1 #!/usr/bin/env perl 2 # 3 # ==================================================================== 4 # Written by Andy Polyakov <appro (at] fy.chalmers.se> for the OpenSSL 5 # project. The module is, however, dual licensed under OpenSSL and 6 # CRYPTOGAMS licenses depending on where you obtain it. For further 7 # details see http://www.openssl.org/~appro/cryptogams/. 8 # ==================================================================== 9 # 10 # This module implements support for Intel AES-NI extension. In 11 # OpenSSL context it's used with Intel engine, but can also be used as 12 # drop-in replacement for crypto/aes/asm/aes-x86_64.pl [see below for 13 # details]. 14 # 15 # Performance. 16 # 17 # Given aes(enc|dec) instructions' latency asymptotic performance for 18 # non-parallelizable modes such as CBC encrypt is 3.75 cycles per byte 19 # processed with 128-bit key. And given their throughput asymptotic 20 # performance for parallelizable modes is 1.25 cycles per byte. Being 21 # asymptotic limit it's not something you commonly achieve in reality, 22 # but how close does one get? Below are results collected for 23 # different modes and block sized. Pairs of numbers are for en-/ 24 # decryption. 25 # 26 # 16-byte 64-byte 256-byte 1-KB 8-KB 27 # ECB 4.25/4.25 1.38/1.38 1.28/1.28 1.26/1.26 1.26/1.26 28 # CTR 5.42/5.42 1.92/1.92 1.44/1.44 1.28/1.28 1.26/1.26 29 # CBC 4.38/4.43 4.15/1.43 4.07/1.32 4.07/1.29 4.06/1.28 30 # CCM 5.66/9.42 4.42/5.41 4.16/4.40 4.09/4.15 4.06/4.07 31 # OFB 5.42/5.42 4.64/4.64 4.44/4.44 4.39/4.39 4.38/4.38 32 # CFB 5.73/5.85 5.56/5.62 5.48/5.56 5.47/5.55 5.47/5.55 33 # 34 # ECB, CTR, CBC and CCM results are free from EVP overhead. This means 35 # that otherwise used 'openssl speed -evp aes-128-??? -engine aesni 36 # [-decrypt]' will exhibit 10-15% worse results for smaller blocks. 37 # The results were collected with specially crafted speed.c benchmark 38 # in order to compare them with results reported in "Intel Advanced 39 # Encryption Standard (AES) New Instruction Set" White Paper Revision 40 # 3.0 dated May 2010. All above results are consistently better. This 41 # module also provides better performance for block sizes smaller than 42 # 128 bytes in points *not* represented in the above table. 43 # 44 # Looking at the results for 8-KB buffer. 45 # 46 # CFB and OFB results are far from the limit, because implementation 47 # uses "generic" CRYPTO_[c|o]fb128_encrypt interfaces relying on 48 # single-block aesni_encrypt, which is not the most optimal way to go. 49 # CBC encrypt result is unexpectedly high and there is no documented 50 # explanation for it. Seemingly there is a small penalty for feeding 51 # the result back to AES unit the way it's done in CBC mode. There is 52 # nothing one can do and the result appears optimal. CCM result is 53 # identical to CBC, because CBC-MAC is essentially CBC encrypt without 54 # saving output. CCM CTR "stays invisible," because it's neatly 55 # interleaved wih CBC-MAC. This provides ~30% improvement over 56 # "straghtforward" CCM implementation with CTR and CBC-MAC performed 57 # disjointly. Parallelizable modes practically achieve the theoretical 58 # limit. 59 # 60 # Looking at how results vary with buffer size. 61 # 62 # Curves are practically saturated at 1-KB buffer size. In most cases 63 # "256-byte" performance is >95%, and "64-byte" is ~90% of "8-KB" one. 64 # CTR curve doesn't follow this pattern and is "slowest" changing one 65 # with "256-byte" result being 87% of "8-KB." This is because overhead 66 # in CTR mode is most computationally intensive. Small-block CCM 67 # decrypt is slower than encrypt, because first CTR and last CBC-MAC 68 # iterations can't be interleaved. 69 # 70 # Results for 192- and 256-bit keys. 71 # 72 # EVP-free results were observed to scale perfectly with number of 73 # rounds for larger block sizes, i.e. 192-bit result being 10/12 times 74 # lower and 256-bit one - 10/14. Well, in CBC encrypt case differences 75 # are a tad smaller, because the above mentioned penalty biases all 76 # results by same constant value. In similar way function call 77 # overhead affects small-block performance, as well as OFB and CFB 78 # results. Differences are not large, most common coefficients are 79 # 10/11.7 and 10/13.4 (as opposite to 10/12.0 and 10/14.0), but one 80 # observe even 10/11.2 and 10/12.4 (CTR, OFB, CFB)... 81 82 # January 2011 83 # 84 # While Westmere processor features 6 cycles latency for aes[enc|dec] 85 # instructions, which can be scheduled every second cycle, Sandy 86 # Bridge spends 8 cycles per instruction, but it can schedule them 87 # every cycle. This means that code targeting Westmere would perform 88 # suboptimally on Sandy Bridge. Therefore this update. 89 # 90 # In addition, non-parallelizable CBC encrypt (as well as CCM) is 91 # optimized. Relative improvement might appear modest, 8% on Westmere, 92 # but in absolute terms it's 3.77 cycles per byte encrypted with 93 # 128-bit key on Westmere, and 5.07 - on Sandy Bridge. These numbers 94 # should be compared to asymptotic limits of 3.75 for Westmere and 95 # 5.00 for Sandy Bridge. Actually, the fact that they get this close 96 # to asymptotic limits is quite amazing. Indeed, the limit is 97 # calculated as latency times number of rounds, 10 for 128-bit key, 98 # and divided by 16, the number of bytes in block, or in other words 99 # it accounts *solely* for aesenc instructions. But there are extra 100 # instructions, and numbers so close to the asymptotic limits mean 101 # that it's as if it takes as little as *one* additional cycle to 102 # execute all of them. How is it possible? It is possible thanks to 103 # out-of-order execution logic, which manages to overlap post- 104 # processing of previous block, things like saving the output, with 105 # actual encryption of current block, as well as pre-processing of 106 # current block, things like fetching input and xor-ing it with 107 # 0-round element of the key schedule, with actual encryption of 108 # previous block. Keep this in mind... 109 # 110 # For parallelizable modes, such as ECB, CBC decrypt, CTR, higher 111 # performance is achieved by interleaving instructions working on 112 # independent blocks. In which case asymptotic limit for such modes 113 # can be obtained by dividing above mentioned numbers by AES 114 # instructions' interleave factor. Westmere can execute at most 3 115 # instructions at a time, meaning that optimal interleave factor is 3, 116 # and that's where the "magic" number of 1.25 come from. "Optimal 117 # interleave factor" means that increase of interleave factor does 118 # not improve performance. The formula has proven to reflect reality 119 # pretty well on Westmere... Sandy Bridge on the other hand can 120 # execute up to 8 AES instructions at a time, so how does varying 121 # interleave factor affect the performance? Here is table for ECB 122 # (numbers are cycles per byte processed with 128-bit key): 123 # 124 # instruction interleave factor 3x 6x 8x 125 # theoretical asymptotic limit 1.67 0.83 0.625 126 # measured performance for 8KB block 1.05 0.86 0.84 127 # 128 # "as if" interleave factor 4.7x 5.8x 6.0x 129 # 130 # Further data for other parallelizable modes: 131 # 132 # CBC decrypt 1.16 0.93 0.93 133 # CTR 1.14 0.91 n/a 134 # 135 # Well, given 3x column it's probably inappropriate to call the limit 136 # asymptotic, if it can be surpassed, isn't it? What happens there? 137 # Rewind to CBC paragraph for the answer. Yes, out-of-order execution 138 # magic is responsible for this. Processor overlaps not only the 139 # additional instructions with AES ones, but even AES instuctions 140 # processing adjacent triplets of independent blocks. In the 6x case 141 # additional instructions still claim disproportionally small amount 142 # of additional cycles, but in 8x case number of instructions must be 143 # a tad too high for out-of-order logic to cope with, and AES unit 144 # remains underutilized... As you can see 8x interleave is hardly 145 # justifiable, so there no need to feel bad that 32-bit aesni-x86.pl 146 # utilizies 6x interleave because of limited register bank capacity. 147 # 148 # Higher interleave factors do have negative impact on Westmere 149 # performance. While for ECB mode it's negligible ~1.5%, other 150 # parallelizables perform ~5% worse, which is outweighed by ~25% 151 # improvement on Sandy Bridge. To balance regression on Westmere 152 # CTR mode was implemented with 6x aesenc interleave factor. 153 154 # April 2011 155 # 156 # Add aesni_xts_[en|de]crypt. Westmere spends 1.33 cycles processing 157 # one byte out of 8KB with 128-bit key, Sandy Bridge - 0.97. Just like 158 # in CTR mode AES instruction interleave factor was chosen to be 6x. 159 160 $PREFIX="aesni"; # if $PREFIX is set to "AES", the script 161 # generates drop-in replacement for 162 # crypto/aes/asm/aes-x86_64.pl:-) 163 164 $flavour = shift; 165 $output = shift; 166 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } 167 168 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); 169 170 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 171 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or 172 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or 173 die "can't locate x86_64-xlate.pl"; 174 175 open STDOUT,"| $^X $xlate $flavour $output"; 176 177 $movkey = $PREFIX eq "aesni" ? "movups" : "movups"; 178 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") : # Win64 order 179 ("%rdi","%rsi","%rdx","%rcx"); # Unix order 180 181 $code=".text\n"; 182 183 $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!! 184 # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ... 185 $inp="%rdi"; 186 $out="%rsi"; 187 $len="%rdx"; 188 $key="%rcx"; # input to and changed by aesni_[en|de]cryptN !!! 189 $ivp="%r8"; # cbc, ctr, ... 190 191 $rnds_="%r10d"; # backup copy for $rounds 192 $key_="%r11"; # backup copy for $key 193 194 # %xmm register layout 195 $rndkey0="%xmm0"; $rndkey1="%xmm1"; 196 $inout0="%xmm2"; $inout1="%xmm3"; 197 $inout2="%xmm4"; $inout3="%xmm5"; 198 $inout4="%xmm6"; $inout5="%xmm7"; 199 $inout6="%xmm8"; $inout7="%xmm9"; 200 201 $in2="%xmm6"; $in1="%xmm7"; # used in CBC decrypt, CTR, ... 202 $in0="%xmm8"; $iv="%xmm9"; 203 205 # Inline version of internal aesni_[en|de]crypt1. 206 # 207 # Why folded loop? Because aes[enc|dec] is slow enough to accommodate 208 # cycles which take care of loop variables... 209 { my $sn; 210 sub aesni_generate1 { 211 my ($p,$key,$rounds,$inout,$ivec)=@_; $inout=$inout0 if (!defined($inout)); 212 ++$sn; 213 $code.=<<___; 214 $movkey ($key),$rndkey0 215 $movkey 16($key),$rndkey1 216 ___ 217 $code.=<<___ if (defined($ivec)); 218 xorps $rndkey0,$ivec 219 lea 32($key),$key 220 xorps $ivec,$inout 221 ___ 222 $code.=<<___ if (!defined($ivec)); 223 lea 32($key),$key 224 xorps $rndkey0,$inout 225 ___ 226 $code.=<<___; 227 .Loop_${p}1_$sn: 228 aes${p} $rndkey1,$inout 229 dec $rounds 230 $movkey ($key),$rndkey1 231 lea 16($key),$key 232 jnz .Loop_${p}1_$sn # loop body is 16 bytes 233 aes${p}last $rndkey1,$inout 234 ___ 235 }} 236 # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key); 237 # 238 { my ($inp,$out,$key) = @_4args; 239 240 $code.=<<___; 241 .globl ${PREFIX}_encrypt 242 .type ${PREFIX}_encrypt,\@abi-omnipotent 243 .align 16 244 ${PREFIX}_encrypt: 245 movups ($inp),$inout0 # load input 246 mov 240($key),$rounds # key->rounds 247 ___ 248 &aesni_generate1("enc",$key,$rounds); 249 $code.=<<___; 250 movups $inout0,($out) # output 251 ret 252 .size ${PREFIX}_encrypt,.-${PREFIX}_encrypt 253 254 .globl ${PREFIX}_decrypt 255 .type ${PREFIX}_decrypt,\@abi-omnipotent 256 .align 16 257 ${PREFIX}_decrypt: 258 movups ($inp),$inout0 # load input 259 mov 240($key),$rounds # key->rounds 260 ___ 261 &aesni_generate1("dec",$key,$rounds); 262 $code.=<<___; 263 movups $inout0,($out) # output 264 ret 265 .size ${PREFIX}_decrypt, .-${PREFIX}_decrypt 266 ___ 267 } 268 270 # _aesni_[en|de]cryptN are private interfaces, N denotes interleave 271 # factor. Why 3x subroutine were originally used in loops? Even though 272 # aes[enc|dec] latency was originally 6, it could be scheduled only 273 # every *2nd* cycle. Thus 3x interleave was the one providing optimal 274 # utilization, i.e. when subroutine's throughput is virtually same as 275 # of non-interleaved subroutine [for number of input blocks up to 3]. 276 # This is why it makes no sense to implement 2x subroutine. 277 # aes[enc|dec] latency in next processor generation is 8, but the 278 # instructions can be scheduled every cycle. Optimal interleave for 279 # new processor is therefore 8x... 280 sub aesni_generate3 { 281 my $dir=shift; 282 # As already mentioned it takes in $key and $rounds, which are *not* 283 # preserved. $inout[0-2] is cipher/clear text... 284 $code.=<<___; 285 .type _aesni_${dir}rypt3,\@abi-omnipotent 286 .align 16 287 _aesni_${dir}rypt3: 288 $movkey ($key),$rndkey0 289 shr \$1,$rounds 290 $movkey 16($key),$rndkey1 291 lea 32($key),$key 292 xorps $rndkey0,$inout0 293 xorps $rndkey0,$inout1 294 xorps $rndkey0,$inout2 295 $movkey ($key),$rndkey0 296 297 .L${dir}_loop3: 298 aes${dir} $rndkey1,$inout0 299 aes${dir} $rndkey1,$inout1 300 dec $rounds 301 aes${dir} $rndkey1,$inout2 302 $movkey 16($key),$rndkey1 303 aes${dir} $rndkey0,$inout0 304 aes${dir} $rndkey0,$inout1 305 lea 32($key),$key 306 aes${dir} $rndkey0,$inout2 307 $movkey ($key),$rndkey0 308 jnz .L${dir}_loop3 309 310 aes${dir} $rndkey1,$inout0 311 aes${dir} $rndkey1,$inout1 312 aes${dir} $rndkey1,$inout2 313 aes${dir}last $rndkey0,$inout0 314 aes${dir}last $rndkey0,$inout1 315 aes${dir}last $rndkey0,$inout2 316 ret 317 .size _aesni_${dir}rypt3,.-_aesni_${dir}rypt3 318 ___ 319 } 320 # 4x interleave is implemented to improve small block performance, 321 # most notably [and naturally] 4 block by ~30%. One can argue that one 322 # should have implemented 5x as well, but improvement would be <20%, 323 # so it's not worth it... 324 sub aesni_generate4 { 325 my $dir=shift; 326 # As already mentioned it takes in $key and $rounds, which are *not* 327 # preserved. $inout[0-3] is cipher/clear text... 328 $code.=<<___; 329 .type _aesni_${dir}rypt4,\@abi-omnipotent 330 .align 16 331 _aesni_${dir}rypt4: 332 $movkey ($key),$rndkey0 333 shr \$1,$rounds 334 $movkey 16($key),$rndkey1 335 lea 32($key),$key 336 xorps $rndkey0,$inout0 337 xorps $rndkey0,$inout1 338 xorps $rndkey0,$inout2 339 xorps $rndkey0,$inout3 340 $movkey ($key),$rndkey0 341 342 .L${dir}_loop4: 343 aes${dir} $rndkey1,$inout0 344 aes${dir} $rndkey1,$inout1 345 dec $rounds 346 aes${dir} $rndkey1,$inout2 347 aes${dir} $rndkey1,$inout3 348 $movkey 16($key),$rndkey1 349 aes${dir} $rndkey0,$inout0 350 aes${dir} $rndkey0,$inout1 351 lea 32($key),$key 352 aes${dir} $rndkey0,$inout2 353 aes${dir} $rndkey0,$inout3 354 $movkey ($key),$rndkey0 355 jnz .L${dir}_loop4 356 357 aes${dir} $rndkey1,$inout0 358 aes${dir} $rndkey1,$inout1 359 aes${dir} $rndkey1,$inout2 360 aes${dir} $rndkey1,$inout3 361 aes${dir}last $rndkey0,$inout0 362 aes${dir}last $rndkey0,$inout1 363 aes${dir}last $rndkey0,$inout2 364 aes${dir}last $rndkey0,$inout3 365 ret 366 .size _aesni_${dir}rypt4,.-_aesni_${dir}rypt4 367 ___ 368 } 369 sub aesni_generate6 { 370 my $dir=shift; 371 # As already mentioned it takes in $key and $rounds, which are *not* 372 # preserved. $inout[0-5] is cipher/clear text... 373 $code.=<<___; 374 .type _aesni_${dir}rypt6,\@abi-omnipotent 375 .align 16 376 _aesni_${dir}rypt6: 377 $movkey ($key),$rndkey0 378 shr \$1,$rounds 379 $movkey 16($key),$rndkey1 380 lea 32($key),$key 381 xorps $rndkey0,$inout0 382 pxor $rndkey0,$inout1 383 aes${dir} $rndkey1,$inout0 384 pxor $rndkey0,$inout2 385 aes${dir} $rndkey1,$inout1 386 pxor $rndkey0,$inout3 387 aes${dir} $rndkey1,$inout2 388 pxor $rndkey0,$inout4 389 aes${dir} $rndkey1,$inout3 390 pxor $rndkey0,$inout5 391 dec $rounds 392 aes${dir} $rndkey1,$inout4 393 $movkey ($key),$rndkey0 394 aes${dir} $rndkey1,$inout5 395 jmp .L${dir}_loop6_enter 396 .align 16 397 .L${dir}_loop6: 398 aes${dir} $rndkey1,$inout0 399 aes${dir} $rndkey1,$inout1 400 dec $rounds 401 aes${dir} $rndkey1,$inout2 402 aes${dir} $rndkey1,$inout3 403 aes${dir} $rndkey1,$inout4 404 aes${dir} $rndkey1,$inout5 405 .L${dir}_loop6_enter: # happens to be 16-byte aligned 406 $movkey 16($key),$rndkey1 407 aes${dir} $rndkey0,$inout0 408 aes${dir} $rndkey0,$inout1 409 lea 32($key),$key 410 aes${dir} $rndkey0,$inout2 411 aes${dir} $rndkey0,$inout3 412 aes${dir} $rndkey0,$inout4 413 aes${dir} $rndkey0,$inout5 414 $movkey ($key),$rndkey0 415 jnz .L${dir}_loop6 416 417 aes${dir} $rndkey1,$inout0 418 aes${dir} $rndkey1,$inout1 419 aes${dir} $rndkey1,$inout2 420 aes${dir} $rndkey1,$inout3 421 aes${dir} $rndkey1,$inout4 422 aes${dir} $rndkey1,$inout5 423 aes${dir}last $rndkey0,$inout0 424 aes${dir}last $rndkey0,$inout1 425 aes${dir}last $rndkey0,$inout2 426 aes${dir}last $rndkey0,$inout3 427 aes${dir}last $rndkey0,$inout4 428 aes${dir}last $rndkey0,$inout5 429 ret 430 .size _aesni_${dir}rypt6,.-_aesni_${dir}rypt6 431 ___ 432 } 433 sub aesni_generate8 { 434 my $dir=shift; 435 # As already mentioned it takes in $key and $rounds, which are *not* 436 # preserved. $inout[0-7] is cipher/clear text... 437 $code.=<<___; 438 .type _aesni_${dir}rypt8,\@abi-omnipotent 439 .align 16 440 _aesni_${dir}rypt8: 441 $movkey ($key),$rndkey0 442 shr \$1,$rounds 443 $movkey 16($key),$rndkey1 444 lea 32($key),$key 445 xorps $rndkey0,$inout0 446 xorps $rndkey0,$inout1 447 aes${dir} $rndkey1,$inout0 448 pxor $rndkey0,$inout2 449 aes${dir} $rndkey1,$inout1 450 pxor $rndkey0,$inout3 451 aes${dir} $rndkey1,$inout2 452 pxor $rndkey0,$inout4 453 aes${dir} $rndkey1,$inout3 454 pxor $rndkey0,$inout5 455 dec $rounds 456 aes${dir} $rndkey1,$inout4 457 pxor $rndkey0,$inout6 458 aes${dir} $rndkey1,$inout5 459 pxor $rndkey0,$inout7 460 $movkey ($key),$rndkey0 461 aes${dir} $rndkey1,$inout6 462 aes${dir} $rndkey1,$inout7 463 $movkey 16($key),$rndkey1 464 jmp .L${dir}_loop8_enter 465 .align 16 466 .L${dir}_loop8: 467 aes${dir} $rndkey1,$inout0 468 aes${dir} $rndkey1,$inout1 469 dec $rounds 470 aes${dir} $rndkey1,$inout2 471 aes${dir} $rndkey1,$inout3 472 aes${dir} $rndkey1,$inout4 473 aes${dir} $rndkey1,$inout5 474 aes${dir} $rndkey1,$inout6 475 aes${dir} $rndkey1,$inout7 476 $movkey 16($key),$rndkey1 477 .L${dir}_loop8_enter: # happens to be 16-byte aligned 478 aes${dir} $rndkey0,$inout0 479 aes${dir} $rndkey0,$inout1 480 lea 32($key),$key 481 aes${dir} $rndkey0,$inout2 482 aes${dir} $rndkey0,$inout3 483 aes${dir} $rndkey0,$inout4 484 aes${dir} $rndkey0,$inout5 485 aes${dir} $rndkey0,$inout6 486 aes${dir} $rndkey0,$inout7 487 $movkey ($key),$rndkey0 488 jnz .L${dir}_loop8 489 490 aes${dir} $rndkey1,$inout0 491 aes${dir} $rndkey1,$inout1 492 aes${dir} $rndkey1,$inout2 493 aes${dir} $rndkey1,$inout3 494 aes${dir} $rndkey1,$inout4 495 aes${dir} $rndkey1,$inout5 496 aes${dir} $rndkey1,$inout6 497 aes${dir} $rndkey1,$inout7 498 aes${dir}last $rndkey0,$inout0 499 aes${dir}last $rndkey0,$inout1 500 aes${dir}last $rndkey0,$inout2 501 aes${dir}last $rndkey0,$inout3 502 aes${dir}last $rndkey0,$inout4 503 aes${dir}last $rndkey0,$inout5 504 aes${dir}last $rndkey0,$inout6 505 aes${dir}last $rndkey0,$inout7 506 ret 507 .size _aesni_${dir}rypt8,.-_aesni_${dir}rypt8 508 ___ 509 } 510 &aesni_generate3("enc") if ($PREFIX eq "aesni"); 511 &aesni_generate3("dec"); 512 &aesni_generate4("enc") if ($PREFIX eq "aesni"); 513 &aesni_generate4("dec"); 514 &aesni_generate6("enc") if ($PREFIX eq "aesni"); 515 &aesni_generate6("dec"); 516 &aesni_generate8("enc") if ($PREFIX eq "aesni"); 517 &aesni_generate8("dec"); 518 520 if ($PREFIX eq "aesni") { 521 ######################################################################## 522 # void aesni_ecb_encrypt (const void *in, void *out, 523 # size_t length, const AES_KEY *key, 524 # int enc); 525 $code.=<<___; 526 .globl aesni_ecb_encrypt 527 .type aesni_ecb_encrypt,\@function,5 528 .align 16 529 aesni_ecb_encrypt: 530 and \$-16,$len 531 jz .Lecb_ret 532 533 mov 240($key),$rounds # key->rounds 534 $movkey ($key),$rndkey0 535 mov $key,$key_ # backup $key 536 mov $rounds,$rnds_ # backup $rounds 537 test %r8d,%r8d # 5th argument 538 jz .Lecb_decrypt 539 #--------------------------- ECB ENCRYPT ------------------------------# 540 cmp \$0x80,$len 541 jb .Lecb_enc_tail 542 543 movdqu ($inp),$inout0 544 movdqu 0x10($inp),$inout1 545 movdqu 0x20($inp),$inout2 546 movdqu 0x30($inp),$inout3 547 movdqu 0x40($inp),$inout4 548 movdqu 0x50($inp),$inout5 549 movdqu 0x60($inp),$inout6 550 movdqu 0x70($inp),$inout7 551 lea 0x80($inp),$inp 552 sub \$0x80,$len 553 jmp .Lecb_enc_loop8_enter 554 .align 16 555 .Lecb_enc_loop8: 556 movups $inout0,($out) 557 mov $key_,$key # restore $key 558 movdqu ($inp),$inout0 559 mov $rnds_,$rounds # restore $rounds 560 movups $inout1,0x10($out) 561 movdqu 0x10($inp),$inout1 562 movups $inout2,0x20($out) 563 movdqu 0x20($inp),$inout2 564 movups $inout3,0x30($out) 565 movdqu 0x30($inp),$inout3 566 movups $inout4,0x40($out) 567 movdqu 0x40($inp),$inout4 568 movups $inout5,0x50($out) 569 movdqu 0x50($inp),$inout5 570 movups $inout6,0x60($out) 571 movdqu 0x60($inp),$inout6 572 movups $inout7,0x70($out) 573 lea 0x80($out),$out 574 movdqu 0x70($inp),$inout7 575 lea 0x80($inp),$inp 576 .Lecb_enc_loop8_enter: 577 578 call _aesni_encrypt8 579 580 sub \$0x80,$len 581 jnc .Lecb_enc_loop8 582 583 movups $inout0,($out) 584 mov $key_,$key # restore $key 585 movups $inout1,0x10($out) 586 mov $rnds_,$rounds # restore $rounds 587 movups $inout2,0x20($out) 588 movups $inout3,0x30($out) 589 movups $inout4,0x40($out) 590 movups $inout5,0x50($out) 591 movups $inout6,0x60($out) 592 movups $inout7,0x70($out) 593 lea 0x80($out),$out 594 add \$0x80,$len 595 jz .Lecb_ret 596 597 .Lecb_enc_tail: 598 movups ($inp),$inout0 599 cmp \$0x20,$len 600 jb .Lecb_enc_one 601 movups 0x10($inp),$inout1 602 je .Lecb_enc_two 603 movups 0x20($inp),$inout2 604 cmp \$0x40,$len 605 jb .Lecb_enc_three 606 movups 0x30($inp),$inout3 607 je .Lecb_enc_four 608 movups 0x40($inp),$inout4 609 cmp \$0x60,$len 610 jb .Lecb_enc_five 611 movups 0x50($inp),$inout5 612 je .Lecb_enc_six 613 movdqu 0x60($inp),$inout6 614 call _aesni_encrypt8 615 movups $inout0,($out) 616 movups $inout1,0x10($out) 617 movups $inout2,0x20($out) 618 movups $inout3,0x30($out) 619 movups $inout4,0x40($out) 620 movups $inout5,0x50($out) 621 movups $inout6,0x60($out) 622 jmp .Lecb_ret 623 .align 16 624 .Lecb_enc_one: 625 ___ 626 &aesni_generate1("enc",$key,$rounds); 627 $code.=<<___; 628 movups $inout0,($out) 629 jmp .Lecb_ret 630 .align 16 631 .Lecb_enc_two: 632 xorps $inout2,$inout2 633 call _aesni_encrypt3 634 movups $inout0,($out) 635 movups $inout1,0x10($out) 636 jmp .Lecb_ret 637 .align 16 638 .Lecb_enc_three: 639 call _aesni_encrypt3 640 movups $inout0,($out) 641 movups $inout1,0x10($out) 642 movups $inout2,0x20($out) 643 jmp .Lecb_ret 644 .align 16 645 .Lecb_enc_four: 646 call _aesni_encrypt4 647 movups $inout0,($out) 648 movups $inout1,0x10($out) 649 movups $inout2,0x20($out) 650 movups $inout3,0x30($out) 651 jmp .Lecb_ret 652 .align 16 653 .Lecb_enc_five: 654 xorps $inout5,$inout5 655 call _aesni_encrypt6 656 movups $inout0,($out) 657 movups $inout1,0x10($out) 658 movups $inout2,0x20($out) 659 movups $inout3,0x30($out) 660 movups $inout4,0x40($out) 661 jmp .Lecb_ret 662 .align 16 663 .Lecb_enc_six: 664 call _aesni_encrypt6 665 movups $inout0,($out) 666 movups $inout1,0x10($out) 667 movups $inout2,0x20($out) 668 movups $inout3,0x30($out) 669 movups $inout4,0x40($out) 670 movups $inout5,0x50($out) 671 jmp .Lecb_ret 672 #--------------------------- ECB DECRYPT ------------------------------# 674 .align 16 675 .Lecb_decrypt: 676 cmp \$0x80,$len 677 jb .Lecb_dec_tail 678 679 movdqu ($inp),$inout0 680 movdqu 0x10($inp),$inout1 681 movdqu 0x20($inp),$inout2 682 movdqu 0x30($inp),$inout3 683 movdqu 0x40($inp),$inout4 684 movdqu 0x50($inp),$inout5 685 movdqu 0x60($inp),$inout6 686 movdqu 0x70($inp),$inout7 687 lea 0x80($inp),$inp 688 sub \$0x80,$len 689 jmp .Lecb_dec_loop8_enter 690 .align 16 691 .Lecb_dec_loop8: 692 movups $inout0,($out) 693 mov $key_,$key # restore $key 694 movdqu ($inp),$inout0 695 mov $rnds_,$rounds # restore $rounds 696 movups $inout1,0x10($out) 697 movdqu 0x10($inp),$inout1 698 movups $inout2,0x20($out) 699 movdqu 0x20($inp),$inout2 700 movups $inout3,0x30($out) 701 movdqu 0x30($inp),$inout3 702 movups $inout4,0x40($out) 703 movdqu 0x40($inp),$inout4 704 movups $inout5,0x50($out) 705 movdqu 0x50($inp),$inout5 706 movups $inout6,0x60($out) 707 movdqu 0x60($inp),$inout6 708 movups $inout7,0x70($out) 709 lea 0x80($out),$out 710 movdqu 0x70($inp),$inout7 711 lea 0x80($inp),$inp 712 .Lecb_dec_loop8_enter: 713 714 call _aesni_decrypt8 715 716 $movkey ($key_),$rndkey0 717 sub \$0x80,$len 718 jnc .Lecb_dec_loop8 719 720 movups $inout0,($out) 721 mov $key_,$key # restore $key 722 movups $inout1,0x10($out) 723 mov $rnds_,$rounds # restore $rounds 724 movups $inout2,0x20($out) 725 movups $inout3,0x30($out) 726 movups $inout4,0x40($out) 727 movups $inout5,0x50($out) 728 movups $inout6,0x60($out) 729 movups $inout7,0x70($out) 730 lea 0x80($out),$out 731 add \$0x80,$len 732 jz .Lecb_ret 733 734 .Lecb_dec_tail: 735 movups ($inp),$inout0 736 cmp \$0x20,$len 737 jb .Lecb_dec_one 738 movups 0x10($inp),$inout1 739 je .Lecb_dec_two 740 movups 0x20($inp),$inout2 741 cmp \$0x40,$len 742 jb .Lecb_dec_three 743 movups 0x30($inp),$inout3 744 je .Lecb_dec_four 745 movups 0x40($inp),$inout4 746 cmp \$0x60,$len 747 jb .Lecb_dec_five 748 movups 0x50($inp),$inout5 749 je .Lecb_dec_six 750 movups 0x60($inp),$inout6 751 $movkey ($key),$rndkey0 752 call _aesni_decrypt8 753 movups $inout0,($out) 754 movups $inout1,0x10($out) 755 movups $inout2,0x20($out) 756 movups $inout3,0x30($out) 757 movups $inout4,0x40($out) 758 movups $inout5,0x50($out) 759 movups $inout6,0x60($out) 760 jmp .Lecb_ret 761 .align 16 762 .Lecb_dec_one: 763 ___ 764 &aesni_generate1("dec",$key,$rounds); 765 $code.=<<___; 766 movups $inout0,($out) 767 jmp .Lecb_ret 768 .align 16 769 .Lecb_dec_two: 770 xorps $inout2,$inout2 771 call _aesni_decrypt3 772 movups $inout0,($out) 773 movups $inout1,0x10($out) 774 jmp .Lecb_ret 775 .align 16 776 .Lecb_dec_three: 777 call _aesni_decrypt3 778 movups $inout0,($out) 779 movups $inout1,0x10($out) 780 movups $inout2,0x20($out) 781 jmp .Lecb_ret 782 .align 16 783 .Lecb_dec_four: 784 call _aesni_decrypt4 785 movups $inout0,($out) 786 movups $inout1,0x10($out) 787 movups $inout2,0x20($out) 788 movups $inout3,0x30($out) 789 jmp .Lecb_ret 790 .align 16 791 .Lecb_dec_five: 792 xorps $inout5,$inout5 793 call _aesni_decrypt6 794 movups $inout0,($out) 795 movups $inout1,0x10($out) 796 movups $inout2,0x20($out) 797 movups $inout3,0x30($out) 798 movups $inout4,0x40($out) 799 jmp .Lecb_ret 800 .align 16 801 .Lecb_dec_six: 802 call _aesni_decrypt6 803 movups $inout0,($out) 804 movups $inout1,0x10($out) 805 movups $inout2,0x20($out) 806 movups $inout3,0x30($out) 807 movups $inout4,0x40($out) 808 movups $inout5,0x50($out) 809 810 .Lecb_ret: 811 ret 812 .size aesni_ecb_encrypt,.-aesni_ecb_encrypt 813 ___ 814 816 { 817 ###################################################################### 818 # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out, 819 # size_t blocks, const AES_KEY *key, 820 # const char *ivec,char *cmac); 821 # 822 # Handles only complete blocks, operates on 64-bit counter and 823 # does not update *ivec! Nor does it finalize CMAC value 824 # (see engine/eng_aesni.c for details) 825 # 826 { 827 my $cmac="%r9"; # 6th argument 828 829 my $increment="%xmm6"; 830 my $bswap_mask="%xmm7"; 831 832 $code.=<<___; 833 .globl aesni_ccm64_encrypt_blocks 834 .type aesni_ccm64_encrypt_blocks,\@function,6 835 .align 16 836 aesni_ccm64_encrypt_blocks: 837 ___ 838 $code.=<<___ if ($win64); 839 lea -0x58(%rsp),%rsp 840 movaps %xmm6,(%rsp) 841 movaps %xmm7,0x10(%rsp) 842 movaps %xmm8,0x20(%rsp) 843 movaps %xmm9,0x30(%rsp) 844 .Lccm64_enc_body: 845 ___ 846 $code.=<<___; 847 mov 240($key),$rounds # key->rounds 848 movdqu ($ivp),$iv 849 movdqa .Lincrement64(%rip),$increment 850 movdqa .Lbswap_mask(%rip),$bswap_mask 851 852 shr \$1,$rounds 853 lea 0($key),$key_ 854 movdqu ($cmac),$inout1 855 movdqa $iv,$inout0 856 mov $rounds,$rnds_ 857 pshufb $bswap_mask,$iv 858 jmp .Lccm64_enc_outer 859 .align 16 860 .Lccm64_enc_outer: 861 $movkey ($key_),$rndkey0 862 mov $rnds_,$rounds 863 movups ($inp),$in0 # load inp 864 865 xorps $rndkey0,$inout0 # counter 866 $movkey 16($key_),$rndkey1 867 xorps $in0,$rndkey0 868 lea 32($key_),$key 869 xorps $rndkey0,$inout1 # cmac^=inp 870 $movkey ($key),$rndkey0 871 872 .Lccm64_enc2_loop: 873 aesenc $rndkey1,$inout0 874 dec $rounds 875 aesenc $rndkey1,$inout1 876 $movkey 16($key),$rndkey1 877 aesenc $rndkey0,$inout0 878 lea 32($key),$key 879 aesenc $rndkey0,$inout1 880 $movkey 0($key),$rndkey0 881 jnz .Lccm64_enc2_loop 882 aesenc $rndkey1,$inout0 883 aesenc $rndkey1,$inout1 884 paddq $increment,$iv 885 aesenclast $rndkey0,$inout0 886 aesenclast $rndkey0,$inout1 887 888 dec $len 889 lea 16($inp),$inp 890 xorps $inout0,$in0 # inp ^= E(iv) 891 movdqa $iv,$inout0 892 movups $in0,($out) # save output 893 lea 16($out),$out 894 pshufb $bswap_mask,$inout0 895 jnz .Lccm64_enc_outer 896 897 movups $inout1,($cmac) 898 ___ 899 $code.=<<___ if ($win64); 900 movaps (%rsp),%xmm6 901 movaps 0x10(%rsp),%xmm7 902 movaps 0x20(%rsp),%xmm8 903 movaps 0x30(%rsp),%xmm9 904 lea 0x58(%rsp),%rsp 905 .Lccm64_enc_ret: 906 ___ 907 $code.=<<___; 908 ret 909 .size aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks 910 ___ 911 ###################################################################### 912 $code.=<<___; 913 .globl aesni_ccm64_decrypt_blocks 914 .type aesni_ccm64_decrypt_blocks,\@function,6 915 .align 16 916 aesni_ccm64_decrypt_blocks: 917 ___ 918 $code.=<<___ if ($win64); 919 lea -0x58(%rsp),%rsp 920 movaps %xmm6,(%rsp) 921 movaps %xmm7,0x10(%rsp) 922 movaps %xmm8,0x20(%rsp) 923 movaps %xmm9,0x30(%rsp) 924 .Lccm64_dec_body: 925 ___ 926 $code.=<<___; 927 mov 240($key),$rounds # key->rounds 928 movups ($ivp),$iv 929 movdqu ($cmac),$inout1 930 movdqa .Lincrement64(%rip),$increment 931 movdqa .Lbswap_mask(%rip),$bswap_mask 932 933 movaps $iv,$inout0 934 mov $rounds,$rnds_ 935 mov $key,$key_ 936 pshufb $bswap_mask,$iv 937 ___ 938 &aesni_generate1("enc",$key,$rounds); 939 $code.=<<___; 940 movups ($inp),$in0 # load inp 941 paddq $increment,$iv 942 lea 16($inp),$inp 943 jmp .Lccm64_dec_outer 944 .align 16 945 .Lccm64_dec_outer: 946 xorps $inout0,$in0 # inp ^= E(iv) 947 movdqa $iv,$inout0 948 mov $rnds_,$rounds 949 movups $in0,($out) # save output 950 lea 16($out),$out 951 pshufb $bswap_mask,$inout0 952 953 sub \$1,$len 954 jz .Lccm64_dec_break 955 956 $movkey ($key_),$rndkey0 957 shr \$1,$rounds 958 $movkey 16($key_),$rndkey1 959 xorps $rndkey0,$in0 960 lea 32($key_),$key 961 xorps $rndkey0,$inout0 962 xorps $in0,$inout1 # cmac^=out 963 $movkey ($key),$rndkey0 964 965 .Lccm64_dec2_loop: 966 aesenc $rndkey1,$inout0 967 dec $rounds 968 aesenc $rndkey1,$inout1 969 $movkey 16($key),$rndkey1 970 aesenc $rndkey0,$inout0 971 lea 32($key),$key 972 aesenc $rndkey0,$inout1 973 $movkey 0($key),$rndkey0 974 jnz .Lccm64_dec2_loop 975 movups ($inp),$in0 # load inp 976 paddq $increment,$iv 977 aesenc $rndkey1,$inout0 978 aesenc $rndkey1,$inout1 979 lea 16($inp),$inp 980 aesenclast $rndkey0,$inout0 981 aesenclast $rndkey0,$inout1 982 jmp .Lccm64_dec_outer 983 984 .align 16 985 .Lccm64_dec_break: 986 #xorps $in0,$inout1 # cmac^=out 987 ___ 988 &aesni_generate1("enc",$key_,$rounds,$inout1,$in0); 989 $code.=<<___; 990 movups $inout1,($cmac) 991 ___ 992 $code.=<<___ if ($win64); 993 movaps (%rsp),%xmm6 994 movaps 0x10(%rsp),%xmm7 995 movaps 0x20(%rsp),%xmm8 996 movaps 0x30(%rsp),%xmm9 997 lea 0x58(%rsp),%rsp 998 .Lccm64_dec_ret: 999 ___ 1000 $code.=<<___; 1001 ret 1002 .size aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks 1003 ___ 1004 } 1006 ###################################################################### 1007 # void aesni_ctr32_encrypt_blocks (const void *in, void *out, 1008 # size_t blocks, const AES_KEY *key, 1009 # const char *ivec); 1010 # 1011 # Handles only complete blocks, operates on 32-bit counter and 1012 # does not update *ivec! (see engine/eng_aesni.c for details) 1013 # 1014 { 1015 my $reserved = $win64?0:-0x28; 1016 my ($in0,$in1,$in2,$in3)=map("%xmm$_",(8..11)); 1017 my ($iv0,$iv1,$ivec)=("%xmm12","%xmm13","%xmm14"); 1018 my $bswap_mask="%xmm15"; 1019 1020 $code.=<<___; 1021 .globl aesni_ctr32_encrypt_blocks 1022 .type aesni_ctr32_encrypt_blocks,\@function,5 1023 .align 16 1024 aesni_ctr32_encrypt_blocks: 1025 ___ 1026 $code.=<<___ if ($win64); 1027 lea -0xc8(%rsp),%rsp 1028 movaps %xmm6,0x20(%rsp) 1029 movaps %xmm7,0x30(%rsp) 1030 movaps %xmm8,0x40(%rsp) 1031 movaps %xmm9,0x50(%rsp) 1032 movaps %xmm10,0x60(%rsp) 1033 movaps %xmm11,0x70(%rsp) 1034 movaps %xmm12,0x80(%rsp) 1035 movaps %xmm13,0x90(%rsp) 1036 movaps %xmm14,0xa0(%rsp) 1037 movaps %xmm15,0xb0(%rsp) 1038 .Lctr32_body: 1039 ___ 1040 $code.=<<___; 1041 cmp \$1,$len 1042 je .Lctr32_one_shortcut 1043 1044 movdqu ($ivp),$ivec 1045 movdqa .Lbswap_mask(%rip),$bswap_mask 1046 xor $rounds,$rounds 1047 pextrd \$3,$ivec,$rnds_ # pull 32-bit counter 1048 pinsrd \$3,$rounds,$ivec # wipe 32-bit counter 1049 1050 mov 240($key),$rounds # key->rounds 1051 bswap $rnds_ 1052 pxor $iv0,$iv0 # vector of 3 32-bit counters 1053 pxor $iv1,$iv1 # vector of 3 32-bit counters 1054 pinsrd \$0,$rnds_,$iv0 1055 lea 3($rnds_),$key_ 1056 pinsrd \$0,$key_,$iv1 1057 inc $rnds_ 1058 pinsrd \$1,$rnds_,$iv0 1059 inc $key_ 1060 pinsrd \$1,$key_,$iv1 1061 inc $rnds_ 1062 pinsrd \$2,$rnds_,$iv0 1063 inc $key_ 1064 pinsrd \$2,$key_,$iv1 1065 movdqa $iv0,$reserved(%rsp) 1066 pshufb $bswap_mask,$iv0 1067 movdqa $iv1,`$reserved+0x10`(%rsp) 1068 pshufb $bswap_mask,$iv1 1069 1070 pshufd \$`3<<6`,$iv0,$inout0 # place counter to upper dword 1071 pshufd \$`2<<6`,$iv0,$inout1 1072 pshufd \$`1<<6`,$iv0,$inout2 1073 cmp \$6,$len 1074 jb .Lctr32_tail 1075 shr \$1,$rounds 1076 mov $key,$key_ # backup $key 1077 mov $rounds,$rnds_ # backup $rounds 1078 sub \$6,$len 1079 jmp .Lctr32_loop6 1080 1081 .align 16 1082 .Lctr32_loop6: 1083 pshufd \$`3<<6`,$iv1,$inout3 1084 por $ivec,$inout0 # merge counter-less ivec 1085 $movkey ($key_),$rndkey0 1086 pshufd \$`2<<6`,$iv1,$inout4 1087 por $ivec,$inout1 1088 $movkey 16($key_),$rndkey1 1089 pshufd \$`1<<6`,$iv1,$inout5 1090 por $ivec,$inout2 1091 por $ivec,$inout3 1092 xorps $rndkey0,$inout0 1093 por $ivec,$inout4 1094 por $ivec,$inout5 1095 1096 # inline _aesni_encrypt6 and interleave last rounds 1097 # with own code... 1098 1099 pxor $rndkey0,$inout1 1100 aesenc $rndkey1,$inout0 1101 lea 32($key_),$key 1102 pxor $rndkey0,$inout2 1103 aesenc $rndkey1,$inout1 1104 movdqa .Lincrement32(%rip),$iv1 1105 pxor $rndkey0,$inout3 1106 aesenc $rndkey1,$inout2 1107 movdqa $reserved(%rsp),$iv0 1108 pxor $rndkey0,$inout4 1109 aesenc $rndkey1,$inout3 1110 pxor $rndkey0,$inout5 1111 $movkey ($key),$rndkey0 1112 dec $rounds 1113 aesenc $rndkey1,$inout4 1114 aesenc $rndkey1,$inout5 1115 jmp .Lctr32_enc_loop6_enter 1116 .align 16 1117 .Lctr32_enc_loop6: 1118 aesenc $rndkey1,$inout0 1119 aesenc $rndkey1,$inout1 1120 dec $rounds 1121 aesenc $rndkey1,$inout2 1122 aesenc $rndkey1,$inout3 1123 aesenc $rndkey1,$inout4 1124 aesenc $rndkey1,$inout5 1125 .Lctr32_enc_loop6_enter: 1126 $movkey 16($key),$rndkey1 1127 aesenc $rndkey0,$inout0 1128 aesenc $rndkey0,$inout1 1129 lea 32($key),$key 1130 aesenc $rndkey0,$inout2 1131 aesenc $rndkey0,$inout3 1132 aesenc $rndkey0,$inout4 1133 aesenc $rndkey0,$inout5 1134 $movkey ($key),$rndkey0 1135 jnz .Lctr32_enc_loop6 1136 1137 aesenc $rndkey1,$inout0 1138 paddd $iv1,$iv0 # increment counter vector 1139 aesenc $rndkey1,$inout1 1140 paddd `$reserved+0x10`(%rsp),$iv1 1141 aesenc $rndkey1,$inout2 1142 movdqa $iv0,$reserved(%rsp) # save counter vector 1143 aesenc $rndkey1,$inout3 1144 movdqa $iv1,`$reserved+0x10`(%rsp) 1145 aesenc $rndkey1,$inout4 1146 pshufb $bswap_mask,$iv0 # byte swap 1147 aesenc $rndkey1,$inout5 1148 pshufb $bswap_mask,$iv1 1149 1150 aesenclast $rndkey0,$inout0 1151 movups ($inp),$in0 # load input 1152 aesenclast $rndkey0,$inout1 1153 movups 0x10($inp),$in1 1154 aesenclast $rndkey0,$inout2 1155 movups 0x20($inp),$in2 1156 aesenclast $rndkey0,$inout3 1157 movups 0x30($inp),$in3 1158 aesenclast $rndkey0,$inout4 1159 movups 0x40($inp),$rndkey1 1160 aesenclast $rndkey0,$inout5 1161 movups 0x50($inp),$rndkey0 1162 lea 0x60($inp),$inp 1163 1164 xorps $inout0,$in0 # xor 1165 pshufd \$`3<<6`,$iv0,$inout0 1166 xorps $inout1,$in1 1167 pshufd \$`2<<6`,$iv0,$inout1 1168 movups $in0,($out) # store output 1169 xorps $inout2,$in2 1170 pshufd \$`1<<6`,$iv0,$inout2 1171 movups $in1,0x10($out) 1172 xorps $inout3,$in3 1173 movups $in2,0x20($out) 1174 xorps $inout4,$rndkey1 1175 movups $in3,0x30($out) 1176 xorps $inout5,$rndkey0 1177 movups $rndkey1,0x40($out) 1178 movups $rndkey0,0x50($out) 1179 lea 0x60($out),$out 1180 mov $rnds_,$rounds 1181 sub \$6,$len 1182 jnc .Lctr32_loop6 1183 1184 add \$6,$len 1185 jz .Lctr32_done 1186 mov $key_,$key # restore $key 1187 lea 1($rounds,$rounds),$rounds # restore original value 1188 1189 .Lctr32_tail: 1190 por $ivec,$inout0 1191 movups ($inp),$in0 1192 cmp \$2,$len 1193 jb .Lctr32_one 1194 1195 por $ivec,$inout1 1196 movups 0x10($inp),$in1 1197 je .Lctr32_two 1198 1199 pshufd \$`3<<6`,$iv1,$inout3 1200 por $ivec,$inout2 1201 movups 0x20($inp),$in2 1202 cmp \$4,$len 1203 jb .Lctr32_three 1204 1205 pshufd \$`2<<6`,$iv1,$inout4 1206 por $ivec,$inout3 1207 movups 0x30($inp),$in3 1208 je .Lctr32_four 1209 1210 por $ivec,$inout4 1211 xorps $inout5,$inout5 1212 1213 call _aesni_encrypt6 1214 1215 movups 0x40($inp),$rndkey1 1216 xorps $inout0,$in0 1217 xorps $inout1,$in1 1218 movups $in0,($out) 1219 xorps $inout2,$in2 1220 movups $in1,0x10($out) 1221 xorps $inout3,$in3 1222 movups $in2,0x20($out) 1223 xorps $inout4,$rndkey1 1224 movups $in3,0x30($out) 1225 movups $rndkey1,0x40($out) 1226 jmp .Lctr32_done 1227 1228 .align 16 1229 .Lctr32_one_shortcut: 1230 movups ($ivp),$inout0 1231 movups ($inp),$in0 1232 mov 240($key),$rounds # key->rounds 1233 .Lctr32_one: 1234 ___ 1235 &aesni_generate1("enc",$key,$rounds); 1236 $code.=<<___; 1237 xorps $inout0,$in0 1238 movups $in0,($out) 1239 jmp .Lctr32_done 1240 1241 .align 16 1242 .Lctr32_two: 1243 xorps $inout2,$inout2 1244 call _aesni_encrypt3 1245 xorps $inout0,$in0 1246 xorps $inout1,$in1 1247 movups $in0,($out) 1248 movups $in1,0x10($out) 1249 jmp .Lctr32_done 1250 1251 .align 16 1252 .Lctr32_three: 1253 call _aesni_encrypt3 1254 xorps $inout0,$in0 1255 xorps $inout1,$in1 1256 movups $in0,($out) 1257 xorps $inout2,$in2 1258 movups $in1,0x10($out) 1259 movups $in2,0x20($out) 1260 jmp .Lctr32_done 1261 1262 .align 16 1263 .Lctr32_four: 1264 call _aesni_encrypt4 1265 xorps $inout0,$in0 1266 xorps $inout1,$in1 1267 movups $in0,($out) 1268 xorps $inout2,$in2 1269 movups $in1,0x10($out) 1270 xorps $inout3,$in3 1271 movups $in2,0x20($out) 1272 movups $in3,0x30($out) 1273 1274 .Lctr32_done: 1275 ___ 1276 $code.=<<___ if ($win64); 1277 movaps 0x20(%rsp),%xmm6 1278 movaps 0x30(%rsp),%xmm7 1279 movaps 0x40(%rsp),%xmm8 1280 movaps 0x50(%rsp),%xmm9 1281 movaps 0x60(%rsp),%xmm10 1282 movaps 0x70(%rsp),%xmm11 1283 movaps 0x80(%rsp),%xmm12 1284 movaps 0x90(%rsp),%xmm13 1285 movaps 0xa0(%rsp),%xmm14 1286 movaps 0xb0(%rsp),%xmm15 1287 lea 0xc8(%rsp),%rsp 1288 .Lctr32_ret: 1289 ___ 1290 $code.=<<___; 1291 ret 1292 .size aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks 1293 ___ 1294 } 1295 1297 ###################################################################### 1298 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len, 1299 # const AES_KEY *key1, const AES_KEY *key2 1300 # const unsigned char iv[16]); 1301 # 1302 { 1303 my @tweak=map("%xmm$_",(10..15)); 1304 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]); 1305 my ($key2,$ivp,$len_)=("%r8","%r9","%r9"); 1306 my $frame_size = 0x68 + ($win64?160:0); 1307 1308 $code.=<<___; 1309 .globl aesni_xts_encrypt 1310 .type aesni_xts_encrypt,\@function,6 1311 .align 16 1312 aesni_xts_encrypt: 1313 lea -$frame_size(%rsp),%rsp 1314 ___ 1315 $code.=<<___ if ($win64); 1316 movaps %xmm6,0x60(%rsp) 1317 movaps %xmm7,0x70(%rsp) 1318 movaps %xmm8,0x80(%rsp) 1319 movaps %xmm9,0x90(%rsp) 1320 movaps %xmm10,0xa0(%rsp) 1321 movaps %xmm11,0xb0(%rsp) 1322 movaps %xmm12,0xc0(%rsp) 1323 movaps %xmm13,0xd0(%rsp) 1324 movaps %xmm14,0xe0(%rsp) 1325 movaps %xmm15,0xf0(%rsp) 1326 .Lxts_enc_body: 1327 ___ 1328 $code.=<<___; 1329 movups ($ivp),@tweak[5] # load clear-text tweak 1330 mov 240(%r8),$rounds # key2->rounds 1331 mov 240($key),$rnds_ # key1->rounds 1332 ___ 1333 # generate the tweak 1334 &aesni_generate1("enc",$key2,$rounds,@tweak[5]); 1335 $code.=<<___; 1336 mov $key,$key_ # backup $key 1337 mov $rnds_,$rounds # backup $rounds 1338 mov $len,$len_ # backup $len 1339 and \$-16,$len 1340 1341 movdqa .Lxts_magic(%rip),$twmask 1342 pxor $twtmp,$twtmp 1343 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1344 ___ 1345 for ($i=0;$i<4;$i++) { 1346 $code.=<<___; 1347 pshufd \$0x13,$twtmp,$twres 1348 pxor $twtmp,$twtmp 1349 movdqa @tweak[5],@tweak[$i] 1350 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1351 pand $twmask,$twres # isolate carry and residue 1352 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1353 pxor $twres,@tweak[5] 1354 ___ 1355 } 1356 $code.=<<___; 1357 sub \$16*6,$len 1358 jc .Lxts_enc_short 1359 1360 shr \$1,$rounds 1361 sub \$1,$rounds 1362 mov $rounds,$rnds_ 1363 jmp .Lxts_enc_grandloop 1364 1365 .align 16 1366 .Lxts_enc_grandloop: 1367 pshufd \$0x13,$twtmp,$twres 1368 movdqa @tweak[5],@tweak[4] 1369 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1370 movdqu `16*0`($inp),$inout0 # load input 1371 pand $twmask,$twres # isolate carry and residue 1372 movdqu `16*1`($inp),$inout1 1373 pxor $twres,@tweak[5] 1374 1375 movdqu `16*2`($inp),$inout2 1376 pxor @tweak[0],$inout0 # input^=tweak 1377 movdqu `16*3`($inp),$inout3 1378 pxor @tweak[1],$inout1 1379 movdqu `16*4`($inp),$inout4 1380 pxor @tweak[2],$inout2 1381 movdqu `16*5`($inp),$inout5 1382 lea `16*6`($inp),$inp 1383 pxor @tweak[3],$inout3 1384 $movkey ($key_),$rndkey0 1385 pxor @tweak[4],$inout4 1386 pxor @tweak[5],$inout5 1387 1388 # inline _aesni_encrypt6 and interleave first and last rounds 1389 # with own code... 1390 $movkey 16($key_),$rndkey1 1391 pxor $rndkey0,$inout0 1392 pxor $rndkey0,$inout1 1393 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks 1394 aesenc $rndkey1,$inout0 1395 lea 32($key_),$key 1396 pxor $rndkey0,$inout2 1397 movdqa @tweak[1],`16*1`(%rsp) 1398 aesenc $rndkey1,$inout1 1399 pxor $rndkey0,$inout3 1400 movdqa @tweak[2],`16*2`(%rsp) 1401 aesenc $rndkey1,$inout2 1402 pxor $rndkey0,$inout4 1403 movdqa @tweak[3],`16*3`(%rsp) 1404 aesenc $rndkey1,$inout3 1405 pxor $rndkey0,$inout5 1406 $movkey ($key),$rndkey0 1407 dec $rounds 1408 movdqa @tweak[4],`16*4`(%rsp) 1409 aesenc $rndkey1,$inout4 1410 movdqa @tweak[5],`16*5`(%rsp) 1411 aesenc $rndkey1,$inout5 1412 pxor $twtmp,$twtmp 1413 pcmpgtd @tweak[5],$twtmp 1414 jmp .Lxts_enc_loop6_enter 1415 1416 .align 16 1417 .Lxts_enc_loop6: 1418 aesenc $rndkey1,$inout0 1419 aesenc $rndkey1,$inout1 1420 dec $rounds 1421 aesenc $rndkey1,$inout2 1422 aesenc $rndkey1,$inout3 1423 aesenc $rndkey1,$inout4 1424 aesenc $rndkey1,$inout5 1425 .Lxts_enc_loop6_enter: 1426 $movkey 16($key),$rndkey1 1427 aesenc $rndkey0,$inout0 1428 aesenc $rndkey0,$inout1 1429 lea 32($key),$key 1430 aesenc $rndkey0,$inout2 1431 aesenc $rndkey0,$inout3 1432 aesenc $rndkey0,$inout4 1433 aesenc $rndkey0,$inout5 1434 $movkey ($key),$rndkey0 1435 jnz .Lxts_enc_loop6 1436 1437 pshufd \$0x13,$twtmp,$twres 1438 pxor $twtmp,$twtmp 1439 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1440 aesenc $rndkey1,$inout0 1441 pand $twmask,$twres # isolate carry and residue 1442 aesenc $rndkey1,$inout1 1443 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1444 aesenc $rndkey1,$inout2 1445 pxor $twres,@tweak[5] 1446 aesenc $rndkey1,$inout3 1447 aesenc $rndkey1,$inout4 1448 aesenc $rndkey1,$inout5 1449 $movkey 16($key),$rndkey1 1450 1451 pshufd \$0x13,$twtmp,$twres 1452 pxor $twtmp,$twtmp 1453 movdqa @tweak[5],@tweak[0] 1454 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1455 aesenc $rndkey0,$inout0 1456 pand $twmask,$twres # isolate carry and residue 1457 aesenc $rndkey0,$inout1 1458 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1459 aesenc $rndkey0,$inout2 1460 pxor $twres,@tweak[5] 1461 aesenc $rndkey0,$inout3 1462 aesenc $rndkey0,$inout4 1463 aesenc $rndkey0,$inout5 1464 $movkey 32($key),$rndkey0 1465 1466 pshufd \$0x13,$twtmp,$twres 1467 pxor $twtmp,$twtmp 1468 movdqa @tweak[5],@tweak[1] 1469 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1470 aesenc $rndkey1,$inout0 1471 pand $twmask,$twres # isolate carry and residue 1472 aesenc $rndkey1,$inout1 1473 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1474 aesenc $rndkey1,$inout2 1475 pxor $twres,@tweak[5] 1476 aesenc $rndkey1,$inout3 1477 aesenc $rndkey1,$inout4 1478 aesenc $rndkey1,$inout5 1479 1480 pshufd \$0x13,$twtmp,$twres 1481 pxor $twtmp,$twtmp 1482 movdqa @tweak[5],@tweak[2] 1483 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1484 aesenclast $rndkey0,$inout0 1485 pand $twmask,$twres # isolate carry and residue 1486 aesenclast $rndkey0,$inout1 1487 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1488 aesenclast $rndkey0,$inout2 1489 pxor $twres,@tweak[5] 1490 aesenclast $rndkey0,$inout3 1491 aesenclast $rndkey0,$inout4 1492 aesenclast $rndkey0,$inout5 1493 1494 pshufd \$0x13,$twtmp,$twres 1495 pxor $twtmp,$twtmp 1496 movdqa @tweak[5],@tweak[3] 1497 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1498 xorps `16*0`(%rsp),$inout0 # output^=tweak 1499 pand $twmask,$twres # isolate carry and residue 1500 xorps `16*1`(%rsp),$inout1 1501 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1502 pxor $twres,@tweak[5] 1503 1504 xorps `16*2`(%rsp),$inout2 1505 movups $inout0,`16*0`($out) # write output 1506 xorps `16*3`(%rsp),$inout3 1507 movups $inout1,`16*1`($out) 1508 xorps `16*4`(%rsp),$inout4 1509 movups $inout2,`16*2`($out) 1510 xorps `16*5`(%rsp),$inout5 1511 movups $inout3,`16*3`($out) 1512 mov $rnds_,$rounds # restore $rounds 1513 movups $inout4,`16*4`($out) 1514 movups $inout5,`16*5`($out) 1515 lea `16*6`($out),$out 1516 sub \$16*6,$len 1517 jnc .Lxts_enc_grandloop 1518 1519 lea 3($rounds,$rounds),$rounds # restore original value 1520 mov $key_,$key # restore $key 1521 mov $rounds,$rnds_ # backup $rounds 1522 1523 .Lxts_enc_short: 1524 add \$16*6,$len 1525 jz .Lxts_enc_done 1526 1527 cmp \$0x20,$len 1528 jb .Lxts_enc_one 1529 je .Lxts_enc_two 1530 1531 cmp \$0x40,$len 1532 jb .Lxts_enc_three 1533 je .Lxts_enc_four 1534 1535 pshufd \$0x13,$twtmp,$twres 1536 movdqa @tweak[5],@tweak[4] 1537 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1538 movdqu ($inp),$inout0 1539 pand $twmask,$twres # isolate carry and residue 1540 movdqu 16*1($inp),$inout1 1541 pxor $twres,@tweak[5] 1542 1543 movdqu 16*2($inp),$inout2 1544 pxor @tweak[0],$inout0 1545 movdqu 16*3($inp),$inout3 1546 pxor @tweak[1],$inout1 1547 movdqu 16*4($inp),$inout4 1548 lea 16*5($inp),$inp 1549 pxor @tweak[2],$inout2 1550 pxor @tweak[3],$inout3 1551 pxor @tweak[4],$inout4 1552 1553 call _aesni_encrypt6 1554 1555 xorps @tweak[0],$inout0 1556 movdqa @tweak[5],@tweak[0] 1557 xorps @tweak[1],$inout1 1558 xorps @tweak[2],$inout2 1559 movdqu $inout0,($out) 1560 xorps @tweak[3],$inout3 1561 movdqu $inout1,16*1($out) 1562 xorps @tweak[4],$inout4 1563 movdqu $inout2,16*2($out) 1564 movdqu $inout3,16*3($out) 1565 movdqu $inout4,16*4($out) 1566 lea 16*5($out),$out 1567 jmp .Lxts_enc_done 1568 1569 .align 16 1570 .Lxts_enc_one: 1571 movups ($inp),$inout0 1572 lea 16*1($inp),$inp 1573 xorps @tweak[0],$inout0 1574 ___ 1575 &aesni_generate1("enc",$key,$rounds); 1576 $code.=<<___; 1577 xorps @tweak[0],$inout0 1578 movdqa @tweak[1],@tweak[0] 1579 movups $inout0,($out) 1580 lea 16*1($out),$out 1581 jmp .Lxts_enc_done 1582 1583 .align 16 1584 .Lxts_enc_two: 1585 movups ($inp),$inout0 1586 movups 16($inp),$inout1 1587 lea 32($inp),$inp 1588 xorps @tweak[0],$inout0 1589 xorps @tweak[1],$inout1 1590 1591 call _aesni_encrypt3 1592 1593 xorps @tweak[0],$inout0 1594 movdqa @tweak[2],@tweak[0] 1595 xorps @tweak[1],$inout1 1596 movups $inout0,($out) 1597 movups $inout1,16*1($out) 1598 lea 16*2($out),$out 1599 jmp .Lxts_enc_done 1600 1601 .align 16 1602 .Lxts_enc_three: 1603 movups ($inp),$inout0 1604 movups 16*1($inp),$inout1 1605 movups 16*2($inp),$inout2 1606 lea 16*3($inp),$inp 1607 xorps @tweak[0],$inout0 1608 xorps @tweak[1],$inout1 1609 xorps @tweak[2],$inout2 1610 1611 call _aesni_encrypt3 1612 1613 xorps @tweak[0],$inout0 1614 movdqa @tweak[3],@tweak[0] 1615 xorps @tweak[1],$inout1 1616 xorps @tweak[2],$inout2 1617 movups $inout0,($out) 1618 movups $inout1,16*1($out) 1619 movups $inout2,16*2($out) 1620 lea 16*3($out),$out 1621 jmp .Lxts_enc_done 1622 1623 .align 16 1624 .Lxts_enc_four: 1625 movups ($inp),$inout0 1626 movups 16*1($inp),$inout1 1627 movups 16*2($inp),$inout2 1628 xorps @tweak[0],$inout0 1629 movups 16*3($inp),$inout3 1630 lea 16*4($inp),$inp 1631 xorps @tweak[1],$inout1 1632 xorps @tweak[2],$inout2 1633 xorps @tweak[3],$inout3 1634 1635 call _aesni_encrypt4 1636 1637 xorps @tweak[0],$inout0 1638 movdqa @tweak[5],@tweak[0] 1639 xorps @tweak[1],$inout1 1640 xorps @tweak[2],$inout2 1641 movups $inout0,($out) 1642 xorps @tweak[3],$inout3 1643 movups $inout1,16*1($out) 1644 movups $inout2,16*2($out) 1645 movups $inout3,16*3($out) 1646 lea 16*4($out),$out 1647 jmp .Lxts_enc_done 1648 1649 .align 16 1650 .Lxts_enc_done: 1651 and \$15,$len_ 1652 jz .Lxts_enc_ret 1653 mov $len_,$len 1654 1655 .Lxts_enc_steal: 1656 movzb ($inp),%eax # borrow $rounds ... 1657 movzb -16($out),%ecx # ... and $key 1658 lea 1($inp),$inp 1659 mov %al,-16($out) 1660 mov %cl,0($out) 1661 lea 1($out),$out 1662 sub \$1,$len 1663 jnz .Lxts_enc_steal 1664 1665 sub $len_,$out # rewind $out 1666 mov $key_,$key # restore $key 1667 mov $rnds_,$rounds # restore $rounds 1668 1669 movups -16($out),$inout0 1670 xorps @tweak[0],$inout0 1671 ___ 1672 &aesni_generate1("enc",$key,$rounds); 1673 $code.=<<___; 1674 xorps @tweak[0],$inout0 1675 movups $inout0,-16($out) 1676 1677 .Lxts_enc_ret: 1678 ___ 1679 $code.=<<___ if ($win64); 1680 movaps 0x60(%rsp),%xmm6 1681 movaps 0x70(%rsp),%xmm7 1682 movaps 0x80(%rsp),%xmm8 1683 movaps 0x90(%rsp),%xmm9 1684 movaps 0xa0(%rsp),%xmm10 1685 movaps 0xb0(%rsp),%xmm11 1686 movaps 0xc0(%rsp),%xmm12 1687 movaps 0xd0(%rsp),%xmm13 1688 movaps 0xe0(%rsp),%xmm14 1689 movaps 0xf0(%rsp),%xmm15 1690 ___ 1691 $code.=<<___; 1692 lea $frame_size(%rsp),%rsp 1693 .Lxts_enc_epilogue: 1694 ret 1695 .size aesni_xts_encrypt,.-aesni_xts_encrypt 1696 ___ 1697 1698 $code.=<<___; 1699 .globl aesni_xts_decrypt 1700 .type aesni_xts_decrypt,\@function,6 1701 .align 16 1702 aesni_xts_decrypt: 1703 lea -$frame_size(%rsp),%rsp 1704 ___ 1705 $code.=<<___ if ($win64); 1706 movaps %xmm6,0x60(%rsp) 1707 movaps %xmm7,0x70(%rsp) 1708 movaps %xmm8,0x80(%rsp) 1709 movaps %xmm9,0x90(%rsp) 1710 movaps %xmm10,0xa0(%rsp) 1711 movaps %xmm11,0xb0(%rsp) 1712 movaps %xmm12,0xc0(%rsp) 1713 movaps %xmm13,0xd0(%rsp) 1714 movaps %xmm14,0xe0(%rsp) 1715 movaps %xmm15,0xf0(%rsp) 1716 .Lxts_dec_body: 1717 ___ 1718 $code.=<<___; 1719 movups ($ivp),@tweak[5] # load clear-text tweak 1720 mov 240($key2),$rounds # key2->rounds 1721 mov 240($key),$rnds_ # key1->rounds 1722 ___ 1723 # generate the tweak 1724 &aesni_generate1("enc",$key2,$rounds,@tweak[5]); 1725 $code.=<<___; 1726 xor %eax,%eax # if ($len%16) len-=16; 1727 test \$15,$len 1728 setnz %al 1729 shl \$4,%rax 1730 sub %rax,$len 1731 1732 mov $key,$key_ # backup $key 1733 mov $rnds_,$rounds # backup $rounds 1734 mov $len,$len_ # backup $len 1735 and \$-16,$len 1736 1737 movdqa .Lxts_magic(%rip),$twmask 1738 pxor $twtmp,$twtmp 1739 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1740 ___ 1741 for ($i=0;$i<4;$i++) { 1742 $code.=<<___; 1743 pshufd \$0x13,$twtmp,$twres 1744 pxor $twtmp,$twtmp 1745 movdqa @tweak[5],@tweak[$i] 1746 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1747 pand $twmask,$twres # isolate carry and residue 1748 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1749 pxor $twres,@tweak[5] 1750 ___ 1751 } 1752 $code.=<<___; 1753 sub \$16*6,$len 1754 jc .Lxts_dec_short 1755 1756 shr \$1,$rounds 1757 sub \$1,$rounds 1758 mov $rounds,$rnds_ 1759 jmp .Lxts_dec_grandloop 1760 1761 .align 16 1762 .Lxts_dec_grandloop: 1763 pshufd \$0x13,$twtmp,$twres 1764 movdqa @tweak[5],@tweak[4] 1765 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1766 movdqu `16*0`($inp),$inout0 # load input 1767 pand $twmask,$twres # isolate carry and residue 1768 movdqu `16*1`($inp),$inout1 1769 pxor $twres,@tweak[5] 1770 1771 movdqu `16*2`($inp),$inout2 1772 pxor @tweak[0],$inout0 # input^=tweak 1773 movdqu `16*3`($inp),$inout3 1774 pxor @tweak[1],$inout1 1775 movdqu `16*4`($inp),$inout4 1776 pxor @tweak[2],$inout2 1777 movdqu `16*5`($inp),$inout5 1778 lea `16*6`($inp),$inp 1779 pxor @tweak[3],$inout3 1780 $movkey ($key_),$rndkey0 1781 pxor @tweak[4],$inout4 1782 pxor @tweak[5],$inout5 1783 1784 # inline _aesni_decrypt6 and interleave first and last rounds 1785 # with own code... 1786 $movkey 16($key_),$rndkey1 1787 pxor $rndkey0,$inout0 1788 pxor $rndkey0,$inout1 1789 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks 1790 aesdec $rndkey1,$inout0 1791 lea 32($key_),$key 1792 pxor $rndkey0,$inout2 1793 movdqa @tweak[1],`16*1`(%rsp) 1794 aesdec $rndkey1,$inout1 1795 pxor $rndkey0,$inout3 1796 movdqa @tweak[2],`16*2`(%rsp) 1797 aesdec $rndkey1,$inout2 1798 pxor $rndkey0,$inout4 1799 movdqa @tweak[3],`16*3`(%rsp) 1800 aesdec $rndkey1,$inout3 1801 pxor $rndkey0,$inout5 1802 $movkey ($key),$rndkey0 1803 dec $rounds 1804 movdqa @tweak[4],`16*4`(%rsp) 1805 aesdec $rndkey1,$inout4 1806 movdqa @tweak[5],`16*5`(%rsp) 1807 aesdec $rndkey1,$inout5 1808 pxor $twtmp,$twtmp 1809 pcmpgtd @tweak[5],$twtmp 1810 jmp .Lxts_dec_loop6_enter 1811 1812 .align 16 1813 .Lxts_dec_loop6: 1814 aesdec $rndkey1,$inout0 1815 aesdec $rndkey1,$inout1 1816 dec $rounds 1817 aesdec $rndkey1,$inout2 1818 aesdec $rndkey1,$inout3 1819 aesdec $rndkey1,$inout4 1820 aesdec $rndkey1,$inout5 1821 .Lxts_dec_loop6_enter: 1822 $movkey 16($key),$rndkey1 1823 aesdec $rndkey0,$inout0 1824 aesdec $rndkey0,$inout1 1825 lea 32($key),$key 1826 aesdec $rndkey0,$inout2 1827 aesdec $rndkey0,$inout3 1828 aesdec $rndkey0,$inout4 1829 aesdec $rndkey0,$inout5 1830 $movkey ($key),$rndkey0 1831 jnz .Lxts_dec_loop6 1832 1833 pshufd \$0x13,$twtmp,$twres 1834 pxor $twtmp,$twtmp 1835 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1836 aesdec $rndkey1,$inout0 1837 pand $twmask,$twres # isolate carry and residue 1838 aesdec $rndkey1,$inout1 1839 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1840 aesdec $rndkey1,$inout2 1841 pxor $twres,@tweak[5] 1842 aesdec $rndkey1,$inout3 1843 aesdec $rndkey1,$inout4 1844 aesdec $rndkey1,$inout5 1845 $movkey 16($key),$rndkey1 1846 1847 pshufd \$0x13,$twtmp,$twres 1848 pxor $twtmp,$twtmp 1849 movdqa @tweak[5],@tweak[0] 1850 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1851 aesdec $rndkey0,$inout0 1852 pand $twmask,$twres # isolate carry and residue 1853 aesdec $rndkey0,$inout1 1854 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1855 aesdec $rndkey0,$inout2 1856 pxor $twres,@tweak[5] 1857 aesdec $rndkey0,$inout3 1858 aesdec $rndkey0,$inout4 1859 aesdec $rndkey0,$inout5 1860 $movkey 32($key),$rndkey0 1861 1862 pshufd \$0x13,$twtmp,$twres 1863 pxor $twtmp,$twtmp 1864 movdqa @tweak[5],@tweak[1] 1865 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1866 aesdec $rndkey1,$inout0 1867 pand $twmask,$twres # isolate carry and residue 1868 aesdec $rndkey1,$inout1 1869 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1870 aesdec $rndkey1,$inout2 1871 pxor $twres,@tweak[5] 1872 aesdec $rndkey1,$inout3 1873 aesdec $rndkey1,$inout4 1874 aesdec $rndkey1,$inout5 1875 1876 pshufd \$0x13,$twtmp,$twres 1877 pxor $twtmp,$twtmp 1878 movdqa @tweak[5],@tweak[2] 1879 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1880 aesdeclast $rndkey0,$inout0 1881 pand $twmask,$twres # isolate carry and residue 1882 aesdeclast $rndkey0,$inout1 1883 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1884 aesdeclast $rndkey0,$inout2 1885 pxor $twres,@tweak[5] 1886 aesdeclast $rndkey0,$inout3 1887 aesdeclast $rndkey0,$inout4 1888 aesdeclast $rndkey0,$inout5 1889 1890 pshufd \$0x13,$twtmp,$twres 1891 pxor $twtmp,$twtmp 1892 movdqa @tweak[5],@tweak[3] 1893 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1894 xorps `16*0`(%rsp),$inout0 # output^=tweak 1895 pand $twmask,$twres # isolate carry and residue 1896 xorps `16*1`(%rsp),$inout1 1897 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1898 pxor $twres,@tweak[5] 1899 1900 xorps `16*2`(%rsp),$inout2 1901 movups $inout0,`16*0`($out) # write output 1902 xorps `16*3`(%rsp),$inout3 1903 movups $inout1,`16*1`($out) 1904 xorps `16*4`(%rsp),$inout4 1905 movups $inout2,`16*2`($out) 1906 xorps `16*5`(%rsp),$inout5 1907 movups $inout3,`16*3`($out) 1908 mov $rnds_,$rounds # restore $rounds 1909 movups $inout4,`16*4`($out) 1910 movups $inout5,`16*5`($out) 1911 lea `16*6`($out),$out 1912 sub \$16*6,$len 1913 jnc .Lxts_dec_grandloop 1914 1915 lea 3($rounds,$rounds),$rounds # restore original value 1916 mov $key_,$key # restore $key 1917 mov $rounds,$rnds_ # backup $rounds 1918 1919 .Lxts_dec_short: 1920 add \$16*6,$len 1921 jz .Lxts_dec_done 1922 1923 cmp \$0x20,$len 1924 jb .Lxts_dec_one 1925 je .Lxts_dec_two 1926 1927 cmp \$0x40,$len 1928 jb .Lxts_dec_three 1929 je .Lxts_dec_four 1930 1931 pshufd \$0x13,$twtmp,$twres 1932 movdqa @tweak[5],@tweak[4] 1933 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1934 movdqu ($inp),$inout0 1935 pand $twmask,$twres # isolate carry and residue 1936 movdqu 16*1($inp),$inout1 1937 pxor $twres,@tweak[5] 1938 1939 movdqu 16*2($inp),$inout2 1940 pxor @tweak[0],$inout0 1941 movdqu 16*3($inp),$inout3 1942 pxor @tweak[1],$inout1 1943 movdqu 16*4($inp),$inout4 1944 lea 16*5($inp),$inp 1945 pxor @tweak[2],$inout2 1946 pxor @tweak[3],$inout3 1947 pxor @tweak[4],$inout4 1948 1949 call _aesni_decrypt6 1950 1951 xorps @tweak[0],$inout0 1952 xorps @tweak[1],$inout1 1953 xorps @tweak[2],$inout2 1954 movdqu $inout0,($out) 1955 xorps @tweak[3],$inout3 1956 movdqu $inout1,16*1($out) 1957 xorps @tweak[4],$inout4 1958 movdqu $inout2,16*2($out) 1959 pxor $twtmp,$twtmp 1960 movdqu $inout3,16*3($out) 1961 pcmpgtd @tweak[5],$twtmp 1962 movdqu $inout4,16*4($out) 1963 lea 16*5($out),$out 1964 pshufd \$0x13,$twtmp,@tweak[1] # $twres 1965 and \$15,$len_ 1966 jz .Lxts_dec_ret 1967 1968 movdqa @tweak[5],@tweak[0] 1969 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1970 pand $twmask,@tweak[1] # isolate carry and residue 1971 pxor @tweak[5],@tweak[1] 1972 jmp .Lxts_dec_done2 1973 1974 .align 16 1975 .Lxts_dec_one: 1976 movups ($inp),$inout0 1977 lea 16*1($inp),$inp 1978 xorps @tweak[0],$inout0 1979 ___ 1980 &aesni_generate1("dec",$key,$rounds); 1981 $code.=<<___; 1982 xorps @tweak[0],$inout0 1983 movdqa @tweak[1],@tweak[0] 1984 movups $inout0,($out) 1985 movdqa @tweak[2],@tweak[1] 1986 lea 16*1($out),$out 1987 jmp .Lxts_dec_done 1988 1989 .align 16 1990 .Lxts_dec_two: 1991 movups ($inp),$inout0 1992 movups 16($inp),$inout1 1993 lea 32($inp),$inp 1994 xorps @tweak[0],$inout0 1995 xorps @tweak[1],$inout1 1996 1997 call _aesni_decrypt3 1998 1999 xorps @tweak[0],$inout0 2000 movdqa @tweak[2],@tweak[0] 2001 xorps @tweak[1],$inout1 2002 movdqa @tweak[3],@tweak[1] 2003 movups $inout0,($out) 2004 movups $inout1,16*1($out) 2005 lea 16*2($out),$out 2006 jmp .Lxts_dec_done 2007 2008 .align 16 2009 .Lxts_dec_three: 2010 movups ($inp),$inout0 2011 movups 16*1($inp),$inout1 2012 movups 16*2($inp),$inout2 2013 lea 16*3($inp),$inp 2014 xorps @tweak[0],$inout0 2015 xorps @tweak[1],$inout1 2016 xorps @tweak[2],$inout2 2017 2018 call _aesni_decrypt3 2019 2020 xorps @tweak[0],$inout0 2021 movdqa @tweak[3],@tweak[0] 2022 xorps @tweak[1],$inout1 2023 movdqa @tweak[5],@tweak[1] 2024 xorps @tweak[2],$inout2 2025 movups $inout0,($out) 2026 movups $inout1,16*1($out) 2027 movups $inout2,16*2($out) 2028 lea 16*3($out),$out 2029 jmp .Lxts_dec_done 2030 2031 .align 16 2032 .Lxts_dec_four: 2033 pshufd \$0x13,$twtmp,$twres 2034 movdqa @tweak[5],@tweak[4] 2035 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 2036 movups ($inp),$inout0 2037 pand $twmask,$twres # isolate carry and residue 2038 movups 16*1($inp),$inout1 2039 pxor $twres,@tweak[5] 2040 2041 movups 16*2($inp),$inout2 2042 xorps @tweak[0],$inout0 2043 movups 16*3($inp),$inout3 2044 lea 16*4($inp),$inp 2045 xorps @tweak[1],$inout1 2046 xorps @tweak[2],$inout2 2047 xorps @tweak[3],$inout3 2048 2049 call _aesni_decrypt4 2050 2051 xorps @tweak[0],$inout0 2052 movdqa @tweak[4],@tweak[0] 2053 xorps @tweak[1],$inout1 2054 movdqa @tweak[5],@tweak[1] 2055 xorps @tweak[2],$inout2 2056 movups $inout0,($out) 2057 xorps @tweak[3],$inout3 2058 movups $inout1,16*1($out) 2059 movups $inout2,16*2($out) 2060 movups $inout3,16*3($out) 2061 lea 16*4($out),$out 2062 jmp .Lxts_dec_done 2063 2064 .align 16 2065 .Lxts_dec_done: 2066 and \$15,$len_ 2067 jz .Lxts_dec_ret 2068 .Lxts_dec_done2: 2069 mov $len_,$len 2070 mov $key_,$key # restore $key 2071 mov $rnds_,$rounds # restore $rounds 2072 2073 movups ($inp),$inout0 2074 xorps @tweak[1],$inout0 2075 ___ 2076 &aesni_generate1("dec",$key,$rounds); 2077 $code.=<<___; 2078 xorps @tweak[1],$inout0 2079 movups $inout0,($out) 2080 2081 .Lxts_dec_steal: 2082 movzb 16($inp),%eax # borrow $rounds ... 2083 movzb ($out),%ecx # ... and $key 2084 lea 1($inp),$inp 2085 mov %al,($out) 2086 mov %cl,16($out) 2087 lea 1($out),$out 2088 sub \$1,$len 2089 jnz .Lxts_dec_steal 2090 2091 sub $len_,$out # rewind $out 2092 mov $key_,$key # restore $key 2093 mov $rnds_,$rounds # restore $rounds 2094 2095 movups ($out),$inout0 2096 xorps @tweak[0],$inout0 2097 ___ 2098 &aesni_generate1("dec",$key,$rounds); 2099 $code.=<<___; 2100 xorps @tweak[0],$inout0 2101 movups $inout0,($out) 2102 2103 .Lxts_dec_ret: 2104 ___ 2105 $code.=<<___ if ($win64); 2106 movaps 0x60(%rsp),%xmm6 2107 movaps 0x70(%rsp),%xmm7 2108 movaps 0x80(%rsp),%xmm8 2109 movaps 0x90(%rsp),%xmm9 2110 movaps 0xa0(%rsp),%xmm10 2111 movaps 0xb0(%rsp),%xmm11 2112 movaps 0xc0(%rsp),%xmm12 2113 movaps 0xd0(%rsp),%xmm13 2114 movaps 0xe0(%rsp),%xmm14 2115 movaps 0xf0(%rsp),%xmm15 2116 ___ 2117 $code.=<<___; 2118 lea $frame_size(%rsp),%rsp 2119 .Lxts_dec_epilogue: 2120 ret 2121 .size aesni_xts_decrypt,.-aesni_xts_decrypt 2122 ___ 2123 } }} 2124 2126 ######################################################################## 2127 # void $PREFIX_cbc_encrypt (const void *inp, void *out, 2128 # size_t length, const AES_KEY *key, 2129 # unsigned char *ivp,const int enc); 2130 { 2131 my $reserved = $win64?0x40:-0x18; # used in decrypt 2132 $code.=<<___; 2133 .globl ${PREFIX}_cbc_encrypt 2134 .type ${PREFIX}_cbc_encrypt,\@function,6 2135 .align 16 2136 ${PREFIX}_cbc_encrypt: 2137 test $len,$len # check length 2138 jz .Lcbc_ret 2139 2140 mov 240($key),$rnds_ # key->rounds 2141 mov $key,$key_ # backup $key 2142 test %r9d,%r9d # 6th argument 2143 jz .Lcbc_decrypt 2144 #--------------------------- CBC ENCRYPT ------------------------------# 2145 movups ($ivp),$inout0 # load iv as initial state 2146 mov $rnds_,$rounds 2147 cmp \$16,$len 2148 jb .Lcbc_enc_tail 2149 sub \$16,$len 2150 jmp .Lcbc_enc_loop 2151 .align 16 2152 .Lcbc_enc_loop: 2153 movups ($inp),$inout1 # load input 2154 lea 16($inp),$inp 2155 #xorps $inout1,$inout0 2156 ___ 2157 &aesni_generate1("enc",$key,$rounds,$inout0,$inout1); 2158 $code.=<<___; 2159 mov $rnds_,$rounds # restore $rounds 2160 mov $key_,$key # restore $key 2161 movups $inout0,0($out) # store output 2162 lea 16($out),$out 2163 sub \$16,$len 2164 jnc .Lcbc_enc_loop 2165 add \$16,$len 2166 jnz .Lcbc_enc_tail 2167 movups $inout0,($ivp) 2168 jmp .Lcbc_ret 2169 2170 .Lcbc_enc_tail: 2171 mov $len,%rcx # zaps $key 2172 xchg $inp,$out # $inp is %rsi and $out is %rdi now 2173 .long 0x9066A4F3 # rep movsb 2174 mov \$16,%ecx # zero tail 2175 sub $len,%rcx 2176 xor %eax,%eax 2177 .long 0x9066AAF3 # rep stosb 2178 lea -16(%rdi),%rdi # rewind $out by 1 block 2179 mov $rnds_,$rounds # restore $rounds 2180 mov %rdi,%rsi # $inp and $out are the same 2181 mov $key_,$key # restore $key 2182 xor $len,$len # len=16 2183 jmp .Lcbc_enc_loop # one more spin 2184 #--------------------------- CBC DECRYPT ------------------------------# 2186 .align 16 2187 .Lcbc_decrypt: 2188 ___ 2189 $code.=<<___ if ($win64); 2190 lea -0x58(%rsp),%rsp 2191 movaps %xmm6,(%rsp) 2192 movaps %xmm7,0x10(%rsp) 2193 movaps %xmm8,0x20(%rsp) 2194 movaps %xmm9,0x30(%rsp) 2195 .Lcbc_decrypt_body: 2196 ___ 2197 $code.=<<___; 2198 movups ($ivp),$iv 2199 mov $rnds_,$rounds 2200 cmp \$0x70,$len 2201 jbe .Lcbc_dec_tail 2202 shr \$1,$rnds_ 2203 sub \$0x70,$len 2204 mov $rnds_,$rounds 2205 movaps $iv,$reserved(%rsp) 2206 jmp .Lcbc_dec_loop8_enter 2207 .align 16 2208 .Lcbc_dec_loop8: 2209 movaps $rndkey0,$reserved(%rsp) # save IV 2210 movups $inout7,($out) 2211 lea 0x10($out),$out 2212 .Lcbc_dec_loop8_enter: 2213 $movkey ($key),$rndkey0 2214 movups ($inp),$inout0 # load input 2215 movups 0x10($inp),$inout1 2216 $movkey 16($key),$rndkey1 2217 2218 lea 32($key),$key 2219 movdqu 0x20($inp),$inout2 2220 xorps $rndkey0,$inout0 2221 movdqu 0x30($inp),$inout3 2222 xorps $rndkey0,$inout1 2223 movdqu 0x40($inp),$inout4 2224 aesdec $rndkey1,$inout0 2225 pxor $rndkey0,$inout2 2226 movdqu 0x50($inp),$inout5 2227 aesdec $rndkey1,$inout1 2228 pxor $rndkey0,$inout3 2229 movdqu 0x60($inp),$inout6 2230 aesdec $rndkey1,$inout2 2231 pxor $rndkey0,$inout4 2232 movdqu 0x70($inp),$inout7 2233 aesdec $rndkey1,$inout3 2234 pxor $rndkey0,$inout5 2235 dec $rounds 2236 aesdec $rndkey1,$inout4 2237 pxor $rndkey0,$inout6 2238 aesdec $rndkey1,$inout5 2239 pxor $rndkey0,$inout7 2240 $movkey ($key),$rndkey0 2241 aesdec $rndkey1,$inout6 2242 aesdec $rndkey1,$inout7 2243 $movkey 16($key),$rndkey1 2244 2245 call .Ldec_loop8_enter 2246 2247 movups ($inp),$rndkey1 # re-load input 2248 movups 0x10($inp),$rndkey0 2249 xorps $reserved(%rsp),$inout0 # ^= IV 2250 xorps $rndkey1,$inout1 2251 movups 0x20($inp),$rndkey1 2252 xorps $rndkey0,$inout2 2253 movups 0x30($inp),$rndkey0 2254 xorps $rndkey1,$inout3 2255 movups 0x40($inp),$rndkey1 2256 xorps $rndkey0,$inout4 2257 movups 0x50($inp),$rndkey0 2258 xorps $rndkey1,$inout5 2259 movups 0x60($inp),$rndkey1 2260 xorps $rndkey0,$inout6 2261 movups 0x70($inp),$rndkey0 # IV 2262 xorps $rndkey1,$inout7 2263 movups $inout0,($out) 2264 movups $inout1,0x10($out) 2265 movups $inout2,0x20($out) 2266 movups $inout3,0x30($out) 2267 mov $rnds_,$rounds # restore $rounds 2268 movups $inout4,0x40($out) 2269 mov $key_,$key # restore $key 2270 movups $inout5,0x50($out) 2271 lea 0x80($inp),$inp 2272 movups $inout6,0x60($out) 2273 lea 0x70($out),$out 2274 sub \$0x80,$len 2275 ja .Lcbc_dec_loop8 2276 2277 movaps $inout7,$inout0 2278 movaps $rndkey0,$iv 2279 add \$0x70,$len 2280 jle .Lcbc_dec_tail_collected 2281 movups $inout0,($out) 2282 lea 1($rnds_,$rnds_),$rounds 2283 lea 0x10($out),$out 2284 .Lcbc_dec_tail: 2285 movups ($inp),$inout0 2286 movaps $inout0,$in0 2287 cmp \$0x10,$len 2288 jbe .Lcbc_dec_one 2289 2290 movups 0x10($inp),$inout1 2291 movaps $inout1,$in1 2292 cmp \$0x20,$len 2293 jbe .Lcbc_dec_two 2294 2295 movups 0x20($inp),$inout2 2296 movaps $inout2,$in2 2297 cmp \$0x30,$len 2298 jbe .Lcbc_dec_three 2299 2300 movups 0x30($inp),$inout3 2301 cmp \$0x40,$len 2302 jbe .Lcbc_dec_four 2303 2304 movups 0x40($inp),$inout4 2305 cmp \$0x50,$len 2306 jbe .Lcbc_dec_five 2307 2308 movups 0x50($inp),$inout5 2309 cmp \$0x60,$len 2310 jbe .Lcbc_dec_six 2311 2312 movups 0x60($inp),$inout6 2313 movaps $iv,$reserved(%rsp) # save IV 2314 call _aesni_decrypt8 2315 movups ($inp),$rndkey1 2316 movups 0x10($inp),$rndkey0 2317 xorps $reserved(%rsp),$inout0 # ^= IV 2318 xorps $rndkey1,$inout1 2319 movups 0x20($inp),$rndkey1 2320 xorps $rndkey0,$inout2 2321 movups 0x30($inp),$rndkey0 2322 xorps $rndkey1,$inout3 2323 movups 0x40($inp),$rndkey1 2324 xorps $rndkey0,$inout4 2325 movups 0x50($inp),$rndkey0 2326 xorps $rndkey1,$inout5 2327 movups 0x60($inp),$iv # IV 2328 xorps $rndkey0,$inout6 2329 movups $inout0,($out) 2330 movups $inout1,0x10($out) 2331 movups $inout2,0x20($out) 2332 movups $inout3,0x30($out) 2333 movups $inout4,0x40($out) 2334 movups $inout5,0x50($out) 2335 lea 0x60($out),$out 2336 movaps $inout6,$inout0 2337 sub \$0x70,$len 2338 jmp .Lcbc_dec_tail_collected 2339 .align 16 2340 .Lcbc_dec_one: 2341 ___ 2342 &aesni_generate1("dec",$key,$rounds); 2343 $code.=<<___; 2344 xorps $iv,$inout0 2345 movaps $in0,$iv 2346 sub \$0x10,$len 2347 jmp .Lcbc_dec_tail_collected 2348 .align 16 2349 .Lcbc_dec_two: 2350 xorps $inout2,$inout2 2351 call _aesni_decrypt3 2352 xorps $iv,$inout0 2353 xorps $in0,$inout1 2354 movups $inout0,($out) 2355 movaps $in1,$iv 2356 movaps $inout1,$inout0 2357 lea 0x10($out),$out 2358 sub \$0x20,$len 2359 jmp .Lcbc_dec_tail_collected 2360 .align 16 2361 .Lcbc_dec_three: 2362 call _aesni_decrypt3 2363 xorps $iv,$inout0 2364 xorps $in0,$inout1 2365 movups $inout0,($out) 2366 xorps $in1,$inout2 2367 movups $inout1,0x10($out) 2368 movaps $in2,$iv 2369 movaps $inout2,$inout0 2370 lea 0x20($out),$out 2371 sub \$0x30,$len 2372 jmp .Lcbc_dec_tail_collected 2373 .align 16 2374 .Lcbc_dec_four: 2375 call _aesni_decrypt4 2376 xorps $iv,$inout0 2377 movups 0x30($inp),$iv 2378 xorps $in0,$inout1 2379 movups $inout0,($out) 2380 xorps $in1,$inout2 2381 movups $inout1,0x10($out) 2382 xorps $in2,$inout3 2383 movups $inout2,0x20($out) 2384 movaps $inout3,$inout0 2385 lea 0x30($out),$out 2386 sub \$0x40,$len 2387 jmp .Lcbc_dec_tail_collected 2388 .align 16 2389 .Lcbc_dec_five: 2390 xorps $inout5,$inout5 2391 call _aesni_decrypt6 2392 movups 0x10($inp),$rndkey1 2393 movups 0x20($inp),$rndkey0 2394 xorps $iv,$inout0 2395 xorps $in0,$inout1 2396 xorps $rndkey1,$inout2 2397 movups 0x30($inp),$rndkey1 2398 xorps $rndkey0,$inout3 2399 movups 0x40($inp),$iv 2400 xorps $rndkey1,$inout4 2401 movups $inout0,($out) 2402 movups $inout1,0x10($out) 2403 movups $inout2,0x20($out) 2404 movups $inout3,0x30($out) 2405 lea 0x40($out),$out 2406 movaps $inout4,$inout0 2407 sub \$0x50,$len 2408 jmp .Lcbc_dec_tail_collected 2409 .align 16 2410 .Lcbc_dec_six: 2411 call _aesni_decrypt6 2412 movups 0x10($inp),$rndkey1 2413 movups 0x20($inp),$rndkey0 2414 xorps $iv,$inout0 2415 xorps $in0,$inout1 2416 xorps $rndkey1,$inout2 2417 movups 0x30($inp),$rndkey1 2418 xorps $rndkey0,$inout3 2419 movups 0x40($inp),$rndkey0 2420 xorps $rndkey1,$inout4 2421 movups 0x50($inp),$iv 2422 xorps $rndkey0,$inout5 2423 movups $inout0,($out) 2424 movups $inout1,0x10($out) 2425 movups $inout2,0x20($out) 2426 movups $inout3,0x30($out) 2427 movups $inout4,0x40($out) 2428 lea 0x50($out),$out 2429 movaps $inout5,$inout0 2430 sub \$0x60,$len 2431 jmp .Lcbc_dec_tail_collected 2432 .align 16 2433 .Lcbc_dec_tail_collected: 2434 and \$15,$len 2435 movups $iv,($ivp) 2436 jnz .Lcbc_dec_tail_partial 2437 movups $inout0,($out) 2438 jmp .Lcbc_dec_ret 2439 .align 16 2440 .Lcbc_dec_tail_partial: 2441 movaps $inout0,$reserved(%rsp) 2442 mov \$16,%rcx 2443 mov $out,%rdi 2444 sub $len,%rcx 2445 lea $reserved(%rsp),%rsi 2446 .long 0x9066A4F3 # rep movsb 2447 2448 .Lcbc_dec_ret: 2449 ___ 2450 $code.=<<___ if ($win64); 2451 movaps (%rsp),%xmm6 2452 movaps 0x10(%rsp),%xmm7 2453 movaps 0x20(%rsp),%xmm8 2454 movaps 0x30(%rsp),%xmm9 2455 lea 0x58(%rsp),%rsp 2456 ___ 2457 $code.=<<___; 2458 .Lcbc_ret: 2459 ret 2460 .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt 2461 ___ 2462 } 2464 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey, 2465 # int bits, AES_KEY *key) 2466 { my ($inp,$bits,$key) = @_4args; 2467 $bits =~ s/%r/%e/; 2468 2469 $code.=<<___; 2470 .globl ${PREFIX}_set_decrypt_key 2471 .type ${PREFIX}_set_decrypt_key,\@abi-omnipotent 2472 .align 16 2473 ${PREFIX}_set_decrypt_key: 2474 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 2475 call __aesni_set_encrypt_key 2476 shl \$4,$bits # rounds-1 after _aesni_set_encrypt_key 2477 test %eax,%eax 2478 jnz .Ldec_key_ret 2479 lea 16($key,$bits),$inp # points at the end of key schedule 2480 2481 $movkey ($key),%xmm0 # just swap 2482 $movkey ($inp),%xmm1 2483 $movkey %xmm0,($inp) 2484 $movkey %xmm1,($key) 2485 lea 16($key),$key 2486 lea -16($inp),$inp 2487 2488 .Ldec_key_inverse: 2489 $movkey ($key),%xmm0 # swap and inverse 2490 $movkey ($inp),%xmm1 2491 aesimc %xmm0,%xmm0 2492 aesimc %xmm1,%xmm1 2493 lea 16($key),$key 2494 lea -16($inp),$inp 2495 $movkey %xmm0,16($inp) 2496 $movkey %xmm1,-16($key) 2497 cmp $key,$inp 2498 ja .Ldec_key_inverse 2499 2500 $movkey ($key),%xmm0 # inverse middle 2501 aesimc %xmm0,%xmm0 2502 $movkey %xmm0,($inp) 2503 .Ldec_key_ret: 2504 add \$8,%rsp 2505 ret 2506 .LSEH_end_set_decrypt_key: 2507 .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key 2508 ___ 2509 2511 # This is based on submission by 2512 # 2513 # Huang Ying <ying.huang (at] intel.com> 2514 # Vinodh Gopal <vinodh.gopal (at] intel.com> 2515 # Kahraman Akdemir 2516 # 2517 # Agressively optimized in respect to aeskeygenassist's critical path 2518 # and is contained in %xmm0-5 to meet Win64 ABI requirement. 2519 # 2520 $code.=<<___; 2521 .globl ${PREFIX}_set_encrypt_key 2522 .type ${PREFIX}_set_encrypt_key,\@abi-omnipotent 2523 .align 16 2524 ${PREFIX}_set_encrypt_key: 2525 __aesni_set_encrypt_key: 2526 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 2527 mov \$-1,%rax 2528 test $inp,$inp 2529 jz .Lenc_key_ret 2530 test $key,$key 2531 jz .Lenc_key_ret 2532 2533 movups ($inp),%xmm0 # pull first 128 bits of *userKey 2534 xorps %xmm4,%xmm4 # low dword of xmm4 is assumed 0 2535 lea 16($key),%rax 2536 cmp \$256,$bits 2537 je .L14rounds 2538 cmp \$192,$bits 2539 je .L12rounds 2540 cmp \$128,$bits 2541 jne .Lbad_keybits 2542 2543 .L10rounds: 2544 mov \$9,$bits # 10 rounds for 128-bit key 2545 $movkey %xmm0,($key) # round 0 2546 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 1 2547 call .Lkey_expansion_128_cold 2548 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 2 2549 call .Lkey_expansion_128 2550 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 3 2551 call .Lkey_expansion_128 2552 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 4 2553 call .Lkey_expansion_128 2554 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 5 2555 call .Lkey_expansion_128 2556 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 6 2557 call .Lkey_expansion_128 2558 aeskeygenassist \$0x40,%xmm0,%xmm1 # round 7 2559 call .Lkey_expansion_128 2560 aeskeygenassist \$0x80,%xmm0,%xmm1 # round 8 2561 call .Lkey_expansion_128 2562 aeskeygenassist \$0x1b,%xmm0,%xmm1 # round 9 2563 call .Lkey_expansion_128 2564 aeskeygenassist \$0x36,%xmm0,%xmm1 # round 10 2565 call .Lkey_expansion_128 2566 $movkey %xmm0,(%rax) 2567 mov $bits,80(%rax) # 240(%rdx) 2568 xor %eax,%eax 2569 jmp .Lenc_key_ret 2570 2571 .align 16 2572 .L12rounds: 2573 movq 16($inp),%xmm2 # remaining 1/3 of *userKey 2574 mov \$11,$bits # 12 rounds for 192 2575 $movkey %xmm0,($key) # round 0 2576 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 1,2 2577 call .Lkey_expansion_192a_cold 2578 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 2,3 2579 call .Lkey_expansion_192b 2580 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 4,5 2581 call .Lkey_expansion_192a 2582 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 5,6 2583 call .Lkey_expansion_192b 2584 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 7,8 2585 call .Lkey_expansion_192a 2586 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 8,9 2587 call .Lkey_expansion_192b 2588 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 10,11 2589 call .Lkey_expansion_192a 2590 aeskeygenassist \$0x80,%xmm2,%xmm1 # round 11,12 2591 call .Lkey_expansion_192b 2592 $movkey %xmm0,(%rax) 2593 mov $bits,48(%rax) # 240(%rdx) 2594 xor %rax, %rax 2595 jmp .Lenc_key_ret 2596 2597 .align 16 2598 .L14rounds: 2599 movups 16($inp),%xmm2 # remaning half of *userKey 2600 mov \$13,$bits # 14 rounds for 256 2601 lea 16(%rax),%rax 2602 $movkey %xmm0,($key) # round 0 2603 $movkey %xmm2,16($key) # round 1 2604 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 2 2605 call .Lkey_expansion_256a_cold 2606 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 3 2607 call .Lkey_expansion_256b 2608 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 4 2609 call .Lkey_expansion_256a 2610 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 5 2611 call .Lkey_expansion_256b 2612 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 6 2613 call .Lkey_expansion_256a 2614 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 7 2615 call .Lkey_expansion_256b 2616 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 8 2617 call .Lkey_expansion_256a 2618 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 9 2619 call .Lkey_expansion_256b 2620 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 10 2621 call .Lkey_expansion_256a 2622 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 11 2623 call .Lkey_expansion_256b 2624 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 12 2625 call .Lkey_expansion_256a 2626 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 13 2627 call .Lkey_expansion_256b 2628 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 14 2629 call .Lkey_expansion_256a 2630 $movkey %xmm0,(%rax) 2631 mov $bits,16(%rax) # 240(%rdx) 2632 xor %rax,%rax 2633 jmp .Lenc_key_ret 2634 2635 .align 16 2636 .Lbad_keybits: 2637 mov \$-2,%rax 2638 .Lenc_key_ret: 2639 add \$8,%rsp 2640 ret 2641 .LSEH_end_set_encrypt_key: 2642 2644 .align 16 2645 .Lkey_expansion_128: 2646 $movkey %xmm0,(%rax) 2647 lea 16(%rax),%rax 2648 .Lkey_expansion_128_cold: 2649 shufps \$0b00010000,%xmm0,%xmm4 2650 xorps %xmm4, %xmm0 2651 shufps \$0b10001100,%xmm0,%xmm4 2652 xorps %xmm4, %xmm0 2653 shufps \$0b11111111,%xmm1,%xmm1 # critical path 2654 xorps %xmm1,%xmm0 2655 ret 2656 2657 .align 16 2658 .Lkey_expansion_192a: 2659 $movkey %xmm0,(%rax) 2660 lea 16(%rax),%rax 2661 .Lkey_expansion_192a_cold: 2662 movaps %xmm2, %xmm5 2663 .Lkey_expansion_192b_warm: 2664 shufps \$0b00010000,%xmm0,%xmm4 2665 movdqa %xmm2,%xmm3 2666 xorps %xmm4,%xmm0 2667 shufps \$0b10001100,%xmm0,%xmm4 2668 pslldq \$4,%xmm3 2669 xorps %xmm4,%xmm0 2670 pshufd \$0b01010101,%xmm1,%xmm1 # critical path 2671 pxor %xmm3,%xmm2 2672 pxor %xmm1,%xmm0 2673 pshufd \$0b11111111,%xmm0,%xmm3 2674 pxor %xmm3,%xmm2 2675 ret 2676 2677 .align 16 2678 .Lkey_expansion_192b: 2679 movaps %xmm0,%xmm3 2680 shufps \$0b01000100,%xmm0,%xmm5 2681 $movkey %xmm5,(%rax) 2682 shufps \$0b01001110,%xmm2,%xmm3 2683 $movkey %xmm3,16(%rax) 2684 lea 32(%rax),%rax 2685 jmp .Lkey_expansion_192b_warm 2686 2687 .align 16 2688 .Lkey_expansion_256a: 2689 $movkey %xmm2,(%rax) 2690 lea 16(%rax),%rax 2691 .Lkey_expansion_256a_cold: 2692 shufps \$0b00010000,%xmm0,%xmm4 2693 xorps %xmm4,%xmm0 2694 shufps \$0b10001100,%xmm0,%xmm4 2695 xorps %xmm4,%xmm0 2696 shufps \$0b11111111,%xmm1,%xmm1 # critical path 2697 xorps %xmm1,%xmm0 2698 ret 2699 2700 .align 16 2701 .Lkey_expansion_256b: 2702 $movkey %xmm0,(%rax) 2703 lea 16(%rax),%rax 2704 2705 shufps \$0b00010000,%xmm2,%xmm4 2706 xorps %xmm4,%xmm2 2707 shufps \$0b10001100,%xmm2,%xmm4 2708 xorps %xmm4,%xmm2 2709 shufps \$0b10101010,%xmm1,%xmm1 # critical path 2710 xorps %xmm1,%xmm2 2711 ret 2712 .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key 2713 .size __aesni_set_encrypt_key,.-__aesni_set_encrypt_key 2714 ___ 2715 } 2716 2718 $code.=<<___; 2719 .align 64 2720 .Lbswap_mask: 2721 .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 2722 .Lincrement32: 2723 .long 6,6,6,0 2724 .Lincrement64: 2725 .long 1,0,0,0 2726 .Lxts_magic: 2727 .long 0x87,0,1,0 2728 2729 .asciz "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>" 2730 .align 64 2731 ___ 2732 2733 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 2734 # CONTEXT *context,DISPATCHER_CONTEXT *disp) 2735 if ($win64) { 2736 $rec="%rcx"; 2737 $frame="%rdx"; 2738 $context="%r8"; 2739 $disp="%r9"; 2740 2741 $code.=<<___; 2742 .extern __imp_RtlVirtualUnwind 2743 ___ 2744 $code.=<<___ if ($PREFIX eq "aesni"); 2745 .type ecb_se_handler,\@abi-omnipotent 2746 .align 16 2747 ecb_se_handler: 2748 push %rsi 2749 push %rdi 2750 push %rbx 2751 push %rbp 2752 push %r12 2753 push %r13 2754 push %r14 2755 push %r15 2756 pushfq 2757 sub \$64,%rsp 2758 2759 mov 152($context),%rax # pull context->Rsp 2760 2761 jmp .Lcommon_seh_tail 2762 .size ecb_se_handler,.-ecb_se_handler 2763 2764 .type ccm64_se_handler,\@abi-omnipotent 2765 .align 16 2766 ccm64_se_handler: 2767 push %rsi 2768 push %rdi 2769 push %rbx 2770 push %rbp 2771 push %r12 2772 push %r13 2773 push %r14 2774 push %r15 2775 pushfq 2776 sub \$64,%rsp 2777 2778 mov 120($context),%rax # pull context->Rax 2779 mov 248($context),%rbx # pull context->Rip 2780 2781 mov 8($disp),%rsi # disp->ImageBase 2782 mov 56($disp),%r11 # disp->HandlerData 2783 2784 mov 0(%r11),%r10d # HandlerData[0] 2785 lea (%rsi,%r10),%r10 # prologue label 2786 cmp %r10,%rbx # context->Rip<prologue label 2787 jb .Lcommon_seh_tail 2788 2789 mov 152($context),%rax # pull context->Rsp 2790 2791 mov 4(%r11),%r10d # HandlerData[1] 2792 lea (%rsi,%r10),%r10 # epilogue label 2793 cmp %r10,%rbx # context->Rip>=epilogue label 2794 jae .Lcommon_seh_tail 2795 2796 lea 0(%rax),%rsi # %xmm save area 2797 lea 512($context),%rdi # &context.Xmm6 2798 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) 2799 .long 0xa548f3fc # cld; rep movsq 2800 lea 0x58(%rax),%rax # adjust stack pointer 2801 2802 jmp .Lcommon_seh_tail 2803 .size ccm64_se_handler,.-ccm64_se_handler 2804 2805 .type ctr32_se_handler,\@abi-omnipotent 2806 .align 16 2807 ctr32_se_handler: 2808 push %rsi 2809 push %rdi 2810 push %rbx 2811 push %rbp 2812 push %r12 2813 push %r13 2814 push %r14 2815 push %r15 2816 pushfq 2817 sub \$64,%rsp 2818 2819 mov 120($context),%rax # pull context->Rax 2820 mov 248($context),%rbx # pull context->Rip 2821 2822 lea .Lctr32_body(%rip),%r10 2823 cmp %r10,%rbx # context->Rip<"prologue" label 2824 jb .Lcommon_seh_tail 2825 2826 mov 152($context),%rax # pull context->Rsp 2827 2828 lea .Lctr32_ret(%rip),%r10 2829 cmp %r10,%rbx 2830 jae .Lcommon_seh_tail 2831 2832 lea 0x20(%rax),%rsi # %xmm save area 2833 lea 512($context),%rdi # &context.Xmm6 2834 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) 2835 .long 0xa548f3fc # cld; rep movsq 2836 lea 0xc8(%rax),%rax # adjust stack pointer 2837 2838 jmp .Lcommon_seh_tail 2839 .size ctr32_se_handler,.-ctr32_se_handler 2840 2841 .type xts_se_handler,\@abi-omnipotent 2842 .align 16 2843 xts_se_handler: 2844 push %rsi 2845 push %rdi 2846 push %rbx 2847 push %rbp 2848 push %r12 2849 push %r13 2850 push %r14 2851 push %r15 2852 pushfq 2853 sub \$64,%rsp 2854 2855 mov 120($context),%rax # pull context->Rax 2856 mov 248($context),%rbx # pull context->Rip 2857 2858 mov 8($disp),%rsi # disp->ImageBase 2859 mov 56($disp),%r11 # disp->HandlerData 2860 2861 mov 0(%r11),%r10d # HandlerData[0] 2862 lea (%rsi,%r10),%r10 # prologue lable 2863 cmp %r10,%rbx # context->Rip<prologue label 2864 jb .Lcommon_seh_tail 2865 2866 mov 152($context),%rax # pull context->Rsp 2867 2868 mov 4(%r11),%r10d # HandlerData[1] 2869 lea (%rsi,%r10),%r10 # epilogue label 2870 cmp %r10,%rbx # context->Rip>=epilogue label 2871 jae .Lcommon_seh_tail 2872 2873 lea 0x60(%rax),%rsi # %xmm save area 2874 lea 512($context),%rdi # & context.Xmm6 2875 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) 2876 .long 0xa548f3fc # cld; rep movsq 2877 lea 0x68+160(%rax),%rax # adjust stack pointer 2878 2879 jmp .Lcommon_seh_tail 2880 .size xts_se_handler,.-xts_se_handler 2881 ___ 2882 $code.=<<___; 2883 .type cbc_se_handler,\@abi-omnipotent 2884 .align 16 2885 cbc_se_handler: 2886 push %rsi 2887 push %rdi 2888 push %rbx 2889 push %rbp 2890 push %r12 2891 push %r13 2892 push %r14 2893 push %r15 2894 pushfq 2895 sub \$64,%rsp 2896 2897 mov 152($context),%rax # pull context->Rsp 2898 mov 248($context),%rbx # pull context->Rip 2899 2900 lea .Lcbc_decrypt(%rip),%r10 2901 cmp %r10,%rbx # context->Rip<"prologue" label 2902 jb .Lcommon_seh_tail 2903 2904 lea .Lcbc_decrypt_body(%rip),%r10 2905 cmp %r10,%rbx # context->Rip<cbc_decrypt_body 2906 jb .Lrestore_cbc_rax 2907 2908 lea .Lcbc_ret(%rip),%r10 2909 cmp %r10,%rbx # context->Rip>="epilogue" label 2910 jae .Lcommon_seh_tail 2911 2912 lea 0(%rax),%rsi # top of stack 2913 lea 512($context),%rdi # &context.Xmm6 2914 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) 2915 .long 0xa548f3fc # cld; rep movsq 2916 lea 0x58(%rax),%rax # adjust stack pointer 2917 jmp .Lcommon_seh_tail 2918 2919 .Lrestore_cbc_rax: 2920 mov 120($context),%rax 2921 2922 .Lcommon_seh_tail: 2923 mov 8(%rax),%rdi 2924 mov 16(%rax),%rsi 2925 mov %rax,152($context) # restore context->Rsp 2926 mov %rsi,168($context) # restore context->Rsi 2927 mov %rdi,176($context) # restore context->Rdi 2928 2929 mov 40($disp),%rdi # disp->ContextRecord 2930 mov $context,%rsi # context 2931 mov \$154,%ecx # sizeof(CONTEXT) 2932 .long 0xa548f3fc # cld; rep movsq 2933 2934 mov $disp,%rsi 2935 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER 2936 mov 8(%rsi),%rdx # arg2, disp->ImageBase 2937 mov 0(%rsi),%r8 # arg3, disp->ControlPc 2938 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry 2939 mov 40(%rsi),%r10 # disp->ContextRecord 2940 lea 56(%rsi),%r11 # &disp->HandlerData 2941 lea 24(%rsi),%r12 # &disp->EstablisherFrame 2942 mov %r10,32(%rsp) # arg5 2943 mov %r11,40(%rsp) # arg6 2944 mov %r12,48(%rsp) # arg7 2945 mov %rcx,56(%rsp) # arg8, (NULL) 2946 call *__imp_RtlVirtualUnwind(%rip) 2947 2948 mov \$1,%eax # ExceptionContinueSearch 2949 add \$64,%rsp 2950 popfq 2951 pop %r15 2952 pop %r14 2953 pop %r13 2954 pop %r12 2955 pop %rbp 2956 pop %rbx 2957 pop %rdi 2958 pop %rsi 2959 ret 2960 .size cbc_se_handler,.-cbc_se_handler 2961 2962 .section .pdata 2963 .align 4 2964 ___ 2965 $code.=<<___ if ($PREFIX eq "aesni"); 2966 .rva .LSEH_begin_aesni_ecb_encrypt 2967 .rva .LSEH_end_aesni_ecb_encrypt 2968 .rva .LSEH_info_ecb 2969 2970 .rva .LSEH_begin_aesni_ccm64_encrypt_blocks 2971 .rva .LSEH_end_aesni_ccm64_encrypt_blocks 2972 .rva .LSEH_info_ccm64_enc 2973 2974 .rva .LSEH_begin_aesni_ccm64_decrypt_blocks 2975 .rva .LSEH_end_aesni_ccm64_decrypt_blocks 2976 .rva .LSEH_info_ccm64_dec 2977 2978 .rva .LSEH_begin_aesni_ctr32_encrypt_blocks 2979 .rva .LSEH_end_aesni_ctr32_encrypt_blocks 2980 .rva .LSEH_info_ctr32 2981 2982 .rva .LSEH_begin_aesni_xts_encrypt 2983 .rva .LSEH_end_aesni_xts_encrypt 2984 .rva .LSEH_info_xts_enc 2985 2986 .rva .LSEH_begin_aesni_xts_decrypt 2987 .rva .LSEH_end_aesni_xts_decrypt 2988 .rva .LSEH_info_xts_dec 2989 ___ 2990 $code.=<<___; 2991 .rva .LSEH_begin_${PREFIX}_cbc_encrypt 2992 .rva .LSEH_end_${PREFIX}_cbc_encrypt 2993 .rva .LSEH_info_cbc 2994 2995 .rva ${PREFIX}_set_decrypt_key 2996 .rva .LSEH_end_set_decrypt_key 2997 .rva .LSEH_info_key 2998 2999 .rva ${PREFIX}_set_encrypt_key 3000 .rva .LSEH_end_set_encrypt_key 3001 .rva .LSEH_info_key 3002 .section .xdata 3003 .align 8 3004 ___ 3005 $code.=<<___ if ($PREFIX eq "aesni"); 3006 .LSEH_info_ecb: 3007 .byte 9,0,0,0 3008 .rva ecb_se_handler 3009 .LSEH_info_ccm64_enc: 3010 .byte 9,0,0,0 3011 .rva ccm64_se_handler 3012 .rva .Lccm64_enc_body,.Lccm64_enc_ret # HandlerData[] 3013 .LSEH_info_ccm64_dec: 3014 .byte 9,0,0,0 3015 .rva ccm64_se_handler 3016 .rva .Lccm64_dec_body,.Lccm64_dec_ret # HandlerData[] 3017 .LSEH_info_ctr32: 3018 .byte 9,0,0,0 3019 .rva ctr32_se_handler 3020 .LSEH_info_xts_enc: 3021 .byte 9,0,0,0 3022 .rva xts_se_handler 3023 .rva .Lxts_enc_body,.Lxts_enc_epilogue # HandlerData[] 3024 .LSEH_info_xts_dec: 3025 .byte 9,0,0,0 3026 .rva xts_se_handler 3027 .rva .Lxts_dec_body,.Lxts_dec_epilogue # HandlerData[] 3028 ___ 3029 $code.=<<___; 3030 .LSEH_info_cbc: 3031 .byte 9,0,0,0 3032 .rva cbc_se_handler 3033 .LSEH_info_key: 3034 .byte 0x01,0x04,0x01,0x00 3035 .byte 0x04,0x02,0x00,0x00 # sub rsp,8 3036 ___ 3037 } 3038 3039 sub rex { 3040 local *opcode=shift; 3041 my ($dst,$src)=@_; 3042 my $rex=0; 3043 3044 $rex|=0x04 if($dst>=8); 3045 $rex|=0x01 if($src>=8); 3046 push @opcode,$rex|0x40 if($rex); 3047 } 3048 3049 sub aesni { 3050 my $line=shift; 3051 my @opcode=(0x66); 3052 3053 if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 3054 rex(\@opcode,$4,$3); 3055 push @opcode,0x0f,0x3a,0xdf; 3056 push @opcode,0xc0|($3&7)|(($4&7)<<3); # ModR/M 3057 my $c=$2; 3058 push @opcode,$c=~/^0/?oct($c):$c; 3059 return ".byte\t".join(',',@opcode); 3060 } 3061 elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) { 3062 my %opcodelet = ( 3063 "aesimc" => 0xdb, 3064 "aesenc" => 0xdc, "aesenclast" => 0xdd, 3065 "aesdec" => 0xde, "aesdeclast" => 0xdf 3066 ); 3067 return undef if (!defined($opcodelet{$1})); 3068 rex(\@opcode,$3,$2); 3069 push @opcode,0x0f,0x38,$opcodelet{$1}; 3070 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 3071 return ".byte\t".join(',',@opcode); 3072 } 3073 return $line; 3074 } 3075 3076 $code =~ s/\`([^\`]*)\`/eval($1)/gem; 3077 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem; 3078 3079 print $code; 3080 3081 close STDOUT; 3082