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.

38 .text

39 .globl padlock_capability

40 .type padlock_capability,@function

41 .align 16

42 padlock_capability:

43 movq %rbx,%r8

44 xorl %eax,%eax

45 cpuid

46 xorl %eax,%eax

47 cmpl $1953391939,%ebx

48 jne .Lnoluck

49 cmpl $1215460705,%edx

50 jne .Lnoluck

51 cmpl $1936487777,%ecx

52 jne .Lnoluck

53 movl $3221225472,%eax

54 cpuid

55 movl %eax,%edx

56 xorl %eax,%eax

57 cmpl $3221225473,%edx

58 jb .Lnoluck

59 movl $3221225473,%eax

60 cpuid

61 movl %edx,%eax

62 andl $4294967279,%eax

63 orl $16,%eax

64 .Lnoluck:

65 movq %r8,%rbx

66 .byte 0xf3,0xc3

67 .size padlock_capability,.-padlock_capability

69 .globl padlock_key_bswap

70 .type padlock_key_bswap,@function

71 .align 16

72 padlock_key_bswap:

73 movl 240(%rdi),%edx

74 .Lbswap_loop:

75 movl (%rdi),%eax

76 bswapl %eax

77 movl %eax,(%rdi)

78 leaq 4(%rdi),%rdi

79 subl $1,%edx

80 jnz .Lbswap_loop

81 .byte 0xf3,0xc3

82 .size padlock_key_bswap,.-padlock_key_bswap

84 .globl padlock_verify_context

85 .type padlock_verify_context,@function

86 .align 16

87 padlock_verify_context:

88 movq %rdi,%rdx

89 pushf

90 leaq .Lpadlock_saved_context(%rip),%rax

91 call _padlock_verify_ctx

92 leaq 8(%rsp),%rsp

93 .byte 0xf3,0xc3

94 .size padlock_verify_context,.-padlock_verify_context

96 .type _padlock_verify_ctx,@function

97 .align 16

98 _padlock_verify_ctx:

99 movq 8(%rsp),%r8

100 btq $30,%r8

101 jnc .Lverified

102 cmpq (%rax),%rdx

103 je .Lverified

104 pushf

105 popf

106 .Lverified:

107 movq %rdx,(%rax)

108 .byte 0xf3,0xc3

109 .size _padlock_verify_ctx,.-_padlock_verify_ctx

110

111 .globl padlock_reload_key

112 .type padlock_reload_key,@function

113 .align 16

114 padlock_reload_key:

115 pushf

116 popf

117 .byte 0xf3,0xc3

118 .size padlock_reload_key,.-padlock_reload_key

119

120 .globl padlock_aes_block

121 .type padlock_aes_block,@function

122 .align 16

123 padlock_aes_block:

124 movq %rbx,%r8

125 movq $1,%rcx

126 leaq 32(%rdx),%rbx

127 leaq 16(%rdx),%rdx

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

129 movq %r8,%rbx

130 .byte 0xf3,0xc3

131 .size padlock_aes_block,.-padlock_aes_block

132

133 .globl padlock_xstore

134 .type padlock_xstore,@function

135 .align 16

136 padlock_xstore:

137 movl %esi,%edx

138 .byte 0x0f,0xa7,0xc0

139 .byte 0xf3,0xc3

140 .size padlock_xstore,.-padlock_xstore

141

142 .globl padlock_sha1_oneshot

143 .type padlock_sha1_oneshot,@function

144 .align 16

145 padlock_sha1_oneshot:

146 movq %rdx,%rcx

147 movq %rdi,%rdx

148 movups (%rdi),%xmm0

149 subq $128+8,%rsp

150 movl 16(%rdi),%eax

151 movaps %xmm0,(%rsp)

152 movq %rsp,%rdi

153 movl %eax,16(%rsp)

154 xorq %rax,%rax

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

156 movaps (%rsp),%xmm0

157 movl 16(%rsp),%eax

158 addq $128+8,%rsp

159 movups %xmm0,(%rdx)

160 movl %eax,16(%rdx)

161 .byte 0xf3,0xc3

162 .size padlock_sha1_oneshot,.-padlock_sha1_oneshot

163

164 .globl padlock_sha1_blocks

165 .type padlock_sha1_blocks,@function

166 .align 16

167 padlock_sha1_blocks:

168 movq %rdx,%rcx

169 movq %rdi,%rdx

170 movups (%rdi),%xmm0

171 subq $128+8,%rsp

172 movl 16(%rdi),%eax

173 movaps %xmm0,(%rsp)

174 movq %rsp,%rdi

175 movl %eax,16(%rsp)

176 movq $-1,%rax

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

178 movaps (%rsp),%xmm0

179 movl 16(%rsp),%eax

180 addq $128+8,%rsp

181 movups %xmm0,(%rdx)

182 movl %eax,16(%rdx)

183 .byte 0xf3,0xc3

184 .size padlock_sha1_blocks,.-padlock_sha1_blocks

185

186 .globl padlock_sha256_oneshot

187 .type padlock_sha256_oneshot,@function

188 .align 16

189 padlock_sha256_oneshot:

190 movq %rdx,%rcx

191 movq %rdi,%rdx

192 movups (%rdi),%xmm0

193 subq $128+8,%rsp

194 movups 16(%rdi),%xmm1

195 movaps %xmm0,(%rsp)

196 movq %rsp,%rdi

197 movaps %xmm1,16(%rsp)

198 xorq %rax,%rax

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

200 movaps (%rsp),%xmm0

201 movaps 16(%rsp),%xmm1

202 addq $128+8,%rsp

203 movups %xmm0,(%rdx)

204 movups %xmm1,16(%rdx)

205 .byte 0xf3,0xc3

206 .size padlock_sha256_oneshot,.-padlock_sha256_oneshot

207

208 .globl padlock_sha256_blocks

209 .type padlock_sha256_blocks,@function

210 .align 16

211 padlock_sha256_blocks:

212 movq %rdx,%rcx

213 movq %rdi,%rdx

214 movups (%rdi),%xmm0

215 subq $128+8,%rsp

216 movups 16(%rdi),%xmm1

217 movaps %xmm0,(%rsp)

218 movq %rsp,%rdi

219 movaps %xmm1,16(%rsp)

220 movq $-1,%rax

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

222 movaps (%rsp),%xmm0

223 movaps 16(%rsp),%xmm1

224 addq $128+8,%rsp

225 movups %xmm0,(%rdx)

226 movups %xmm1,16(%rdx)

227 .byte 0xf3,0xc3

228 .size padlock_sha256_blocks,.-padlock_sha256_blocks

229

230 .globl padlock_sha512_blocks

231 .type padlock_sha512_blocks,@function

232 .align 16

233 padlock_sha512_blocks:

234 movq %rdx,%rcx

235 movq %rdi,%rdx

236 movups (%rdi),%xmm0

237 subq $128+8,%rsp

238 movups 16(%rdi),%xmm1

239 movups 32(%rdi),%xmm2

240 movups 48(%rdi),%xmm3

241 movaps %xmm0,(%rsp)

242 movq %rsp,%rdi

243 movaps %xmm1,16(%rsp)

244 movaps %xmm2,32(%rsp)

245 movaps %xmm3,48(%rsp)

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

247 movaps (%rsp),%xmm0

248 movaps 16(%rsp),%xmm1

249 movaps 32(%rsp),%xmm2

250 movaps 48(%rsp),%xmm3

251 addq $128+8,%rsp

252 movups %xmm0,(%rdx)

253 movups %xmm1,16(%rdx)

254 movups %xmm2,32(%rdx)

255 movups %xmm3,48(%rdx)

256 .byte 0xf3,0xc3

257 .size padlock_sha512_blocks,.-padlock_sha512_blocks

258 .globl padlock_ecb_encrypt

259 .type padlock_ecb_encrypt,@function

260 .align 16

261 padlock_ecb_encrypt:

262 pushq %rbp

263 pushq %rbx

264

265 xorl %eax,%eax

266 testq $15,%rdx

267 jnz .Lecb_abort

268 testq $15,%rcx

269 jnz .Lecb_abort

270 leaq .Lpadlock_saved_context(%rip),%rax

271 pushf

272 cld

273 call _padlock_verify_ctx

274 leaq 16(%rdx),%rdx

275 xorl %eax,%eax

276 xorl %ebx,%ebx

277 cmpq $128,%rcx

278 jbe .Lecb_short

279 testl $32,(%rdx)

280 jnz .Lecb_aligned

281 testq $15,%rdi

282 setz %al

283 testq $15,%rsi

284 setz %bl

285 testl %ebx,%eax

286 jnz .Lecb_aligned

287 negq %rax

288 movq $512,%rbx

289 notq %rax

290 leaq (%rsp),%rbp

291 cmpq %rbx,%rcx

292 cmovcq %rcx,%rbx

293 andq %rbx,%rax

294 movq %rcx,%rbx

295 negq %rax

296 andq $512-1,%rbx

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

298 jmp .Lecb_loop

299 .align 16

300 .Lecb_loop:

301 cmpq %rcx,%rbx

302 cmovaq %rcx,%rbx

303 movq %rdi,%r8

304 movq %rsi,%r9

305 movq %rcx,%r10

306 movq %rbx,%rcx

307 movq %rbx,%r11

308 testq $15,%rdi

309 cmovnzq %rsp,%rdi

310 testq $15,%rsi

311 jz .Lecb_inp_aligned

312 shrq $3,%rcx

313 .byte 0xf3,0x48,0xa5

314 subq %rbx,%rdi

315 movq %rbx,%rcx

316 movq %rdi,%rsi

317 .Lecb_inp_aligned:

318 leaq -16(%rdx),%rax

319 leaq 16(%rdx),%rbx

320 shrq $4,%rcx

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

322 movq %r8,%rdi

323 movq %r11,%rbx

324 testq $15,%rdi

325 jz .Lecb_out_aligned

326 movq %rbx,%rcx

327 shrq $3,%rcx

328 leaq (%rsp),%rsi

329 .byte 0xf3,0x48,0xa5

330 subq %rbx,%rdi

331 .Lecb_out_aligned:

332 movq %r9,%rsi

333 movq %r10,%rcx

334 addq %rbx,%rdi

335 addq %rbx,%rsi

336 subq %rbx,%rcx

337 movq $512,%rbx

338 jnz .Lecb_loop

339

340 cmpq %rsp,%rbp

341 je .Lecb_done

342

343 pxor %xmm0,%xmm0

344 leaq (%rsp),%rax

345 .Lecb_bzero:

346 movaps %xmm0,(%rax)

347 leaq 16(%rax),%rax

348 cmpq %rax,%rbp

349 ja .Lecb_bzero

350

351 .Lecb_done:

352 leaq (%rbp),%rsp

353 jmp .Lecb_exit

354 .align 16

355 .Lecb_short:

356 movq %rsp,%rbp

357 subq %rcx,%rsp

358 xorq %rbx,%rbx

359 .Lecb_short_copy:

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

361 leaq 16(%rbx),%rbx

362 cmpq %rbx,%rcx

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

364 ja .Lecb_short_copy

365 movq %rsp,%rsi

366 movq %rcx,%rbx

367 jmp .Lecb_loop

368 .align 16

369 .Lecb_aligned:

370 leaq -16(%rdx),%rax

371 leaq 16(%rdx),%rbx

372 shrq $4,%rcx

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

374 .Lecb_exit:

375 movl $1,%eax

376 leaq 8(%rsp),%rsp

377 .Lecb_abort:

378 popq %rbx

379 popq %rbp

380 .byte 0xf3,0xc3

381 .size padlock_ecb_encrypt,.-padlock_ecb_encrypt

382 .globl padlock_cbc_encrypt

383 .type padlock_cbc_encrypt,@function

384 .align 16

385 padlock_cbc_encrypt:

386 pushq %rbp

387 pushq %rbx

388

389 xorl %eax,%eax

390 testq $15,%rdx

391 jnz .Lcbc_abort

392 testq $15,%rcx

393 jnz .Lcbc_abort

394 leaq .Lpadlock_saved_context(%rip),%rax

395 pushf

396 cld

397 call _padlock_verify_ctx

398 leaq 16(%rdx),%rdx

399 xorl %eax,%eax

400 xorl %ebx,%ebx

401 cmpq $64,%rcx

402 jbe .Lcbc_short

403 testl $32,(%rdx)

404 jnz .Lcbc_aligned

405 testq $15,%rdi

406 setz %al

407 testq $15,%rsi

408 setz %bl

409 testl %ebx,%eax

410 jnz .Lcbc_aligned

411 negq %rax

412 movq $512,%rbx

413 notq %rax

414 leaq (%rsp),%rbp

415 cmpq %rbx,%rcx

416 cmovcq %rcx,%rbx

417 andq %rbx,%rax

418 movq %rcx,%rbx

419 negq %rax

420 andq $512-1,%rbx

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

422 jmp .Lcbc_loop

423 .align 16

424 .Lcbc_loop:

425 cmpq %rcx,%rbx

426 cmovaq %rcx,%rbx

427 movq %rdi,%r8

428 movq %rsi,%r9

429 movq %rcx,%r10

430 movq %rbx,%rcx

431 movq %rbx,%r11

432 testq $15,%rdi

433 cmovnzq %rsp,%rdi

434 testq $15,%rsi

435 jz .Lcbc_inp_aligned

436 shrq $3,%rcx

437 .byte 0xf3,0x48,0xa5

438 subq %rbx,%rdi

439 movq %rbx,%rcx

440 movq %rdi,%rsi

441 .Lcbc_inp_aligned:

442 leaq -16(%rdx),%rax

443 leaq 16(%rdx),%rbx

444 shrq $4,%rcx

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

446 movdqa (%rax),%xmm0

447 movdqa %xmm0,-16(%rdx)

448 movq %r8,%rdi

449 movq %r11,%rbx

450 testq $15,%rdi

451 jz .Lcbc_out_aligned

452 movq %rbx,%rcx

453 shrq $3,%rcx

454 leaq (%rsp),%rsi

455 .byte 0xf3,0x48,0xa5

456 subq %rbx,%rdi

457 .Lcbc_out_aligned:

458 movq %r9,%rsi

459 movq %r10,%rcx

460 addq %rbx,%rdi

461 addq %rbx,%rsi

462 subq %rbx,%rcx

463 movq $512,%rbx

464 jnz .Lcbc_loop

465

466 cmpq %rsp,%rbp

467 je .Lcbc_done

468

469 pxor %xmm0,%xmm0

470 leaq (%rsp),%rax

471 .Lcbc_bzero:

472 movaps %xmm0,(%rax)

473 leaq 16(%rax),%rax

474 cmpq %rax,%rbp

475 ja .Lcbc_bzero

476

477 .Lcbc_done:

478 leaq (%rbp),%rsp

479 jmp .Lcbc_exit

480 .align 16

481 .Lcbc_short:

482 movq %rsp,%rbp

483 subq %rcx,%rsp

484 xorq %rbx,%rbx

485 .Lcbc_short_copy:

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

487 leaq 16(%rbx),%rbx

488 cmpq %rbx,%rcx

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

490 ja .Lcbc_short_copy

491 movq %rsp,%rsi

492 movq %rcx,%rbx

493 jmp .Lcbc_loop

494 .align 16

495 .Lcbc_aligned:

496 leaq -16(%rdx),%rax

497 leaq 16(%rdx),%rbx

498 shrq $4,%rcx

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

500 movdqa (%rax),%xmm0

501 movdqa %xmm0,-16(%rdx)

502 .Lcbc_exit:

503 movl $1,%eax

504 leaq 8(%rsp),%rsp

505 .Lcbc_abort:

506 popq %rbx

507 popq %rbp

508 .byte 0xf3,0xc3

509 .size padlock_cbc_encrypt,.-padlock_cbc_encrypt

510 .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

511 .align 16

512 .data

513 .align 8

514 .Lpadlock_saved_context:

515 .quad 0

516

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