www.gitorious.org Git - gnutls:gnutls.git/blob - lib/accelerated/x86/elf/padlock-x86-64.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.

37 #

38 # *** This file is auto-generated ***

39 #

40 .text

41 .globl padlock_capability

42 .type padlock_capability,@function

43 .align 16

44 padlock_capability:

45 movq %rbx,%r8

46 xorl %eax,%eax

47 cpuid

48 xorl %eax,%eax

49 cmpl $1953391939,%ebx

50 jne .Lnoluck

51 cmpl $1215460705,%edx

52 jne .Lnoluck

53 cmpl $1936487777,%ecx

54 jne .Lnoluck

55 movl $3221225472,%eax

56 cpuid

57 movl %eax,%edx

58 xorl %eax,%eax

59 cmpl $3221225473,%edx

60 jb .Lnoluck

61 movl $3221225473,%eax

62 cpuid

63 movl %edx,%eax

64 andl $4294967279,%eax

65 orl $16,%eax

66 .Lnoluck:

67 movq %r8,%rbx

68 .byte 0xf3,0xc3

69 .size padlock_capability,.-padlock_capability

71 .globl padlock_key_bswap

72 .type padlock_key_bswap,@function

73 .align 16

74 padlock_key_bswap:

75 movl 240(%rdi),%edx

76 .Lbswap_loop:

77 movl (%rdi),%eax

78 bswapl %eax

79 movl %eax,(%rdi)

80 leaq 4(%rdi),%rdi

81 subl $1,%edx

82 jnz .Lbswap_loop

83 .byte 0xf3,0xc3

84 .size padlock_key_bswap,.-padlock_key_bswap

86 .globl padlock_verify_context

87 .type padlock_verify_context,@function

88 .align 16

89 padlock_verify_context:

90 movq %rdi,%rdx

91 pushf

92 leaq .Lpadlock_saved_context(%rip),%rax

93 call _padlock_verify_ctx

94 leaq 8(%rsp),%rsp

95 .byte 0xf3,0xc3

96 .size padlock_verify_context,.-padlock_verify_context

98 .type _padlock_verify_ctx,@function

99 .align 16

100 _padlock_verify_ctx:

101 movq 8(%rsp),%r8

102 btq $30,%r8

103 jnc .Lverified

104 cmpq (%rax),%rdx

105 je .Lverified

106 pushf

107 popf

108 .Lverified:

109 movq %rdx,(%rax)

110 .byte 0xf3,0xc3

111 .size _padlock_verify_ctx,.-_padlock_verify_ctx

112

113 .globl padlock_reload_key

114 .type padlock_reload_key,@function

115 .align 16

116 padlock_reload_key:

117 pushf

118 popf

119 .byte 0xf3,0xc3

120 .size padlock_reload_key,.-padlock_reload_key

121

122 .globl padlock_aes_block

123 .type padlock_aes_block,@function

124 .align 16

125 padlock_aes_block:

126 movq %rbx,%r8

127 movq $1,%rcx

128 leaq 32(%rdx),%rbx

129 leaq 16(%rdx),%rdx

130 .byte 0xf3,0x0f,0xa7,0xc8

131 movq %r8,%rbx

132 .byte 0xf3,0xc3

133 .size padlock_aes_block,.-padlock_aes_block

134

135 .globl padlock_xstore

136 .type padlock_xstore,@function

137 .align 16

138 padlock_xstore:

139 movl %esi,%edx

140 .byte 0x0f,0xa7,0xc0

141 .byte 0xf3,0xc3

142 .size padlock_xstore,.-padlock_xstore

143

144 .globl padlock_sha1_oneshot

145 .type padlock_sha1_oneshot,@function

146 .align 16

147 padlock_sha1_oneshot:

148 movq %rdx,%rcx

149 movq %rdi,%rdx

150 movups (%rdi),%xmm0

151 subq $128+8,%rsp

152 movl 16(%rdi),%eax

153 movaps %xmm0,(%rsp)

154 movq %rsp,%rdi

155 movl %eax,16(%rsp)

156 xorq %rax,%rax

157 .byte 0xf3,0x0f,0xa6,0xc8

158 movaps (%rsp),%xmm0

159 movl 16(%rsp),%eax

160 addq $128+8,%rsp

161 movups %xmm0,(%rdx)

162 movl %eax,16(%rdx)

163 .byte 0xf3,0xc3

164 .size padlock_sha1_oneshot,.-padlock_sha1_oneshot

165

166 .globl padlock_sha1_blocks

167 .type padlock_sha1_blocks,@function

168 .align 16

169 padlock_sha1_blocks:

170 movq %rdx,%rcx

171 movq %rdi,%rdx

172 movups (%rdi),%xmm0

173 subq $128+8,%rsp

174 movl 16(%rdi),%eax

175 movaps %xmm0,(%rsp)

176 movq %rsp,%rdi

177 movl %eax,16(%rsp)

178 movq $-1,%rax

179 .byte 0xf3,0x0f,0xa6,0xc8

180 movaps (%rsp),%xmm0

181 movl 16(%rsp),%eax

182 addq $128+8,%rsp

183 movups %xmm0,(%rdx)

184 movl %eax,16(%rdx)

185 .byte 0xf3,0xc3

186 .size padlock_sha1_blocks,.-padlock_sha1_blocks

187

188 .globl padlock_sha256_oneshot

189 .type padlock_sha256_oneshot,@function

190 .align 16

191 padlock_sha256_oneshot:

192 movq %rdx,%rcx

193 movq %rdi,%rdx

194 movups (%rdi),%xmm0

195 subq $128+8,%rsp

196 movups 16(%rdi),%xmm1

197 movaps %xmm0,(%rsp)

198 movq %rsp,%rdi

199 movaps %xmm1,16(%rsp)

200 xorq %rax,%rax

201 .byte 0xf3,0x0f,0xa6,0xd0

202 movaps (%rsp),%xmm0

203 movaps 16(%rsp),%xmm1

204 addq $128+8,%rsp

205 movups %xmm0,(%rdx)

206 movups %xmm1,16(%rdx)

207 .byte 0xf3,0xc3

208 .size padlock_sha256_oneshot,.-padlock_sha256_oneshot

209

210 .globl padlock_sha256_blocks

211 .type padlock_sha256_blocks,@function

212 .align 16

213 padlock_sha256_blocks:

214 movq %rdx,%rcx

215 movq %rdi,%rdx

216 movups (%rdi),%xmm0

217 subq $128+8,%rsp

218 movups 16(%rdi),%xmm1

219 movaps %xmm0,(%rsp)

220 movq %rsp,%rdi

221 movaps %xmm1,16(%rsp)

222 movq $-1,%rax

223 .byte 0xf3,0x0f,0xa6,0xd0

224 movaps (%rsp),%xmm0

225 movaps 16(%rsp),%xmm1

226 addq $128+8,%rsp

227 movups %xmm0,(%rdx)

228 movups %xmm1,16(%rdx)

229 .byte 0xf3,0xc3

230 .size padlock_sha256_blocks,.-padlock_sha256_blocks

231

232 .globl padlock_sha512_blocks

233 .type padlock_sha512_blocks,@function

234 .align 16

235 padlock_sha512_blocks:

236 movq %rdx,%rcx

237 movq %rdi,%rdx

238 movups (%rdi),%xmm0

239 subq $128+8,%rsp

240 movups 16(%rdi),%xmm1

241 movups 32(%rdi),%xmm2

242 movups 48(%rdi),%xmm3

243 movaps %xmm0,(%rsp)

244 movq %rsp,%rdi

245 movaps %xmm1,16(%rsp)

246 movaps %xmm2,32(%rsp)

247 movaps %xmm3,48(%rsp)

248 .byte 0xf3,0x0f,0xa6,0xe0

249 movaps (%rsp),%xmm0

250 movaps 16(%rsp),%xmm1

251 movaps 32(%rsp),%xmm2

252 movaps 48(%rsp),%xmm3

253 addq $128+8,%rsp

254 movups %xmm0,(%rdx)

255 movups %xmm1,16(%rdx)

256 movups %xmm2,32(%rdx)

257 movups %xmm3,48(%rdx)

258 .byte 0xf3,0xc3

259 .size padlock_sha512_blocks,.-padlock_sha512_blocks

260 .globl padlock_ecb_encrypt

261 .type padlock_ecb_encrypt,@function

262 .align 16

263 padlock_ecb_encrypt:

264 pushq %rbp

265 pushq %rbx

266

267 xorl %eax,%eax

268 testq $15,%rdx

269 jnz .Lecb_abort

270 testq $15,%rcx

271 jnz .Lecb_abort

272 leaq .Lpadlock_saved_context(%rip),%rax

273 pushf

274 cld

275 call _padlock_verify_ctx

276 leaq 16(%rdx),%rdx

277 xorl %eax,%eax

278 xorl %ebx,%ebx

279 cmpq $128,%rcx

280 jbe .Lecb_short

281 testl $32,(%rdx)

282 jnz .Lecb_aligned

283 testq $15,%rdi

284 setz %al

285 testq $15,%rsi

286 setz %bl

287 testl %ebx,%eax

288 jnz .Lecb_aligned

289 negq %rax

290 movq $512,%rbx

291 notq %rax

292 leaq (%rsp),%rbp

293 cmpq %rbx,%rcx

294 cmovcq %rcx,%rbx

295 andq %rbx,%rax

296 movq %rcx,%rbx

297 negq %rax

298 andq $512-1,%rbx

299 leaq (%rax,%rbp,1),%rsp

300 jmp .Lecb_loop

301 .align 16

302 .Lecb_loop:

303 cmpq %rcx,%rbx

304 cmovaq %rcx,%rbx

305 movq %rdi,%r8

306 movq %rsi,%r9

307 movq %rcx,%r10

308 movq %rbx,%rcx

309 movq %rbx,%r11

310 testq $15,%rdi

311 cmovnzq %rsp,%rdi

312 testq $15,%rsi

313 jz .Lecb_inp_aligned

314 shrq $3,%rcx

315 .byte 0xf3,0x48,0xa5

316 subq %rbx,%rdi

317 movq %rbx,%rcx

318 movq %rdi,%rsi

319 .Lecb_inp_aligned:

320 leaq -16(%rdx),%rax

321 leaq 16(%rdx),%rbx

322 shrq $4,%rcx

323 .byte 0xf3,0x0f,0xa7,200

324 movq %r8,%rdi

325 movq %r11,%rbx

326 testq $15,%rdi

327 jz .Lecb_out_aligned

328 movq %rbx,%rcx

329 shrq $3,%rcx

330 leaq (%rsp),%rsi

331 .byte 0xf3,0x48,0xa5

332 subq %rbx,%rdi

333 .Lecb_out_aligned:

334 movq %r9,%rsi

335 movq %r10,%rcx

336 addq %rbx,%rdi

337 addq %rbx,%rsi

338 subq %rbx,%rcx

339 movq $512,%rbx

340 jnz .Lecb_loop

341

342 cmpq %rsp,%rbp

343 je .Lecb_done

344

345 pxor %xmm0,%xmm0

346 leaq (%rsp),%rax

347 .Lecb_bzero:

348 movaps %xmm0,(%rax)

349 leaq 16(%rax),%rax

350 cmpq %rax,%rbp

351 ja .Lecb_bzero

352

353 .Lecb_done:

354 leaq (%rbp),%rsp

355 jmp .Lecb_exit

356 .align 16

357 .Lecb_short:

358 movq %rsp,%rbp

359 subq %rcx,%rsp

360 xorq %rbx,%rbx

361 .Lecb_short_copy:

362 movups (%rsi,%rbx,1),%xmm0

363 leaq 16(%rbx),%rbx

364 cmpq %rbx,%rcx

365 movaps %xmm0,-16(%rsp,%rbx,1)

366 ja .Lecb_short_copy

367 movq %rsp,%rsi

368 movq %rcx,%rbx

369 jmp .Lecb_loop

370 .align 16

371 .Lecb_aligned:

372 leaq -16(%rdx),%rax

373 leaq 16(%rdx),%rbx

374 shrq $4,%rcx

375 .byte 0xf3,0x0f,0xa7,200

376 .Lecb_exit:

377 movl $1,%eax

378 leaq 8(%rsp),%rsp

379 .Lecb_abort:

380 popq %rbx

381 popq %rbp

382 .byte 0xf3,0xc3

383 .size padlock_ecb_encrypt,.-padlock_ecb_encrypt

384 .globl padlock_cbc_encrypt

385 .type padlock_cbc_encrypt,@function

386 .align 16

387 padlock_cbc_encrypt:

388 pushq %rbp

389 pushq %rbx

390

391 xorl %eax,%eax

392 testq $15,%rdx

393 jnz .Lcbc_abort

394 testq $15,%rcx

395 jnz .Lcbc_abort

396 leaq .Lpadlock_saved_context(%rip),%rax

397 pushf

398 cld

399 call _padlock_verify_ctx

400 leaq 16(%rdx),%rdx

401 xorl %eax,%eax

402 xorl %ebx,%ebx

403 cmpq $64,%rcx

404 jbe .Lcbc_short

405 testl $32,(%rdx)

406 jnz .Lcbc_aligned

407 testq $15,%rdi

408 setz %al

409 testq $15,%rsi

410 setz %bl

411 testl %ebx,%eax

412 jnz .Lcbc_aligned

413 negq %rax

414 movq $512,%rbx

415 notq %rax

416 leaq (%rsp),%rbp

417 cmpq %rbx,%rcx

418 cmovcq %rcx,%rbx

419 andq %rbx,%rax

420 movq %rcx,%rbx

421 negq %rax

422 andq $512-1,%rbx

423 leaq (%rax,%rbp,1),%rsp

424 jmp .Lcbc_loop

425 .align 16

426 .Lcbc_loop:

427 cmpq %rcx,%rbx

428 cmovaq %rcx,%rbx

429 movq %rdi,%r8

430 movq %rsi,%r9

431 movq %rcx,%r10

432 movq %rbx,%rcx

433 movq %rbx,%r11

434 testq $15,%rdi

435 cmovnzq %rsp,%rdi

436 testq $15,%rsi

437 jz .Lcbc_inp_aligned

438 shrq $3,%rcx

439 .byte 0xf3,0x48,0xa5

440 subq %rbx,%rdi

441 movq %rbx,%rcx

442 movq %rdi,%rsi

443 .Lcbc_inp_aligned:

444 leaq -16(%rdx),%rax

445 leaq 16(%rdx),%rbx

446 shrq $4,%rcx

447 .byte 0xf3,0x0f,0xa7,208

448 movdqa (%rax),%xmm0

449 movdqa %xmm0,-16(%rdx)

450 movq %r8,%rdi

451 movq %r11,%rbx

452 testq $15,%rdi

453 jz .Lcbc_out_aligned

454 movq %rbx,%rcx

455 shrq $3,%rcx

456 leaq (%rsp),%rsi

457 .byte 0xf3,0x48,0xa5

458 subq %rbx,%rdi

459 .Lcbc_out_aligned:

460 movq %r9,%rsi

461 movq %r10,%rcx

462 addq %rbx,%rdi

463 addq %rbx,%rsi

464 subq %rbx,%rcx

465 movq $512,%rbx

466 jnz .Lcbc_loop

467

468 cmpq %rsp,%rbp

469 je .Lcbc_done

470

471 pxor %xmm0,%xmm0

472 leaq (%rsp),%rax

473 .Lcbc_bzero:

474 movaps %xmm0,(%rax)

475 leaq 16(%rax),%rax

476 cmpq %rax,%rbp

477 ja .Lcbc_bzero

478

479 .Lcbc_done:

480 leaq (%rbp),%rsp

481 jmp .Lcbc_exit

482 .align 16

483 .Lcbc_short:

484 movq %rsp,%rbp

485 subq %rcx,%rsp

486 xorq %rbx,%rbx

487 .Lcbc_short_copy:

488 movups (%rsi,%rbx,1),%xmm0

489 leaq 16(%rbx),%rbx

490 cmpq %rbx,%rcx

491 movaps %xmm0,-16(%rsp,%rbx,1)

492 ja .Lcbc_short_copy

493 movq %rsp,%rsi

494 movq %rcx,%rbx

495 jmp .Lcbc_loop

496 .align 16

497 .Lcbc_aligned:

498 leaq -16(%rdx),%rax

499 leaq 16(%rdx),%rbx

500 shrq $4,%rcx

501 .byte 0xf3,0x0f,0xa7,208

502 movdqa (%rax),%xmm0

503 movdqa %xmm0,-16(%rdx)

504 .Lcbc_exit:

505 movl $1,%eax

506 leaq 8(%rsp),%rsp

507 .Lcbc_abort:

508 popq %rbx

509 popq %rbp

510 .byte 0xf3,0xc3

511 .size padlock_cbc_encrypt,.-padlock_cbc_encrypt

512 .byte 86,73,65,32,80,97,100,108,111,99,107,32,120,56,54,95,54,52,32,109,111,100,117,108,101,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0

513 .align 16

514 .data

515 .align 8

516 .Lpadlock_saved_context:

517 .quad 0

518

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