lib/accelerated/x86/coff/padlock-x86-coff.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 .def    _padlock_capability;    .scl    2;      .type   32;     .endef
  44 .align  16
  45 _padlock_capability:
  46 .L_padlock_capability_begin:
  47         pushl   %ebx
  48         pushfl
  49         popl    %eax
  50         movl    %eax,%ecx
  51         xorl    $2097152,%eax
  52         pushl   %eax
  53         popfl
  54         pushfl
  55         popl    %eax
  56         xorl    %eax,%ecx
  57         xorl    %eax,%eax
  58         btl     $21,%ecx
  59         jnc     .L000noluck
  60         .byte   0x0f,0xa2
  61         xorl    %eax,%eax
  62         cmpl    $0x746e6543,%ebx
  63         jne     .L000noluck
  64         cmpl    $0x48727561,%edx
  65         jne     .L000noluck
  66         cmpl    $0x736c7561,%ecx
  67         jne     .L000noluck
  68         movl    $3221225472,%eax
  69         .byte   0x0f,0xa2
  70         movl    %eax,%edx
  71         xorl    %eax,%eax
  72         cmpl    $3221225473,%edx
  73         jb      .L000noluck
  74         movl    $1,%eax
  75         .byte   0x0f,0xa2
  76         orl     $15,%eax
  77         xorl    %ebx,%ebx
  78         andl    $4095,%eax
  79         cmpl    $1791,%eax
  80         sete    %bl
  81         movl    $3221225473,%eax
  82         pushl   %ebx
  83         .byte   0x0f,0xa2
  84         popl    %ebx
  85         movl    %edx,%eax
  86         shll    $4,%ebx
  87         andl    $4294967279,%eax
  88         orl     %ebx,%eax
  89 .L000noluck:
  90         popl    %ebx
  91         ret
  92 .globl  _padlock_key_bswap
  93 .def    _padlock_key_bswap;     .scl    2;      .type   32;     .endef
  94 .align  16
  95 _padlock_key_bswap:
  96 .L_padlock_key_bswap_begin:
  97         movl    4(%esp),%edx
  98         movl    240(%edx),%ecx
  99 .L001bswap_loop:
 100         movl    (%edx),%eax
 101         bswap   %eax
 102         movl    %eax,(%edx)
 103         leal    4(%edx),%edx
 104         subl    $1,%ecx
 105         jnz     .L001bswap_loop
 106         ret
 107 .globl  _padlock_verify_context
 108 .def    _padlock_verify_context;        .scl    2;      .type   32;     .endef
 109 .align  16
 110 _padlock_verify_context:
 111 .L_padlock_verify_context_begin:
 112         movl    4(%esp),%edx
 113         leal    .Lpadlock_saved_context,%eax
 114         pushfl
 115         call    __padlock_verify_ctx
 116 .L002verify_pic_point:
 117         leal    4(%esp),%esp
 118         ret
 119 .def    __padlock_verify_ctx;   .scl    3;      .type   32;     .endef
 120 .align  16
 121 __padlock_verify_ctx:
 122         btl     $30,4(%esp)
 123         jnc     .L003verified
 124         cmpl    (%eax),%edx
 125         je      .L003verified
 126         pushfl
 127         popfl
 128 .L003verified:
 129         movl    %edx,(%eax)
 130         ret
 131 .globl  _padlock_reload_key
 132 .def    _padlock_reload_key;    .scl    2;      .type   32;     .endef
 133 .align  16
 134 _padlock_reload_key:
 135 .L_padlock_reload_key_begin:
 136         pushfl
 137         popfl
 138         ret
 139 .globl  _padlock_aes_block
 140 .def    _padlock_aes_block;     .scl    2;      .type   32;     .endef
 141 .align  16
 142 _padlock_aes_block:
 143 .L_padlock_aes_block_begin:
 144         pushl   %edi
 145         pushl   %esi
 146         pushl   %ebx
 147         movl    16(%esp),%edi
 148         movl    20(%esp),%esi
 149         movl    24(%esp),%edx
 150         movl    $1,%ecx
 151         leal    32(%edx),%ebx
 152         leal    16(%edx),%edx
 153 .byte   243,15,167,200
 154         popl    %ebx
 155         popl    %esi
 156         popl    %edi
 157         ret
 158 .globl  _padlock_ecb_encrypt
 159 .def    _padlock_ecb_encrypt;   .scl    2;      .type   32;     .endef
 160 .align  16
 161 _padlock_ecb_encrypt:
 162 .L_padlock_ecb_encrypt_begin:
 163         pushl   %ebp
 164         pushl   %ebx
 165         pushl   %esi
 166         pushl   %edi
 167         movl    20(%esp),%edi
 168         movl    24(%esp),%esi
 169         movl    28(%esp),%edx
 170         movl    32(%esp),%ecx
 171         testl   $15,%edx
 172         jnz     .L004ecb_abort
 173         testl   $15,%ecx
 174         jnz     .L004ecb_abort
 175         leal    .Lpadlock_saved_context,%eax
 176         pushfl
 177         cld
 178         call    __padlock_verify_ctx
 179 .L005ecb_pic_point:
 180         leal    16(%edx),%edx
 181         xorl    %eax,%eax
 182         xorl    %ebx,%ebx
 183         cmpl    $128,%ecx
 184         jbe     .L006ecb_short
 185         testl   $32,(%edx)
 186         jnz     .L007ecb_aligned
 187         testl   $15,%edi
 188         setz    %al
 189         testl   $15,%esi
 190         setz    %bl
 191         testl   %ebx,%eax
 192         jnz     .L007ecb_aligned
 193         negl    %eax
 194         movl    $512,%ebx
 195         notl    %eax
 196         leal    -24(%esp),%ebp
 197         cmpl    %ebx,%ecx
 198         cmovcl  %ecx,%ebx
 199         andl    %ebx,%eax
 200         movl    %ecx,%ebx
 201         negl    %eax
 202         andl    $511,%ebx
 203         leal    (%eax,%ebp,1),%esp
 204         andl    $-16,%esp
 205         jmp     .L008ecb_loop
 206 .align  16
 207 .L008ecb_loop:
 208         movl    %edi,(%ebp)
 209         movl    %esi,4(%ebp)
 210         movl    %ecx,8(%ebp)
 211         movl    %ebx,%ecx
 212         movl    %ebx,12(%ebp)
 213         testl   $15,%edi
 214         cmovnzl %esp,%edi
 215         testl   $15,%esi
 216         jz      .L009ecb_inp_aligned
 217         shrl    $2,%ecx
 218 .byte   243,165
 219         subl    %ebx,%edi
 220         movl    %ebx,%ecx
 221         movl    %edi,%esi
 222 .L009ecb_inp_aligned:
 223         leal    -16(%edx),%eax
 224         leal    16(%edx),%ebx
 225         shrl    $4,%ecx
 226 .byte   243,15,167,200
 227         movl    (%ebp),%edi
 228         movl    12(%ebp),%ebx
 229         testl   $15,%edi
 230         jz      .L010ecb_out_aligned
 231         movl    %ebx,%ecx
 232         shrl    $2,%ecx
 233         leal    (%esp),%esi
 234 .byte   243,165
 235         subl    %ebx,%edi
 236 .L010ecb_out_aligned:
 237         movl    4(%ebp),%esi
 238         movl    8(%ebp),%ecx
 239         addl    %ebx,%edi
 240         addl    %ebx,%esi
 241         subl    %ebx,%ecx
 242         movl    $512,%ebx
 243         jnz     .L008ecb_loop
 244         cmpl    %ebp,%esp
 245         je      .L011ecb_done
 246         pxor    %xmm0,%xmm0
 247         leal    (%esp),%eax
 248 .L012ecb_bzero:
 249         movaps  %xmm0,(%eax)
 250         leal    16(%eax),%eax
 251         cmpl    %eax,%ebp
 252         ja      .L012ecb_bzero
 253 .L011ecb_done:
 254         leal    24(%ebp),%esp
 255         jmp     .L013ecb_exit
 256 .align  16
 257 .L006ecb_short:
 258         xorl    %eax,%eax
 259         leal    -24(%esp),%ebp
 260         subl    %ecx,%eax
 261         leal    (%eax,%ebp,1),%esp
 262         andl    $-16,%esp
 263         xorl    %ebx,%ebx
 264 .L014ecb_short_copy:
 265         movups  (%esi,%ebx,1),%xmm0
 266         leal    16(%ebx),%ebx
 267         cmpl    %ebx,%ecx
 268         movaps  %xmm0,-16(%esp,%ebx,1)
 269         ja      .L014ecb_short_copy
 270         movl    %esp,%esi
 271         movl    %ecx,%ebx
 272         jmp     .L008ecb_loop
 273 .align  16
 274 .L007ecb_aligned:
 275         leal    -16(%edx),%eax
 276         leal    16(%edx),%ebx
 277         shrl    $4,%ecx
 278 .byte   243,15,167,200
 279 .L013ecb_exit:
 280         movl    $1,%eax
 281         leal    4(%esp),%esp
 282 .L004ecb_abort:
 283         popl    %edi
 284         popl    %esi
 285         popl    %ebx
 286         popl    %ebp
 287         ret
 288 .globl  _padlock_cbc_encrypt
 289 .def    _padlock_cbc_encrypt;   .scl    2;      .type   32;     .endef
 290 .align  16
 291 _padlock_cbc_encrypt:
 292 .L_padlock_cbc_encrypt_begin:
 293         pushl   %ebp
 294         pushl   %ebx
 295         pushl   %esi
 296         pushl   %edi
 297         movl    20(%esp),%edi
 298         movl    24(%esp),%esi
 299         movl    28(%esp),%edx
 300         movl    32(%esp),%ecx
 301         testl   $15,%edx
 302         jnz     .L015cbc_abort
 303         testl   $15,%ecx
 304         jnz     .L015cbc_abort
 305         leal    .Lpadlock_saved_context,%eax
 306         pushfl
 307         cld
 308         call    __padlock_verify_ctx
 309 .L016cbc_pic_point:
 310         leal    16(%edx),%edx
 311         xorl    %eax,%eax
 312         xorl    %ebx,%ebx
 313         cmpl    $64,%ecx
 314         jbe     .L017cbc_short
 315         testl   $32,(%edx)
 316         jnz     .L018cbc_aligned
 317         testl   $15,%edi
 318         setz    %al
 319         testl   $15,%esi
 320         setz    %bl
 321         testl   %ebx,%eax
 322         jnz     .L018cbc_aligned
 323         negl    %eax
 324         movl    $512,%ebx
 325         notl    %eax
 326         leal    -24(%esp),%ebp
 327         cmpl    %ebx,%ecx
 328         cmovcl  %ecx,%ebx
 329         andl    %ebx,%eax
 330         movl    %ecx,%ebx
 331         negl    %eax
 332         andl    $511,%ebx
 333         leal    (%eax,%ebp,1),%esp
 334         andl    $-16,%esp
 335         jmp     .L019cbc_loop
 336 .align  16
 337 .L019cbc_loop:
 338         movl    %edi,(%ebp)
 339         movl    %esi,4(%ebp)
 340         movl    %ecx,8(%ebp)
 341         movl    %ebx,%ecx
 342         movl    %ebx,12(%ebp)
 343         testl   $15,%edi
 344         cmovnzl %esp,%edi
 345         testl   $15,%esi
 346         jz      .L020cbc_inp_aligned
 347         shrl    $2,%ecx
 348 .byte   243,165
 349         subl    %ebx,%edi
 350         movl    %ebx,%ecx
 351         movl    %edi,%esi
 352 .L020cbc_inp_aligned:
 353         leal    -16(%edx),%eax
 354         leal    16(%edx),%ebx
 355         shrl    $4,%ecx
 356 .byte   243,15,167,208
 357         movaps  (%eax),%xmm0
 358         movaps  %xmm0,-16(%edx)
 359         movl    (%ebp),%edi
 360         movl    12(%ebp),%ebx
 361         testl   $15,%edi
 362         jz      .L021cbc_out_aligned
 363         movl    %ebx,%ecx
 364         shrl    $2,%ecx
 365         leal    (%esp),%esi
 366 .byte   243,165
 367         subl    %ebx,%edi
 368 .L021cbc_out_aligned:
 369         movl    4(%ebp),%esi
 370         movl    8(%ebp),%ecx
 371         addl    %ebx,%edi
 372         addl    %ebx,%esi
 373         subl    %ebx,%ecx
 374         movl    $512,%ebx
 375         jnz     .L019cbc_loop
 376         cmpl    %ebp,%esp
 377         je      .L022cbc_done
 378         pxor    %xmm0,%xmm0
 379         leal    (%esp),%eax
 380 .L023cbc_bzero:
 381         movaps  %xmm0,(%eax)
 382         leal    16(%eax),%eax
 383         cmpl    %eax,%ebp
 384         ja      .L023cbc_bzero
 385 .L022cbc_done:
 386         leal    24(%ebp),%esp
 387         jmp     .L024cbc_exit
 388 .align  16
 389 .L017cbc_short:
 390         xorl    %eax,%eax
 391         leal    -24(%esp),%ebp
 392         subl    %ecx,%eax
 393         leal    (%eax,%ebp,1),%esp
 394         andl    $-16,%esp
 395         xorl    %ebx,%ebx
 396 .L025cbc_short_copy:
 397         movups  (%esi,%ebx,1),%xmm0
 398         leal    16(%ebx),%ebx
 399         cmpl    %ebx,%ecx
 400         movaps  %xmm0,-16(%esp,%ebx,1)
 401         ja      .L025cbc_short_copy
 402         movl    %esp,%esi
 403         movl    %ecx,%ebx
 404         jmp     .L019cbc_loop
 405 .align  16
 406 .L018cbc_aligned:
 407         leal    -16(%edx),%eax
 408         leal    16(%edx),%ebx
 409         shrl    $4,%ecx
 410 .byte   243,15,167,208
 411         movaps  (%eax),%xmm0
 412         movaps  %xmm0,-16(%edx)
 413 .L024cbc_exit:
 414         movl    $1,%eax
 415         leal    4(%esp),%esp
 416 .L015cbc_abort:
 417         popl    %edi
 418         popl    %esi
 419         popl    %ebx
 420         popl    %ebp
 421         ret
 422 .globl  _padlock_xstore
 423 .def    _padlock_xstore;        .scl    2;      .type   32;     .endef
 424 .align  16
 425 _padlock_xstore:
 426 .L_padlock_xstore_begin:
 427         pushl   %edi
 428         movl    8(%esp),%edi
 429         movl    12(%esp),%edx
 430 .byte   15,167,192
 431         popl    %edi
 432         ret
 433 .def    __win32_segv_handler;   .scl    3;      .type   32;     .endef
 434 .align  16
 435 __win32_segv_handler:
 436         movl    $1,%eax
 437         movl    4(%esp),%edx
 438         movl    12(%esp),%ecx
 439         cmpl    $3221225477,(%edx)
 440         jne     .L026ret
 441         addl    $4,184(%ecx)
 442         movl    $0,%eax
 443 .L026ret:
 444         ret
 445 .globl  _padlock_sha1_oneshot
 446 .def    _padlock_sha1_oneshot;  .scl    2;      .type   32;     .endef
 447 .align  16
 448 _padlock_sha1_oneshot:
 449 .L_padlock_sha1_oneshot_begin:
 450         pushl   %edi
 451         pushl   %esi
 452         xorl    %eax,%eax
 453         movl    12(%esp),%edi
 454         movl    16(%esp),%esi
 455         movl    20(%esp),%ecx
 456         pushl   __win32_segv_handler
 457 .byte   100,255,48
 458 .byte   100,137,32
 459         movl    %esp,%edx
 460         addl    $-128,%esp
 461         movups  (%edi),%xmm0
 462         andl    $-16,%esp
 463         movl    16(%edi),%eax
 464         movaps  %xmm0,(%esp)
 465         movl    %esp,%edi
 466         movl    %eax,16(%esp)
 467         xorl    %eax,%eax
 468 .byte   243,15,166,200
 469         movaps  (%esp),%xmm0
 470         movl    16(%esp),%eax
 471         movl    %edx,%esp
 472 .byte   100,143,5,0,0,0,0
 473         leal    4(%esp),%esp
 474         movl    16(%esp),%edi
 475         movups  %xmm0,(%edi)
 476         movl    %eax,16(%edi)
 477         popl    %esi
 478         popl    %edi
 479         ret
 480 .globl  _padlock_sha1_blocks
 481 .def    _padlock_sha1_blocks;   .scl    2;      .type   32;     .endef
 482 .align  16
 483 _padlock_sha1_blocks:
 484 .L_padlock_sha1_blocks_begin:
 485         pushl   %edi
 486         pushl   %esi
 487         movl    12(%esp),%edi
 488         movl    16(%esp),%esi
 489         movl    %esp,%edx
 490         movl    20(%esp),%ecx
 491         addl    $-128,%esp
 492         movups  (%edi),%xmm0
 493         andl    $-16,%esp
 494         movl    16(%edi),%eax
 495         movaps  %xmm0,(%esp)
 496         movl    %esp,%edi
 497         movl    %eax,16(%esp)
 498         movl    $-1,%eax
 499 .byte   243,15,166,200
 500         movaps  (%esp),%xmm0
 501         movl    16(%esp),%eax
 502         movl    %edx,%esp
 503         movl    12(%esp),%edi
 504         movups  %xmm0,(%edi)
 505         movl    %eax,16(%edi)
 506         popl    %esi
 507         popl    %edi
 508         ret
 509 .globl  _padlock_sha256_oneshot
 510 .def    _padlock_sha256_oneshot;        .scl    2;      .type   32;     .endef
 511 .align  16
 512 _padlock_sha256_oneshot:
 513 .L_padlock_sha256_oneshot_begin:
 514         pushl   %edi
 515         pushl   %esi
 516         xorl    %eax,%eax
 517         movl    12(%esp),%edi
 518         movl    16(%esp),%esi
 519         movl    20(%esp),%ecx
 520         pushl   __win32_segv_handler
 521 .byte   100,255,48
 522 .byte   100,137,32
 523         movl    %esp,%edx
 524         addl    $-128,%esp
 525         movups  (%edi),%xmm0
 526         andl    $-16,%esp
 527         movups  16(%edi),%xmm1
 528         movaps  %xmm0,(%esp)
 529         movl    %esp,%edi
 530         movaps  %xmm1,16(%esp)
 531         xorl    %eax,%eax
 532 .byte   243,15,166,208
 533         movaps  (%esp),%xmm0
 534         movaps  16(%esp),%xmm1
 535         movl    %edx,%esp
 536 .byte   100,143,5,0,0,0,0
 537         leal    4(%esp),%esp
 538         movl    16(%esp),%edi
 539         movups  %xmm0,(%edi)
 540         movups  %xmm1,16(%edi)
 541         popl    %esi
 542         popl    %edi
 543         ret
 544 .globl  _padlock_sha256_blocks
 545 .def    _padlock_sha256_blocks; .scl    2;      .type   32;     .endef
 546 .align  16
 547 _padlock_sha256_blocks:
 548 .L_padlock_sha256_blocks_begin:
 549         pushl   %edi
 550         pushl   %esi
 551         movl    12(%esp),%edi
 552         movl    16(%esp),%esi
 553         movl    20(%esp),%ecx
 554         movl    %esp,%edx
 555         addl    $-128,%esp
 556         movups  (%edi),%xmm0
 557         andl    $-16,%esp
 558         movups  16(%edi),%xmm1
 559         movaps  %xmm0,(%esp)
 560         movl    %esp,%edi
 561         movaps  %xmm1,16(%esp)
 562         movl    $-1,%eax
 563 .byte   243,15,166,208
 564         movaps  (%esp),%xmm0
 565         movaps  16(%esp),%xmm1
 566         movl    %edx,%esp
 567         movl    12(%esp),%edi
 568         movups  %xmm0,(%edi)
 569         movups  %xmm1,16(%edi)
 570         popl    %esi
 571         popl    %edi
 572         ret
 573 .globl  _padlock_sha512_blocks
 574 .def    _padlock_sha512_blocks; .scl    2;      .type   32;     .endef
 575 .align  16
 576 _padlock_sha512_blocks:
 577 .L_padlock_sha512_blocks_begin:
 578         pushl   %edi
 579         pushl   %esi
 580         movl    12(%esp),%edi
 581         movl    16(%esp),%esi
 582         movl    20(%esp),%ecx
 583         movl    %esp,%edx
 584         addl    $-128,%esp
 585         movups  (%edi),%xmm0
 586         andl    $-16,%esp
 587         movups  16(%edi),%xmm1
 588         movups  32(%edi),%xmm2
 589         movups  48(%edi),%xmm3
 590         movaps  %xmm0,(%esp)
 591         movl    %esp,%edi
 592         movaps  %xmm1,16(%esp)
 593         movaps  %xmm2,32(%esp)
 594         movaps  %xmm3,48(%esp)
 595 .byte   243,15,166,224
 596         movaps  (%esp),%xmm0
 597         movaps  16(%esp),%xmm1
 598         movaps  32(%esp),%xmm2
 599         movaps  48(%esp),%xmm3
 600         movl    %edx,%esp
 601         movl    12(%esp),%edi
 602         movups  %xmm0,(%edi)
 603         movups  %xmm1,16(%edi)
 604         movups  %xmm2,32(%edi)
 605         movups  %xmm3,48(%edi)
 606         popl    %esi
 607         popl    %edi
 608         ret
 609 .byte   86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,32
 610 .byte   109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65
 611 .byte   77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101
 612 .byte   110,115,115,108,46,111,114,103,62,0
 613 .align  16
 614 .data
 615 .align  4
 616 .Lpadlock_saved_context:
 617 .long   0