl    24(%esp),%edx
 145         movl    $1,%ecx
 146         leal    32(%edx),%ebx
 147         leal    16(%edx),%edx
 148 .byte   243,15,167,200
 149         popl    %ebx
 150         popl    %esi
 151         popl    %edi
 152         ret
 153 .globl  _padlock_ecb_encrypt
 154 .align  4
 155 _padlock_ecb_encrypt:
 156 L_padlock_ecb_encrypt_begin:
 157         pushl   %ebp
 158         pushl   %ebx
 159         pushl   %esi
 160         pushl   %edi
 161         movl    20(%esp),%edi
 162         movl    24(%esp),%esi
 163         movl    28(%esp),%edx
 164         movl    32(%esp),%ecx
 165         testl   $15,%edx
 166         jnz     L004ecb_abort
 167         testl   $15,%ecx
 168         jnz     L004ecb_abort
 169         leal    Lpadlock_saved_context-L005ecb_pic_point,%eax
 170         pushfl
 171         cld
 172         call    __padlock_verify_ctx
 173 L005ecb_pic_point:
 174         leal    16(%edx),%edx
 175         xorl    %eax,%eax
 176         xorl    %ebx,%ebx
 177         cmpl    $128,%ecx
 178         jbe     L006ecb_short
 179         testl   $32,(%edx)
 180         jnz     L007ecb_aligned
 181         testl   $15,%edi
 182         setz    %al
 183         testl   $15,%esi
 184         setz    %bl
 185         testl   %ebx,%eax
 186         jnz     L007ecb_aligned
 187         negl    %eax
 188         movl    $512,%ebx
 189         notl    %eax
 190         leal    -24(%esp),%ebp
 191         cmpl    %ebx,%ecx
 192         cmovcl  %ecx,%ebx
 193         andl    %ebx,%eax
 194         movl    %ecx,%ebx
 195         negl    %eax
 196         andl    $511,%ebx
 197         leal    (%eax,%ebp,1),%esp
 198         andl    $-16,%esp
 199         jmp     L008ecb_loop
 200 .align  4,0x90
 201 L008ecb_loop:
 202         movl    %edi,(%ebp)
 203         movl    %esi,4(%ebp)
 204         movl    %ecx,8(%ebp)
 205         movl    %ebx,%ecx
 206         movl    %ebx,12(%ebp)
 207         testl   $15,%edi
 208         cmovnzl %esp,%edi
 209         testl   $15,%esi
 210         jz      L009ecb_inp_aligned
 211         shrl    $2,%ecx
 212 .byte   243,165
 213         subl    %ebx,%edi
 214         movl    %ebx,%ecx
 215         movl    %edi,%esi
 216 L009ecb_inp_aligned:
 217         leal    -16(%edx),%eax
 218         leal    16(%edx),%ebx
 219         shrl    $4,%ecx
 220 .byte   243,15,167,200
 221         movl    (%ebp),%edi
 222         movl    12(%ebp),%ebx
 223         testl   $15,%edi
 224         jz      L010ecb_out_aligned
 225         movl    %ebx,%ecx
 226         shrl    $2,%ecx
 227         leal    (%esp),%esi
 228 .byte   243,165
 229         subl    %ebx,%edi
 230 L010ecb_out_aligned:
 231         movl    4(%ebp),%esi
 232         movl    8(%ebp),%ecx
 233         addl    %ebx,%edi
 234         addl    %ebx,%esi
 235         subl    %ebx,%ecx
 236         movl    $512,%ebx
 237         jnz     L008ecb_loop
 238         cmpl    %ebp,%esp
 239         je      L011ecb_done
 240         pxor    %xmm0,%xmm0
 241         leal    (%esp),%eax
 242 L012ecb_bzero:
 243         movaps  %xmm0,(%eax)
 244         leal    16(%eax),%eax
 245         cmpl    %eax,%ebp
 246         ja      L012ecb_bzero
 247 L011ecb_done:
 248         leal    24(%ebp),%esp
 249         jmp     L013ecb_exit
 250 .align  4,0x90
 251 L006ecb_short:
 252         xorl    %eax,%eax
 253         leal    -24(%esp),%ebp
 254         subl    %ecx,%eax
 255         leal    (%eax,%ebp,1),%esp
 256         andl    $-16,%esp
 257         xorl    %ebx,%ebx
 258 L014ecb_short_copy:
 259         movups  (%esi,%ebx,1),%xmm0
 260         leal    16(%ebx),%ebx
 261         cmpl    %ebx,%ecx
 262         movaps  %xmm0,-16(%esp,%ebx,1)
 263         ja      L014ecb_short_copy
 264         movl    %esp,%esi
 265         movl    %ecx,%ebx
 266         jmp     L008ecb_loop
 267 .align  4,0x90
 268 L007ecb_aligned:
 269         leal    -16(%edx),%eax
 270         leal    16(%edx),%ebx
 271         shrl    $4,%ecx
 272 .byte   243,15,167,200
 273 L013ecb_exit:
 274         movl    $1,%eax
 275         leal    4(%esp),%esp
 276 L004ecb_abort:
 277         popl    %edi
 278         popl    %esi
 279         popl    %ebx
 280         popl    %ebp
 281         ret
 282 .globl  _padlock_cbc_encrypt
 283 .align  4
 284 _padlock_cbc_encrypt:
 285 L_padlock_cbc_encrypt_begin:
 286         pushl   %ebp
 287         pushl   %ebx
 288         pushl   %esi
 289         pushl   %edi
 290         movl    20(%esp),%edi
 291         movl    24(%esp),%esi
 292         movl    28(%esp),%edx
 293         movl    32(%esp),%ecx
 294         testl   $15,%edx
 295         jnz     L015cbc_abort
 296         testl   $15,%ecx
 297         jnz     L015cbc_abort
 298         leal    Lpadlock_saved_context-L016cbc_pic_point,%eax
 299         pushfl
 300         cld
 301         call    __padlock_verify_ctx
 302 L016cbc_pic_point:
 303         leal    16(%edx),%edx
 304         xorl    %eax,%eax
 305         xorl    %ebx,%ebx
 306         cmpl    $64,%ecx
 307         jbe     L017cbc_short
 308         testl   $32,(%edx)
 309         jnz     L018cbc_aligned
 310         testl   $15,%edi
 311         setz    %al
 312         testl   $15,%esi
 313         setz    %bl
 314         testl   %ebx,%eax
 315         jnz     L018cbc_aligned
 316         negl    %eax
 317         movl    $512,%ebx
 318         notl    %eax
 319         leal    -24(%esp),%ebp
 320         cmpl    %ebx,%ecx
 321         cmovcl  %ecx,%ebx
 322         andl    %ebx,%eax
 323         movl    %ecx,%ebx
 324         negl    %eax
 325         andl    $511,%ebx
 326         leal    (%eax,%ebp,1),%esp
 327         andl    $-16,%esp
 328         jmp     L019cbc_loop
 329 .align  4,0x90
 330 L019cbc_loop:
 331         movl    %edi,(%ebp)
 332         movl    %esi,4(%ebp)
 333         movl    %ecx,8(%ebp)
 334         movl    %ebx,%ecx
 335         movl    %ebx,12(%ebp)
 336         testl   $15,%edi
 337         cmovnzl %esp,%edi
 338         testl   $15,%esi
 339         jz      L020cbc_inp_aligned
 340         shrl    $2,%ecx
 341 .byte   243,165
 342         subl    %ebx,%edi
 343         movl    %ebx,%ecx
 344         movl    %edi,%esi
 345 L020cbc_inp_aligned:
 346         leal    -16(%edx),%eax
 347         leal    16(%edx),%ebx
 348         shrl    $4,%ecx
 349 .byte   243,15,167,208
 350         movaps  (%eax),%xmm0
 351         movaps  %xmm0,-16(%edx)
 352         movl    (%ebp),%edi
 353         movl    12(%ebp),%ebx
 354         testl   $15,%edi
 355         jz      L021cbc_out_aligned
 356         movl    %ebx,%ecx
 357         shrl    $2,%ecx
 358         leal    (%esp),%esi
 359 .byte   243,165
 360         subl    %ebx,%edi
 361 L021cbc_out_aligned:
 362         movl    4(%ebp),%esi
 363         movl    8(%ebp),%ecx
 364         addl    %ebx,%edi
 365         addl    %ebx,%esi
 366         subl    %ebx,%ecx
 367         movl    $512,%ebx
 368         jnz     L019cbc_loop
 369         cmpl    %ebp,%esp
 370         je      L022cbc_done
 371         pxor    %xmm0,%xmm0
 372         leal    (%esp),%eax
 373 L023cbc_bzero:
 374         movaps  %xmm0,(%eax)
 375         leal    16(%eax),%eax
 376         cmpl    %eax,%ebp
 377         ja      L023cbc_bzero
 378 L022cbc_done:
 379         leal    24(%ebp),%esp
 380         jmp     L024cbc_exit
 381 .align  4,0x90
 382 L017cbc_short:
 383         xorl    %eax,%eax
 384         leal    -24(%esp),%ebp
 385         subl    %ecx,%eax
 386         leal    (%eax,%ebp,1),%esp
 387         andl    $-16,%esp
 388         xorl    %ebx,%ebx
 389 L025cbc_short_copy:
 390         movups  (%esi,%ebx,1),%xmm0
 391         leal    16(%ebx),%ebx
 392         cmpl    %ebx,%ecx
 393         movaps  %xmm0,-16(%esp,%ebx,1)
 394         ja      L025cbc_short_copy
 395         movl    %esp,%esi
 396         movl    %ecx,%ebx
 397         jmp     L019cbc_loop
 398 .align  4,0x90
 399 L018cbc_aligned:
 400         leal    -16(%edx),%eax
 401         leal    16(%edx),%ebx
 402         shrl    $4,%ecx
 403 .byte   243,15,167,208
 404         movaps  (%eax),%xmm0
 405         movaps  %xmm0,-16(%edx)
 406 L024cbc_exit:
 407         movl    $1,%eax
 408         leal    4(%esp),%esp
 409 L015cbc_abort:
 410         popl    %edi
 411         popl    %esi
 412         popl    %ebx
 413         popl    %ebp
 414         ret
 415 .globl  _padlock_xstore
 416 .align  4
 417 _padlock_xstore:
 418 L_padlock_xstore_begin:
 419         pushl   %edi
 420         movl    8(%esp),%edi
 421         movl    12(%esp),%edx
 422 .byte   15,167,192
 423         popl    %edi
 424         ret
 425 .align  4
 426 __win32_segv_handler:
 427         movl    $1,%eax
 428         movl    4(%esp),%edx
 429         movl    12(%esp),%ecx
 430         cmpl    $3221225477,(%edx)
 431         jne     L026ret
 432         addl    $4,184(%ecx)
 433         movl    $0,%eax
 434 L026ret:
 435         ret
 436 .globl  _padlock_sha1_oneshot
 437 .align  4
 438 _padlock_sha1_oneshot:
 439 L_padlock_sha1_oneshot_begin:
 440         pushl   %edi
 441         pushl   %esi
 442         xorl    %eax,%eax
 443         movl    12(%esp),%edi
 444         movl    16(%esp),%esi
 445         movl    20(%esp),%ecx
 446         movl    %esp,%edx
 447         addl    $-128,%esp
 448         movups  (%edi),%xmm0
 449         andl    $-16,%esp
 450         movl    16(%edi),%eax
 451         movaps  %xmm0,(%esp)
 452         movl    %esp,%edi
 453         movl    %eax,16(%esp)
 454         xorl    %eax,%eax
 455 .byte   243,15,166,200
 456         movaps  (%esp),%xmm0
 457         movl    16(%esp),%eax
 458         movl    %edx,%esp
 459         movl    12(%esp),%edi
 460         movups  %xmm0,(%edi)
 461         movl    %eax,16(%edi)
 462         popl    %esi
 463         popl    %edi
 464         ret
 465 .globl  _padlock_sha1_blocks
 466 .align  4
 467 _padlock_sha1_blocks:
 468 L_padlock_sha1_blocks_begin:
 469         pushl   %edi
 470         pushl   %esi
 471         movl    12(%esp),%edi
 472         movl    16(%esp),%esi
 473         movl    %esp,%edx
 474         movl    20(%esp),%ecx
 475         addl    $-128,%esp
 476         movups  (%edi),%xmm0
 477         andl    $-16,%esp
 478         movl    16(%edi),%eax
 479         movaps  %xmm0,(%esp)
 480         movl    %esp,%edi
 481         movl    %eax,16(%esp)
 482         movl    $-1,%eax
 483 .byte   243,15,166,200
 484         movaps  (%esp),%xmm0
 485         movl    16(%esp),%eax
 486         movl    %edx,%esp
 487         movl    12(%esp),%edi
 488         movups  %xmm0,(%edi)
 489         movl    %eax,16(%edi)
 490         popl    %esi
 491         popl    %edi
 492         ret
 493 .globl  _padlock_sha256_oneshot
 494 .align  4
 495 _padlock_sha256_oneshot:
 496 L_padlock_sha256_oneshot_begin:
 497         pushl   %edi
 498         pushl   %esi
 499         xorl    %eax,%eax
 500         movl    12(%esp),%edi
 501         movl    16(%esp),%esi
 502         movl    20(%esp),%ecx
 503         movl    %esp,%edx
 504         addl    $-128,%esp
 505         movups  (%edi),%xmm0
 506         andl    $-16,%esp
 507         movups  16(%edi),%xmm1
 508         movaps  %xmm0,(%esp)
 509         movl    %esp,%edi
 510         movaps  %xmm1,16(%esp)
 511         xorl    %eax,%eax
 512 .byte   243,15,166,208
 513         movaps  (%esp),%xmm0
 514         movaps  16(%esp),%xmm1
 515         movl    %edx,%esp
 516         movl    12(%esp),%edi
 517         movups  %xmm0,(%edi)
 518         movups  %xmm1,16(%edi)
 519         popl    %esi
 520         popl    %edi
 521         ret
 522 .globl  _padlock_sha256_blocks
 523 .align  4
 524 _padlock_sha256_blocks:
 525 L_padlock_sha256_blocks_begin:
 526         pushl   %edi
 527         pushl   %esi
 528         movl    12(%esp),%edi
 529         movl    16(%esp),%esi
 530         movl    20(%esp),%ecx
 531         movl    %esp,%edx
 532         addl    $-128,%esp
 533         movups  (%edi),%xmm0
 534         andl    $-16,%esp
 535         movups  16(%edi),%xmm1
 536         movaps  %xmm0,(%esp)
 537         movl    %esp,%edi
 538         movaps  %xmm1,16(%esp)
 539         movl    $-1,%eax
 540 .byte   243,15,166,208
 541         movaps  (%esp),%xmm0
 542         movaps  16(%esp),%xmm1
 543         movl    %edx,%esp
 544         movl    12(%esp),%edi
 545         movups  %xmm0,(%edi)
 546         movups  %xmm1,16(%edi)
 547         popl    %esi
 548         popl    %edi
 549         ret
 550 .globl  _padlock_sha512_blocks
 551 .align  4
 552 _padlock_sha512_blocks:
 553 L_padlock_sha512_blocks_begin:
 554         pushl   %edi
 555         pushl   %esi
 556         movl    12(%esp),%edi
 557         movl    16(%esp),%esi
 558         movl    20(%esp),%ecx
 559         movl    %esp,%edx
 560         addl    $-128,%esp
 561         movups  (%edi),%xmm0
 562         andl    $-16,%esp
 563         movups  16(%edi),%xmm1
 564         movups  32(%edi),%xmm2
 565         movups  48(%edi),%xmm3
 566         movaps  %xmm0,(%esp)
 567         movl    %esp,%edi
 568         movaps  %xmm1,16(%esp)
 569         movaps  %xmm2,32(%esp)
 570         movaps  %xmm3,48(%esp)
 571 .byte   243,15,166,224
 572         movaps  (%esp),%xmm0
 573         movaps  16(%esp),%xmm1
 574         movaps  32(%esp),%xmm2
 575         movaps  48(%esp),%xmm3
 576         movl    %edx,%esp
 577         movl    12(%esp),%edi
 578         movups  %xmm0,(%edi)
 579         movups  %xmm1,16(%edi)
 580         movups  %xmm2,32(%edi)
 581         movups  %xmm3,48(%edi)
 582         popl    %esi
 583         popl    %edi
 584         ret
 585 .byte   86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,32
 586 .byte   109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65
 587 .byte   77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101
 588 .byte   110,115,115,108,46,111,114,103,62,0
 589 .align  4,0x90
 590 .data
 591 .align  2,0x90
 592 Lpadlock_saved_context:
 593 .long   0