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 $xlate) or 173 die "can't locate x86_64-xlate.pl"; 174 175 open OUT,"| \"$^X\" $xlate $flavour $output"; 176 *STDOUT=*OUT; 177 178 $movkey = $PREFIX eq "aesni" ? "movups" : "movups"; 179 @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") : # Win64 order 180 ("%rdi","%rsi","%rdx","%rcx"); # Unix order 181 182 $code=".text\n"; 183 184 $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!! 185 # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ... 186 $inp="%rdi"; 187 $out="%rsi"; 188 $len="%rdx"; 189 $key="%rcx"; # input to and changed by aesni_[en|de]cryptN !!! 190 $ivp="%r8"; # cbc, ctr, ... 191 192 $rnds_="%r10d"; # backup copy for $rounds 193 $key_="%r11"; # backup copy for $key 194 195 # %xmm register layout 196 $rndkey0="%xmm0"; $rndkey1="%xmm1"; 197 $inout0="%xmm2"; $inout1="%xmm3"; 198 $inout2="%xmm4"; $inout3="%xmm5"; 199 $inout4="%xmm6"; $inout5="%xmm7"; 200 $inout6="%xmm8"; $inout7="%xmm9"; 201 202 $in2="%xmm6"; $in1="%xmm7"; # used in CBC decrypt, CTR, ... 203 $in0="%xmm8"; $iv="%xmm9"; 204 206 # Inline version of internal aesni_[en|de]crypt1. 207 # 208 # Why folded loop? Because aes[enc|dec] is slow enough to accommodate 209 # cycles which take care of loop variables... 210 { my $sn; 211 sub aesni_generate1 { 212 my ($p,$key,$rounds,$inout,$ivec)=@_; $inout=$inout0 if (!defined($inout)); 213 ++$sn; 214 $code.=<<___; 215 $movkey ($key),$rndkey0 216 $movkey 16($key),$rndkey1 217 ___ 218 $code.=<<___ if (defined($ivec)); 219 xorps $rndkey0,$ivec 220 lea 32($key),$key 221 xorps $ivec,$inout 222 ___ 223 $code.=<<___ if (!defined($ivec)); 224 lea 32($key),$key 225 xorps $rndkey0,$inout 226 ___ 227 $code.=<<___; 228 .Loop_${p}1_$sn: 229 aes${p} $rndkey1,$inout 230 dec $rounds 231 $movkey ($key),$rndkey1 232 lea 16($key),$key 233 jnz .Loop_${p}1_$sn # loop body is 16 bytes 234 aes${p}last $rndkey1,$inout 235 ___ 236 }} 237 # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key); 238 # 239 { my ($inp,$out,$key) = @_4args; 240 241 $code.=<<___; 242 .globl ${PREFIX}_encrypt 243 .type ${PREFIX}_encrypt,\@abi-omnipotent 244 .align 16 245 ${PREFIX}_encrypt: 246 movups ($inp),$inout0 # load input 247 mov 240($key),$rounds # key->rounds 248 ___ 249 &aesni_generate1("enc",$key,$rounds); 250 $code.=<<___; 251 movups $inout0,($out) # output 252 ret 253 .size ${PREFIX}_encrypt,.-${PREFIX}_encrypt 254 255 .globl ${PREFIX}_decrypt 256 .type ${PREFIX}_decrypt,\@abi-omnipotent 257 .align 16 258 ${PREFIX}_decrypt: 259 movups ($inp),$inout0 # load input 260 mov 240($key),$rounds # key->rounds 261 ___ 262 &aesni_generate1("dec",$key,$rounds); 263 $code.=<<___; 264 movups $inout0,($out) # output 265 ret 266 .size ${PREFIX}_decrypt, .-${PREFIX}_decrypt 267 ___ 268 } 269 271 # _aesni_[en|de]cryptN are private interfaces, N denotes interleave 272 # factor. Why 3x subroutine were originally used in loops? Even though 273 # aes[enc|dec] latency was originally 6, it could be scheduled only 274 # every *2nd* cycle. Thus 3x interleave was the one providing optimal 275 # utilization, i.e. when subroutine's throughput is virtually same as 276 # of non-interleaved subroutine [for number of input blocks up to 3]. 277 # This is why it makes no sense to implement 2x subroutine. 278 # aes[enc|dec] latency in next processor generation is 8, but the 279 # instructions can be scheduled every cycle. Optimal interleave for 280 # new processor is therefore 8x... 281 sub aesni_generate3 { 282 my $dir=shift; 283 # As already mentioned it takes in $key and $rounds, which are *not* 284 # preserved. $inout[0-2] is cipher/clear text... 285 $code.=<<___; 286 .type _aesni_${dir}rypt3,\@abi-omnipotent 287 .align 16 288 _aesni_${dir}rypt3: 289 $movkey ($key),$rndkey0 290 shr \$1,$rounds 291 $movkey 16($key),$rndkey1 292 lea 32($key),$key 293 xorps $rndkey0,$inout0 294 xorps $rndkey0,$inout1 295 xorps $rndkey0,$inout2 296 $movkey ($key),$rndkey0 297 298 .L${dir}_loop3: 299 aes${dir} $rndkey1,$inout0 300 aes${dir} $rndkey1,$inout1 301 dec $rounds 302 aes${dir} $rndkey1,$inout2 303 $movkey 16($key),$rndkey1 304 aes${dir} $rndkey0,$inout0 305 aes${dir} $rndkey0,$inout1 306 lea 32($key),$key 307 aes${dir} $rndkey0,$inout2 308 $movkey ($key),$rndkey0 309 jnz .L${dir}_loop3 310 311 aes${dir} $rndkey1,$inout0 312 aes${dir} $rndkey1,$inout1 313 aes${dir} $rndkey1,$inout2 314 aes${dir}last $rndkey0,$inout0 315 aes${dir}last $rndkey0,$inout1 316 aes${dir}last $rndkey0,$inout2 317 ret 318 .size _aesni_${dir}rypt3,.-_aesni_${dir}rypt3 319 ___ 320 } 321 # 4x interleave is implemented to improve small block performance, 322 # most notably [and naturally] 4 block by ~30%. One can argue that one 323 # should have implemented 5x as well, but improvement would be <20%, 324 # so it's not worth it... 325 sub aesni_generate4 { 326 my $dir=shift; 327 # As already mentioned it takes in $key and $rounds, which are *not* 328 # preserved. $inout[0-3] is cipher/clear text... 329 $code.=<<___; 330 .type _aesni_${dir}rypt4,\@abi-omnipotent 331 .align 16 332 _aesni_${dir}rypt4: 333 $movkey ($key),$rndkey0 334 shr \$1,$rounds 335 $movkey 16($key),$rndkey1 336 lea 32($key),$key 337 xorps $rndkey0,$inout0 338 xorps $rndkey0,$inout1 339 xorps $rndkey0,$inout2 340 xorps $rndkey0,$inout3 341 $movkey ($key),$rndkey0 342 343 .L${dir}_loop4: 344 aes${dir} $rndkey1,$inout0 345 aes${dir} $rndkey1,$inout1 346 dec $rounds 347 aes${dir} $rndkey1,$inout2 348 aes${dir} $rndkey1,$inout3 349 $movkey 16($key),$rndkey1 350 aes${dir} $rndkey0,$inout0 351 aes${dir} $rndkey0,$inout1 352 lea 32($key),$key 353 aes${dir} $rndkey0,$inout2 354 aes${dir} $rndkey0,$inout3 355 $movkey ($key),$rndkey0 356 jnz .L${dir}_loop4 357 358 aes${dir} $rndkey1,$inout0 359 aes${dir} $rndkey1,$inout1 360 aes${dir} $rndkey1,$inout2 361 aes${dir} $rndkey1,$inout3 362 aes${dir}last $rndkey0,$inout0 363 aes${dir}last $rndkey0,$inout1 364 aes${dir}last $rndkey0,$inout2 365 aes${dir}last $rndkey0,$inout3 366 ret 367 .size _aesni_${dir}rypt4,.-_aesni_${dir}rypt4 368 ___ 369 } 370 sub aesni_generate6 { 371 my $dir=shift; 372 # As already mentioned it takes in $key and $rounds, which are *not* 373 # preserved. $inout[0-5] is cipher/clear text... 374 $code.=<<___; 375 .type _aesni_${dir}rypt6,\@abi-omnipotent 376 .align 16 377 _aesni_${dir}rypt6: 378 $movkey ($key),$rndkey0 379 shr \$1,$rounds 380 $movkey 16($key),$rndkey1 381 lea 32($key),$key 382 xorps $rndkey0,$inout0 383 pxor $rndkey0,$inout1 384 aes${dir} $rndkey1,$inout0 385 pxor $rndkey0,$inout2 386 aes${dir} $rndkey1,$inout1 387 pxor $rndkey0,$inout3 388 aes${dir} $rndkey1,$inout2 389 pxor $rndkey0,$inout4 390 aes${dir} $rndkey1,$inout3 391 pxor $rndkey0,$inout5 392 dec $rounds 393 aes${dir} $rndkey1,$inout4 394 $movkey ($key),$rndkey0 395 aes${dir} $rndkey1,$inout5 396 jmp .L${dir}_loop6_enter 397 .align 16 398 .L${dir}_loop6: 399 aes${dir} $rndkey1,$inout0 400 aes${dir} $rndkey1,$inout1 401 dec $rounds 402 aes${dir} $rndkey1,$inout2 403 aes${dir} $rndkey1,$inout3 404 aes${dir} $rndkey1,$inout4 405 aes${dir} $rndkey1,$inout5 406 .L${dir}_loop6_enter: # happens to be 16-byte aligned 407 $movkey 16($key),$rndkey1 408 aes${dir} $rndkey0,$inout0 409 aes${dir} $rndkey0,$inout1 410 lea 32($key),$key 411 aes${dir} $rndkey0,$inout2 412 aes${dir} $rndkey0,$inout3 413 aes${dir} $rndkey0,$inout4 414 aes${dir} $rndkey0,$inout5 415 $movkey ($key),$rndkey0 416 jnz .L${dir}_loop6 417 418 aes${dir} $rndkey1,$inout0 419 aes${dir} $rndkey1,$inout1 420 aes${dir} $rndkey1,$inout2 421 aes${dir} $rndkey1,$inout3 422 aes${dir} $rndkey1,$inout4 423 aes${dir} $rndkey1,$inout5 424 aes${dir}last $rndkey0,$inout0 425 aes${dir}last $rndkey0,$inout1 426 aes${dir}last $rndkey0,$inout2 427 aes${dir}last $rndkey0,$inout3 428 aes${dir}last $rndkey0,$inout4 429 aes${dir}last $rndkey0,$inout5 430 ret 431 .size _aesni_${dir}rypt6,.-_aesni_${dir}rypt6 432 ___ 433 } 434 sub aesni_generate8 { 435 my $dir=shift; 436 # As already mentioned it takes in $key and $rounds, which are *not* 437 # preserved. $inout[0-7] is cipher/clear text... 438 $code.=<<___; 439 .type _aesni_${dir}rypt8,\@abi-omnipotent 440 .align 16 441 _aesni_${dir}rypt8: 442 $movkey ($key),$rndkey0 443 shr \$1,$rounds 444 $movkey 16($key),$rndkey1 445 lea 32($key),$key 446 xorps $rndkey0,$inout0 447 xorps $rndkey0,$inout1 448 aes${dir} $rndkey1,$inout0 449 pxor $rndkey0,$inout2 450 aes${dir} $rndkey1,$inout1 451 pxor $rndkey0,$inout3 452 aes${dir} $rndkey1,$inout2 453 pxor $rndkey0,$inout4 454 aes${dir} $rndkey1,$inout3 455 pxor $rndkey0,$inout5 456 dec $rounds 457 aes${dir} $rndkey1,$inout4 458 pxor $rndkey0,$inout6 459 aes${dir} $rndkey1,$inout5 460 pxor $rndkey0,$inout7 461 $movkey ($key),$rndkey0 462 aes${dir} $rndkey1,$inout6 463 aes${dir} $rndkey1,$inout7 464 $movkey 16($key),$rndkey1 465 jmp .L${dir}_loop8_enter 466 .align 16 467 .L${dir}_loop8: 468 aes${dir} $rndkey1,$inout0 469 aes${dir} $rndkey1,$inout1 470 dec $rounds 471 aes${dir} $rndkey1,$inout2 472 aes${dir} $rndkey1,$inout3 473 aes${dir} $rndkey1,$inout4 474 aes${dir} $rndkey1,$inout5 475 aes${dir} $rndkey1,$inout6 476 aes${dir} $rndkey1,$inout7 477 $movkey 16($key),$rndkey1 478 .L${dir}_loop8_enter: # happens to be 16-byte aligned 479 aes${dir} $rndkey0,$inout0 480 aes${dir} $rndkey0,$inout1 481 lea 32($key),$key 482 aes${dir} $rndkey0,$inout2 483 aes${dir} $rndkey0,$inout3 484 aes${dir} $rndkey0,$inout4 485 aes${dir} $rndkey0,$inout5 486 aes${dir} $rndkey0,$inout6 487 aes${dir} $rndkey0,$inout7 488 $movkey ($key),$rndkey0 489 jnz .L${dir}_loop8 490 491 aes${dir} $rndkey1,$inout0 492 aes${dir} $rndkey1,$inout1 493 aes${dir} $rndkey1,$inout2 494 aes${dir} $rndkey1,$inout3 495 aes${dir} $rndkey1,$inout4 496 aes${dir} $rndkey1,$inout5 497 aes${dir} $rndkey1,$inout6 498 aes${dir} $rndkey1,$inout7 499 aes${dir}last $rndkey0,$inout0 500 aes${dir}last $rndkey0,$inout1 501 aes${dir}last $rndkey0,$inout2 502 aes${dir}last $rndkey0,$inout3 503 aes${dir}last $rndkey0,$inout4 504 aes${dir}last $rndkey0,$inout5 505 aes${dir}last $rndkey0,$inout6 506 aes${dir}last $rndkey0,$inout7 507 ret 508 .size _aesni_${dir}rypt8,.-_aesni_${dir}rypt8 509 ___ 510 } 511 &aesni_generate3("enc") if ($PREFIX eq "aesni"); 512 &aesni_generate3("dec"); 513 &aesni_generate4("enc") if ($PREFIX eq "aesni"); 514 &aesni_generate4("dec"); 515 &aesni_generate6("enc") if ($PREFIX eq "aesni"); 516 &aesni_generate6("dec"); 517 &aesni_generate8("enc") if ($PREFIX eq "aesni"); 518 &aesni_generate8("dec"); 519 521 if ($PREFIX eq "aesni") { 522 ######################################################################## 523 # void aesni_ecb_encrypt (const void *in, void *out, 524 # size_t length, const AES_KEY *key, 525 # int enc); 526 $code.=<<___; 527 .globl aesni_ecb_encrypt 528 .type aesni_ecb_encrypt,\@function,5 529 .align 16 530 aesni_ecb_encrypt: 531 and \$-16,$len 532 jz .Lecb_ret 533 534 mov 240($key),$rounds # key->rounds 535 $movkey ($key),$rndkey0 536 mov $key,$key_ # backup $key 537 mov $rounds,$rnds_ # backup $rounds 538 test %r8d,%r8d # 5th argument 539 jz .Lecb_decrypt 540 #--------------------------- ECB ENCRYPT ------------------------------# 541 cmp \$0x80,$len 542 jb .Lecb_enc_tail 543 544 movdqu ($inp),$inout0 545 movdqu 0x10($inp),$inout1 546 movdqu 0x20($inp),$inout2 547 movdqu 0x30($inp),$inout3 548 movdqu 0x40($inp),$inout4 549 movdqu 0x50($inp),$inout5 550 movdqu 0x60($inp),$inout6 551 movdqu 0x70($inp),$inout7 552 lea 0x80($inp),$inp 553 sub \$0x80,$len 554 jmp .Lecb_enc_loop8_enter 555 .align 16 556 .Lecb_enc_loop8: 557 movups $inout0,($out) 558 mov $key_,$key # restore $key 559 movdqu ($inp),$inout0 560 mov $rnds_,$rounds # restore $rounds 561 movups $inout1,0x10($out) 562 movdqu 0x10($inp),$inout1 563 movups $inout2,0x20($out) 564 movdqu 0x20($inp),$inout2 565 movups $inout3,0x30($out) 566 movdqu 0x30($inp),$inout3 567 movups $inout4,0x40($out) 568 movdqu 0x40($inp),$inout4 569 movups $inout5,0x50($out) 570 movdqu 0x50($inp),$inout5 571 movups $inout6,0x60($out) 572 movdqu 0x60($inp),$inout6 573 movups $inout7,0x70($out) 574 lea 0x80($out),$out 575 movdqu 0x70($inp),$inout7 576 lea 0x80($inp),$inp 577 .Lecb_enc_loop8_enter: 578 579 call _aesni_encrypt8 580 581 sub \$0x80,$len 582 jnc .Lecb_enc_loop8 583 584 movups $inout0,($out) 585 mov $key_,$key # restore $key 586 movups $inout1,0x10($out) 587 mov $rnds_,$rounds # restore $rounds 588 movups $inout2,0x20($out) 589 movups $inout3,0x30($out) 590 movups $inout4,0x40($out) 591 movups $inout5,0x50($out) 592 movups $inout6,0x60($out) 593 movups $inout7,0x70($out) 594 lea 0x80($out),$out 595 add \$0x80,$len 596 jz .Lecb_ret 597 598 .Lecb_enc_tail: 599 movups ($inp),$inout0 600 cmp \$0x20,$len 601 jb .Lecb_enc_one 602 movups 0x10($inp),$inout1 603 je .Lecb_enc_two 604 movups 0x20($inp),$inout2 605 cmp \$0x40,$len 606 jb .Lecb_enc_three 607 movups 0x30($inp),$inout3 608 je .Lecb_enc_four 609 movups 0x40($inp),$inout4 610 cmp \$0x60,$len 611 jb .Lecb_enc_five 612 movups 0x50($inp),$inout5 613 je .Lecb_enc_six 614 movdqu 0x60($inp),$inout6 615 call _aesni_encrypt8 616 movups $inout0,($out) 617 movups $inout1,0x10($out) 618 movups $inout2,0x20($out) 619 movups $inout3,0x30($out) 620 movups $inout4,0x40($out) 621 movups $inout5,0x50($out) 622 movups $inout6,0x60($out) 623 jmp .Lecb_ret 624 .align 16 625 .Lecb_enc_one: 626 ___ 627 &aesni_generate1("enc",$key,$rounds); 628 $code.=<<___; 629 movups $inout0,($out) 630 jmp .Lecb_ret 631 .align 16 632 .Lecb_enc_two: 633 xorps $inout2,$inout2 634 call _aesni_encrypt3 635 movups $inout0,($out) 636 movups $inout1,0x10($out) 637 jmp .Lecb_ret 638 .align 16 639 .Lecb_enc_three: 640 call _aesni_encrypt3 641 movups $inout0,($out) 642 movups $inout1,0x10($out) 643 movups $inout2,0x20($out) 644 jmp .Lecb_ret 645 .align 16 646 .Lecb_enc_four: 647 call _aesni_encrypt4 648 movups $inout0,($out) 649 movups $inout1,0x10($out) 650 movups $inout2,0x20($out) 651 movups $inout3,0x30($out) 652 jmp .Lecb_ret 653 .align 16 654 .Lecb_enc_five: 655 xorps $inout5,$inout5 656 call _aesni_encrypt6 657 movups $inout0,($out) 658 movups $inout1,0x10($out) 659 movups $inout2,0x20($out) 660 movups $inout3,0x30($out) 661 movups $inout4,0x40($out) 662 jmp .Lecb_ret 663 .align 16 664 .Lecb_enc_six: 665 call _aesni_encrypt6 666 movups $inout0,($out) 667 movups $inout1,0x10($out) 668 movups $inout2,0x20($out) 669 movups $inout3,0x30($out) 670 movups $inout4,0x40($out) 671 movups $inout5,0x50($out) 672 jmp .Lecb_ret 673 #--------------------------- ECB DECRYPT ------------------------------# 675 .align 16 676 .Lecb_decrypt: 677 cmp \$0x80,$len 678 jb .Lecb_dec_tail 679 680 movdqu ($inp),$inout0 681 movdqu 0x10($inp),$inout1 682 movdqu 0x20($inp),$inout2 683 movdqu 0x30($inp),$inout3 684 movdqu 0x40($inp),$inout4 685 movdqu 0x50($inp),$inout5 686 movdqu 0x60($inp),$inout6 687 movdqu 0x70($inp),$inout7 688 lea 0x80($inp),$inp 689 sub \$0x80,$len 690 jmp .Lecb_dec_loop8_enter 691 .align 16 692 .Lecb_dec_loop8: 693 movups $inout0,($out) 694 mov $key_,$key # restore $key 695 movdqu ($inp),$inout0 696 mov $rnds_,$rounds # restore $rounds 697 movups $inout1,0x10($out) 698 movdqu 0x10($inp),$inout1 699 movups $inout2,0x20($out) 700 movdqu 0x20($inp),$inout2 701 movups $inout3,0x30($out) 702 movdqu 0x30($inp),$inout3 703 movups $inout4,0x40($out) 704 movdqu 0x40($inp),$inout4 705 movups $inout5,0x50($out) 706 movdqu 0x50($inp),$inout5 707 movups $inout6,0x60($out) 708 movdqu 0x60($inp),$inout6 709 movups $inout7,0x70($out) 710 lea 0x80($out),$out 711 movdqu 0x70($inp),$inout7 712 lea 0x80($inp),$inp 713 .Lecb_dec_loop8_enter: 714 715 call _aesni_decrypt8 716 717 $movkey ($key_),$rndkey0 718 sub \$0x80,$len 719 jnc .Lecb_dec_loop8 720 721 movups $inout0,($out) 722 mov $key_,$key # restore $key 723 movups $inout1,0x10($out) 724 mov $rnds_,$rounds # restore $rounds 725 movups $inout2,0x20($out) 726 movups $inout3,0x30($out) 727 movups $inout4,0x40($out) 728 movups $inout5,0x50($out) 729 movups $inout6,0x60($out) 730 movups $inout7,0x70($out) 731 lea 0x80($out),$out 732 add \$0x80,$len 733 jz .Lecb_ret 734 735 .Lecb_dec_tail: 736 movups ($inp),$inout0 737 cmp \$0x20,$len 738 jb .Lecb_dec_one 739 movups 0x10($inp),$inout1 740 je .Lecb_dec_two 741 movups 0x20($inp),$inout2 742 cmp \$0x40,$len 743 jb .Lecb_dec_three 744 movups 0x30($inp),$inout3 745 je .Lecb_dec_four 746 movups 0x40($inp),$inout4 747 cmp \$0x60,$len 748 jb .Lecb_dec_five 749 movups 0x50($inp),$inout5 750 je .Lecb_dec_six 751 movups 0x60($inp),$inout6 752 $movkey ($key),$rndkey0 753 call _aesni_decrypt8 754 movups $inout0,($out) 755 movups $inout1,0x10($out) 756 movups $inout2,0x20($out) 757 movups $inout3,0x30($out) 758 movups $inout4,0x40($out) 759 movups $inout5,0x50($out) 760 movups $inout6,0x60($out) 761 jmp .Lecb_ret 762 .align 16 763 .Lecb_dec_one: 764 ___ 765 &aesni_generate1("dec",$key,$rounds); 766 $code.=<<___; 767 movups $inout0,($out) 768 jmp .Lecb_ret 769 .align 16 770 .Lecb_dec_two: 771 xorps $inout2,$inout2 772 call _aesni_decrypt3 773 movups $inout0,($out) 774 movups $inout1,0x10($out) 775 jmp .Lecb_ret 776 .align 16 777 .Lecb_dec_three: 778 call _aesni_decrypt3 779 movups $inout0,($out) 780 movups $inout1,0x10($out) 781 movups $inout2,0x20($out) 782 jmp .Lecb_ret 783 .align 16 784 .Lecb_dec_four: 785 call _aesni_decrypt4 786 movups $inout0,($out) 787 movups $inout1,0x10($out) 788 movups $inout2,0x20($out) 789 movups $inout3,0x30($out) 790 jmp .Lecb_ret 791 .align 16 792 .Lecb_dec_five: 793 xorps $inout5,$inout5 794 call _aesni_decrypt6 795 movups $inout0,($out) 796 movups $inout1,0x10($out) 797 movups $inout2,0x20($out) 798 movups $inout3,0x30($out) 799 movups $inout4,0x40($out) 800 jmp .Lecb_ret 801 .align 16 802 .Lecb_dec_six: 803 call _aesni_decrypt6 804 movups $inout0,($out) 805 movups $inout1,0x10($out) 806 movups $inout2,0x20($out) 807 movups $inout3,0x30($out) 808 movups $inout4,0x40($out) 809 movups $inout5,0x50($out) 810 811 .Lecb_ret: 812 ret 813 .size aesni_ecb_encrypt,.-aesni_ecb_encrypt 814 ___ 815 817 { 818 ###################################################################### 819 # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out, 820 # size_t blocks, const AES_KEY *key, 821 # const char *ivec,char *cmac); 822 # 823 # Handles only complete blocks, operates on 64-bit counter and 824 # does not update *ivec! Nor does it finalize CMAC value 825 # (see engine/eng_aesni.c for details) 826 # 827 { 828 my $cmac="%r9"; # 6th argument 829 830 my $increment="%xmm6"; 831 my $bswap_mask="%xmm7"; 832 833 $code.=<<___; 834 .globl aesni_ccm64_encrypt_blocks 835 .type aesni_ccm64_encrypt_blocks,\@function,6 836 .align 16 837 aesni_ccm64_encrypt_blocks: 838 ___ 839 $code.=<<___ if ($win64); 840 lea -0x58(%rsp),%rsp 841 movaps %xmm6,(%rsp) 842 movaps %xmm7,0x10(%rsp) 843 movaps %xmm8,0x20(%rsp) 844 movaps %xmm9,0x30(%rsp) 845 .Lccm64_enc_body: 846 ___ 847 $code.=<<___; 848 mov 240($key),$rounds # key->rounds 849 movdqu ($ivp),$iv 850 movdqa .Lincrement64(%rip),$increment 851 movdqa .Lbswap_mask(%rip),$bswap_mask 852 853 shr \$1,$rounds 854 lea 0($key),$key_ 855 movdqu ($cmac),$inout1 856 movdqa $iv,$inout0 857 mov $rounds,$rnds_ 858 pshufb $bswap_mask,$iv 859 jmp .Lccm64_enc_outer 860 .align 16 861 .Lccm64_enc_outer: 862 $movkey ($key_),$rndkey0 863 mov $rnds_,$rounds 864 movups ($inp),$in0 # load inp 865 866 xorps $rndkey0,$inout0 # counter 867 $movkey 16($key_),$rndkey1 868 xorps $in0,$rndkey0 869 lea 32($key_),$key 870 xorps $rndkey0,$inout1 # cmac^=inp 871 $movkey ($key),$rndkey0 872 873 .Lccm64_enc2_loop: 874 aesenc $rndkey1,$inout0 875 dec $rounds 876 aesenc $rndkey1,$inout1 877 $movkey 16($key),$rndkey1 878 aesenc $rndkey0,$inout0 879 lea 32($key),$key 880 aesenc $rndkey0,$inout1 881 $movkey 0($key),$rndkey0 882 jnz .Lccm64_enc2_loop 883 aesenc $rndkey1,$inout0 884 aesenc $rndkey1,$inout1 885 paddq $increment,$iv 886 aesenclast $rndkey0,$inout0 887 aesenclast $rndkey0,$inout1 888 889 dec $len 890 lea 16($inp),$inp 891 xorps $inout0,$in0 # inp ^= E(iv) 892 movdqa $iv,$inout0 893 movups $in0,($out) # save output 894 lea 16($out),$out 895 pshufb $bswap_mask,$inout0 896 jnz .Lccm64_enc_outer 897 898 movups $inout1,($cmac) 899 ___ 900 $code.=<<___ if ($win64); 901 movaps (%rsp),%xmm6 902 movaps 0x10(%rsp),%xmm7 903 movaps 0x20(%rsp),%xmm8 904 movaps 0x30(%rsp),%xmm9 905 lea 0x58(%rsp),%rsp 906 .Lccm64_enc_ret: 907 ___ 908 $code.=<<___; 909 ret 910 .size aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks 911 ___ 912 ###################################################################### 913 $code.=<<___; 914 .globl aesni_ccm64_decrypt_blocks 915 .type aesni_ccm64_decrypt_blocks,\@function,6 916 .align 16 917 aesni_ccm64_decrypt_blocks: 918 ___ 919 $code.=<<___ if ($win64); 920 lea -0x58(%rsp),%rsp 921 movaps %xmm6,(%rsp) 922 movaps %xmm7,0x10(%rsp) 923 movaps %xmm8,0x20(%rsp) 924 movaps %xmm9,0x30(%rsp) 925 .Lccm64_dec_body: 926 ___ 927 $code.=<<___; 928 mov 240($key),$rounds # key->rounds 929 movups ($ivp),$iv 930 movdqu ($cmac),$inout1 931 movdqa .Lincrement64(%rip),$increment 932 movdqa .Lbswap_mask(%rip),$bswap_mask 933 934 movaps $iv,$inout0 935 mov $rounds,$rnds_ 936 mov $key,$key_ 937 pshufb $bswap_mask,$iv 938 ___ 939 &aesni_generate1("enc",$key,$rounds); 940 $code.=<<___; 941 movups ($inp),$in0 # load inp 942 paddq $increment,$iv 943 lea 16($inp),$inp 944 jmp .Lccm64_dec_outer 945 .align 16 946 .Lccm64_dec_outer: 947 xorps $inout0,$in0 # inp ^= E(iv) 948 movdqa $iv,$inout0 949 mov $rnds_,$rounds 950 movups $in0,($out) # save output 951 lea 16($out),$out 952 pshufb $bswap_mask,$inout0 953 954 sub \$1,$len 955 jz .Lccm64_dec_break 956 957 $movkey ($key_),$rndkey0 958 shr \$1,$rounds 959 $movkey 16($key_),$rndkey1 960 xorps $rndkey0,$in0 961 lea 32($key_),$key 962 xorps $rndkey0,$inout0 963 xorps $in0,$inout1 # cmac^=out 964 $movkey ($key),$rndkey0 965 966 .Lccm64_dec2_loop: 967 aesenc $rndkey1,$inout0 968 dec $rounds 969 aesenc $rndkey1,$inout1 970 $movkey 16($key),$rndkey1 971 aesenc $rndkey0,$inout0 972 lea 32($key),$key 973 aesenc $rndkey0,$inout1 974 $movkey 0($key),$rndkey0 975 jnz .Lccm64_dec2_loop 976 movups ($inp),$in0 # load inp 977 paddq $increment,$iv 978 aesenc $rndkey1,$inout0 979 aesenc $rndkey1,$inout1 980 lea 16($inp),$inp 981 aesenclast $rndkey0,$inout0 982 aesenclast $rndkey0,$inout1 983 jmp .Lccm64_dec_outer 984 985 .align 16 986 .Lccm64_dec_break: 987 #xorps $in0,$inout1 # cmac^=out 988 ___ 989 &aesni_generate1("enc",$key_,$rounds,$inout1,$in0); 990 $code.=<<___; 991 movups $inout1,($cmac) 992 ___ 993 $code.=<<___ if ($win64); 994 movaps (%rsp),%xmm6 995 movaps 0x10(%rsp),%xmm7 996 movaps 0x20(%rsp),%xmm8 997 movaps 0x30(%rsp),%xmm9 998 lea 0x58(%rsp),%rsp 999 .Lccm64_dec_ret: 1000 ___ 1001 $code.=<<___; 1002 ret 1003 .size aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks 1004 ___ 1005 } 1007 ###################################################################### 1008 # void aesni_ctr32_encrypt_blocks (const void *in, void *out, 1009 # size_t blocks, const AES_KEY *key, 1010 # const char *ivec); 1011 # 1012 # Handles only complete blocks, operates on 32-bit counter and 1013 # does not update *ivec! (see engine/eng_aesni.c for details) 1014 # 1015 { 1016 my $reserved = $win64?0:-0x28; 1017 my ($in0,$in1,$in2,$in3)=map("%xmm$_",(8..11)); 1018 my ($iv0,$iv1,$ivec)=("%xmm12","%xmm13","%xmm14"); 1019 my $bswap_mask="%xmm15"; 1020 1021 $code.=<<___; 1022 .globl aesni_ctr32_encrypt_blocks 1023 .type aesni_ctr32_encrypt_blocks,\@function,5 1024 .align 16 1025 aesni_ctr32_encrypt_blocks: 1026 ___ 1027 $code.=<<___ if ($win64); 1028 lea -0xc8(%rsp),%rsp 1029 movaps %xmm6,0x20(%rsp) 1030 movaps %xmm7,0x30(%rsp) 1031 movaps %xmm8,0x40(%rsp) 1032 movaps %xmm9,0x50(%rsp) 1033 movaps %xmm10,0x60(%rsp) 1034 movaps %xmm11,0x70(%rsp) 1035 movaps %xmm12,0x80(%rsp) 1036 movaps %xmm13,0x90(%rsp) 1037 movaps %xmm14,0xa0(%rsp) 1038 movaps %xmm15,0xb0(%rsp) 1039 .Lctr32_body: 1040 ___ 1041 $code.=<<___; 1042 cmp \$1,$len 1043 je .Lctr32_one_shortcut 1044 1045 movdqu ($ivp),$ivec 1046 movdqa .Lbswap_mask(%rip),$bswap_mask 1047 xor $rounds,$rounds 1048 pextrd \$3,$ivec,$rnds_ # pull 32-bit counter 1049 pinsrd \$3,$rounds,$ivec # wipe 32-bit counter 1050 1051 mov 240($key),$rounds # key->rounds 1052 bswap $rnds_ 1053 pxor $iv0,$iv0 # vector of 3 32-bit counters 1054 pxor $iv1,$iv1 # vector of 3 32-bit counters 1055 pinsrd \$0,$rnds_,$iv0 1056 lea 3($rnds_),$key_ 1057 pinsrd \$0,$key_,$iv1 1058 inc $rnds_ 1059 pinsrd \$1,$rnds_,$iv0 1060 inc $key_ 1061 pinsrd \$1,$key_,$iv1 1062 inc $rnds_ 1063 pinsrd \$2,$rnds_,$iv0 1064 inc $key_ 1065 pinsrd \$2,$key_,$iv1 1066 movdqa $iv0,$reserved(%rsp) 1067 pshufb $bswap_mask,$iv0 1068 movdqa $iv1,`$reserved+0x10`(%rsp) 1069 pshufb $bswap_mask,$iv1 1070 1071 pshufd \$`3<<6`,$iv0,$inout0 # place counter to upper dword 1072 pshufd \$`2<<6`,$iv0,$inout1 1073 pshufd \$`1<<6`,$iv0,$inout2 1074 cmp \$6,$len 1075 jb .Lctr32_tail 1076 shr \$1,$rounds 1077 mov $key,$key_ # backup $key 1078 mov $rounds,$rnds_ # backup $rounds 1079 sub \$6,$len 1080 jmp .Lctr32_loop6 1081 1082 .align 16 1083 .Lctr32_loop6: 1084 pshufd \$`3<<6`,$iv1,$inout3 1085 por $ivec,$inout0 # merge counter-less ivec 1086 $movkey ($key_),$rndkey0 1087 pshufd \$`2<<6`,$iv1,$inout4 1088 por $ivec,$inout1 1089 $movkey 16($key_),$rndkey1 1090 pshufd \$`1<<6`,$iv1,$inout5 1091 por $ivec,$inout2 1092 por $ivec,$inout3 1093 xorps $rndkey0,$inout0 1094 por $ivec,$inout4 1095 por $ivec,$inout5 1096 1097 # inline _aesni_encrypt6 and interleave last rounds 1098 # with own code... 1099 1100 pxor $rndkey0,$inout1 1101 aesenc $rndkey1,$inout0 1102 lea 32($key_),$key 1103 pxor $rndkey0,$inout2 1104 aesenc $rndkey1,$inout1 1105 movdqa .Lincrement32(%rip),$iv1 1106 pxor $rndkey0,$inout3 1107 aesenc $rndkey1,$inout2 1108 movdqa $reserved(%rsp),$iv0 1109 pxor $rndkey0,$inout4 1110 aesenc $rndkey1,$inout3 1111 pxor $rndkey0,$inout5 1112 $movkey ($key),$rndkey0 1113 dec $rounds 1114 aesenc $rndkey1,$inout4 1115 aesenc $rndkey1,$inout5 1116 jmp .Lctr32_enc_loop6_enter 1117 .align 16 1118 .Lctr32_enc_loop6: 1119 aesenc $rndkey1,$inout0 1120 aesenc $rndkey1,$inout1 1121 dec $rounds 1122 aesenc $rndkey1,$inout2 1123 aesenc $rndkey1,$inout3 1124 aesenc $rndkey1,$inout4 1125 aesenc $rndkey1,$inout5 1126 .Lctr32_enc_loop6_enter: 1127 $movkey 16($key),$rndkey1 1128 aesenc $rndkey0,$inout0 1129 aesenc $rndkey0,$inout1 1130 lea 32($key),$key 1131 aesenc $rndkey0,$inout2 1132 aesenc $rndkey0,$inout3 1133 aesenc $rndkey0,$inout4 1134 aesenc $rndkey0,$inout5 1135 $movkey ($key),$rndkey0 1136 jnz .Lctr32_enc_loop6 1137 1138 aesenc $rndkey1,$inout0 1139 paddd $iv1,$iv0 # increment counter vector 1140 aesenc $rndkey1,$inout1 1141 paddd `$reserved+0x10`(%rsp),$iv1 1142 aesenc $rndkey1,$inout2 1143 movdqa $iv0,$reserved(%rsp) # save counter vector 1144 aesenc $rndkey1,$inout3 1145 movdqa $iv1,`$reserved+0x10`(%rsp) 1146 aesenc $rndkey1,$inout4 1147 pshufb $bswap_mask,$iv0 # byte swap 1148 aesenc $rndkey1,$inout5 1149 pshufb $bswap_mask,$iv1 1150 1151 aesenclast $rndkey0,$inout0 1152 movups ($inp),$in0 # load input 1153 aesenclast $rndkey0,$inout1 1154 movups 0x10($inp),$in1 1155 aesenclast $rndkey0,$inout2 1156 movups 0x20($inp),$in2 1157 aesenclast $rndkey0,$inout3 1158 movups 0x30($inp),$in3 1159 aesenclast $rndkey0,$inout4 1160 movups 0x40($inp),$rndkey1 1161 aesenclast $rndkey0,$inout5 1162 movups 0x50($inp),$rndkey0 1163 lea 0x60($inp),$inp 1164 1165 xorps $inout0,$in0 # xor 1166 pshufd \$`3<<6`,$iv0,$inout0 1167 xorps $inout1,$in1 1168 pshufd \$`2<<6`,$iv0,$inout1 1169 movups $in0,($out) # store output 1170 xorps $inout2,$in2 1171 pshufd \$`1<<6`,$iv0,$inout2 1172 movups $in1,0x10($out) 1173 xorps $inout3,$in3 1174 movups $in2,0x20($out) 1175 xorps $inout4,$rndkey1 1176 movups $in3,0x30($out) 1177 xorps $inout5,$rndkey0 1178 movups $rndkey1,0x40($out) 1179 movups $rndkey0,0x50($out) 1180 lea 0x60($out),$out 1181 mov $rnds_,$rounds 1182 sub \$6,$len 1183 jnc .Lctr32_loop6 1184 1185 add \$6,$len 1186 jz .Lctr32_done 1187 mov $key_,$key # restore $key 1188 lea 1($rounds,$rounds),$rounds # restore original value 1189 1190 .Lctr32_tail: 1191 por $ivec,$inout0 1192 movups ($inp),$in0 1193 cmp \$2,$len 1194 jb .Lctr32_one 1195 1196 por $ivec,$inout1 1197 movups 0x10($inp),$in1 1198 je .Lctr32_two 1199 1200 pshufd \$`3<<6`,$iv1,$inout3 1201 por $ivec,$inout2 1202 movups 0x20($inp),$in2 1203 cmp \$4,$len 1204 jb .Lctr32_three 1205 1206 pshufd \$`2<<6`,$iv1,$inout4 1207 por $ivec,$inout3 1208 movups 0x30($inp),$in3 1209 je .Lctr32_four 1210 1211 por $ivec,$inout4 1212 xorps $inout5,$inout5 1213 1214 call _aesni_encrypt6 1215 1216 movups 0x40($inp),$rndkey1 1217 xorps $inout0,$in0 1218 xorps $inout1,$in1 1219 movups $in0,($out) 1220 xorps $inout2,$in2 1221 movups $in1,0x10($out) 1222 xorps $inout3,$in3 1223 movups $in2,0x20($out) 1224 xorps $inout4,$rndkey1 1225 movups $in3,0x30($out) 1226 movups $rndkey1,0x40($out) 1227 jmp .Lctr32_done 1228 1229 .align 16 1230 .Lctr32_one_shortcut: 1231 movups ($ivp),$inout0 1232 movups ($inp),$in0 1233 mov 240($key),$rounds # key->rounds 1234 .Lctr32_one: 1235 ___ 1236 &aesni_generate1("enc",$key,$rounds); 1237 $code.=<<___; 1238 xorps $inout0,$in0 1239 movups $in0,($out) 1240 jmp .Lctr32_done 1241 1242 .align 16 1243 .Lctr32_two: 1244 xorps $inout2,$inout2 1245 call _aesni_encrypt3 1246 xorps $inout0,$in0 1247 xorps $inout1,$in1 1248 movups $in0,($out) 1249 movups $in1,0x10($out) 1250 jmp .Lctr32_done 1251 1252 .align 16 1253 .Lctr32_three: 1254 call _aesni_encrypt3 1255 xorps $inout0,$in0 1256 xorps $inout1,$in1 1257 movups $in0,($out) 1258 xorps $inout2,$in2 1259 movups $in1,0x10($out) 1260 movups $in2,0x20($out) 1261 jmp .Lctr32_done 1262 1263 .align 16 1264 .Lctr32_four: 1265 call _aesni_encrypt4 1266 xorps $inout0,$in0 1267 xorps $inout1,$in1 1268 movups $in0,($out) 1269 xorps $inout2,$in2 1270 movups $in1,0x10($out) 1271 xorps $inout3,$in3 1272 movups $in2,0x20($out) 1273 movups $in3,0x30($out) 1274 1275 .Lctr32_done: 1276 ___ 1277 $code.=<<___ if ($win64); 1278 movaps 0x20(%rsp),%xmm6 1279 movaps 0x30(%rsp),%xmm7 1280 movaps 0x40(%rsp),%xmm8 1281 movaps 0x50(%rsp),%xmm9 1282 movaps 0x60(%rsp),%xmm10 1283 movaps 0x70(%rsp),%xmm11 1284 movaps 0x80(%rsp),%xmm12 1285 movaps 0x90(%rsp),%xmm13 1286 movaps 0xa0(%rsp),%xmm14 1287 movaps 0xb0(%rsp),%xmm15 1288 lea 0xc8(%rsp),%rsp 1289 .Lctr32_ret: 1290 ___ 1291 $code.=<<___; 1292 ret 1293 .size aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks 1294 ___ 1295 } 1296 1298 ###################################################################### 1299 # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len, 1300 # const AES_KEY *key1, const AES_KEY *key2 1301 # const unsigned char iv[16]); 1302 # 1303 { 1304 my @tweak=map("%xmm$_",(10..15)); 1305 my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]); 1306 my ($key2,$ivp,$len_)=("%r8","%r9","%r9"); 1307 my $frame_size = 0x68 + ($win64?160:0); 1308 1309 $code.=<<___; 1310 .globl aesni_xts_encrypt 1311 .type aesni_xts_encrypt,\@function,6 1312 .align 16 1313 aesni_xts_encrypt: 1314 lea -$frame_size(%rsp),%rsp 1315 ___ 1316 $code.=<<___ if ($win64); 1317 movaps %xmm6,0x60(%rsp) 1318 movaps %xmm7,0x70(%rsp) 1319 movaps %xmm8,0x80(%rsp) 1320 movaps %xmm9,0x90(%rsp) 1321 movaps %xmm10,0xa0(%rsp) 1322 movaps %xmm11,0xb0(%rsp) 1323 movaps %xmm12,0xc0(%rsp) 1324 movaps %xmm13,0xd0(%rsp) 1325 movaps %xmm14,0xe0(%rsp) 1326 movaps %xmm15,0xf0(%rsp) 1327 .Lxts_enc_body: 1328 ___ 1329 $code.=<<___; 1330 movups ($ivp),@tweak[5] # load clear-text tweak 1331 mov 240(%r8),$rounds # key2->rounds 1332 mov 240($key),$rnds_ # key1->rounds 1333 ___ 1334 # generate the tweak 1335 &aesni_generate1("enc",$key2,$rounds,@tweak[5]); 1336 $code.=<<___; 1337 mov $key,$key_ # backup $key 1338 mov $rnds_,$rounds # backup $rounds 1339 mov $len,$len_ # backup $len 1340 and \$-16,$len 1341 1342 movdqa .Lxts_magic(%rip),$twmask 1343 pxor $twtmp,$twtmp 1344 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1345 ___ 1346 for ($i=0;$i<4;$i++) { 1347 $code.=<<___; 1348 pshufd \$0x13,$twtmp,$twres 1349 pxor $twtmp,$twtmp 1350 movdqa @tweak[5],@tweak[$i] 1351 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1352 pand $twmask,$twres # isolate carry and residue 1353 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1354 pxor $twres,@tweak[5] 1355 ___ 1356 } 1357 $code.=<<___; 1358 sub \$16*6,$len 1359 jc .Lxts_enc_short 1360 1361 shr \$1,$rounds 1362 sub \$1,$rounds 1363 mov $rounds,$rnds_ 1364 jmp .Lxts_enc_grandloop 1365 1366 .align 16 1367 .Lxts_enc_grandloop: 1368 pshufd \$0x13,$twtmp,$twres 1369 movdqa @tweak[5],@tweak[4] 1370 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1371 movdqu `16*0`($inp),$inout0 # load input 1372 pand $twmask,$twres # isolate carry and residue 1373 movdqu `16*1`($inp),$inout1 1374 pxor $twres,@tweak[5] 1375 1376 movdqu `16*2`($inp),$inout2 1377 pxor @tweak[0],$inout0 # input^=tweak 1378 movdqu `16*3`($inp),$inout3 1379 pxor @tweak[1],$inout1 1380 movdqu `16*4`($inp),$inout4 1381 pxor @tweak[2],$inout2 1382 movdqu `16*5`($inp),$inout5 1383 lea `16*6`($inp),$inp 1384 pxor @tweak[3],$inout3 1385 $movkey ($key_),$rndkey0 1386 pxor @tweak[4],$inout4 1387 pxor @tweak[5],$inout5 1388 1389 # inline _aesni_encrypt6 and interleave first and last rounds 1390 # with own code... 1391 $movkey 16($key_),$rndkey1 1392 pxor $rndkey0,$inout0 1393 pxor $rndkey0,$inout1 1394 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks 1395 aesenc $rndkey1,$inout0 1396 lea 32($key_),$key 1397 pxor $rndkey0,$inout2 1398 movdqa @tweak[1],`16*1`(%rsp) 1399 aesenc $rndkey1,$inout1 1400 pxor $rndkey0,$inout3 1401 movdqa @tweak[2],`16*2`(%rsp) 1402 aesenc $rndkey1,$inout2 1403 pxor $rndkey0,$inout4 1404 movdqa @tweak[3],`16*3`(%rsp) 1405 aesenc $rndkey1,$inout3 1406 pxor $rndkey0,$inout5 1407 $movkey ($key),$rndkey0 1408 dec $rounds 1409 movdqa @tweak[4],`16*4`(%rsp) 1410 aesenc $rndkey1,$inout4 1411 movdqa @tweak[5],`16*5`(%rsp) 1412 aesenc $rndkey1,$inout5 1413 pxor $twtmp,$twtmp 1414 pcmpgtd @tweak[5],$twtmp 1415 jmp .Lxts_enc_loop6_enter 1416 1417 .align 16 1418 .Lxts_enc_loop6: 1419 aesenc $rndkey1,$inout0 1420 aesenc $rndkey1,$inout1 1421 dec $rounds 1422 aesenc $rndkey1,$inout2 1423 aesenc $rndkey1,$inout3 1424 aesenc $rndkey1,$inout4 1425 aesenc $rndkey1,$inout5 1426 .Lxts_enc_loop6_enter: 1427 $movkey 16($key),$rndkey1 1428 aesenc $rndkey0,$inout0 1429 aesenc $rndkey0,$inout1 1430 lea 32($key),$key 1431 aesenc $rndkey0,$inout2 1432 aesenc $rndkey0,$inout3 1433 aesenc $rndkey0,$inout4 1434 aesenc $rndkey0,$inout5 1435 $movkey ($key),$rndkey0 1436 jnz .Lxts_enc_loop6 1437 1438 pshufd \$0x13,$twtmp,$twres 1439 pxor $twtmp,$twtmp 1440 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1441 aesenc $rndkey1,$inout0 1442 pand $twmask,$twres # isolate carry and residue 1443 aesenc $rndkey1,$inout1 1444 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1445 aesenc $rndkey1,$inout2 1446 pxor $twres,@tweak[5] 1447 aesenc $rndkey1,$inout3 1448 aesenc $rndkey1,$inout4 1449 aesenc $rndkey1,$inout5 1450 $movkey 16($key),$rndkey1 1451 1452 pshufd \$0x13,$twtmp,$twres 1453 pxor $twtmp,$twtmp 1454 movdqa @tweak[5],@tweak[0] 1455 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1456 aesenc $rndkey0,$inout0 1457 pand $twmask,$twres # isolate carry and residue 1458 aesenc $rndkey0,$inout1 1459 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1460 aesenc $rndkey0,$inout2 1461 pxor $twres,@tweak[5] 1462 aesenc $rndkey0,$inout3 1463 aesenc $rndkey0,$inout4 1464 aesenc $rndkey0,$inout5 1465 $movkey 32($key),$rndkey0 1466 1467 pshufd \$0x13,$twtmp,$twres 1468 pxor $twtmp,$twtmp 1469 movdqa @tweak[5],@tweak[1] 1470 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1471 aesenc $rndkey1,$inout0 1472 pand $twmask,$twres # isolate carry and residue 1473 aesenc $rndkey1,$inout1 1474 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1475 aesenc $rndkey1,$inout2 1476 pxor $twres,@tweak[5] 1477 aesenc $rndkey1,$inout3 1478 aesenc $rndkey1,$inout4 1479 aesenc $rndkey1,$inout5 1480 1481 pshufd \$0x13,$twtmp,$twres 1482 pxor $twtmp,$twtmp 1483 movdqa @tweak[5],@tweak[2] 1484 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1485 aesenclast $rndkey0,$inout0 1486 pand $twmask,$twres # isolate carry and residue 1487 aesenclast $rndkey0,$inout1 1488 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1489 aesenclast $rndkey0,$inout2 1490 pxor $twres,@tweak[5] 1491 aesenclast $rndkey0,$inout3 1492 aesenclast $rndkey0,$inout4 1493 aesenclast $rndkey0,$inout5 1494 1495 pshufd \$0x13,$twtmp,$twres 1496 pxor $twtmp,$twtmp 1497 movdqa @tweak[5],@tweak[3] 1498 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1499 xorps `16*0`(%rsp),$inout0 # output^=tweak 1500 pand $twmask,$twres # isolate carry and residue 1501 xorps `16*1`(%rsp),$inout1 1502 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1503 pxor $twres,@tweak[5] 1504 1505 xorps `16*2`(%rsp),$inout2 1506 movups $inout0,`16*0`($out) # write output 1507 xorps `16*3`(%rsp),$inout3 1508 movups $inout1,`16*1`($out) 1509 xorps `16*4`(%rsp),$inout4 1510 movups $inout2,`16*2`($out) 1511 xorps `16*5`(%rsp),$inout5 1512 movups $inout3,`16*3`($out) 1513 mov $rnds_,$rounds # restore $rounds 1514 movups $inout4,`16*4`($out) 1515 movups $inout5,`16*5`($out) 1516 lea `16*6`($out),$out 1517 sub \$16*6,$len 1518 jnc .Lxts_enc_grandloop 1519 1520 lea 3($rounds,$rounds),$rounds # restore original value 1521 mov $key_,$key # restore $key 1522 mov $rounds,$rnds_ # backup $rounds 1523 1524 .Lxts_enc_short: 1525 add \$16*6,$len 1526 jz .Lxts_enc_done 1527 1528 cmp \$0x20,$len 1529 jb .Lxts_enc_one 1530 je .Lxts_enc_two 1531 1532 cmp \$0x40,$len 1533 jb .Lxts_enc_three 1534 je .Lxts_enc_four 1535 1536 pshufd \$0x13,$twtmp,$twres 1537 movdqa @tweak[5],@tweak[4] 1538 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1539 movdqu ($inp),$inout0 1540 pand $twmask,$twres # isolate carry and residue 1541 movdqu 16*1($inp),$inout1 1542 pxor $twres,@tweak[5] 1543 1544 movdqu 16*2($inp),$inout2 1545 pxor @tweak[0],$inout0 1546 movdqu 16*3($inp),$inout3 1547 pxor @tweak[1],$inout1 1548 movdqu 16*4($inp),$inout4 1549 lea 16*5($inp),$inp 1550 pxor @tweak[2],$inout2 1551 pxor @tweak[3],$inout3 1552 pxor @tweak[4],$inout4 1553 1554 call _aesni_encrypt6 1555 1556 xorps @tweak[0],$inout0 1557 movdqa @tweak[5],@tweak[0] 1558 xorps @tweak[1],$inout1 1559 xorps @tweak[2],$inout2 1560 movdqu $inout0,($out) 1561 xorps @tweak[3],$inout3 1562 movdqu $inout1,16*1($out) 1563 xorps @tweak[4],$inout4 1564 movdqu $inout2,16*2($out) 1565 movdqu $inout3,16*3($out) 1566 movdqu $inout4,16*4($out) 1567 lea 16*5($out),$out 1568 jmp .Lxts_enc_done 1569 1570 .align 16 1571 .Lxts_enc_one: 1572 movups ($inp),$inout0 1573 lea 16*1($inp),$inp 1574 xorps @tweak[0],$inout0 1575 ___ 1576 &aesni_generate1("enc",$key,$rounds); 1577 $code.=<<___; 1578 xorps @tweak[0],$inout0 1579 movdqa @tweak[1],@tweak[0] 1580 movups $inout0,($out) 1581 lea 16*1($out),$out 1582 jmp .Lxts_enc_done 1583 1584 .align 16 1585 .Lxts_enc_two: 1586 movups ($inp),$inout0 1587 movups 16($inp),$inout1 1588 lea 32($inp),$inp 1589 xorps @tweak[0],$inout0 1590 xorps @tweak[1],$inout1 1591 1592 call _aesni_encrypt3 1593 1594 xorps @tweak[0],$inout0 1595 movdqa @tweak[2],@tweak[0] 1596 xorps @tweak[1],$inout1 1597 movups $inout0,($out) 1598 movups $inout1,16*1($out) 1599 lea 16*2($out),$out 1600 jmp .Lxts_enc_done 1601 1602 .align 16 1603 .Lxts_enc_three: 1604 movups ($inp),$inout0 1605 movups 16*1($inp),$inout1 1606 movups 16*2($inp),$inout2 1607 lea 16*3($inp),$inp 1608 xorps @tweak[0],$inout0 1609 xorps @tweak[1],$inout1 1610 xorps @tweak[2],$inout2 1611 1612 call _aesni_encrypt3 1613 1614 xorps @tweak[0],$inout0 1615 movdqa @tweak[3],@tweak[0] 1616 xorps @tweak[1],$inout1 1617 xorps @tweak[2],$inout2 1618 movups $inout0,($out) 1619 movups $inout1,16*1($out) 1620 movups $inout2,16*2($out) 1621 lea 16*3($out),$out 1622 jmp .Lxts_enc_done 1623 1624 .align 16 1625 .Lxts_enc_four: 1626 movups ($inp),$inout0 1627 movups 16*1($inp),$inout1 1628 movups 16*2($inp),$inout2 1629 xorps @tweak[0],$inout0 1630 movups 16*3($inp),$inout3 1631 lea 16*4($inp),$inp 1632 xorps @tweak[1],$inout1 1633 xorps @tweak[2],$inout2 1634 xorps @tweak[3],$inout3 1635 1636 call _aesni_encrypt4 1637 1638 xorps @tweak[0],$inout0 1639 movdqa @tweak[5],@tweak[0] 1640 xorps @tweak[1],$inout1 1641 xorps @tweak[2],$inout2 1642 movups $inout0,($out) 1643 xorps @tweak[3],$inout3 1644 movups $inout1,16*1($out) 1645 movups $inout2,16*2($out) 1646 movups $inout3,16*3($out) 1647 lea 16*4($out),$out 1648 jmp .Lxts_enc_done 1649 1650 .align 16 1651 .Lxts_enc_done: 1652 and \$15,$len_ 1653 jz .Lxts_enc_ret 1654 mov $len_,$len 1655 1656 .Lxts_enc_steal: 1657 movzb ($inp),%eax # borrow $rounds ... 1658 movzb -16($out),%ecx # ... and $key 1659 lea 1($inp),$inp 1660 mov %al,-16($out) 1661 mov %cl,0($out) 1662 lea 1($out),$out 1663 sub \$1,$len 1664 jnz .Lxts_enc_steal 1665 1666 sub $len_,$out # rewind $out 1667 mov $key_,$key # restore $key 1668 mov $rnds_,$rounds # restore $rounds 1669 1670 movups -16($out),$inout0 1671 xorps @tweak[0],$inout0 1672 ___ 1673 &aesni_generate1("enc",$key,$rounds); 1674 $code.=<<___; 1675 xorps @tweak[0],$inout0 1676 movups $inout0,-16($out) 1677 1678 .Lxts_enc_ret: 1679 ___ 1680 $code.=<<___ if ($win64); 1681 movaps 0x60(%rsp),%xmm6 1682 movaps 0x70(%rsp),%xmm7 1683 movaps 0x80(%rsp),%xmm8 1684 movaps 0x90(%rsp),%xmm9 1685 movaps 0xa0(%rsp),%xmm10 1686 movaps 0xb0(%rsp),%xmm11 1687 movaps 0xc0(%rsp),%xmm12 1688 movaps 0xd0(%rsp),%xmm13 1689 movaps 0xe0(%rsp),%xmm14 1690 movaps 0xf0(%rsp),%xmm15 1691 ___ 1692 $code.=<<___; 1693 lea $frame_size(%rsp),%rsp 1694 .Lxts_enc_epilogue: 1695 ret 1696 .size aesni_xts_encrypt,.-aesni_xts_encrypt 1697 ___ 1698 1699 $code.=<<___; 1700 .globl aesni_xts_decrypt 1701 .type aesni_xts_decrypt,\@function,6 1702 .align 16 1703 aesni_xts_decrypt: 1704 lea -$frame_size(%rsp),%rsp 1705 ___ 1706 $code.=<<___ if ($win64); 1707 movaps %xmm6,0x60(%rsp) 1708 movaps %xmm7,0x70(%rsp) 1709 movaps %xmm8,0x80(%rsp) 1710 movaps %xmm9,0x90(%rsp) 1711 movaps %xmm10,0xa0(%rsp) 1712 movaps %xmm11,0xb0(%rsp) 1713 movaps %xmm12,0xc0(%rsp) 1714 movaps %xmm13,0xd0(%rsp) 1715 movaps %xmm14,0xe0(%rsp) 1716 movaps %xmm15,0xf0(%rsp) 1717 .Lxts_dec_body: 1718 ___ 1719 $code.=<<___; 1720 movups ($ivp),@tweak[5] # load clear-text tweak 1721 mov 240($key2),$rounds # key2->rounds 1722 mov 240($key),$rnds_ # key1->rounds 1723 ___ 1724 # generate the tweak 1725 &aesni_generate1("enc",$key2,$rounds,@tweak[5]); 1726 $code.=<<___; 1727 xor %eax,%eax # if ($len%16) len-=16; 1728 test \$15,$len 1729 setnz %al 1730 shl \$4,%rax 1731 sub %rax,$len 1732 1733 mov $key,$key_ # backup $key 1734 mov $rnds_,$rounds # backup $rounds 1735 mov $len,$len_ # backup $len 1736 and \$-16,$len 1737 1738 movdqa .Lxts_magic(%rip),$twmask 1739 pxor $twtmp,$twtmp 1740 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1741 ___ 1742 for ($i=0;$i<4;$i++) { 1743 $code.=<<___; 1744 pshufd \$0x13,$twtmp,$twres 1745 pxor $twtmp,$twtmp 1746 movdqa @tweak[5],@tweak[$i] 1747 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1748 pand $twmask,$twres # isolate carry and residue 1749 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1750 pxor $twres,@tweak[5] 1751 ___ 1752 } 1753 $code.=<<___; 1754 sub \$16*6,$len 1755 jc .Lxts_dec_short 1756 1757 shr \$1,$rounds 1758 sub \$1,$rounds 1759 mov $rounds,$rnds_ 1760 jmp .Lxts_dec_grandloop 1761 1762 .align 16 1763 .Lxts_dec_grandloop: 1764 pshufd \$0x13,$twtmp,$twres 1765 movdqa @tweak[5],@tweak[4] 1766 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1767 movdqu `16*0`($inp),$inout0 # load input 1768 pand $twmask,$twres # isolate carry and residue 1769 movdqu `16*1`($inp),$inout1 1770 pxor $twres,@tweak[5] 1771 1772 movdqu `16*2`($inp),$inout2 1773 pxor @tweak[0],$inout0 # input^=tweak 1774 movdqu `16*3`($inp),$inout3 1775 pxor @tweak[1],$inout1 1776 movdqu `16*4`($inp),$inout4 1777 pxor @tweak[2],$inout2 1778 movdqu `16*5`($inp),$inout5 1779 lea `16*6`($inp),$inp 1780 pxor @tweak[3],$inout3 1781 $movkey ($key_),$rndkey0 1782 pxor @tweak[4],$inout4 1783 pxor @tweak[5],$inout5 1784 1785 # inline _aesni_decrypt6 and interleave first and last rounds 1786 # with own code... 1787 $movkey 16($key_),$rndkey1 1788 pxor $rndkey0,$inout0 1789 pxor $rndkey0,$inout1 1790 movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks 1791 aesdec $rndkey1,$inout0 1792 lea 32($key_),$key 1793 pxor $rndkey0,$inout2 1794 movdqa @tweak[1],`16*1`(%rsp) 1795 aesdec $rndkey1,$inout1 1796 pxor $rndkey0,$inout3 1797 movdqa @tweak[2],`16*2`(%rsp) 1798 aesdec $rndkey1,$inout2 1799 pxor $rndkey0,$inout4 1800 movdqa @tweak[3],`16*3`(%rsp) 1801 aesdec $rndkey1,$inout3 1802 pxor $rndkey0,$inout5 1803 $movkey ($key),$rndkey0 1804 dec $rounds 1805 movdqa @tweak[4],`16*4`(%rsp) 1806 aesdec $rndkey1,$inout4 1807 movdqa @tweak[5],`16*5`(%rsp) 1808 aesdec $rndkey1,$inout5 1809 pxor $twtmp,$twtmp 1810 pcmpgtd @tweak[5],$twtmp 1811 jmp .Lxts_dec_loop6_enter 1812 1813 .align 16 1814 .Lxts_dec_loop6: 1815 aesdec $rndkey1,$inout0 1816 aesdec $rndkey1,$inout1 1817 dec $rounds 1818 aesdec $rndkey1,$inout2 1819 aesdec $rndkey1,$inout3 1820 aesdec $rndkey1,$inout4 1821 aesdec $rndkey1,$inout5 1822 .Lxts_dec_loop6_enter: 1823 $movkey 16($key),$rndkey1 1824 aesdec $rndkey0,$inout0 1825 aesdec $rndkey0,$inout1 1826 lea 32($key),$key 1827 aesdec $rndkey0,$inout2 1828 aesdec $rndkey0,$inout3 1829 aesdec $rndkey0,$inout4 1830 aesdec $rndkey0,$inout5 1831 $movkey ($key),$rndkey0 1832 jnz .Lxts_dec_loop6 1833 1834 pshufd \$0x13,$twtmp,$twres 1835 pxor $twtmp,$twtmp 1836 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1837 aesdec $rndkey1,$inout0 1838 pand $twmask,$twres # isolate carry and residue 1839 aesdec $rndkey1,$inout1 1840 pcmpgtd @tweak[5],$twtmp # broadcast upper bits 1841 aesdec $rndkey1,$inout2 1842 pxor $twres,@tweak[5] 1843 aesdec $rndkey1,$inout3 1844 aesdec $rndkey1,$inout4 1845 aesdec $rndkey1,$inout5 1846 $movkey 16($key),$rndkey1 1847 1848 pshufd \$0x13,$twtmp,$twres 1849 pxor $twtmp,$twtmp 1850 movdqa @tweak[5],@tweak[0] 1851 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1852 aesdec $rndkey0,$inout0 1853 pand $twmask,$twres # isolate carry and residue 1854 aesdec $rndkey0,$inout1 1855 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1856 aesdec $rndkey0,$inout2 1857 pxor $twres,@tweak[5] 1858 aesdec $rndkey0,$inout3 1859 aesdec $rndkey0,$inout4 1860 aesdec $rndkey0,$inout5 1861 $movkey 32($key),$rndkey0 1862 1863 pshufd \$0x13,$twtmp,$twres 1864 pxor $twtmp,$twtmp 1865 movdqa @tweak[5],@tweak[1] 1866 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1867 aesdec $rndkey1,$inout0 1868 pand $twmask,$twres # isolate carry and residue 1869 aesdec $rndkey1,$inout1 1870 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1871 aesdec $rndkey1,$inout2 1872 pxor $twres,@tweak[5] 1873 aesdec $rndkey1,$inout3 1874 aesdec $rndkey1,$inout4 1875 aesdec $rndkey1,$inout5 1876 1877 pshufd \$0x13,$twtmp,$twres 1878 pxor $twtmp,$twtmp 1879 movdqa @tweak[5],@tweak[2] 1880 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1881 aesdeclast $rndkey0,$inout0 1882 pand $twmask,$twres # isolate carry and residue 1883 aesdeclast $rndkey0,$inout1 1884 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1885 aesdeclast $rndkey0,$inout2 1886 pxor $twres,@tweak[5] 1887 aesdeclast $rndkey0,$inout3 1888 aesdeclast $rndkey0,$inout4 1889 aesdeclast $rndkey0,$inout5 1890 1891 pshufd \$0x13,$twtmp,$twres 1892 pxor $twtmp,$twtmp 1893 movdqa @tweak[5],@tweak[3] 1894 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1895 xorps `16*0`(%rsp),$inout0 # output^=tweak 1896 pand $twmask,$twres # isolate carry and residue 1897 xorps `16*1`(%rsp),$inout1 1898 pcmpgtd @tweak[5],$twtmp # broadcat upper bits 1899 pxor $twres,@tweak[5] 1900 1901 xorps `16*2`(%rsp),$inout2 1902 movups $inout0,`16*0`($out) # write output 1903 xorps `16*3`(%rsp),$inout3 1904 movups $inout1,`16*1`($out) 1905 xorps `16*4`(%rsp),$inout4 1906 movups $inout2,`16*2`($out) 1907 xorps `16*5`(%rsp),$inout5 1908 movups $inout3,`16*3`($out) 1909 mov $rnds_,$rounds # restore $rounds 1910 movups $inout4,`16*4`($out) 1911 movups $inout5,`16*5`($out) 1912 lea `16*6`($out),$out 1913 sub \$16*6,$len 1914 jnc .Lxts_dec_grandloop 1915 1916 lea 3($rounds,$rounds),$rounds # restore original value 1917 mov $key_,$key # restore $key 1918 mov $rounds,$rnds_ # backup $rounds 1919 1920 .Lxts_dec_short: 1921 add \$16*6,$len 1922 jz .Lxts_dec_done 1923 1924 cmp \$0x20,$len 1925 jb .Lxts_dec_one 1926 je .Lxts_dec_two 1927 1928 cmp \$0x40,$len 1929 jb .Lxts_dec_three 1930 je .Lxts_dec_four 1931 1932 pshufd \$0x13,$twtmp,$twres 1933 movdqa @tweak[5],@tweak[4] 1934 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1935 movdqu ($inp),$inout0 1936 pand $twmask,$twres # isolate carry and residue 1937 movdqu 16*1($inp),$inout1 1938 pxor $twres,@tweak[5] 1939 1940 movdqu 16*2($inp),$inout2 1941 pxor @tweak[0],$inout0 1942 movdqu 16*3($inp),$inout3 1943 pxor @tweak[1],$inout1 1944 movdqu 16*4($inp),$inout4 1945 lea 16*5($inp),$inp 1946 pxor @tweak[2],$inout2 1947 pxor @tweak[3],$inout3 1948 pxor @tweak[4],$inout4 1949 1950 call _aesni_decrypt6 1951 1952 xorps @tweak[0],$inout0 1953 xorps @tweak[1],$inout1 1954 xorps @tweak[2],$inout2 1955 movdqu $inout0,($out) 1956 xorps @tweak[3],$inout3 1957 movdqu $inout1,16*1($out) 1958 xorps @tweak[4],$inout4 1959 movdqu $inout2,16*2($out) 1960 pxor $twtmp,$twtmp 1961 movdqu $inout3,16*3($out) 1962 pcmpgtd @tweak[5],$twtmp 1963 movdqu $inout4,16*4($out) 1964 lea 16*5($out),$out 1965 pshufd \$0x13,$twtmp,@tweak[1] # $twres 1966 and \$15,$len_ 1967 jz .Lxts_dec_ret 1968 1969 movdqa @tweak[5],@tweak[0] 1970 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 1971 pand $twmask,@tweak[1] # isolate carry and residue 1972 pxor @tweak[5],@tweak[1] 1973 jmp .Lxts_dec_done2 1974 1975 .align 16 1976 .Lxts_dec_one: 1977 movups ($inp),$inout0 1978 lea 16*1($inp),$inp 1979 xorps @tweak[0],$inout0 1980 ___ 1981 &aesni_generate1("dec",$key,$rounds); 1982 $code.=<<___; 1983 xorps @tweak[0],$inout0 1984 movdqa @tweak[1],@tweak[0] 1985 movups $inout0,($out) 1986 movdqa @tweak[2],@tweak[1] 1987 lea 16*1($out),$out 1988 jmp .Lxts_dec_done 1989 1990 .align 16 1991 .Lxts_dec_two: 1992 movups ($inp),$inout0 1993 movups 16($inp),$inout1 1994 lea 32($inp),$inp 1995 xorps @tweak[0],$inout0 1996 xorps @tweak[1],$inout1 1997 1998 call _aesni_decrypt3 1999 2000 xorps @tweak[0],$inout0 2001 movdqa @tweak[2],@tweak[0] 2002 xorps @tweak[1],$inout1 2003 movdqa @tweak[3],@tweak[1] 2004 movups $inout0,($out) 2005 movups $inout1,16*1($out) 2006 lea 16*2($out),$out 2007 jmp .Lxts_dec_done 2008 2009 .align 16 2010 .Lxts_dec_three: 2011 movups ($inp),$inout0 2012 movups 16*1($inp),$inout1 2013 movups 16*2($inp),$inout2 2014 lea 16*3($inp),$inp 2015 xorps @tweak[0],$inout0 2016 xorps @tweak[1],$inout1 2017 xorps @tweak[2],$inout2 2018 2019 call _aesni_decrypt3 2020 2021 xorps @tweak[0],$inout0 2022 movdqa @tweak[3],@tweak[0] 2023 xorps @tweak[1],$inout1 2024 movdqa @tweak[5],@tweak[1] 2025 xorps @tweak[2],$inout2 2026 movups $inout0,($out) 2027 movups $inout1,16*1($out) 2028 movups $inout2,16*2($out) 2029 lea 16*3($out),$out 2030 jmp .Lxts_dec_done 2031 2032 .align 16 2033 .Lxts_dec_four: 2034 pshufd \$0x13,$twtmp,$twres 2035 movdqa @tweak[5],@tweak[4] 2036 paddq @tweak[5],@tweak[5] # psllq 1,$tweak 2037 movups ($inp),$inout0 2038 pand $twmask,$twres # isolate carry and residue 2039 movups 16*1($inp),$inout1 2040 pxor $twres,@tweak[5] 2041 2042 movups 16*2($inp),$inout2 2043 xorps @tweak[0],$inout0 2044 movups 16*3($inp),$inout3 2045 lea 16*4($inp),$inp 2046 xorps @tweak[1],$inout1 2047 xorps @tweak[2],$inout2 2048 xorps @tweak[3],$inout3 2049 2050 call _aesni_decrypt4 2051 2052 xorps @tweak[0],$inout0 2053 movdqa @tweak[4],@tweak[0] 2054 xorps @tweak[1],$inout1 2055 movdqa @tweak[5],@tweak[1] 2056 xorps @tweak[2],$inout2 2057 movups $inout0,($out) 2058 xorps @tweak[3],$inout3 2059 movups $inout1,16*1($out) 2060 movups $inout2,16*2($out) 2061 movups $inout3,16*3($out) 2062 lea 16*4($out),$out 2063 jmp .Lxts_dec_done 2064 2065 .align 16 2066 .Lxts_dec_done: 2067 and \$15,$len_ 2068 jz .Lxts_dec_ret 2069 .Lxts_dec_done2: 2070 mov $len_,$len 2071 mov $key_,$key # restore $key 2072 mov $rnds_,$rounds # restore $rounds 2073 2074 movups ($inp),$inout0 2075 xorps @tweak[1],$inout0 2076 ___ 2077 &aesni_generate1("dec",$key,$rounds); 2078 $code.=<<___; 2079 xorps @tweak[1],$inout0 2080 movups $inout0,($out) 2081 2082 .Lxts_dec_steal: 2083 movzb 16($inp),%eax # borrow $rounds ... 2084 movzb ($out),%ecx # ... and $key 2085 lea 1($inp),$inp 2086 mov %al,($out) 2087 mov %cl,16($out) 2088 lea 1($out),$out 2089 sub \$1,$len 2090 jnz .Lxts_dec_steal 2091 2092 sub $len_,$out # rewind $out 2093 mov $key_,$key # restore $key 2094 mov $rnds_,$rounds # restore $rounds 2095 2096 movups ($out),$inout0 2097 xorps @tweak[0],$inout0 2098 ___ 2099 &aesni_generate1("dec",$key,$rounds); 2100 $code.=<<___; 2101 xorps @tweak[0],$inout0 2102 movups $inout0,($out) 2103 2104 .Lxts_dec_ret: 2105 ___ 2106 $code.=<<___ if ($win64); 2107 movaps 0x60(%rsp),%xmm6 2108 movaps 0x70(%rsp),%xmm7 2109 movaps 0x80(%rsp),%xmm8 2110 movaps 0x90(%rsp),%xmm9 2111 movaps 0xa0(%rsp),%xmm10 2112 movaps 0xb0(%rsp),%xmm11 2113 movaps 0xc0(%rsp),%xmm12 2114 movaps 0xd0(%rsp),%xmm13 2115 movaps 0xe0(%rsp),%xmm14 2116 movaps 0xf0(%rsp),%xmm15 2117 ___ 2118 $code.=<<___; 2119 lea $frame_size(%rsp),%rsp 2120 .Lxts_dec_epilogue: 2121 ret 2122 .size aesni_xts_decrypt,.-aesni_xts_decrypt 2123 ___ 2124 } }} 2125 2127 ######################################################################## 2128 # void $PREFIX_cbc_encrypt (const void *inp, void *out, 2129 # size_t length, const AES_KEY *key, 2130 # unsigned char *ivp,const int enc); 2131 { 2132 my $reserved = $win64?0x40:-0x18; # used in decrypt 2133 $code.=<<___; 2134 .globl ${PREFIX}_cbc_encrypt 2135 .type ${PREFIX}_cbc_encrypt,\@function,6 2136 .align 16 2137 ${PREFIX}_cbc_encrypt: 2138 test $len,$len # check length 2139 jz .Lcbc_ret 2140 2141 mov 240($key),$rnds_ # key->rounds 2142 mov $key,$key_ # backup $key 2143 test %r9d,%r9d # 6th argument 2144 jz .Lcbc_decrypt 2145 #--------------------------- CBC ENCRYPT ------------------------------# 2146 movups ($ivp),$inout0 # load iv as initial state 2147 mov $rnds_,$rounds 2148 cmp \$16,$len 2149 jb .Lcbc_enc_tail 2150 sub \$16,$len 2151 jmp .Lcbc_enc_loop 2152 .align 16 2153 .Lcbc_enc_loop: 2154 movups ($inp),$inout1 # load input 2155 lea 16($inp),$inp 2156 #xorps $inout1,$inout0 2157 ___ 2158 &aesni_generate1("enc",$key,$rounds,$inout0,$inout1); 2159 $code.=<<___; 2160 mov $rnds_,$rounds # restore $rounds 2161 mov $key_,$key # restore $key 2162 movups $inout0,0($out) # store output 2163 lea 16($out),$out 2164 sub \$16,$len 2165 jnc .Lcbc_enc_loop 2166 add \$16,$len 2167 jnz .Lcbc_enc_tail 2168 movups $inout0,($ivp) 2169 jmp .Lcbc_ret 2170 2171 .Lcbc_enc_tail: 2172 mov $len,%rcx # zaps $key 2173 xchg $inp,$out # $inp is %rsi and $out is %rdi now 2174 .long 0x9066A4F3 # rep movsb 2175 mov \$16,%ecx # zero tail 2176 sub $len,%rcx 2177 xor %eax,%eax 2178 .long 0x9066AAF3 # rep stosb 2179 lea -16(%rdi),%rdi # rewind $out by 1 block 2180 mov $rnds_,$rounds # restore $rounds 2181 mov %rdi,%rsi # $inp and $out are the same 2182 mov $key_,$key # restore $key 2183 xor $len,$len # len=16 2184 jmp .Lcbc_enc_loop # one more spin 2185 #--------------------------- CBC DECRYPT ------------------------------# 2187 .align 16 2188 .Lcbc_decrypt: 2189 ___ 2190 $code.=<<___ if ($win64); 2191 lea -0x58(%rsp),%rsp 2192 movaps %xmm6,(%rsp) 2193 movaps %xmm7,0x10(%rsp) 2194 movaps %xmm8,0x20(%rsp) 2195 movaps %xmm9,0x30(%rsp) 2196 .Lcbc_decrypt_body: 2197 ___ 2198 $code.=<<___; 2199 movups ($ivp),$iv 2200 mov $rnds_,$rounds 2201 cmp \$0x70,$len 2202 jbe .Lcbc_dec_tail 2203 shr \$1,$rnds_ 2204 sub \$0x70,$len 2205 mov $rnds_,$rounds 2206 movaps $iv,$reserved(%rsp) 2207 jmp .Lcbc_dec_loop8_enter 2208 .align 16 2209 .Lcbc_dec_loop8: 2210 movaps $rndkey0,$reserved(%rsp) # save IV 2211 movups $inout7,($out) 2212 lea 0x10($out),$out 2213 .Lcbc_dec_loop8_enter: 2214 $movkey ($key),$rndkey0 2215 movups ($inp),$inout0 # load input 2216 movups 0x10($inp),$inout1 2217 $movkey 16($key),$rndkey1 2218 2219 lea 32($key),$key 2220 movdqu 0x20($inp),$inout2 2221 xorps $rndkey0,$inout0 2222 movdqu 0x30($inp),$inout3 2223 xorps $rndkey0,$inout1 2224 movdqu 0x40($inp),$inout4 2225 aesdec $rndkey1,$inout0 2226 pxor $rndkey0,$inout2 2227 movdqu 0x50($inp),$inout5 2228 aesdec $rndkey1,$inout1 2229 pxor $rndkey0,$inout3 2230 movdqu 0x60($inp),$inout6 2231 aesdec $rndkey1,$inout2 2232 pxor $rndkey0,$inout4 2233 movdqu 0x70($inp),$inout7 2234 aesdec $rndkey1,$inout3 2235 pxor $rndkey0,$inout5 2236 dec $rounds 2237 aesdec $rndkey1,$inout4 2238 pxor $rndkey0,$inout6 2239 aesdec $rndkey1,$inout5 2240 pxor $rndkey0,$inout7 2241 $movkey ($key),$rndkey0 2242 aesdec $rndkey1,$inout6 2243 aesdec $rndkey1,$inout7 2244 $movkey 16($key),$rndkey1 2245 2246 call .Ldec_loop8_enter 2247 2248 movups ($inp),$rndkey1 # re-load input 2249 movups 0x10($inp),$rndkey0 2250 xorps $reserved(%rsp),$inout0 # ^= IV 2251 xorps $rndkey1,$inout1 2252 movups 0x20($inp),$rndkey1 2253 xorps $rndkey0,$inout2 2254 movups 0x30($inp),$rndkey0 2255 xorps $rndkey1,$inout3 2256 movups 0x40($inp),$rndkey1 2257 xorps $rndkey0,$inout4 2258 movups 0x50($inp),$rndkey0 2259 xorps $rndkey1,$inout5 2260 movups 0x60($inp),$rndkey1 2261 xorps $rndkey0,$inout6 2262 movups 0x70($inp),$rndkey0 # IV 2263 xorps $rndkey1,$inout7 2264 movups $inout0,($out) 2265 movups $inout1,0x10($out) 2266 movups $inout2,0x20($out) 2267 movups $inout3,0x30($out) 2268 mov $rnds_,$rounds # restore $rounds 2269 movups $inout4,0x40($out) 2270 mov $key_,$key # restore $key 2271 movups $inout5,0x50($out) 2272 lea 0x80($inp),$inp 2273 movups $inout6,0x60($out) 2274 lea 0x70($out),$out 2275 sub \$0x80,$len 2276 ja .Lcbc_dec_loop8 2277 2278 movaps $inout7,$inout0 2279 movaps $rndkey0,$iv 2280 add \$0x70,$len 2281 jle .Lcbc_dec_tail_collected 2282 movups $inout0,($out) 2283 lea 1($rnds_,$rnds_),$rounds 2284 lea 0x10($out),$out 2285 .Lcbc_dec_tail: 2286 movups ($inp),$inout0 2287 movaps $inout0,$in0 2288 cmp \$0x10,$len 2289 jbe .Lcbc_dec_one 2290 2291 movups 0x10($inp),$inout1 2292 movaps $inout1,$in1 2293 cmp \$0x20,$len 2294 jbe .Lcbc_dec_two 2295 2296 movups 0x20($inp),$inout2 2297 movaps $inout2,$in2 2298 cmp \$0x30,$len 2299 jbe .Lcbc_dec_three 2300 2301 movups 0x30($inp),$inout3 2302 cmp \$0x40,$len 2303 jbe .Lcbc_dec_four 2304 2305 movups 0x40($inp),$inout4 2306 cmp \$0x50,$len 2307 jbe .Lcbc_dec_five 2308 2309 movups 0x50($inp),$inout5 2310 cmp \$0x60,$len 2311 jbe .Lcbc_dec_six 2312 2313 movups 0x60($inp),$inout6 2314 movaps $iv,$reserved(%rsp) # save IV 2315 call _aesni_decrypt8 2316 movups ($inp),$rndkey1 2317 movups 0x10($inp),$rndkey0 2318 xorps $reserved(%rsp),$inout0 # ^= IV 2319 xorps $rndkey1,$inout1 2320 movups 0x20($inp),$rndkey1 2321 xorps $rndkey0,$inout2 2322 movups 0x30($inp),$rndkey0 2323 xorps $rndkey1,$inout3 2324 movups 0x40($inp),$rndkey1 2325 xorps $rndkey0,$inout4 2326 movups 0x50($inp),$rndkey0 2327 xorps $rndkey1,$inout5 2328 movups 0x60($inp),$iv # IV 2329 xorps $rndkey0,$inout6 2330 movups $inout0,($out) 2331 movups $inout1,0x10($out) 2332 movups $inout2,0x20($out) 2333 movups $inout3,0x30($out) 2334 movups $inout4,0x40($out) 2335 movups $inout5,0x50($out) 2336 lea 0x60($out),$out 2337 movaps $inout6,$inout0 2338 sub \$0x70,$len 2339 jmp .Lcbc_dec_tail_collected 2340 .align 16 2341 .Lcbc_dec_one: 2342 ___ 2343 &aesni_generate1("dec",$key,$rounds); 2344 $code.=<<___; 2345 xorps $iv,$inout0 2346 movaps $in0,$iv 2347 sub \$0x10,$len 2348 jmp .Lcbc_dec_tail_collected 2349 .align 16 2350 .Lcbc_dec_two: 2351 xorps $inout2,$inout2 2352 call _aesni_decrypt3 2353 xorps $iv,$inout0 2354 xorps $in0,$inout1 2355 movups $inout0,($out) 2356 movaps $in1,$iv 2357 movaps $inout1,$inout0 2358 lea 0x10($out),$out 2359 sub \$0x20,$len 2360 jmp .Lcbc_dec_tail_collected 2361 .align 16 2362 .Lcbc_dec_three: 2363 call _aesni_decrypt3 2364 xorps $iv,$inout0 2365 xorps $in0,$inout1 2366 movups $inout0,($out) 2367 xorps $in1,$inout2 2368 movups $inout1,0x10($out) 2369 movaps $in2,$iv 2370 movaps $inout2,$inout0 2371 lea 0x20($out),$out 2372 sub \$0x30,$len 2373 jmp .Lcbc_dec_tail_collected 2374 .align 16 2375 .Lcbc_dec_four: 2376 call _aesni_decrypt4 2377 xorps $iv,$inout0 2378 movups 0x30($inp),$iv 2379 xorps $in0,$inout1 2380 movups $inout0,($out) 2381 xorps $in1,$inout2 2382 movups $inout1,0x10($out) 2383 xorps $in2,$inout3 2384 movups $inout2,0x20($out) 2385 movaps $inout3,$inout0 2386 lea 0x30($out),$out 2387 sub \$0x40,$len 2388 jmp .Lcbc_dec_tail_collected 2389 .align 16 2390 .Lcbc_dec_five: 2391 xorps $inout5,$inout5 2392 call _aesni_decrypt6 2393 movups 0x10($inp),$rndkey1 2394 movups 0x20($inp),$rndkey0 2395 xorps $iv,$inout0 2396 xorps $in0,$inout1 2397 xorps $rndkey1,$inout2 2398 movups 0x30($inp),$rndkey1 2399 xorps $rndkey0,$inout3 2400 movups 0x40($inp),$iv 2401 xorps $rndkey1,$inout4 2402 movups $inout0,($out) 2403 movups $inout1,0x10($out) 2404 movups $inout2,0x20($out) 2405 movups $inout3,0x30($out) 2406 lea 0x40($out),$out 2407 movaps $inout4,$inout0 2408 sub \$0x50,$len 2409 jmp .Lcbc_dec_tail_collected 2410 .align 16 2411 .Lcbc_dec_six: 2412 call _aesni_decrypt6 2413 movups 0x10($inp),$rndkey1 2414 movups 0x20($inp),$rndkey0 2415 xorps $iv,$inout0 2416 xorps $in0,$inout1 2417 xorps $rndkey1,$inout2 2418 movups 0x30($inp),$rndkey1 2419 xorps $rndkey0,$inout3 2420 movups 0x40($inp),$rndkey0 2421 xorps $rndkey1,$inout4 2422 movups 0x50($inp),$iv 2423 xorps $rndkey0,$inout5 2424 movups $inout0,($out) 2425 movups $inout1,0x10($out) 2426 movups $inout2,0x20($out) 2427 movups $inout3,0x30($out) 2428 movups $inout4,0x40($out) 2429 lea 0x50($out),$out 2430 movaps $inout5,$inout0 2431 sub \$0x60,$len 2432 jmp .Lcbc_dec_tail_collected 2433 .align 16 2434 .Lcbc_dec_tail_collected: 2435 and \$15,$len 2436 movups $iv,($ivp) 2437 jnz .Lcbc_dec_tail_partial 2438 movups $inout0,($out) 2439 jmp .Lcbc_dec_ret 2440 .align 16 2441 .Lcbc_dec_tail_partial: 2442 movaps $inout0,$reserved(%rsp) 2443 mov \$16,%rcx 2444 mov $out,%rdi 2445 sub $len,%rcx 2446 lea $reserved(%rsp),%rsi 2447 .long 0x9066A4F3 # rep movsb 2448 2449 .Lcbc_dec_ret: 2450 ___ 2451 $code.=<<___ if ($win64); 2452 movaps (%rsp),%xmm6 2453 movaps 0x10(%rsp),%xmm7 2454 movaps 0x20(%rsp),%xmm8 2455 movaps 0x30(%rsp),%xmm9 2456 lea 0x58(%rsp),%rsp 2457 ___ 2458 $code.=<<___; 2459 .Lcbc_ret: 2460 ret 2461 .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt 2462 ___ 2463 } 2465 # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey, 2466 # int bits, AES_KEY *key) 2467 { my ($inp,$bits,$key) = @_4args; 2468 $bits =~ s/%r/%e/; 2469 2470 $code.=<<___; 2471 .globl ${PREFIX}_set_decrypt_key 2472 .type ${PREFIX}_set_decrypt_key,\@abi-omnipotent 2473 .align 16 2474 ${PREFIX}_set_decrypt_key: 2475 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 2476 call __aesni_set_encrypt_key 2477 shl \$4,$bits # rounds-1 after _aesni_set_encrypt_key 2478 test %eax,%eax 2479 jnz .Ldec_key_ret 2480 lea 16($key,$bits),$inp # points at the end of key schedule 2481 2482 $movkey ($key),%xmm0 # just swap 2483 $movkey ($inp),%xmm1 2484 $movkey %xmm0,($inp) 2485 $movkey %xmm1,($key) 2486 lea 16($key),$key 2487 lea -16($inp),$inp 2488 2489 .Ldec_key_inverse: 2490 $movkey ($key),%xmm0 # swap and inverse 2491 $movkey ($inp),%xmm1 2492 aesimc %xmm0,%xmm0 2493 aesimc %xmm1,%xmm1 2494 lea 16($key),$key 2495 lea -16($inp),$inp 2496 $movkey %xmm0,16($inp) 2497 $movkey %xmm1,-16($key) 2498 cmp $key,$inp 2499 ja .Ldec_key_inverse 2500 2501 $movkey ($key),%xmm0 # inverse middle 2502 aesimc %xmm0,%xmm0 2503 $movkey %xmm0,($inp) 2504 .Ldec_key_ret: 2505 add \$8,%rsp 2506 ret 2507 .LSEH_end_set_decrypt_key: 2508 .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key 2509 ___ 2510 2512 # This is based on submission by 2513 # 2514 # Huang Ying <ying.huang (at] intel.com> 2515 # Vinodh Gopal <vinodh.gopal (at] intel.com> 2516 # Kahraman Akdemir 2517 # 2518 # Agressively optimized in respect to aeskeygenassist's critical path 2519 # and is contained in %xmm0-5 to meet Win64 ABI requirement. 2520 # 2521 $code.=<<___; 2522 .globl ${PREFIX}_set_encrypt_key 2523 .type ${PREFIX}_set_encrypt_key,\@abi-omnipotent 2524 .align 16 2525 ${PREFIX}_set_encrypt_key: 2526 __aesni_set_encrypt_key: 2527 .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 2528 mov \$-1,%rax 2529 test $inp,$inp 2530 jz .Lenc_key_ret 2531 test $key,$key 2532 jz .Lenc_key_ret 2533 2534 movups ($inp),%xmm0 # pull first 128 bits of *userKey 2535 xorps %xmm4,%xmm4 # low dword of xmm4 is assumed 0 2536 lea 16($key),%rax 2537 cmp \$256,$bits 2538 je .L14rounds 2539 cmp \$192,$bits 2540 je .L12rounds 2541 cmp \$128,$bits 2542 jne .Lbad_keybits 2543 2544 .L10rounds: 2545 mov \$9,$bits # 10 rounds for 128-bit key 2546 $movkey %xmm0,($key) # round 0 2547 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 1 2548 call .Lkey_expansion_128_cold 2549 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 2 2550 call .Lkey_expansion_128 2551 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 3 2552 call .Lkey_expansion_128 2553 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 4 2554 call .Lkey_expansion_128 2555 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 5 2556 call .Lkey_expansion_128 2557 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 6 2558 call .Lkey_expansion_128 2559 aeskeygenassist \$0x40,%xmm0,%xmm1 # round 7 2560 call .Lkey_expansion_128 2561 aeskeygenassist \$0x80,%xmm0,%xmm1 # round 8 2562 call .Lkey_expansion_128 2563 aeskeygenassist \$0x1b,%xmm0,%xmm1 # round 9 2564 call .Lkey_expansion_128 2565 aeskeygenassist \$0x36,%xmm0,%xmm1 # round 10 2566 call .Lkey_expansion_128 2567 $movkey %xmm0,(%rax) 2568 mov $bits,80(%rax) # 240(%rdx) 2569 xor %eax,%eax 2570 jmp .Lenc_key_ret 2571 2572 .align 16 2573 .L12rounds: 2574 movq 16($inp),%xmm2 # remaining 1/3 of *userKey 2575 mov \$11,$bits # 12 rounds for 192 2576 $movkey %xmm0,($key) # round 0 2577 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 1,2 2578 call .Lkey_expansion_192a_cold 2579 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 2,3 2580 call .Lkey_expansion_192b 2581 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 4,5 2582 call .Lkey_expansion_192a 2583 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 5,6 2584 call .Lkey_expansion_192b 2585 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 7,8 2586 call .Lkey_expansion_192a 2587 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 8,9 2588 call .Lkey_expansion_192b 2589 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 10,11 2590 call .Lkey_expansion_192a 2591 aeskeygenassist \$0x80,%xmm2,%xmm1 # round 11,12 2592 call .Lkey_expansion_192b 2593 $movkey %xmm0,(%rax) 2594 mov $bits,48(%rax) # 240(%rdx) 2595 xor %rax, %rax 2596 jmp .Lenc_key_ret 2597 2598 .align 16 2599 .L14rounds: 2600 movups 16($inp),%xmm2 # remaning half of *userKey 2601 mov \$13,$bits # 14 rounds for 256 2602 lea 16(%rax),%rax 2603 $movkey %xmm0,($key) # round 0 2604 $movkey %xmm2,16($key) # round 1 2605 aeskeygenassist \$0x1,%xmm2,%xmm1 # round 2 2606 call .Lkey_expansion_256a_cold 2607 aeskeygenassist \$0x1,%xmm0,%xmm1 # round 3 2608 call .Lkey_expansion_256b 2609 aeskeygenassist \$0x2,%xmm2,%xmm1 # round 4 2610 call .Lkey_expansion_256a 2611 aeskeygenassist \$0x2,%xmm0,%xmm1 # round 5 2612 call .Lkey_expansion_256b 2613 aeskeygenassist \$0x4,%xmm2,%xmm1 # round 6 2614 call .Lkey_expansion_256a 2615 aeskeygenassist \$0x4,%xmm0,%xmm1 # round 7 2616 call .Lkey_expansion_256b 2617 aeskeygenassist \$0x8,%xmm2,%xmm1 # round 8 2618 call .Lkey_expansion_256a 2619 aeskeygenassist \$0x8,%xmm0,%xmm1 # round 9 2620 call .Lkey_expansion_256b 2621 aeskeygenassist \$0x10,%xmm2,%xmm1 # round 10 2622 call .Lkey_expansion_256a 2623 aeskeygenassist \$0x10,%xmm0,%xmm1 # round 11 2624 call .Lkey_expansion_256b 2625 aeskeygenassist \$0x20,%xmm2,%xmm1 # round 12 2626 call .Lkey_expansion_256a 2627 aeskeygenassist \$0x20,%xmm0,%xmm1 # round 13 2628 call .Lkey_expansion_256b 2629 aeskeygenassist \$0x40,%xmm2,%xmm1 # round 14 2630 call .Lkey_expansion_256a 2631 $movkey %xmm0,(%rax) 2632 mov $bits,16(%rax) # 240(%rdx) 2633 xor %rax,%rax 2634 jmp .Lenc_key_ret 2635 2636 .align 16 2637 .Lbad_keybits: 2638 mov \$-2,%rax 2639 .Lenc_key_ret: 2640 add \$8,%rsp 2641 ret 2642 .LSEH_end_set_encrypt_key: 2643 2645 .align 16 2646 .Lkey_expansion_128: 2647 $movkey %xmm0,(%rax) 2648 lea 16(%rax),%rax 2649 .Lkey_expansion_128_cold: 2650 shufps \$0b00010000,%xmm0,%xmm4 2651 xorps %xmm4, %xmm0 2652 shufps \$0b10001100,%xmm0,%xmm4 2653 xorps %xmm4, %xmm0 2654 shufps \$0b11111111,%xmm1,%xmm1 # critical path 2655 xorps %xmm1,%xmm0 2656 ret 2657 2658 .align 16 2659 .Lkey_expansion_192a: 2660 $movkey %xmm0,(%rax) 2661 lea 16(%rax),%rax 2662 .Lkey_expansion_192a_cold: 2663 movaps %xmm2, %xmm5 2664 .Lkey_expansion_192b_warm: 2665 shufps \$0b00010000,%xmm0,%xmm4 2666 movdqa %xmm2,%xmm3 2667 xorps %xmm4,%xmm0 2668 shufps \$0b10001100,%xmm0,%xmm4 2669 pslldq \$4,%xmm3 2670 xorps %xmm4,%xmm0 2671 pshufd \$0b01010101,%xmm1,%xmm1 # critical path 2672 pxor %xmm3,%xmm2 2673 pxor %xmm1,%xmm0 2674 pshufd \$0b11111111,%xmm0,%xmm3 2675 pxor %xmm3,%xmm2 2676 ret 2677 2678 .align 16 2679 .Lkey_expansion_192b: 2680 movaps %xmm0,%xmm3 2681 shufps \$0b01000100,%xmm0,%xmm5 2682 $movkey %xmm5,(%rax) 2683 shufps \$0b01001110,%xmm2,%xmm3 2684 $movkey %xmm3,16(%rax) 2685 lea 32(%rax),%rax 2686 jmp .Lkey_expansion_192b_warm 2687 2688 .align 16 2689 .Lkey_expansion_256a: 2690 $movkey %xmm2,(%rax) 2691 lea 16(%rax),%rax 2692 .Lkey_expansion_256a_cold: 2693 shufps \$0b00010000,%xmm0,%xmm4 2694 xorps %xmm4,%xmm0 2695 shufps \$0b10001100,%xmm0,%xmm4 2696 xorps %xmm4,%xmm0 2697 shufps \$0b11111111,%xmm1,%xmm1 # critical path 2698 xorps %xmm1,%xmm0 2699 ret 2700 2701 .align 16 2702 .Lkey_expansion_256b: 2703 $movkey %xmm0,(%rax) 2704 lea 16(%rax),%rax 2705 2706 shufps \$0b00010000,%xmm2,%xmm4 2707 xorps %xmm4,%xmm2 2708 shufps \$0b10001100,%xmm2,%xmm4 2709 xorps %xmm4,%xmm2 2710 shufps \$0b10101010,%xmm1,%xmm1 # critical path 2711 xorps %xmm1,%xmm2 2712 ret 2713 .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key 2714 .size __aesni_set_encrypt_key,.-__aesni_set_encrypt_key 2715 ___ 2716 } 2717 2719 $code.=<<___; 2720 .align 64 2721 .Lbswap_mask: 2722 .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 2723 .Lincrement32: 2724 .long 6,6,6,0 2725 .Lincrement64: 2726 .long 1,0,0,0 2727 .Lxts_magic: 2728 .long 0x87,0,1,0 2729 2730 .asciz "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>" 2731 .align 64 2732 ___ 2733 2734 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, 2735 # CONTEXT *context,DISPATCHER_CONTEXT *disp) 2736 if ($win64) { 2737 $rec="%rcx"; 2738 $frame="%rdx"; 2739 $context="%r8"; 2740 $disp="%r9"; 2741 2742 $code.=<<___; 2743 .extern __imp_RtlVirtualUnwind 2744 ___ 2745 $code.=<<___ if ($PREFIX eq "aesni"); 2746 .type ecb_se_handler,\@abi-omnipotent 2747 .align 16 2748 ecb_se_handler: 2749 push %rsi 2750 push %rdi 2751 push %rbx 2752 push %rbp 2753 push %r12 2754 push %r13 2755 push %r14 2756 push %r15 2757 pushfq 2758 sub \$64,%rsp 2759 2760 mov 152($context),%rax # pull context->Rsp 2761 2762 jmp .Lcommon_seh_tail 2763 .size ecb_se_handler,.-ecb_se_handler 2764 2765 .type ccm64_se_handler,\@abi-omnipotent 2766 .align 16 2767 ccm64_se_handler: 2768 push %rsi 2769 push %rdi 2770 push %rbx 2771 push %rbp 2772 push %r12 2773 push %r13 2774 push %r14 2775 push %r15 2776 pushfq 2777 sub \$64,%rsp 2778 2779 mov 120($context),%rax # pull context->Rax 2780 mov 248($context),%rbx # pull context->Rip 2781 2782 mov 8($disp),%rsi # disp->ImageBase 2783 mov 56($disp),%r11 # disp->HandlerData 2784 2785 mov 0(%r11),%r10d # HandlerData[0] 2786 lea (%rsi,%r10),%r10 # prologue label 2787 cmp %r10,%rbx # context->Rip<prologue label 2788 jb .Lcommon_seh_tail 2789 2790 mov 152($context),%rax # pull context->Rsp 2791 2792 mov 4(%r11),%r10d # HandlerData[1] 2793 lea (%rsi,%r10),%r10 # epilogue label 2794 cmp %r10,%rbx # context->Rip>=epilogue label 2795 jae .Lcommon_seh_tail 2796 2797 lea 0(%rax),%rsi # %xmm save area 2798 lea 512($context),%rdi # &context.Xmm6 2799 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) 2800 .long 0xa548f3fc # cld; rep movsq 2801 lea 0x58(%rax),%rax # adjust stack pointer 2802 2803 jmp .Lcommon_seh_tail 2804 .size ccm64_se_handler,.-ccm64_se_handler 2805 2806 .type ctr32_se_handler,\@abi-omnipotent 2807 .align 16 2808 ctr32_se_handler: 2809 push %rsi 2810 push %rdi 2811 push %rbx 2812 push %rbp 2813 push %r12 2814 push %r13 2815 push %r14 2816 push %r15 2817 pushfq 2818 sub \$64,%rsp 2819 2820 mov 120($context),%rax # pull context->Rax 2821 mov 248($context),%rbx # pull context->Rip 2822 2823 lea .Lctr32_body(%rip),%r10 2824 cmp %r10,%rbx # context->Rip<"prologue" label 2825 jb .Lcommon_seh_tail 2826 2827 mov 152($context),%rax # pull context->Rsp 2828 2829 lea .Lctr32_ret(%rip),%r10 2830 cmp %r10,%rbx 2831 jae .Lcommon_seh_tail 2832 2833 lea 0x20(%rax),%rsi # %xmm save area 2834 lea 512($context),%rdi # &context.Xmm6 2835 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) 2836 .long 0xa548f3fc # cld; rep movsq 2837 lea 0xc8(%rax),%rax # adjust stack pointer 2838 2839 jmp .Lcommon_seh_tail 2840 .size ctr32_se_handler,.-ctr32_se_handler 2841 2842 .type xts_se_handler,\@abi-omnipotent 2843 .align 16 2844 xts_se_handler: 2845 push %rsi 2846 push %rdi 2847 push %rbx 2848 push %rbp 2849 push %r12 2850 push %r13 2851 push %r14 2852 push %r15 2853 pushfq 2854 sub \$64,%rsp 2855 2856 mov 120($context),%rax # pull context->Rax 2857 mov 248($context),%rbx # pull context->Rip 2858 2859 mov 8($disp),%rsi # disp->ImageBase 2860 mov 56($disp),%r11 # disp->HandlerData 2861 2862 mov 0(%r11),%r10d # HandlerData[0] 2863 lea (%rsi,%r10),%r10 # prologue lable 2864 cmp %r10,%rbx # context->Rip<prologue label 2865 jb .Lcommon_seh_tail 2866 2867 mov 152($context),%rax # pull context->Rsp 2868 2869 mov 4(%r11),%r10d # HandlerData[1] 2870 lea (%rsi,%r10),%r10 # epilogue label 2871 cmp %r10,%rbx # context->Rip>=epilogue label 2872 jae .Lcommon_seh_tail 2873 2874 lea 0x60(%rax),%rsi # %xmm save area 2875 lea 512($context),%rdi # & context.Xmm6 2876 mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) 2877 .long 0xa548f3fc # cld; rep movsq 2878 lea 0x68+160(%rax),%rax # adjust stack pointer 2879 2880 jmp .Lcommon_seh_tail 2881 .size xts_se_handler,.-xts_se_handler 2882 ___ 2883 $code.=<<___; 2884 .type cbc_se_handler,\@abi-omnipotent 2885 .align 16 2886 cbc_se_handler: 2887 push %rsi 2888 push %rdi 2889 push %rbx 2890 push %rbp 2891 push %r12 2892 push %r13 2893 push %r14 2894 push %r15 2895 pushfq 2896 sub \$64,%rsp 2897 2898 mov 152($context),%rax # pull context->Rsp 2899 mov 248($context),%rbx # pull context->Rip 2900 2901 lea .Lcbc_decrypt(%rip),%r10 2902 cmp %r10,%rbx # context->Rip<"prologue" label 2903 jb .Lcommon_seh_tail 2904 2905 lea .Lcbc_decrypt_body(%rip),%r10 2906 cmp %r10,%rbx # context->Rip<cbc_decrypt_body 2907 jb .Lrestore_cbc_rax 2908 2909 lea .Lcbc_ret(%rip),%r10 2910 cmp %r10,%rbx # context->Rip>="epilogue" label 2911 jae .Lcommon_seh_tail 2912 2913 lea 0(%rax),%rsi # top of stack 2914 lea 512($context),%rdi # &context.Xmm6 2915 mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) 2916 .long 0xa548f3fc # cld; rep movsq 2917 lea 0x58(%rax),%rax # adjust stack pointer 2918 jmp .Lcommon_seh_tail 2919 2920 .Lrestore_cbc_rax: 2921 mov 120($context),%rax 2922 2923 .Lcommon_seh_tail: 2924 mov 8(%rax),%rdi 2925 mov 16(%rax),%rsi 2926 mov %rax,152($context) # restore context->Rsp 2927 mov %rsi,168($context) # restore context->Rsi 2928 mov %rdi,176($context) # restore context->Rdi 2929 2930 mov 40($disp),%rdi # disp->ContextRecord 2931 mov $context,%rsi # context 2932 mov \$154,%ecx # sizeof(CONTEXT) 2933 .long 0xa548f3fc # cld; rep movsq 2934 2935 mov $disp,%rsi 2936 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER 2937 mov 8(%rsi),%rdx # arg2, disp->ImageBase 2938 mov 0(%rsi),%r8 # arg3, disp->ControlPc 2939 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry 2940 mov 40(%rsi),%r10 # disp->ContextRecord 2941 lea 56(%rsi),%r11 # &disp->HandlerData 2942 lea 24(%rsi),%r12 # &disp->EstablisherFrame 2943 mov %r10,32(%rsp) # arg5 2944 mov %r11,40(%rsp) # arg6 2945 mov %r12,48(%rsp) # arg7 2946 mov %rcx,56(%rsp) # arg8, (NULL) 2947 call *__imp_RtlVirtualUnwind(%rip) 2948 2949 mov \$1,%eax # ExceptionContinueSearch 2950 add \$64,%rsp 2951 popfq 2952 pop %r15 2953 pop %r14 2954 pop %r13 2955 pop %r12 2956 pop %rbp 2957 pop %rbx 2958 pop %rdi 2959 pop %rsi 2960 ret 2961 .size cbc_se_handler,.-cbc_se_handler 2962 2963 .section .pdata 2964 .align 4 2965 ___ 2966 $code.=<<___ if ($PREFIX eq "aesni"); 2967 .rva .LSEH_begin_aesni_ecb_encrypt 2968 .rva .LSEH_end_aesni_ecb_encrypt 2969 .rva .LSEH_info_ecb 2970 2971 .rva .LSEH_begin_aesni_ccm64_encrypt_blocks 2972 .rva .LSEH_end_aesni_ccm64_encrypt_blocks 2973 .rva .LSEH_info_ccm64_enc 2974 2975 .rva .LSEH_begin_aesni_ccm64_decrypt_blocks 2976 .rva .LSEH_end_aesni_ccm64_decrypt_blocks 2977 .rva .LSEH_info_ccm64_dec 2978 2979 .rva .LSEH_begin_aesni_ctr32_encrypt_blocks 2980 .rva .LSEH_end_aesni_ctr32_encrypt_blocks 2981 .rva .LSEH_info_ctr32 2982 2983 .rva .LSEH_begin_aesni_xts_encrypt 2984 .rva .LSEH_end_aesni_xts_encrypt 2985 .rva .LSEH_info_xts_enc 2986 2987 .rva .LSEH_begin_aesni_xts_decrypt 2988 .rva .LSEH_end_aesni_xts_decrypt 2989 .rva .LSEH_info_xts_dec 2990 ___ 2991 $code.=<<___; 2992 .rva .LSEH_begin_${PREFIX}_cbc_encrypt 2993 .rva .LSEH_end_${PREFIX}_cbc_encrypt 2994 .rva .LSEH_info_cbc 2995 2996 .rva ${PREFIX}_set_decrypt_key 2997 .rva .LSEH_end_set_decrypt_key 2998 .rva .LSEH_info_key 2999 3000 .rva ${PREFIX}_set_encrypt_key 3001 .rva .LSEH_end_set_encrypt_key 3002 .rva .LSEH_info_key 3003 .section .xdata 3004 .align 8 3005 ___ 3006 $code.=<<___ if ($PREFIX eq "aesni"); 3007 .LSEH_info_ecb: 3008 .byte 9,0,0,0 3009 .rva ecb_se_handler 3010 .LSEH_info_ccm64_enc: 3011 .byte 9,0,0,0 3012 .rva ccm64_se_handler 3013 .rva .Lccm64_enc_body,.Lccm64_enc_ret # HandlerData[] 3014 .LSEH_info_ccm64_dec: 3015 .byte 9,0,0,0 3016 .rva ccm64_se_handler 3017 .rva .Lccm64_dec_body,.Lccm64_dec_ret # HandlerData[] 3018 .LSEH_info_ctr32: 3019 .byte 9,0,0,0 3020 .rva ctr32_se_handler 3021 .LSEH_info_xts_enc: 3022 .byte 9,0,0,0 3023 .rva xts_se_handler 3024 .rva .Lxts_enc_body,.Lxts_enc_epilogue # HandlerData[] 3025 .LSEH_info_xts_dec: 3026 .byte 9,0,0,0 3027 .rva xts_se_handler 3028 .rva .Lxts_dec_body,.Lxts_dec_epilogue # HandlerData[] 3029 ___ 3030 $code.=<<___; 3031 .LSEH_info_cbc: 3032 .byte 9,0,0,0 3033 .rva cbc_se_handler 3034 .LSEH_info_key: 3035 .byte 0x01,0x04,0x01,0x00 3036 .byte 0x04,0x02,0x00,0x00 # sub rsp,8 3037 ___ 3038 } 3039 3040 sub rex { 3041 local *opcode=shift; 3042 my ($dst,$src)=@_; 3043 my $rex=0; 3044 3045 $rex|=0x04 if($dst>=8); 3046 $rex|=0x01 if($src>=8); 3047 push @opcode,$rex|0x40 if($rex); 3048 } 3049 3050 sub aesni { 3051 my $line=shift; 3052 my @opcode=(0x66); 3053 3054 if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { 3055 rex(\@opcode,$4,$3); 3056 push @opcode,0x0f,0x3a,0xdf; 3057 push @opcode,0xc0|($3&7)|(($4&7)<<3); # ModR/M 3058 my $c=$2; 3059 push @opcode,$c=~/^0/?oct($c):$c; 3060 return ".byte\t".join(',',@opcode); 3061 } 3062 elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) { 3063 my %opcodelet = ( 3064 "aesimc" => 0xdb, 3065 "aesenc" => 0xdc, "aesenclast" => 0xdd, 3066 "aesdec" => 0xde, "aesdeclast" => 0xdf 3067 ); 3068 return undef if (!defined($opcodelet{$1})); 3069 rex(\@opcode,$3,$2); 3070 push @opcode,0x0f,0x38,$opcodelet{$1}; 3071 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M 3072 return ".byte\t".join(',',@opcode); 3073 } 3074 return $line; 3075 } 3076 3077 $code =~ s/\`([^\`]*)\`/eval($1)/gem; 3078 $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem; 3079 3080 print $code; 3081 3082 close STDOUT; 3083