www.gitorious.org Git - gnutls:gnutls.git/blob - lib/accelerated/x86/elf/padlock-x86.s

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.

38 .file "devel/perlasm/e_padlock-x86.s"

39 .text

40 .globl padlock_capability

41 .type padlock_capability,@function

42 .align 16

43 padlock_capability:

44 .L_padlock_capability_begin:

45 pushl %ebx

46 pushfl

47 popl %eax

48 movl %eax,%ecx

49 xorl $2097152,%eax

50 pushl %eax

51 popfl

52 pushfl

53 popl %eax

54 xorl %eax,%ecx

55 xorl %eax,%eax

56 btl $21,%ecx

57 jnc .L000noluck

58 .byte 0x0f,0xa2

59 xorl %eax,%eax

60 cmpl $0x746e6543,%ebx

61 jne .L000noluck

62 cmpl $0x48727561,%edx

63 jne .L000noluck

64 cmpl $0x736c7561,%ecx

65 jne .L000noluck

66 movl $3221225472,%eax

67 .byte 0x0f,0xa2

68 movl %eax,%edx

69 xorl %eax,%eax

70 cmpl $3221225473,%edx

71 jb .L000noluck

72 movl $1,%eax

73 .byte 0x0f,0xa2

74 orl $15,%eax

75 xorl %ebx,%ebx

76 andl $4095,%eax

77 cmpl $1791,%eax

78 sete %bl

79 movl $3221225473,%eax

80 pushl %ebx

81 .byte 0x0f,0xa2

82 popl %ebx

83 movl %edx,%eax

84 shll $4,%ebx

85 andl $4294967279,%eax

86 orl %ebx,%eax

87 .L000noluck:

88 popl %ebx

89 ret

90 .size padlock_capability,.-.L_padlock_capability_begin

91 .globl padlock_key_bswap

92 .type padlock_key_bswap,@function

93 .align 16

94 padlock_key_bswap:

95 .L_padlock_key_bswap_begin:

96 movl 4(%esp),%edx

97 movl 240(%edx),%ecx

98 .L001bswap_loop:

99 movl (%edx),%eax

100 bswap %eax

101 movl %eax,(%edx)

102 leal 4(%edx),%edx

103 subl $1,%ecx

104 jnz .L001bswap_loop

105 ret

106 .size padlock_key_bswap,.-.L_padlock_key_bswap_begin

107 .globl padlock_verify_context

108 .type padlock_verify_context,@function

109 .align 16

110 padlock_verify_context:

111 .L_padlock_verify_context_begin:

112 movl 4(%esp),%edx

113 leal .Lpadlock_saved_context-.L002verify_pic_point,%eax

114 pushfl

115 call _padlock_verify_ctx

116 .L002verify_pic_point:

117 leal 4(%esp),%esp

118 ret

119 .size padlock_verify_context,.-.L_padlock_verify_context_begin

120 .type _padlock_verify_ctx,@function

121 .align 16

122 _padlock_verify_ctx:

123 addl (%esp),%eax

124 btl $30,4(%esp)

125 jnc .L003verified

126 cmpl (%eax),%edx

127 je .L003verified

128 pushfl

129 popfl

130 .L003verified:

131 movl %edx,(%eax)

132 ret

133 .size _padlock_verify_ctx,.-_padlock_verify_ctx

134 .globl padlock_reload_key

135 .type padlock_reload_key,@function

136 .align 16

137 padlock_reload_key:

138 .L_padlock_reload_key_begin:

139 pushfl

140 popfl

141 ret

142 .size padlock_reload_key,.-.L_padlock_reload_key_begin

143 .globl padlock_aes_block

144 .type padlock_aes_block,@function

145 .align 16

146 padlock_aes_block:

147 .L_padlock_aes_block_begin:

148 pushl %edi

149 pushl %esi

150 pushl %ebx

151 movl 16(%esp),%edi

152 movl 20(%esp),%esi

153 movl 24(%esp),%edx

154 movl $1,%ecx

155 leal 32(%edx),%ebx

156 leal 16(%edx),%edx

157 .byte 243,15,167,200

158 popl %ebx

159 popl %esi

160 popl %edi

161 ret

162 .size padlock_aes_block,.-.L_padlock_aes_block_begin

163 .globl padlock_ecb_encrypt

164 .type padlock_ecb_encrypt,@function

165 .align 16

166 padlock_ecb_encrypt:

167 .L_padlock_ecb_encrypt_begin:

168 pushl %ebp

169 pushl %ebx

170 pushl %esi

171 pushl %edi

172 movl 20(%esp),%edi

173 movl 24(%esp),%esi

174 movl 28(%esp),%edx

175 movl 32(%esp),%ecx

176 testl $15,%edx

177 jnz .L004ecb_abort

178 testl $15,%ecx

179 jnz .L004ecb_abort

180 leal .Lpadlock_saved_context-.L005ecb_pic_point,%eax

181 pushfl

182 cld

183 call _padlock_verify_ctx

184 .L005ecb_pic_point:

185 leal 16(%edx),%edx

186 xorl %eax,%eax

187 xorl %ebx,%ebx

188 cmpl $128,%ecx

189 jbe .L006ecb_short

190 testl $32,(%edx)

191 jnz .L007ecb_aligned

192 testl $15,%edi

193 setz %al

194 testl $15,%esi

195 setz %bl

196 testl %ebx,%eax

197 jnz .L007ecb_aligned

198 negl %eax

199 movl $512,%ebx

200 notl %eax

201 leal -24(%esp),%ebp

202 cmpl %ebx,%ecx

203 cmovcl %ecx,%ebx

204 andl %ebx,%eax

205 movl %ecx,%ebx

206 negl %eax

207 andl $511,%ebx

208 leal (%eax,%ebp,1),%esp

209 andl $-16,%esp

210 jmp .L008ecb_loop

211 .align 16

212 .L008ecb_loop:

213 movl %edi,(%ebp)

214 movl %esi,4(%ebp)

215 movl %ecx,8(%ebp)

216 movl %ebx,%ecx

217 movl %ebx,12(%ebp)

218 testl $15,%edi

219 cmovnzl %esp,%edi

220 testl $15,%esi

221 jz .L009ecb_inp_aligned

222 shrl $2,%ecx

223 .byte 243,165

224 subl %ebx,%edi

225 movl %ebx,%ecx

226 movl %edi,%esi

227 .L009ecb_inp_aligned:

228 leal -16(%edx),%eax

229 leal 16(%edx),%ebx

230 shrl $4,%ecx

231 .byte 243,15,167,200

232 movl (%ebp),%edi

233 movl 12(%ebp),%ebx

234 testl $15,%edi

235 jz .L010ecb_out_aligned

236 movl %ebx,%ecx

237 shrl $2,%ecx

238 leal (%esp),%esi

239 .byte 243,165

240 subl %ebx,%edi

241 .L010ecb_out_aligned:

242 movl 4(%ebp),%esi

243 movl 8(%ebp),%ecx

244 addl %ebx,%edi

245 addl %ebx,%esi

246 subl %ebx,%ecx

247 movl $512,%ebx

248 jnz .L008ecb_loop

249 cmpl %ebp,%esp

250 je .L011ecb_done

251 pxor %xmm0,%xmm0

252 leal (%esp),%eax

253 .L012ecb_bzero:

254 movaps %xmm0,(%eax)

255 leal 16(%eax),%eax

256 cmpl %eax,%ebp

257 ja .L012ecb_bzero

258 .L011ecb_done:

259 leal 24(%ebp),%esp

260 jmp .L013ecb_exit

261 .align 16

262 .L006ecb_short:

263 xorl %eax,%eax

264 leal -24(%esp),%ebp

265 subl %ecx,%eax

266 leal (%eax,%ebp,1),%esp

267 andl $-16,%esp

268 xorl %ebx,%ebx

269 .L014ecb_short_copy:

270 movups (%esi,%ebx,1),%xmm0

271 leal 16(%ebx),%ebx

272 cmpl %ebx,%ecx

273 movaps %xmm0,-16(%esp,%ebx,1)

274 ja .L014ecb_short_copy

275 movl %esp,%esi

276 movl %ecx,%ebx

277 jmp .L008ecb_loop

278 .align 16

279 .L007ecb_aligned:

280 leal -16(%edx),%eax

281 leal 16(%edx),%ebx

282 shrl $4,%ecx

283 .byte 243,15,167,200

284 .L013ecb_exit:

285 movl $1,%eax

286 leal 4(%esp),%esp

287 .L004ecb_abort:

288 popl %edi

289 popl %esi

290 popl %ebx

291 popl %ebp

292 ret

293 .size padlock_ecb_encrypt,.-.L_padlock_ecb_encrypt_begin

294 .globl padlock_cbc_encrypt

295 .type padlock_cbc_encrypt,@function

296 .align 16

297 padlock_cbc_encrypt:

298 .L_padlock_cbc_encrypt_begin:

299 pushl %ebp

300 pushl %ebx

301 pushl %esi

302 pushl %edi

303 movl 20(%esp),%edi

304 movl 24(%esp),%esi

305 movl 28(%esp),%edx

306 movl 32(%esp),%ecx

307 testl $15,%edx

308 jnz .L015cbc_abort

309 testl $15,%ecx

310 jnz .L015cbc_abort

311 leal .Lpadlock_saved_context-.L016cbc_pic_point,%eax

312 pushfl

313 cld

314 call _padlock_verify_ctx

315 .L016cbc_pic_point:

316 leal 16(%edx),%edx

317 xorl %eax,%eax

318 xorl %ebx,%ebx

319 cmpl $64,%ecx

320 jbe .L017cbc_short

321 testl $32,(%edx)

322 jnz .L018cbc_aligned

323 testl $15,%edi

324 setz %al

325 testl $15,%esi

326 setz %bl

327 testl %ebx,%eax

328 jnz .L018cbc_aligned

329 negl %eax

330 movl $512,%ebx

331 notl %eax

332 leal -24(%esp),%ebp

333 cmpl %ebx,%ecx

334 cmovcl %ecx,%ebx

335 andl %ebx,%eax

336 movl %ecx,%ebx

337 negl %eax

338 andl $511,%ebx

339 leal (%eax,%ebp,1),%esp

340 andl $-16,%esp

341 jmp .L019cbc_loop

342 .align 16

343 .L019cbc_loop:

344 movl %edi,(%ebp)

345 movl %esi,4(%ebp)

346 movl %ecx,8(%ebp)

347 movl %ebx,%ecx

348 movl %ebx,12(%ebp)

349 testl $15,%edi

350 cmovnzl %esp,%edi

351 testl $15,%esi

352 jz .L020cbc_inp_aligned

353 shrl $2,%ecx

354 .byte 243,165

355 subl %ebx,%edi

356 movl %ebx,%ecx

357 movl %edi,%esi

358 .L020cbc_inp_aligned:

359 leal -16(%edx),%eax

360 leal 16(%edx),%ebx

361 shrl $4,%ecx

362 .byte 243,15,167,208

363 movaps (%eax),%xmm0

364 movaps %xmm0,-16(%edx)

365 movl (%ebp),%edi

366 movl 12(%ebp),%ebx

367 testl $15,%edi

368 jz .L021cbc_out_aligned

369 movl %ebx,%ecx

370 shrl $2,%ecx

371 leal (%esp),%esi

372 .byte 243,165

373 subl %ebx,%edi

374 .L021cbc_out_aligned:

375 movl 4(%ebp),%esi

376 movl 8(%ebp),%ecx

377 addl %ebx,%edi

378 addl %ebx,%esi

379 subl %ebx,%ecx

380 movl $512,%ebx

381 jnz .L019cbc_loop

382 cmpl %ebp,%esp

383 je .L022cbc_done

384 pxor %xmm0,%xmm0

385 leal (%esp),%eax

386 .L023cbc_bzero:

387 movaps %xmm0,(%eax)

388 leal 16(%eax),%eax

389 cmpl %eax,%ebp

390 ja .L023cbc_bzero

391 .L022cbc_done:

392 leal 24(%ebp),%esp

393 jmp .L024cbc_exit

394 .align 16

395 .L017cbc_short:

396 xorl %eax,%eax

397 leal -24(%esp),%ebp

398 subl %ecx,%eax

399 leal (%eax,%ebp,1),%esp

400 andl $-16,%esp

401 xorl %ebx,%ebx

402 .L025cbc_short_copy:

403 movups (%esi,%ebx,1),%xmm0

404 leal 16(%ebx),%ebx

405 cmpl %ebx,%ecx

406 movaps %xmm0,-16(%esp,%ebx,1)

407 ja .L025cbc_short_copy

408 movl %esp,%esi

409 movl %ecx,%ebx

410 jmp .L019cbc_loop

411 .align 16

412 .L018cbc_aligned:

413 leal -16(%edx),%eax

414 leal 16(%edx),%ebx

415 shrl $4,%ecx

416 .byte 243,15,167,208

417 movaps (%eax),%xmm0

418 movaps %xmm0,-16(%edx)

419 .L024cbc_exit:

420 movl $1,%eax

421 leal 4(%esp),%esp

422 .L015cbc_abort:

423 popl %edi

424 popl %esi

425 popl %ebx

426 popl %ebp

427 ret

428 .size padlock_cbc_encrypt,.-.L_padlock_cbc_encrypt_begin

429 .globl padlock_xstore

430 .type padlock_xstore,@function

431 .align 16

432 padlock_xstore:

433 .L_padlock_xstore_begin:

434 pushl %edi

435 movl 8(%esp),%edi

436 movl 12(%esp),%edx

437 .byte 15,167,192

438 popl %edi

439 ret

440 .size padlock_xstore,.-.L_padlock_xstore_begin

441 .type _win32_segv_handler,@function

442 .align 16

443 _win32_segv_handler:

444 movl $1,%eax

445 movl 4(%esp),%edx

446 movl 12(%esp),%ecx

447 cmpl $3221225477,(%edx)

448 jne .L026ret

449 addl $4,184(%ecx)

450 movl $0,%eax

451 .L026ret:

452 ret

453 .size _win32_segv_handler,.-_win32_segv_handler

454 .globl padlock_sha1_oneshot

455 .type padlock_sha1_oneshot,@function

456 .align 16

457 padlock_sha1_oneshot:

458 .L_padlock_sha1_oneshot_begin:

459 pushl %edi

460 pushl %esi

461 xorl %eax,%eax

462 movl 12(%esp),%edi

463 movl 16(%esp),%esi

464 movl 20(%esp),%ecx

465 movl %esp,%edx

466 addl $-128,%esp

467 movups (%edi),%xmm0

468 andl $-16,%esp

469 movl 16(%edi),%eax

470 movaps %xmm0,(%esp)

471 movl %esp,%edi

472 movl %eax,16(%esp)

473 xorl %eax,%eax

474 .byte 243,15,166,200

475 movaps (%esp),%xmm0

476 movl 16(%esp),%eax

477 movl %edx,%esp

478 movl 12(%esp),%edi

479 movups %xmm0,(%edi)

480 movl %eax,16(%edi)

481 popl %esi

482 popl %edi

483 ret

484 .size padlock_sha1_oneshot,.-.L_padlock_sha1_oneshot_begin

485 .globl padlock_sha1_blocks

486 .type padlock_sha1_blocks,@function

487 .align 16

488 padlock_sha1_blocks:

489 .L_padlock_sha1_blocks_begin:

490 pushl %edi

491 pushl %esi

492 movl 12(%esp),%edi

493 movl 16(%esp),%esi

494 movl %esp,%edx

495 movl 20(%esp),%ecx

496 addl $-128,%esp

497 movups (%edi),%xmm0

498 andl $-16,%esp

499 movl 16(%edi),%eax

500 movaps %xmm0,(%esp)

501 movl %esp,%edi

502 movl %eax,16(%esp)

503 movl $-1,%eax

504 .byte 243,15,166,200

505 movaps (%esp),%xmm0

506 movl 16(%esp),%eax

507 movl %edx,%esp

508 movl 12(%esp),%edi

509 movups %xmm0,(%edi)

510 movl %eax,16(%edi)

511 popl %esi

512 popl %edi

513 ret

514 .size padlock_sha1_blocks,.-.L_padlock_sha1_blocks_begin

515 .globl padlock_sha256_oneshot

516 .type padlock_sha256_oneshot,@function

517 .align 16

518 padlock_sha256_oneshot:

519 .L_padlock_sha256_oneshot_begin:

520 pushl %edi

521 pushl %esi

522 xorl %eax,%eax

523 movl 12(%esp),%edi

524 movl 16(%esp),%esi

525 movl 20(%esp),%ecx

526 movl %esp,%edx

527 addl $-128,%esp

528 movups (%edi),%xmm0

529 andl $-16,%esp

530 movups 16(%edi),%xmm1

531 movaps %xmm0,(%esp)

532 movl %esp,%edi

533 movaps %xmm1,16(%esp)

534 xorl %eax,%eax

535 .byte 243,15,166,208

536 movaps (%esp),%xmm0

537 movaps 16(%esp),%xmm1

538 movl %edx,%esp

539 movl 12(%esp),%edi

540 movups %xmm0,(%edi)

541 movups %xmm1,16(%edi)

542 popl %esi

543 popl %edi

544 ret

545 .size padlock_sha256_oneshot,.-.L_padlock_sha256_oneshot_begin

546 .globl padlock_sha256_blocks

547 .type padlock_sha256_blocks,@function

548 .align 16

549 padlock_sha256_blocks:

550 .L_padlock_sha256_blocks_begin:

551 pushl %edi

552 pushl %esi

553 movl 12(%esp),%edi

554 movl 16(%esp),%esi

555 movl 20(%esp),%ecx

556 movl %esp,%edx

557 addl $-128,%esp

558 movups (%edi),%xmm0

559 andl $-16,%esp

560 movups 16(%edi),%xmm1

561 movaps %xmm0,(%esp)

562 movl %esp,%edi

563 movaps %xmm1,16(%esp)

564 movl $-1,%eax

565 .byte 243,15,166,208

566 movaps (%esp),%xmm0

567 movaps 16(%esp),%xmm1

568 movl %edx,%esp

569 movl 12(%esp),%edi

570 movups %xmm0,(%edi)

571 movups %xmm1,16(%edi)

572 popl %esi

573 popl %edi

574 ret

575 .size padlock_sha256_blocks,.-.L_padlock_sha256_blocks_begin

576 .globl padlock_sha512_blocks

577 .type padlock_sha512_blocks,@function

578 .align 16

579 padlock_sha512_blocks:

580 .L_padlock_sha512_blocks_begin:

581 pushl %edi

582 pushl %esi

583 movl 12(%esp),%edi

584 movl 16(%esp),%esi

585 movl 20(%esp),%ecx

586 movl %esp,%edx

587 addl $-128,%esp

588 movups (%edi),%xmm0

589 andl $-16,%esp

590 movups 16(%edi),%xmm1

591 movups 32(%edi),%xmm2

592 movups 48(%edi),%xmm3

593 movaps %xmm0,(%esp)

594 movl %esp,%edi

595 movaps %xmm1,16(%esp)

596 movaps %xmm2,32(%esp)

597 movaps %xmm3,48(%esp)

598 .byte 243,15,166,224

599 movaps (%esp),%xmm0

600 movaps 16(%esp),%xmm1

601 movaps 32(%esp),%xmm2

602 movaps 48(%esp),%xmm3

603 movl %edx,%esp

604 movl 12(%esp),%edi

605 movups %xmm0,(%edi)

606 movups %xmm1,16(%edi)

607 movups %xmm2,32(%edi)

608 movups %xmm3,48(%edi)

609 popl %esi

610 popl %edi

611 ret

612 .size padlock_sha512_blocks,.-.L_padlock_sha512_blocks_begin

613 .byte 86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,32

614 .byte 109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65

615 .byte 77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101

616 .byte 110,115,115,108,46,111,114,103,62,0

617 .align 16

618 .data

619 .align 4

620 .Lpadlock_saved_context:

621 .long 0

622

623 .section .note.GNU-stack,"",%progbits