]>
Commit | Line | Data |
---|---|---|
d64a7232 AP |
1 | #!/usr/bin/env perl |
2 | # | |
3 | # ==================================================================== | |
4 | # Written by Andy Polyakov <appro@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]. | |
d7d119a3 AP |
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 | |
f8501464 | 21 | # asymptotic limit it's not something you commonly achieve in reality, |
d7d119a3 AP |
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 | |
02f358da | 80 | # observe even 10/11.2 and 10/12.4 (CTR, OFB, CFB)... |
d64a7232 | 81 | |
f8501464 AP |
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 | |
9282c335 | 133 | # CTR 1.14 0.91 0.90 |
f8501464 AP |
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 | ||
d2e18031 AP |
160 | ###################################################################### |
161 | # For reference, AMD Bulldozer spends 5.77 cycles per byte processed | |
162 | # with 128-bit key in CBC encrypt and 0.76 cycles in CBC decrypt, 0.70 | |
9282c335 | 163 | # in ECB, 0.76 in CTR, 0.95 in XTS... This means that aes[enc|dec] |
d2e18031 AP |
164 | # instruction latency is 9 cycles and that they can be issued every |
165 | # cycle. | |
166 | ||
d64a7232 AP |
167 | $PREFIX="aesni"; # if $PREFIX is set to "AES", the script |
168 | # generates drop-in replacement for | |
169 | # crypto/aes/asm/aes-x86_64.pl:-) | |
170 | ||
171 | $flavour = shift; | |
172 | $output = shift; | |
173 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } | |
174 | ||
175 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); | |
176 | ||
177 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; | |
178 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or | |
179 | ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or | |
180 | die "can't locate x86_64-xlate.pl"; | |
181 | ||
6251989e | 182 | open STDOUT,"| \"$^X\" $xlate $flavour $output"; |
d64a7232 | 183 | |
8da721ee | 184 | $movkey = $PREFIX eq "aesni" ? "movups" : "movups"; |
d608b4d6 AP |
185 | @_4args=$win64? ("%rcx","%rdx","%r8", "%r9") : # Win64 order |
186 | ("%rdi","%rsi","%rdx","%rcx"); # Unix order | |
d64a7232 AP |
187 | |
188 | $code=".text\n"; | |
189 | ||
190 | $rounds="%eax"; # input to and changed by aesni_[en|de]cryptN !!! | |
d608b4d6 | 191 | # this is natural Unix argument order for public $PREFIX_[ecb|cbc]_encrypt ... |
d64a7232 AP |
192 | $inp="%rdi"; |
193 | $out="%rsi"; | |
d64a7232 AP |
194 | $len="%rdx"; |
195 | $key="%rcx"; # input to and changed by aesni_[en|de]cryptN !!! | |
d7d119a3 | 196 | $ivp="%r8"; # cbc, ctr, ... |
d64a7232 AP |
197 | |
198 | $rnds_="%r10d"; # backup copy for $rounds | |
199 | $key_="%r11"; # backup copy for $key | |
200 | ||
201 | # %xmm register layout | |
f8501464 AP |
202 | $rndkey0="%xmm0"; $rndkey1="%xmm1"; |
203 | $inout0="%xmm2"; $inout1="%xmm3"; | |
204 | $inout2="%xmm4"; $inout3="%xmm5"; | |
205 | $inout4="%xmm6"; $inout5="%xmm7"; | |
206 | $inout6="%xmm8"; $inout7="%xmm9"; | |
207 | ||
208 | $in2="%xmm6"; $in1="%xmm7"; # used in CBC decrypt, CTR, ... | |
209 | $in0="%xmm8"; $iv="%xmm9"; | |
d64a7232 AP |
210 | \f |
211 | # Inline version of internal aesni_[en|de]crypt1. | |
212 | # | |
213 | # Why folded loop? Because aes[enc|dec] is slow enough to accommodate | |
214 | # cycles which take care of loop variables... | |
215 | { my $sn; | |
d608b4d6 | 216 | sub aesni_generate1 { |
f8501464 | 217 | my ($p,$key,$rounds,$inout,$ivec)=@_; $inout=$inout0 if (!defined($inout)); |
d64a7232 AP |
218 | ++$sn; |
219 | $code.=<<___; | |
f8501464 | 220 | $movkey ($key),$rndkey0 |
d64a7232 | 221 | $movkey 16($key),$rndkey1 |
f8501464 AP |
222 | ___ |
223 | $code.=<<___ if (defined($ivec)); | |
224 | xorps $rndkey0,$ivec | |
225 | lea 32($key),$key | |
226 | xorps $ivec,$inout | |
227 | ___ | |
228 | $code.=<<___ if (!defined($ivec)); | |
d608b4d6 | 229 | lea 32($key),$key |
f8501464 AP |
230 | xorps $rndkey0,$inout |
231 | ___ | |
232 | $code.=<<___; | |
d608b4d6 | 233 | .Loop_${p}1_$sn: |
d7d119a3 | 234 | aes${p} $rndkey1,$inout |
d64a7232 | 235 | dec $rounds |
d64a7232 | 236 | $movkey ($key),$rndkey1 |
d64a7232 | 237 | lea 16($key),$key |
d608b4d6 | 238 | jnz .Loop_${p}1_$sn # loop body is 16 bytes |
d7d119a3 | 239 | aes${p}last $rndkey1,$inout |
d64a7232 AP |
240 | ___ |
241 | }} | |
d608b4d6 | 242 | # void $PREFIX_[en|de]crypt (const void *inp,void *out,const AES_KEY *key); |
d64a7232 | 243 | # |
d608b4d6 AP |
244 | { my ($inp,$out,$key) = @_4args; |
245 | ||
d64a7232 AP |
246 | $code.=<<___; |
247 | .globl ${PREFIX}_encrypt | |
d608b4d6 | 248 | .type ${PREFIX}_encrypt,\@abi-omnipotent |
d64a7232 AP |
249 | .align 16 |
250 | ${PREFIX}_encrypt: | |
f8501464 AP |
251 | movups ($inp),$inout0 # load input |
252 | mov 240($key),$rounds # key->rounds | |
d64a7232 | 253 | ___ |
d608b4d6 | 254 | &aesni_generate1("enc",$key,$rounds); |
d64a7232 | 255 | $code.=<<___; |
d608b4d6 | 256 | movups $inout0,($out) # output |
d64a7232 AP |
257 | ret |
258 | .size ${PREFIX}_encrypt,.-${PREFIX}_encrypt | |
d64a7232 | 259 | |
d64a7232 | 260 | .globl ${PREFIX}_decrypt |
d608b4d6 | 261 | .type ${PREFIX}_decrypt,\@abi-omnipotent |
d64a7232 AP |
262 | .align 16 |
263 | ${PREFIX}_decrypt: | |
f8501464 AP |
264 | movups ($inp),$inout0 # load input |
265 | mov 240($key),$rounds # key->rounds | |
d64a7232 | 266 | ___ |
d608b4d6 | 267 | &aesni_generate1("dec",$key,$rounds); |
d64a7232 | 268 | $code.=<<___; |
d608b4d6 | 269 | movups $inout0,($out) # output |
d64a7232 AP |
270 | ret |
271 | .size ${PREFIX}_decrypt, .-${PREFIX}_decrypt | |
272 | ___ | |
d608b4d6 | 273 | } |
d64a7232 | 274 | \f |
f8501464 AP |
275 | # _aesni_[en|de]cryptN are private interfaces, N denotes interleave |
276 | # factor. Why 3x subroutine were originally used in loops? Even though | |
277 | # aes[enc|dec] latency was originally 6, it could be scheduled only | |
278 | # every *2nd* cycle. Thus 3x interleave was the one providing optimal | |
d608b4d6 AP |
279 | # utilization, i.e. when subroutine's throughput is virtually same as |
280 | # of non-interleaved subroutine [for number of input blocks up to 3]. | |
f8501464 AP |
281 | # This is why it makes no sense to implement 2x subroutine. |
282 | # aes[enc|dec] latency in next processor generation is 8, but the | |
283 | # instructions can be scheduled every cycle. Optimal interleave for | |
284 | # new processor is therefore 8x... | |
d608b4d6 | 285 | sub aesni_generate3 { |
d64a7232 AP |
286 | my $dir=shift; |
287 | # As already mentioned it takes in $key and $rounds, which are *not* | |
d608b4d6 | 288 | # preserved. $inout[0-2] is cipher/clear text... |
d64a7232 | 289 | $code.=<<___; |
d608b4d6 | 290 | .type _aesni_${dir}rypt3,\@abi-omnipotent |
d64a7232 | 291 | .align 16 |
d608b4d6 | 292 | _aesni_${dir}rypt3: |
d64a7232 | 293 | $movkey ($key),$rndkey0 |
d608b4d6 | 294 | shr \$1,$rounds |
d64a7232 | 295 | $movkey 16($key),$rndkey1 |
d608b4d6 | 296 | lea 32($key),$key |
f8501464 AP |
297 | xorps $rndkey0,$inout0 |
298 | xorps $rndkey0,$inout1 | |
299 | xorps $rndkey0,$inout2 | |
d7d119a3 | 300 | $movkey ($key),$rndkey0 |
d608b4d6 AP |
301 | |
302 | .L${dir}_loop3: | |
303 | aes${dir} $rndkey1,$inout0 | |
d608b4d6 AP |
304 | aes${dir} $rndkey1,$inout1 |
305 | dec $rounds | |
306 | aes${dir} $rndkey1,$inout2 | |
d608b4d6 | 307 | $movkey 16($key),$rndkey1 |
d7d119a3 | 308 | aes${dir} $rndkey0,$inout0 |
d608b4d6 AP |
309 | aes${dir} $rndkey0,$inout1 |
310 | lea 32($key),$key | |
311 | aes${dir} $rndkey0,$inout2 | |
d7d119a3 | 312 | $movkey ($key),$rndkey0 |
d608b4d6 AP |
313 | jnz .L${dir}_loop3 |
314 | ||
315 | aes${dir} $rndkey1,$inout0 | |
d608b4d6 AP |
316 | aes${dir} $rndkey1,$inout1 |
317 | aes${dir} $rndkey1,$inout2 | |
318 | aes${dir}last $rndkey0,$inout0 | |
319 | aes${dir}last $rndkey0,$inout1 | |
320 | aes${dir}last $rndkey0,$inout2 | |
321 | ret | |
322 | .size _aesni_${dir}rypt3,.-_aesni_${dir}rypt3 | |
323 | ___ | |
324 | } | |
325 | # 4x interleave is implemented to improve small block performance, | |
326 | # most notably [and naturally] 4 block by ~30%. One can argue that one | |
327 | # should have implemented 5x as well, but improvement would be <20%, | |
328 | # so it's not worth it... | |
329 | sub aesni_generate4 { | |
330 | my $dir=shift; | |
331 | # As already mentioned it takes in $key and $rounds, which are *not* | |
332 | # preserved. $inout[0-3] is cipher/clear text... | |
333 | $code.=<<___; | |
334 | .type _aesni_${dir}rypt4,\@abi-omnipotent | |
335 | .align 16 | |
336 | _aesni_${dir}rypt4: | |
337 | $movkey ($key),$rndkey0 | |
d64a7232 | 338 | shr \$1,$rounds |
d608b4d6 | 339 | $movkey 16($key),$rndkey1 |
d64a7232 | 340 | lea 32($key),$key |
f8501464 AP |
341 | xorps $rndkey0,$inout0 |
342 | xorps $rndkey0,$inout1 | |
343 | xorps $rndkey0,$inout2 | |
344 | xorps $rndkey0,$inout3 | |
345 | $movkey ($key),$rndkey0 | |
d608b4d6 AP |
346 | |
347 | .L${dir}_loop4: | |
d64a7232 | 348 | aes${dir} $rndkey1,$inout0 |
d64a7232 AP |
349 | aes${dir} $rndkey1,$inout1 |
350 | dec $rounds | |
351 | aes${dir} $rndkey1,$inout2 | |
352 | aes${dir} $rndkey1,$inout3 | |
d64a7232 | 353 | $movkey 16($key),$rndkey1 |
d7d119a3 | 354 | aes${dir} $rndkey0,$inout0 |
d64a7232 AP |
355 | aes${dir} $rndkey0,$inout1 |
356 | lea 32($key),$key | |
357 | aes${dir} $rndkey0,$inout2 | |
358 | aes${dir} $rndkey0,$inout3 | |
d7d119a3 | 359 | $movkey ($key),$rndkey0 |
d608b4d6 AP |
360 | jnz .L${dir}_loop4 |
361 | ||
d64a7232 | 362 | aes${dir} $rndkey1,$inout0 |
d64a7232 AP |
363 | aes${dir} $rndkey1,$inout1 |
364 | aes${dir} $rndkey1,$inout2 | |
365 | aes${dir} $rndkey1,$inout3 | |
d64a7232 AP |
366 | aes${dir}last $rndkey0,$inout0 |
367 | aes${dir}last $rndkey0,$inout1 | |
368 | aes${dir}last $rndkey0,$inout2 | |
369 | aes${dir}last $rndkey0,$inout3 | |
d64a7232 | 370 | ret |
d608b4d6 | 371 | .size _aesni_${dir}rypt4,.-_aesni_${dir}rypt4 |
d64a7232 AP |
372 | ___ |
373 | } | |
f8501464 AP |
374 | sub aesni_generate6 { |
375 | my $dir=shift; | |
376 | # As already mentioned it takes in $key and $rounds, which are *not* | |
377 | # preserved. $inout[0-5] is cipher/clear text... | |
378 | $code.=<<___; | |
379 | .type _aesni_${dir}rypt6,\@abi-omnipotent | |
380 | .align 16 | |
381 | _aesni_${dir}rypt6: | |
382 | $movkey ($key),$rndkey0 | |
383 | shr \$1,$rounds | |
384 | $movkey 16($key),$rndkey1 | |
385 | lea 32($key),$key | |
386 | xorps $rndkey0,$inout0 | |
387 | pxor $rndkey0,$inout1 | |
388 | aes${dir} $rndkey1,$inout0 | |
389 | pxor $rndkey0,$inout2 | |
390 | aes${dir} $rndkey1,$inout1 | |
391 | pxor $rndkey0,$inout3 | |
392 | aes${dir} $rndkey1,$inout2 | |
393 | pxor $rndkey0,$inout4 | |
394 | aes${dir} $rndkey1,$inout3 | |
395 | pxor $rndkey0,$inout5 | |
396 | dec $rounds | |
397 | aes${dir} $rndkey1,$inout4 | |
398 | $movkey ($key),$rndkey0 | |
399 | aes${dir} $rndkey1,$inout5 | |
400 | jmp .L${dir}_loop6_enter | |
401 | .align 16 | |
402 | .L${dir}_loop6: | |
403 | aes${dir} $rndkey1,$inout0 | |
404 | aes${dir} $rndkey1,$inout1 | |
405 | dec $rounds | |
406 | aes${dir} $rndkey1,$inout2 | |
407 | aes${dir} $rndkey1,$inout3 | |
408 | aes${dir} $rndkey1,$inout4 | |
409 | aes${dir} $rndkey1,$inout5 | |
410 | .L${dir}_loop6_enter: # happens to be 16-byte aligned | |
411 | $movkey 16($key),$rndkey1 | |
412 | aes${dir} $rndkey0,$inout0 | |
413 | aes${dir} $rndkey0,$inout1 | |
414 | lea 32($key),$key | |
415 | aes${dir} $rndkey0,$inout2 | |
416 | aes${dir} $rndkey0,$inout3 | |
417 | aes${dir} $rndkey0,$inout4 | |
418 | aes${dir} $rndkey0,$inout5 | |
419 | $movkey ($key),$rndkey0 | |
420 | jnz .L${dir}_loop6 | |
421 | ||
422 | aes${dir} $rndkey1,$inout0 | |
423 | aes${dir} $rndkey1,$inout1 | |
424 | aes${dir} $rndkey1,$inout2 | |
425 | aes${dir} $rndkey1,$inout3 | |
426 | aes${dir} $rndkey1,$inout4 | |
427 | aes${dir} $rndkey1,$inout5 | |
428 | aes${dir}last $rndkey0,$inout0 | |
429 | aes${dir}last $rndkey0,$inout1 | |
430 | aes${dir}last $rndkey0,$inout2 | |
431 | aes${dir}last $rndkey0,$inout3 | |
432 | aes${dir}last $rndkey0,$inout4 | |
433 | aes${dir}last $rndkey0,$inout5 | |
434 | ret | |
435 | .size _aesni_${dir}rypt6,.-_aesni_${dir}rypt6 | |
436 | ___ | |
437 | } | |
438 | sub aesni_generate8 { | |
439 | my $dir=shift; | |
440 | # As already mentioned it takes in $key and $rounds, which are *not* | |
441 | # preserved. $inout[0-7] is cipher/clear text... | |
442 | $code.=<<___; | |
443 | .type _aesni_${dir}rypt8,\@abi-omnipotent | |
444 | .align 16 | |
445 | _aesni_${dir}rypt8: | |
446 | $movkey ($key),$rndkey0 | |
447 | shr \$1,$rounds | |
448 | $movkey 16($key),$rndkey1 | |
449 | lea 32($key),$key | |
450 | xorps $rndkey0,$inout0 | |
451 | xorps $rndkey0,$inout1 | |
452 | aes${dir} $rndkey1,$inout0 | |
453 | pxor $rndkey0,$inout2 | |
454 | aes${dir} $rndkey1,$inout1 | |
455 | pxor $rndkey0,$inout3 | |
456 | aes${dir} $rndkey1,$inout2 | |
457 | pxor $rndkey0,$inout4 | |
458 | aes${dir} $rndkey1,$inout3 | |
459 | pxor $rndkey0,$inout5 | |
460 | dec $rounds | |
461 | aes${dir} $rndkey1,$inout4 | |
462 | pxor $rndkey0,$inout6 | |
463 | aes${dir} $rndkey1,$inout5 | |
464 | pxor $rndkey0,$inout7 | |
465 | $movkey ($key),$rndkey0 | |
466 | aes${dir} $rndkey1,$inout6 | |
467 | aes${dir} $rndkey1,$inout7 | |
468 | $movkey 16($key),$rndkey1 | |
469 | jmp .L${dir}_loop8_enter | |
470 | .align 16 | |
471 | .L${dir}_loop8: | |
472 | aes${dir} $rndkey1,$inout0 | |
473 | aes${dir} $rndkey1,$inout1 | |
474 | dec $rounds | |
475 | aes${dir} $rndkey1,$inout2 | |
476 | aes${dir} $rndkey1,$inout3 | |
477 | aes${dir} $rndkey1,$inout4 | |
478 | aes${dir} $rndkey1,$inout5 | |
479 | aes${dir} $rndkey1,$inout6 | |
480 | aes${dir} $rndkey1,$inout7 | |
481 | $movkey 16($key),$rndkey1 | |
482 | .L${dir}_loop8_enter: # happens to be 16-byte aligned | |
483 | aes${dir} $rndkey0,$inout0 | |
484 | aes${dir} $rndkey0,$inout1 | |
485 | lea 32($key),$key | |
486 | aes${dir} $rndkey0,$inout2 | |
487 | aes${dir} $rndkey0,$inout3 | |
488 | aes${dir} $rndkey0,$inout4 | |
489 | aes${dir} $rndkey0,$inout5 | |
490 | aes${dir} $rndkey0,$inout6 | |
491 | aes${dir} $rndkey0,$inout7 | |
492 | $movkey ($key),$rndkey0 | |
493 | jnz .L${dir}_loop8 | |
494 | ||
495 | aes${dir} $rndkey1,$inout0 | |
496 | aes${dir} $rndkey1,$inout1 | |
497 | aes${dir} $rndkey1,$inout2 | |
498 | aes${dir} $rndkey1,$inout3 | |
499 | aes${dir} $rndkey1,$inout4 | |
500 | aes${dir} $rndkey1,$inout5 | |
501 | aes${dir} $rndkey1,$inout6 | |
502 | aes${dir} $rndkey1,$inout7 | |
503 | aes${dir}last $rndkey0,$inout0 | |
504 | aes${dir}last $rndkey0,$inout1 | |
505 | aes${dir}last $rndkey0,$inout2 | |
506 | aes${dir}last $rndkey0,$inout3 | |
507 | aes${dir}last $rndkey0,$inout4 | |
508 | aes${dir}last $rndkey0,$inout5 | |
509 | aes${dir}last $rndkey0,$inout6 | |
510 | aes${dir}last $rndkey0,$inout7 | |
511 | ret | |
512 | .size _aesni_${dir}rypt8,.-_aesni_${dir}rypt8 | |
513 | ___ | |
514 | } | |
d608b4d6 AP |
515 | &aesni_generate3("enc") if ($PREFIX eq "aesni"); |
516 | &aesni_generate3("dec"); | |
517 | &aesni_generate4("enc") if ($PREFIX eq "aesni"); | |
518 | &aesni_generate4("dec"); | |
f8501464 AP |
519 | &aesni_generate6("enc") if ($PREFIX eq "aesni"); |
520 | &aesni_generate6("dec"); | |
521 | &aesni_generate8("enc") if ($PREFIX eq "aesni"); | |
522 | &aesni_generate8("dec"); | |
d64a7232 AP |
523 | \f |
524 | if ($PREFIX eq "aesni") { | |
6c83629b | 525 | ######################################################################## |
d64a7232 AP |
526 | # void aesni_ecb_encrypt (const void *in, void *out, |
527 | # size_t length, const AES_KEY *key, | |
528 | # int enc); | |
529 | $code.=<<___; | |
530 | .globl aesni_ecb_encrypt | |
531 | .type aesni_ecb_encrypt,\@function,5 | |
532 | .align 16 | |
533 | aesni_ecb_encrypt: | |
d64a7232 | 534 | and \$-16,$len |
f8501464 AP |
535 | jz .Lecb_ret |
536 | ||
537 | mov 240($key),$rounds # key->rounds | |
538 | $movkey ($key),$rndkey0 | |
d64a7232 | 539 | mov $key,$key_ # backup $key |
d64a7232 | 540 | mov $rounds,$rnds_ # backup $rounds |
d7d119a3 | 541 | test %r8d,%r8d # 5th argument |
d64a7232 AP |
542 | jz .Lecb_decrypt |
543 | #--------------------------- ECB ENCRYPT ------------------------------# | |
f8501464 AP |
544 | cmp \$0x80,$len |
545 | jb .Lecb_enc_tail | |
546 | ||
547 | movdqu ($inp),$inout0 | |
548 | movdqu 0x10($inp),$inout1 | |
549 | movdqu 0x20($inp),$inout2 | |
550 | movdqu 0x30($inp),$inout3 | |
551 | movdqu 0x40($inp),$inout4 | |
552 | movdqu 0x50($inp),$inout5 | |
553 | movdqu 0x60($inp),$inout6 | |
554 | movdqu 0x70($inp),$inout7 | |
555 | lea 0x80($inp),$inp | |
556 | sub \$0x80,$len | |
557 | jmp .Lecb_enc_loop8_enter | |
d64a7232 | 558 | .align 16 |
f8501464 | 559 | .Lecb_enc_loop8: |
d7d119a3 | 560 | movups $inout0,($out) |
f8501464 AP |
561 | mov $key_,$key # restore $key |
562 | movdqu ($inp),$inout0 | |
d64a7232 | 563 | mov $rnds_,$rounds # restore $rounds |
d7d119a3 | 564 | movups $inout1,0x10($out) |
f8501464 AP |
565 | movdqu 0x10($inp),$inout1 |
566 | movups $inout2,0x20($out) | |
567 | movdqu 0x20($inp),$inout2 | |
568 | movups $inout3,0x30($out) | |
569 | movdqu 0x30($inp),$inout3 | |
570 | movups $inout4,0x40($out) | |
571 | movdqu 0x40($inp),$inout4 | |
572 | movups $inout5,0x50($out) | |
573 | movdqu 0x50($inp),$inout5 | |
574 | movups $inout6,0x60($out) | |
575 | movdqu 0x60($inp),$inout6 | |
576 | movups $inout7,0x70($out) | |
577 | lea 0x80($out),$out | |
578 | movdqu 0x70($inp),$inout7 | |
579 | lea 0x80($inp),$inp | |
580 | .Lecb_enc_loop8_enter: | |
581 | ||
582 | call _aesni_encrypt8 | |
583 | ||
584 | sub \$0x80,$len | |
585 | jnc .Lecb_enc_loop8 | |
586 | ||
587 | movups $inout0,($out) | |
d64a7232 | 588 | mov $key_,$key # restore $key |
f8501464 AP |
589 | movups $inout1,0x10($out) |
590 | mov $rnds_,$rounds # restore $rounds | |
d7d119a3 | 591 | movups $inout2,0x20($out) |
f8501464 AP |
592 | movups $inout3,0x30($out) |
593 | movups $inout4,0x40($out) | |
594 | movups $inout5,0x50($out) | |
595 | movups $inout6,0x60($out) | |
596 | movups $inout7,0x70($out) | |
597 | lea 0x80($out),$out | |
598 | add \$0x80,$len | |
599 | jz .Lecb_ret | |
d64a7232 | 600 | |
6c83629b | 601 | .Lecb_enc_tail: |
6c83629b | 602 | movups ($inp),$inout0 |
d7d119a3 | 603 | cmp \$0x20,$len |
6c83629b | 604 | jb .Lecb_enc_one |
d64a7232 AP |
605 | movups 0x10($inp),$inout1 |
606 | je .Lecb_enc_two | |
d64a7232 | 607 | movups 0x20($inp),$inout2 |
f8501464 AP |
608 | cmp \$0x40,$len |
609 | jb .Lecb_enc_three | |
d64a7232 | 610 | movups 0x30($inp),$inout3 |
f8501464 AP |
611 | je .Lecb_enc_four |
612 | movups 0x40($inp),$inout4 | |
613 | cmp \$0x60,$len | |
614 | jb .Lecb_enc_five | |
615 | movups 0x50($inp),$inout5 | |
616 | je .Lecb_enc_six | |
617 | movdqu 0x60($inp),$inout6 | |
618 | call _aesni_encrypt8 | |
d64a7232 AP |
619 | movups $inout0,($out) |
620 | movups $inout1,0x10($out) | |
621 | movups $inout2,0x20($out) | |
622 | movups $inout3,0x30($out) | |
f8501464 AP |
623 | movups $inout4,0x40($out) |
624 | movups $inout5,0x50($out) | |
625 | movups $inout6,0x60($out) | |
d64a7232 AP |
626 | jmp .Lecb_ret |
627 | .align 16 | |
628 | .Lecb_enc_one: | |
629 | ___ | |
d608b4d6 | 630 | &aesni_generate1("enc",$key,$rounds); |
d64a7232 AP |
631 | $code.=<<___; |
632 | movups $inout0,($out) | |
633 | jmp .Lecb_ret | |
634 | .align 16 | |
635 | .Lecb_enc_two: | |
f8501464 | 636 | xorps $inout2,$inout2 |
d608b4d6 | 637 | call _aesni_encrypt3 |
d64a7232 AP |
638 | movups $inout0,($out) |
639 | movups $inout1,0x10($out) | |
640 | jmp .Lecb_ret | |
641 | .align 16 | |
642 | .Lecb_enc_three: | |
d608b4d6 | 643 | call _aesni_encrypt3 |
d64a7232 AP |
644 | movups $inout0,($out) |
645 | movups $inout1,0x10($out) | |
646 | movups $inout2,0x20($out) | |
647 | jmp .Lecb_ret | |
f8501464 AP |
648 | .align 16 |
649 | .Lecb_enc_four: | |
650 | call _aesni_encrypt4 | |
651 | movups $inout0,($out) | |
652 | movups $inout1,0x10($out) | |
653 | movups $inout2,0x20($out) | |
654 | movups $inout3,0x30($out) | |
655 | jmp .Lecb_ret | |
656 | .align 16 | |
657 | .Lecb_enc_five: | |
658 | xorps $inout5,$inout5 | |
659 | call _aesni_encrypt6 | |
660 | movups $inout0,($out) | |
661 | movups $inout1,0x10($out) | |
662 | movups $inout2,0x20($out) | |
663 | movups $inout3,0x30($out) | |
664 | movups $inout4,0x40($out) | |
665 | jmp .Lecb_ret | |
666 | .align 16 | |
667 | .Lecb_enc_six: | |
668 | call _aesni_encrypt6 | |
669 | movups $inout0,($out) | |
670 | movups $inout1,0x10($out) | |
671 | movups $inout2,0x20($out) | |
672 | movups $inout3,0x30($out) | |
673 | movups $inout4,0x40($out) | |
674 | movups $inout5,0x50($out) | |
675 | jmp .Lecb_ret | |
d64a7232 AP |
676 | \f#--------------------------- ECB DECRYPT ------------------------------# |
677 | .align 16 | |
678 | .Lecb_decrypt: | |
f8501464 AP |
679 | cmp \$0x80,$len |
680 | jb .Lecb_dec_tail | |
681 | ||
682 | movdqu ($inp),$inout0 | |
683 | movdqu 0x10($inp),$inout1 | |
684 | movdqu 0x20($inp),$inout2 | |
685 | movdqu 0x30($inp),$inout3 | |
686 | movdqu 0x40($inp),$inout4 | |
687 | movdqu 0x50($inp),$inout5 | |
688 | movdqu 0x60($inp),$inout6 | |
689 | movdqu 0x70($inp),$inout7 | |
690 | lea 0x80($inp),$inp | |
691 | sub \$0x80,$len | |
692 | jmp .Lecb_dec_loop8_enter | |
d64a7232 | 693 | .align 16 |
f8501464 | 694 | .Lecb_dec_loop8: |
d7d119a3 | 695 | movups $inout0,($out) |
f8501464 AP |
696 | mov $key_,$key # restore $key |
697 | movdqu ($inp),$inout0 | |
d64a7232 | 698 | mov $rnds_,$rounds # restore $rounds |
d7d119a3 | 699 | movups $inout1,0x10($out) |
f8501464 AP |
700 | movdqu 0x10($inp),$inout1 |
701 | movups $inout2,0x20($out) | |
702 | movdqu 0x20($inp),$inout2 | |
703 | movups $inout3,0x30($out) | |
704 | movdqu 0x30($inp),$inout3 | |
705 | movups $inout4,0x40($out) | |
706 | movdqu 0x40($inp),$inout4 | |
707 | movups $inout5,0x50($out) | |
708 | movdqu 0x50($inp),$inout5 | |
709 | movups $inout6,0x60($out) | |
710 | movdqu 0x60($inp),$inout6 | |
711 | movups $inout7,0x70($out) | |
712 | lea 0x80($out),$out | |
713 | movdqu 0x70($inp),$inout7 | |
714 | lea 0x80($inp),$inp | |
715 | .Lecb_dec_loop8_enter: | |
716 | ||
717 | call _aesni_decrypt8 | |
718 | ||
719 | $movkey ($key_),$rndkey0 | |
720 | sub \$0x80,$len | |
721 | jnc .Lecb_dec_loop8 | |
722 | ||
723 | movups $inout0,($out) | |
d64a7232 | 724 | mov $key_,$key # restore $key |
f8501464 AP |
725 | movups $inout1,0x10($out) |
726 | mov $rnds_,$rounds # restore $rounds | |
d7d119a3 | 727 | movups $inout2,0x20($out) |
f8501464 AP |
728 | movups $inout3,0x30($out) |
729 | movups $inout4,0x40($out) | |
730 | movups $inout5,0x50($out) | |
731 | movups $inout6,0x60($out) | |
732 | movups $inout7,0x70($out) | |
733 | lea 0x80($out),$out | |
734 | add \$0x80,$len | |
735 | jz .Lecb_ret | |
d64a7232 | 736 | |
6c83629b | 737 | .Lecb_dec_tail: |
6c83629b | 738 | movups ($inp),$inout0 |
d7d119a3 | 739 | cmp \$0x20,$len |
6c83629b | 740 | jb .Lecb_dec_one |
d64a7232 AP |
741 | movups 0x10($inp),$inout1 |
742 | je .Lecb_dec_two | |
d64a7232 | 743 | movups 0x20($inp),$inout2 |
f8501464 AP |
744 | cmp \$0x40,$len |
745 | jb .Lecb_dec_three | |
d64a7232 | 746 | movups 0x30($inp),$inout3 |
f8501464 AP |
747 | je .Lecb_dec_four |
748 | movups 0x40($inp),$inout4 | |
749 | cmp \$0x60,$len | |
750 | jb .Lecb_dec_five | |
751 | movups 0x50($inp),$inout5 | |
752 | je .Lecb_dec_six | |
753 | movups 0x60($inp),$inout6 | |
754 | $movkey ($key),$rndkey0 | |
755 | call _aesni_decrypt8 | |
d64a7232 AP |
756 | movups $inout0,($out) |
757 | movups $inout1,0x10($out) | |
758 | movups $inout2,0x20($out) | |
759 | movups $inout3,0x30($out) | |
f8501464 AP |
760 | movups $inout4,0x40($out) |
761 | movups $inout5,0x50($out) | |
762 | movups $inout6,0x60($out) | |
d64a7232 AP |
763 | jmp .Lecb_ret |
764 | .align 16 | |
765 | .Lecb_dec_one: | |
766 | ___ | |
d608b4d6 | 767 | &aesni_generate1("dec",$key,$rounds); |
d64a7232 AP |
768 | $code.=<<___; |
769 | movups $inout0,($out) | |
770 | jmp .Lecb_ret | |
771 | .align 16 | |
772 | .Lecb_dec_two: | |
f8501464 | 773 | xorps $inout2,$inout2 |
d608b4d6 | 774 | call _aesni_decrypt3 |
d64a7232 AP |
775 | movups $inout0,($out) |
776 | movups $inout1,0x10($out) | |
777 | jmp .Lecb_ret | |
778 | .align 16 | |
779 | .Lecb_dec_three: | |
d608b4d6 | 780 | call _aesni_decrypt3 |
d64a7232 AP |
781 | movups $inout0,($out) |
782 | movups $inout1,0x10($out) | |
783 | movups $inout2,0x20($out) | |
f8501464 AP |
784 | jmp .Lecb_ret |
785 | .align 16 | |
786 | .Lecb_dec_four: | |
787 | call _aesni_decrypt4 | |
788 | movups $inout0,($out) | |
789 | movups $inout1,0x10($out) | |
790 | movups $inout2,0x20($out) | |
791 | movups $inout3,0x30($out) | |
792 | jmp .Lecb_ret | |
793 | .align 16 | |
794 | .Lecb_dec_five: | |
795 | xorps $inout5,$inout5 | |
796 | call _aesni_decrypt6 | |
797 | movups $inout0,($out) | |
798 | movups $inout1,0x10($out) | |
799 | movups $inout2,0x20($out) | |
800 | movups $inout3,0x30($out) | |
801 | movups $inout4,0x40($out) | |
802 | jmp .Lecb_ret | |
803 | .align 16 | |
804 | .Lecb_dec_six: | |
805 | call _aesni_decrypt6 | |
806 | movups $inout0,($out) | |
807 | movups $inout1,0x10($out) | |
808 | movups $inout2,0x20($out) | |
809 | movups $inout3,0x30($out) | |
810 | movups $inout4,0x40($out) | |
811 | movups $inout5,0x50($out) | |
d64a7232 AP |
812 | |
813 | .Lecb_ret: | |
d64a7232 AP |
814 | ret |
815 | .size aesni_ecb_encrypt,.-aesni_ecb_encrypt | |
816 | ___ | |
d7d119a3 AP |
817 | \f |
818 | { | |
6c83629b | 819 | ###################################################################### |
d7d119a3 AP |
820 | # void aesni_ccm64_[en|de]crypt_blocks (const void *in, void *out, |
821 | # size_t blocks, const AES_KEY *key, | |
822 | # const char *ivec,char *cmac); | |
6c83629b | 823 | # |
d7d119a3 AP |
824 | # Handles only complete blocks, operates on 64-bit counter and |
825 | # does not update *ivec! Nor does it finalize CMAC value | |
826 | # (see engine/eng_aesni.c for details) | |
827 | # | |
828 | { | |
829 | my $cmac="%r9"; # 6th argument | |
830 | ||
267b481c AP |
831 | my $increment="%xmm6"; |
832 | my $bswap_mask="%xmm7"; | |
d7d119a3 AP |
833 | |
834 | $code.=<<___; | |
835 | .globl aesni_ccm64_encrypt_blocks | |
836 | .type aesni_ccm64_encrypt_blocks,\@function,6 | |
837 | .align 16 | |
838 | aesni_ccm64_encrypt_blocks: | |
839 | ___ | |
840 | $code.=<<___ if ($win64); | |
841 | lea -0x58(%rsp),%rsp | |
842 | movaps %xmm6,(%rsp) | |
843 | movaps %xmm7,0x10(%rsp) | |
844 | movaps %xmm8,0x20(%rsp) | |
845 | movaps %xmm9,0x30(%rsp) | |
846 | .Lccm64_enc_body: | |
847 | ___ | |
848 | $code.=<<___; | |
267b481c | 849 | mov 240($key),$rounds # key->rounds |
d7d119a3 | 850 | movdqu ($ivp),$iv |
d7d119a3 AP |
851 | movdqa .Lincrement64(%rip),$increment |
852 | movdqa .Lbswap_mask(%rip),$bswap_mask | |
d7d119a3 | 853 | |
267b481c AP |
854 | shr \$1,$rounds |
855 | lea 0($key),$key_ | |
856 | movdqu ($cmac),$inout1 | |
d7d119a3 | 857 | movdqa $iv,$inout0 |
267b481c | 858 | mov $rounds,$rnds_ |
9ee5916d | 859 | pshufb $bswap_mask,$iv |
267b481c AP |
860 | jmp .Lccm64_enc_outer |
861 | .align 16 | |
d7d119a3 | 862 | .Lccm64_enc_outer: |
267b481c | 863 | $movkey ($key_),$rndkey0 |
d7d119a3 | 864 | mov $rnds_,$rounds |
267b481c | 865 | movups ($inp),$in0 # load inp |
d7d119a3 | 866 | |
267b481c AP |
867 | xorps $rndkey0,$inout0 # counter |
868 | $movkey 16($key_),$rndkey1 | |
869 | xorps $in0,$rndkey0 | |
870 | lea 32($key_),$key | |
871 | xorps $rndkey0,$inout1 # cmac^=inp | |
f8501464 AP |
872 | $movkey ($key),$rndkey0 |
873 | ||
874 | .Lccm64_enc2_loop: | |
875 | aesenc $rndkey1,$inout0 | |
876 | dec $rounds | |
877 | aesenc $rndkey1,$inout1 | |
878 | $movkey 16($key),$rndkey1 | |
879 | aesenc $rndkey0,$inout0 | |
880 | lea 32($key),$key | |
881 | aesenc $rndkey0,$inout1 | |
882 | $movkey 0($key),$rndkey0 | |
883 | jnz .Lccm64_enc2_loop | |
884 | aesenc $rndkey1,$inout0 | |
885 | aesenc $rndkey1,$inout1 | |
267b481c | 886 | paddq $increment,$iv |
f8501464 AP |
887 | aesenclast $rndkey0,$inout0 |
888 | aesenclast $rndkey0,$inout1 | |
d7d119a3 | 889 | |
d7d119a3 AP |
890 | dec $len |
891 | lea 16($inp),$inp | |
f8501464 | 892 | xorps $inout0,$in0 # inp ^= E(iv) |
d7d119a3 | 893 | movdqa $iv,$inout0 |
f8501464 | 894 | movups $in0,($out) # save output |
d7d119a3 | 895 | lea 16($out),$out |
9ee5916d | 896 | pshufb $bswap_mask,$inout0 |
d7d119a3 AP |
897 | jnz .Lccm64_enc_outer |
898 | ||
f8501464 | 899 | movups $inout1,($cmac) |
d7d119a3 AP |
900 | ___ |
901 | $code.=<<___ if ($win64); | |
902 | movaps (%rsp),%xmm6 | |
903 | movaps 0x10(%rsp),%xmm7 | |
904 | movaps 0x20(%rsp),%xmm8 | |
905 | movaps 0x30(%rsp),%xmm9 | |
906 | lea 0x58(%rsp),%rsp | |
907 | .Lccm64_enc_ret: | |
908 | ___ | |
909 | $code.=<<___; | |
910 | ret | |
911 | .size aesni_ccm64_encrypt_blocks,.-aesni_ccm64_encrypt_blocks | |
912 | ___ | |
913 | ###################################################################### | |
914 | $code.=<<___; | |
915 | .globl aesni_ccm64_decrypt_blocks | |
916 | .type aesni_ccm64_decrypt_blocks,\@function,6 | |
917 | .align 16 | |
918 | aesni_ccm64_decrypt_blocks: | |
919 | ___ | |
920 | $code.=<<___ if ($win64); | |
921 | lea -0x58(%rsp),%rsp | |
922 | movaps %xmm6,(%rsp) | |
923 | movaps %xmm7,0x10(%rsp) | |
924 | movaps %xmm8,0x20(%rsp) | |
925 | movaps %xmm9,0x30(%rsp) | |
926 | .Lccm64_dec_body: | |
927 | ___ | |
928 | $code.=<<___; | |
267b481c AP |
929 | mov 240($key),$rounds # key->rounds |
930 | movups ($ivp),$iv | |
d7d119a3 AP |
931 | movdqu ($cmac),$inout1 |
932 | movdqa .Lincrement64(%rip),$increment | |
933 | movdqa .Lbswap_mask(%rip),$bswap_mask | |
934 | ||
267b481c | 935 | movaps $iv,$inout0 |
d7d119a3 AP |
936 | mov $rounds,$rnds_ |
937 | mov $key,$key_ | |
267b481c | 938 | pshufb $bswap_mask,$iv |
d7d119a3 AP |
939 | ___ |
940 | &aesni_generate1("enc",$key,$rounds); | |
941 | $code.=<<___; | |
f8501464 | 942 | movups ($inp),$in0 # load inp |
267b481c | 943 | paddq $increment,$iv |
f8501464 | 944 | lea 16($inp),$inp |
267b481c AP |
945 | jmp .Lccm64_dec_outer |
946 | .align 16 | |
947 | .Lccm64_dec_outer: | |
948 | xorps $inout0,$in0 # inp ^= E(iv) | |
949 | movdqa $iv,$inout0 | |
d7d119a3 | 950 | mov $rnds_,$rounds |
267b481c | 951 | movups $in0,($out) # save output |
d7d119a3 | 952 | lea 16($out),$out |
9ee5916d | 953 | pshufb $bswap_mask,$inout0 |
d7d119a3 | 954 | |
f8501464 | 955 | sub \$1,$len |
d7d119a3 AP |
956 | jz .Lccm64_dec_break |
957 | ||
267b481c | 958 | $movkey ($key_),$rndkey0 |
f8501464 | 959 | shr \$1,$rounds |
267b481c | 960 | $movkey 16($key_),$rndkey1 |
f8501464 | 961 | xorps $rndkey0,$in0 |
267b481c | 962 | lea 32($key_),$key |
f8501464 AP |
963 | xorps $rndkey0,$inout0 |
964 | xorps $in0,$inout1 # cmac^=out | |
965 | $movkey ($key),$rndkey0 | |
d7d119a3 | 966 | |
f8501464 AP |
967 | .Lccm64_dec2_loop: |
968 | aesenc $rndkey1,$inout0 | |
969 | dec $rounds | |
970 | aesenc $rndkey1,$inout1 | |
971 | $movkey 16($key),$rndkey1 | |
972 | aesenc $rndkey0,$inout0 | |
973 | lea 32($key),$key | |
974 | aesenc $rndkey0,$inout1 | |
975 | $movkey 0($key),$rndkey0 | |
976 | jnz .Lccm64_dec2_loop | |
267b481c AP |
977 | movups ($inp),$in0 # load inp |
978 | paddq $increment,$iv | |
f8501464 AP |
979 | aesenc $rndkey1,$inout0 |
980 | aesenc $rndkey1,$inout1 | |
267b481c | 981 | lea 16($inp),$inp |
f8501464 | 982 | aesenclast $rndkey0,$inout0 |
267b481c | 983 | aesenclast $rndkey0,$inout1 |
d7d119a3 AP |
984 | jmp .Lccm64_dec_outer |
985 | ||
986 | .align 16 | |
987 | .Lccm64_dec_break: | |
267b481c | 988 | #xorps $in0,$inout1 # cmac^=out |
d7d119a3 | 989 | ___ |
267b481c | 990 | &aesni_generate1("enc",$key_,$rounds,$inout1,$in0); |
d7d119a3 | 991 | $code.=<<___; |
f8501464 | 992 | movups $inout1,($cmac) |
d7d119a3 AP |
993 | ___ |
994 | $code.=<<___ if ($win64); | |
995 | movaps (%rsp),%xmm6 | |
996 | movaps 0x10(%rsp),%xmm7 | |
997 | movaps 0x20(%rsp),%xmm8 | |
998 | movaps 0x30(%rsp),%xmm9 | |
999 | lea 0x58(%rsp),%rsp | |
1000 | .Lccm64_dec_ret: | |
1001 | ___ | |
1002 | $code.=<<___; | |
1003 | ret | |
f8501464 AP |
1004 | .size aesni_ccm64_decrypt_blocks,.-aesni_ccm64_decrypt_blocks |
1005 | ___ | |
1006 | }\f | |
1007 | ###################################################################### | |
1008 | # void aesni_ctr32_encrypt_blocks (const void *in, void *out, | |
1009 | # size_t blocks, const AES_KEY *key, | |
1010 | # const char *ivec); | |
1011 | # | |
1012 | # Handles only complete blocks, operates on 32-bit counter and | |
1013 | # does not update *ivec! (see engine/eng_aesni.c for details) | |
1014 | # | |
1015 | { | |
9282c335 AP |
1016 | my ($in0,$in1,$in2,$in3,$one,$ivec)=map("%xmm$_",(10..15)); |
1017 | my $len_="%r9"; | |
f8501464 AP |
1018 | |
1019 | $code.=<<___; | |
1020 | .globl aesni_ctr32_encrypt_blocks | |
1021 | .type aesni_ctr32_encrypt_blocks,\@function,5 | |
1022 | .align 16 | |
1023 | aesni_ctr32_encrypt_blocks: | |
1024 | ___ | |
1025 | $code.=<<___ if ($win64); | |
9282c335 AP |
1026 | lea -0xa8(%rsp),%rsp |
1027 | movaps %xmm6,0x00(%rsp) | |
1028 | movaps %xmm7,0x10(%rsp) | |
1029 | movaps %xmm8,0x20(%rsp) | |
1030 | movaps %xmm9,0x30(%rsp) | |
1031 | movaps %xmm10,0x40(%rsp) | |
1032 | movaps %xmm11,0x50(%rsp) | |
1033 | movaps %xmm12,0x60(%rsp) | |
1034 | movaps %xmm13,0x70(%rsp) | |
1035 | movaps %xmm14,0x80(%rsp) | |
1036 | movaps %xmm15,0x90(%rsp) | |
f8501464 AP |
1037 | .Lctr32_body: |
1038 | ___ | |
1039 | $code.=<<___; | |
1040 | cmp \$1,$len | |
1041 | je .Lctr32_one_shortcut | |
1042 | ||
9282c335 AP |
1043 | movzb 15($ivp),%rax # counter LSB |
1044 | mov $len,$len_ # backup $len | |
1045 | mov 240($key),$rnds_ # key->rounds | |
1046 | mov $key,$key_ # backup $key | |
f8501464 | 1047 | movdqu ($ivp),$ivec |
9282c335 AP |
1048 | neg %rax |
1049 | movdqa .Lincrement1(%rip),$one | |
1050 | add \$256,%rax # steps to closest overflow | |
f8501464 | 1051 | |
9282c335 AP |
1052 | .Lctr32_grandloop: |
1053 | cmp %rax,$len | |
1054 | cmova %rax,$len | |
1055 | mov $rnds_,$rounds # restore $rounds | |
1056 | sub $len,$len_ | |
1057 | ||
1058 | cmp \$8,$len | |
f8501464 | 1059 | jb .Lctr32_tail |
9282c335 AP |
1060 | |
1061 | $movkey ($key_),$rndkey0 | |
f8501464 | 1062 | shr \$1,$rounds |
9282c335 AP |
1063 | shr \$1,$rnds_ |
1064 | sub \$8,$len | |
1065 | jmp .Lctr32_loop8 | |
f8501464 AP |
1066 | |
1067 | .align 16 | |
9282c335 | 1068 | .Lctr32_loop8: |
f8501464 | 1069 | $movkey 16($key_),$rndkey1 |
9282c335 AP |
1070 | movdqa $rndkey0,$inout0 |
1071 | movdqa $rndkey0,$inout1 | |
1072 | pxor $ivec,$inout0 | |
1073 | paddb $one,$ivec | |
1074 | movdqa $rndkey0,$inout2 | |
1075 | aesenc $rndkey1,$inout0 | |
1076 | pxor $ivec,$inout1 | |
1077 | paddb $one,$ivec | |
1078 | lea 32($key_),$key | |
1079 | movdqa $rndkey0,$inout3 | |
1080 | aesenc $rndkey1,$inout1 | |
1081 | pxor $ivec,$inout2 | |
1082 | paddb $one,$ivec | |
1083 | movdqa $rndkey0,$inout4 | |
1084 | aesenc $rndkey1,$inout2 | |
1085 | pxor $ivec,$inout3 | |
1086 | paddb $one,$ivec | |
1087 | movdqa $rndkey0,$inout5 | |
1088 | aesenc $rndkey1,$inout3 | |
1089 | pxor $ivec,$inout4 | |
1090 | paddb $one,$ivec | |
1091 | movdqa $rndkey0,$inout6 | |
1092 | aesenc $rndkey1,$inout4 | |
1093 | pxor $ivec,$inout5 | |
1094 | paddb $one,$ivec | |
1095 | movdqa $rndkey0,$inout7 | |
1096 | aesenc $rndkey1,$inout5 | |
1097 | pxor $ivec,$inout6 | |
1098 | paddb $one,$ivec | |
1099 | $movkey ($key),$rndkey0 | |
1100 | aesenc $rndkey1,$inout6 | |
1101 | pxor $ivec,$inout7 | |
1102 | paddb $one,$ivec | |
1103 | dec $rounds | |
1104 | aesenc $rndkey1,$inout7 | |
1105 | $movkey 16($key),$rndkey1 | |
1106 | movups ($inp),$in0 # load input | |
1107 | movups 0x10($inp),$in1 | |
1108 | movups 0x20($inp),$in2 | |
1109 | movups 0x30($inp),$in3 | |
1110 | ||
1111 | call .Lenc_loop8_enter | |
1112 | ||
1113 | xorps $in0,$inout0 # xor | |
1114 | movups 0x40($inp),$in0 | |
1115 | xorps $in1,$inout1 | |
1116 | movups 0x50($inp),$in1 | |
1117 | xorps $in2,$inout2 | |
1118 | movups 0x60($inp),$in2 | |
1119 | xorps $in3,$inout3 | |
1120 | movups 0x70($inp),$in3 | |
1121 | lea 0x80($inp),$inp | |
1122 | xorps $in0,$inout4 | |
1123 | movups $inout0,($out) # store output | |
1124 | xorps $in1,$inout5 | |
1125 | movups $inout1,0x10($out) | |
1126 | xorps $in2,$inout6 | |
1127 | movups $inout2,0x20($out) | |
1128 | xorps $in3,$inout7 | |
1129 | movups $inout3,0x30($out) | |
1130 | movups $inout4,0x40($out) | |
1131 | movups $inout5,0x50($out) | |
1132 | movups $inout6,0x60($out) | |
1133 | movups $inout7,0x70($out) | |
1134 | lea 0x80($out),$out | |
1135 | ||
1136 | $movkey ($key_),$rndkey0 | |
f8501464 | 1137 | mov $rnds_,$rounds |
9282c335 AP |
1138 | sub \$8,$len |
1139 | jnc .Lctr32_loop8 | |
f8501464 | 1140 | |
f8501464 | 1141 | lea 1($rounds,$rounds),$rounds # restore original value |
9282c335 AP |
1142 | lea 1($rnds_,$rnds_),$rnds_ # restore original value |
1143 | add \$8,$len | |
1144 | jz .Lctr32_done | |
f8501464 AP |
1145 | |
1146 | .Lctr32_tail: | |
9282c335 AP |
1147 | mov $key_,$key # restore $key |
1148 | movdqa $ivec,$inout0 | |
1149 | paddb $one,$ivec | |
f8501464 AP |
1150 | movups ($inp),$in0 |
1151 | cmp \$2,$len | |
1152 | jb .Lctr32_one | |
1153 | ||
9282c335 AP |
1154 | movdqa $ivec,$inout1 |
1155 | paddb $one,$ivec | |
f8501464 AP |
1156 | movups 0x10($inp),$in1 |
1157 | je .Lctr32_two | |
1158 | ||
9282c335 AP |
1159 | movdqa $ivec,$inout2 |
1160 | paddb $one,$ivec | |
f8501464 AP |
1161 | movups 0x20($inp),$in2 |
1162 | cmp \$4,$len | |
1163 | jb .Lctr32_three | |
1164 | ||
9282c335 AP |
1165 | movdqa $ivec,$inout3 |
1166 | paddb $one,$ivec | |
f8501464 AP |
1167 | movups 0x30($inp),$in3 |
1168 | je .Lctr32_four | |
1169 | ||
9282c335 AP |
1170 | movdqa $ivec,$inout4 |
1171 | paddb $one,$ivec | |
1172 | cmp \$6,$len | |
1173 | jb .Lctr32_five | |
f8501464 | 1174 | |
9282c335 AP |
1175 | movdqa $ivec,$inout5 |
1176 | paddb $one,$ivec | |
1177 | je .Lctr32_six | |
f8501464 | 1178 | |
9282c335 AP |
1179 | movdqa $ivec,$inout6 |
1180 | paddb $one,$ivec | |
1181 | xorps $inout7,$inout7 | |
1182 | ||
1183 | call _aesni_encrypt8 | |
1184 | ||
1185 | xorps $in0,$inout0 # xor | |
1186 | movups 0x40($inp),$in0 | |
1187 | xorps $in1,$inout1 | |
1188 | movups 0x50($inp),$in1 | |
1189 | xorps $in2,$inout2 | |
1190 | movups 0x60($inp),$in2 | |
1191 | lea 0x70($inp),$inp | |
1192 | xorps $in3,$inout3 | |
1193 | movups $inout0,($out) # store output | |
1194 | xorps $in0,$inout4 | |
1195 | movups $inout1,0x10($out) | |
1196 | xorps $in1,$inout5 | |
1197 | movups $inout2,0x20($out) | |
1198 | xorps $in2,$inout6 | |
1199 | movups $inout3,0x30($out) | |
1200 | movups $inout4,0x40($out) | |
1201 | movups $inout5,0x50($out) | |
1202 | movups $inout6,0x60($out) | |
1203 | lea 0x70($out),$out | |
f8501464 AP |
1204 | jmp .Lctr32_done |
1205 | ||
1206 | .align 16 | |
1207 | .Lctr32_one_shortcut: | |
1208 | movups ($ivp),$inout0 | |
9282c335 | 1209 | xor $len_,$len_ |
f8501464 AP |
1210 | movups ($inp),$in0 |
1211 | mov 240($key),$rounds # key->rounds | |
1212 | .Lctr32_one: | |
1213 | ___ | |
1214 | &aesni_generate1("enc",$key,$rounds); | |
1215 | $code.=<<___; | |
9282c335 AP |
1216 | xorps $in0,$inout0 |
1217 | lea 0x10($inp),$inp | |
1218 | movups $inout0,($out) | |
1219 | lea 0x10($out),$out | |
f8501464 AP |
1220 | jmp .Lctr32_done |
1221 | ||
1222 | .align 16 | |
1223 | .Lctr32_two: | |
1224 | xorps $inout2,$inout2 | |
1225 | call _aesni_encrypt3 | |
9282c335 AP |
1226 | xorps $in0,$inout0 # xor |
1227 | lea 0x20($inp),$inp | |
1228 | xorps $in1,$inout1 | |
1229 | movups $inout0,($out) # store output | |
1230 | movups $inout1,0x10($out) | |
1231 | lea 0x20($out),$out | |
f8501464 AP |
1232 | jmp .Lctr32_done |
1233 | ||
1234 | .align 16 | |
1235 | .Lctr32_three: | |
1236 | call _aesni_encrypt3 | |
9282c335 AP |
1237 | xorps $in0,$inout0 # xor |
1238 | lea 0x30($inp),$inp | |
1239 | xorps $in1,$inout1 | |
1240 | movups $inout0,($out) # store output | |
1241 | xorps $in2,$inout2 | |
1242 | movups $inout1,0x10($out) | |
1243 | movups $inout2,0x20($out) | |
1244 | lea 0x30($out),$out | |
f8501464 AP |
1245 | jmp .Lctr32_done |
1246 | ||
1247 | .align 16 | |
1248 | .Lctr32_four: | |
1249 | call _aesni_encrypt4 | |
9282c335 AP |
1250 | xorps $in0,$inout0 # xor |
1251 | lea 0x40($inp),$inp | |
1252 | xorps $in1,$inout1 | |
1253 | movups $inout0,($out) # store output | |
1254 | xorps $in2,$inout2 | |
1255 | movups $inout1,0x10($out) | |
1256 | xorps $in3,$inout3 | |
1257 | movups $inout2,0x20($out) | |
1258 | movups $inout3,0x30($out) | |
1259 | lea 0x40($out),$out | |
1260 | jmp .Lctr32_done | |
1261 | ||
1262 | .align 16 | |
1263 | .Lctr32_five: | |
1264 | xorps $inout5,$inout5 | |
1265 | call _aesni_encrypt6 | |
1266 | xorps $in0,$inout0 # xor | |
1267 | movups 0x40($inp),$in0 | |
1268 | lea 0x50($inp),$inp | |
1269 | xorps $in1,$inout1 | |
1270 | movups $inout0,($out) # store output | |
1271 | xorps $in2,$inout2 | |
1272 | movups $inout1,0x10($out) | |
1273 | xorps $in3,$inout3 | |
1274 | movups $inout2,0x20($out) | |
1275 | xorps $in0,$inout4 | |
1276 | movups $inout3,0x30($out) | |
1277 | movups $inout4,0x40($out) | |
1278 | lea 0x50($out),$out | |
1279 | jmp .Lctr32_done | |
1280 | ||
1281 | .align 16 | |
1282 | .Lctr32_six: | |
1283 | call _aesni_encrypt6 | |
1284 | xorps $in0,$inout0 # xor | |
1285 | movups 0x40($inp),$in0 | |
1286 | xorps $in1,$inout1 | |
1287 | movups 0x50($inp),$in1 | |
1288 | lea 0x60($inp),$inp | |
1289 | xorps $in2,$inout2 | |
1290 | movups $inout0,($out) # store output | |
1291 | xorps $in3,$inout3 | |
1292 | movups $inout1,0x10($out) | |
1293 | xorps $in0,$inout4 | |
1294 | movups $inout2,0x20($out) | |
1295 | xorps $in1,$inout5 | |
1296 | movups $inout3,0x30($out) | |
1297 | movups $inout4,0x40($out) | |
1298 | movups $inout5,0x50($out) | |
1299 | lea 0x60($out),$out | |
f8501464 AP |
1300 | |
1301 | .Lctr32_done: | |
9282c335 AP |
1302 | test $len_,$len_ |
1303 | jz .Lctr32_really_done | |
1304 | ||
1305 | movdqa .Lbswap_mask(%rip),$rndkey1 | |
1306 | pshufb $rndkey1,$ivec | |
1307 | psrldq \$14,$one # 256 | |
1308 | paddd $one,$ivec | |
1309 | pslldq \$14,$one | |
1310 | pshufb $rndkey1,$ivec | |
1311 | mov $len_,$len | |
1312 | mov \$256,%rax | |
1313 | jmp .Lctr32_grandloop | |
1314 | ||
1315 | .Lctr32_really_done: | |
f8501464 AP |
1316 | ___ |
1317 | $code.=<<___ if ($win64); | |
9282c335 AP |
1318 | movaps 0x00(%rsp),%xmm6 |
1319 | movaps 0x10(%rsp),%xmm7 | |
1320 | movaps 0x20(%rsp),%xmm8 | |
1321 | movaps 0x30(%rsp),%xmm9 | |
1322 | movaps 0x40(%rsp),%xmm10 | |
1323 | movaps 0x50(%rsp),%xmm11 | |
1324 | movaps 0x60(%rsp),%xmm12 | |
1325 | movaps 0x70(%rsp),%xmm13 | |
1326 | movaps 0x80(%rsp),%xmm14 | |
1327 | movaps 0x90(%rsp),%xmm15 | |
1328 | lea 0xa8(%rsp),%rsp | |
f8501464 AP |
1329 | ___ |
1330 | $code.=<<___; | |
6a40ebe8 | 1331 | .Lctr32_ret: |
f8501464 AP |
1332 | ret |
1333 | .size aesni_ctr32_encrypt_blocks,.-aesni_ctr32_encrypt_blocks | |
1334 | ___ | |
1335 | } | |
1336 | \f | |
1337 | ###################################################################### | |
1338 | # void aesni_xts_[en|de]crypt(const char *inp,char *out,size_t len, | |
1339 | # const AES_KEY *key1, const AES_KEY *key2 | |
1340 | # const unsigned char iv[16]); | |
1341 | # | |
1342 | { | |
1343 | my @tweak=map("%xmm$_",(10..15)); | |
1344 | my ($twmask,$twres,$twtmp)=("%xmm8","%xmm9",@tweak[4]); | |
1345 | my ($key2,$ivp,$len_)=("%r8","%r9","%r9"); | |
6a40ebe8 | 1346 | my $frame_size = 0x60 + ($win64?160:0); |
f8501464 AP |
1347 | |
1348 | $code.=<<___; | |
1349 | .globl aesni_xts_encrypt | |
1350 | .type aesni_xts_encrypt,\@function,6 | |
1351 | .align 16 | |
1352 | aesni_xts_encrypt: | |
6a40ebe8 AP |
1353 | lea (%rsp),%rax |
1354 | push %rbp | |
1355 | sub \$$frame_size,%rsp | |
1356 | and \$-16,%rsp # Linux kernel stack can be incorrectly seeded | |
f8501464 AP |
1357 | ___ |
1358 | $code.=<<___ if ($win64); | |
1359 | movaps %xmm6,0x60(%rsp) | |
1360 | movaps %xmm7,0x70(%rsp) | |
1361 | movaps %xmm8,0x80(%rsp) | |
1362 | movaps %xmm9,0x90(%rsp) | |
1363 | movaps %xmm10,0xa0(%rsp) | |
1364 | movaps %xmm11,0xb0(%rsp) | |
1365 | movaps %xmm12,0xc0(%rsp) | |
1366 | movaps %xmm13,0xd0(%rsp) | |
1367 | movaps %xmm14,0xe0(%rsp) | |
1368 | movaps %xmm15,0xf0(%rsp) | |
1369 | .Lxts_enc_body: | |
1370 | ___ | |
1371 | $code.=<<___; | |
6a40ebe8 | 1372 | lea -8(%rax),%rbp |
f8501464 AP |
1373 | movups ($ivp),@tweak[5] # load clear-text tweak |
1374 | mov 240(%r8),$rounds # key2->rounds | |
1375 | mov 240($key),$rnds_ # key1->rounds | |
1376 | ___ | |
1377 | # generate the tweak | |
1378 | &aesni_generate1("enc",$key2,$rounds,@tweak[5]); | |
1379 | $code.=<<___; | |
1380 | mov $key,$key_ # backup $key | |
1381 | mov $rnds_,$rounds # backup $rounds | |
1382 | mov $len,$len_ # backup $len | |
1383 | and \$-16,$len | |
1384 | ||
1385 | movdqa .Lxts_magic(%rip),$twmask | |
1386 | pxor $twtmp,$twtmp | |
1387 | pcmpgtd @tweak[5],$twtmp # broadcast upper bits | |
1388 | ___ | |
1389 | for ($i=0;$i<4;$i++) { | |
1390 | $code.=<<___; | |
1391 | pshufd \$0x13,$twtmp,$twres | |
1392 | pxor $twtmp,$twtmp | |
1393 | movdqa @tweak[5],@tweak[$i] | |
1394 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1395 | pand $twmask,$twres # isolate carry and residue | |
1396 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1397 | pxor $twres,@tweak[5] | |
1398 | ___ | |
1399 | } | |
1400 | $code.=<<___; | |
1401 | sub \$16*6,$len | |
1402 | jc .Lxts_enc_short | |
1403 | ||
1404 | shr \$1,$rounds | |
1405 | sub \$1,$rounds | |
1406 | mov $rounds,$rnds_ | |
1407 | jmp .Lxts_enc_grandloop | |
1408 | ||
1409 | .align 16 | |
1410 | .Lxts_enc_grandloop: | |
1411 | pshufd \$0x13,$twtmp,$twres | |
1412 | movdqa @tweak[5],@tweak[4] | |
1413 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1414 | movdqu `16*0`($inp),$inout0 # load input | |
1415 | pand $twmask,$twres # isolate carry and residue | |
1416 | movdqu `16*1`($inp),$inout1 | |
1417 | pxor $twres,@tweak[5] | |
1418 | ||
1419 | movdqu `16*2`($inp),$inout2 | |
1420 | pxor @tweak[0],$inout0 # input^=tweak | |
1421 | movdqu `16*3`($inp),$inout3 | |
1422 | pxor @tweak[1],$inout1 | |
1423 | movdqu `16*4`($inp),$inout4 | |
1424 | pxor @tweak[2],$inout2 | |
1425 | movdqu `16*5`($inp),$inout5 | |
1426 | lea `16*6`($inp),$inp | |
1427 | pxor @tweak[3],$inout3 | |
1428 | $movkey ($key_),$rndkey0 | |
1429 | pxor @tweak[4],$inout4 | |
1430 | pxor @tweak[5],$inout5 | |
1431 | ||
1432 | # inline _aesni_encrypt6 and interleave first and last rounds | |
1433 | # with own code... | |
1434 | $movkey 16($key_),$rndkey1 | |
1435 | pxor $rndkey0,$inout0 | |
1436 | pxor $rndkey0,$inout1 | |
1437 | movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks | |
1438 | aesenc $rndkey1,$inout0 | |
1439 | lea 32($key_),$key | |
1440 | pxor $rndkey0,$inout2 | |
1441 | movdqa @tweak[1],`16*1`(%rsp) | |
1442 | aesenc $rndkey1,$inout1 | |
1443 | pxor $rndkey0,$inout3 | |
1444 | movdqa @tweak[2],`16*2`(%rsp) | |
1445 | aesenc $rndkey1,$inout2 | |
1446 | pxor $rndkey0,$inout4 | |
1447 | movdqa @tweak[3],`16*3`(%rsp) | |
1448 | aesenc $rndkey1,$inout3 | |
1449 | pxor $rndkey0,$inout5 | |
1450 | $movkey ($key),$rndkey0 | |
1451 | dec $rounds | |
1452 | movdqa @tweak[4],`16*4`(%rsp) | |
1453 | aesenc $rndkey1,$inout4 | |
1454 | movdqa @tweak[5],`16*5`(%rsp) | |
1455 | aesenc $rndkey1,$inout5 | |
1456 | pxor $twtmp,$twtmp | |
1457 | pcmpgtd @tweak[5],$twtmp | |
1458 | jmp .Lxts_enc_loop6_enter | |
1459 | ||
1460 | .align 16 | |
1461 | .Lxts_enc_loop6: | |
1462 | aesenc $rndkey1,$inout0 | |
1463 | aesenc $rndkey1,$inout1 | |
1464 | dec $rounds | |
1465 | aesenc $rndkey1,$inout2 | |
1466 | aesenc $rndkey1,$inout3 | |
1467 | aesenc $rndkey1,$inout4 | |
1468 | aesenc $rndkey1,$inout5 | |
1469 | .Lxts_enc_loop6_enter: | |
1470 | $movkey 16($key),$rndkey1 | |
1471 | aesenc $rndkey0,$inout0 | |
1472 | aesenc $rndkey0,$inout1 | |
1473 | lea 32($key),$key | |
1474 | aesenc $rndkey0,$inout2 | |
1475 | aesenc $rndkey0,$inout3 | |
1476 | aesenc $rndkey0,$inout4 | |
1477 | aesenc $rndkey0,$inout5 | |
1478 | $movkey ($key),$rndkey0 | |
1479 | jnz .Lxts_enc_loop6 | |
1480 | ||
1481 | pshufd \$0x13,$twtmp,$twres | |
1482 | pxor $twtmp,$twtmp | |
1483 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1484 | aesenc $rndkey1,$inout0 | |
1485 | pand $twmask,$twres # isolate carry and residue | |
1486 | aesenc $rndkey1,$inout1 | |
1487 | pcmpgtd @tweak[5],$twtmp # broadcast upper bits | |
1488 | aesenc $rndkey1,$inout2 | |
1489 | pxor $twres,@tweak[5] | |
1490 | aesenc $rndkey1,$inout3 | |
1491 | aesenc $rndkey1,$inout4 | |
1492 | aesenc $rndkey1,$inout5 | |
1493 | $movkey 16($key),$rndkey1 | |
1494 | ||
1495 | pshufd \$0x13,$twtmp,$twres | |
1496 | pxor $twtmp,$twtmp | |
1497 | movdqa @tweak[5],@tweak[0] | |
1498 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1499 | aesenc $rndkey0,$inout0 | |
1500 | pand $twmask,$twres # isolate carry and residue | |
1501 | aesenc $rndkey0,$inout1 | |
1502 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1503 | aesenc $rndkey0,$inout2 | |
1504 | pxor $twres,@tweak[5] | |
1505 | aesenc $rndkey0,$inout3 | |
1506 | aesenc $rndkey0,$inout4 | |
1507 | aesenc $rndkey0,$inout5 | |
1508 | $movkey 32($key),$rndkey0 | |
1509 | ||
1510 | pshufd \$0x13,$twtmp,$twres | |
1511 | pxor $twtmp,$twtmp | |
1512 | movdqa @tweak[5],@tweak[1] | |
1513 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1514 | aesenc $rndkey1,$inout0 | |
1515 | pand $twmask,$twres # isolate carry and residue | |
1516 | aesenc $rndkey1,$inout1 | |
1517 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1518 | aesenc $rndkey1,$inout2 | |
1519 | pxor $twres,@tweak[5] | |
1520 | aesenc $rndkey1,$inout3 | |
1521 | aesenc $rndkey1,$inout4 | |
1522 | aesenc $rndkey1,$inout5 | |
1523 | ||
1524 | pshufd \$0x13,$twtmp,$twres | |
1525 | pxor $twtmp,$twtmp | |
1526 | movdqa @tweak[5],@tweak[2] | |
1527 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1528 | aesenclast $rndkey0,$inout0 | |
1529 | pand $twmask,$twres # isolate carry and residue | |
1530 | aesenclast $rndkey0,$inout1 | |
1531 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1532 | aesenclast $rndkey0,$inout2 | |
1533 | pxor $twres,@tweak[5] | |
1534 | aesenclast $rndkey0,$inout3 | |
1535 | aesenclast $rndkey0,$inout4 | |
1536 | aesenclast $rndkey0,$inout5 | |
1537 | ||
1538 | pshufd \$0x13,$twtmp,$twres | |
1539 | pxor $twtmp,$twtmp | |
1540 | movdqa @tweak[5],@tweak[3] | |
1541 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1542 | xorps `16*0`(%rsp),$inout0 # output^=tweak | |
1543 | pand $twmask,$twres # isolate carry and residue | |
1544 | xorps `16*1`(%rsp),$inout1 | |
1545 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1546 | pxor $twres,@tweak[5] | |
1547 | ||
1548 | xorps `16*2`(%rsp),$inout2 | |
1549 | movups $inout0,`16*0`($out) # write output | |
1550 | xorps `16*3`(%rsp),$inout3 | |
1551 | movups $inout1,`16*1`($out) | |
1552 | xorps `16*4`(%rsp),$inout4 | |
1553 | movups $inout2,`16*2`($out) | |
1554 | xorps `16*5`(%rsp),$inout5 | |
1555 | movups $inout3,`16*3`($out) | |
1556 | mov $rnds_,$rounds # restore $rounds | |
1557 | movups $inout4,`16*4`($out) | |
1558 | movups $inout5,`16*5`($out) | |
1559 | lea `16*6`($out),$out | |
1560 | sub \$16*6,$len | |
1561 | jnc .Lxts_enc_grandloop | |
1562 | ||
1563 | lea 3($rounds,$rounds),$rounds # restore original value | |
1564 | mov $key_,$key # restore $key | |
1565 | mov $rounds,$rnds_ # backup $rounds | |
1566 | ||
1567 | .Lxts_enc_short: | |
1568 | add \$16*6,$len | |
1569 | jz .Lxts_enc_done | |
1570 | ||
1571 | cmp \$0x20,$len | |
1572 | jb .Lxts_enc_one | |
1573 | je .Lxts_enc_two | |
1574 | ||
1575 | cmp \$0x40,$len | |
1576 | jb .Lxts_enc_three | |
1577 | je .Lxts_enc_four | |
1578 | ||
1579 | pshufd \$0x13,$twtmp,$twres | |
1580 | movdqa @tweak[5],@tweak[4] | |
1581 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1582 | movdqu ($inp),$inout0 | |
1583 | pand $twmask,$twres # isolate carry and residue | |
1584 | movdqu 16*1($inp),$inout1 | |
1585 | pxor $twres,@tweak[5] | |
1586 | ||
1587 | movdqu 16*2($inp),$inout2 | |
1588 | pxor @tweak[0],$inout0 | |
1589 | movdqu 16*3($inp),$inout3 | |
1590 | pxor @tweak[1],$inout1 | |
1591 | movdqu 16*4($inp),$inout4 | |
1592 | lea 16*5($inp),$inp | |
1593 | pxor @tweak[2],$inout2 | |
1594 | pxor @tweak[3],$inout3 | |
1595 | pxor @tweak[4],$inout4 | |
1596 | ||
1597 | call _aesni_encrypt6 | |
1598 | ||
1599 | xorps @tweak[0],$inout0 | |
1600 | movdqa @tweak[5],@tweak[0] | |
1601 | xorps @tweak[1],$inout1 | |
1602 | xorps @tweak[2],$inout2 | |
1603 | movdqu $inout0,($out) | |
1604 | xorps @tweak[3],$inout3 | |
1605 | movdqu $inout1,16*1($out) | |
1606 | xorps @tweak[4],$inout4 | |
1607 | movdqu $inout2,16*2($out) | |
1608 | movdqu $inout3,16*3($out) | |
1609 | movdqu $inout4,16*4($out) | |
1610 | lea 16*5($out),$out | |
1611 | jmp .Lxts_enc_done | |
1612 | ||
1613 | .align 16 | |
1614 | .Lxts_enc_one: | |
1615 | movups ($inp),$inout0 | |
1616 | lea 16*1($inp),$inp | |
1617 | xorps @tweak[0],$inout0 | |
1618 | ___ | |
1619 | &aesni_generate1("enc",$key,$rounds); | |
1620 | $code.=<<___; | |
1621 | xorps @tweak[0],$inout0 | |
1622 | movdqa @tweak[1],@tweak[0] | |
1623 | movups $inout0,($out) | |
1624 | lea 16*1($out),$out | |
1625 | jmp .Lxts_enc_done | |
1626 | ||
1627 | .align 16 | |
1628 | .Lxts_enc_two: | |
1629 | movups ($inp),$inout0 | |
1630 | movups 16($inp),$inout1 | |
1631 | lea 32($inp),$inp | |
1632 | xorps @tweak[0],$inout0 | |
1633 | xorps @tweak[1],$inout1 | |
1634 | ||
1635 | call _aesni_encrypt3 | |
1636 | ||
1637 | xorps @tweak[0],$inout0 | |
1638 | movdqa @tweak[2],@tweak[0] | |
1639 | xorps @tweak[1],$inout1 | |
1640 | movups $inout0,($out) | |
1641 | movups $inout1,16*1($out) | |
1642 | lea 16*2($out),$out | |
1643 | jmp .Lxts_enc_done | |
1644 | ||
1645 | .align 16 | |
1646 | .Lxts_enc_three: | |
1647 | movups ($inp),$inout0 | |
1648 | movups 16*1($inp),$inout1 | |
1649 | movups 16*2($inp),$inout2 | |
1650 | lea 16*3($inp),$inp | |
1651 | xorps @tweak[0],$inout0 | |
1652 | xorps @tweak[1],$inout1 | |
1653 | xorps @tweak[2],$inout2 | |
1654 | ||
1655 | call _aesni_encrypt3 | |
1656 | ||
1657 | xorps @tweak[0],$inout0 | |
1658 | movdqa @tweak[3],@tweak[0] | |
1659 | xorps @tweak[1],$inout1 | |
1660 | xorps @tweak[2],$inout2 | |
1661 | movups $inout0,($out) | |
1662 | movups $inout1,16*1($out) | |
1663 | movups $inout2,16*2($out) | |
1664 | lea 16*3($out),$out | |
1665 | jmp .Lxts_enc_done | |
1666 | ||
1667 | .align 16 | |
1668 | .Lxts_enc_four: | |
1669 | movups ($inp),$inout0 | |
1670 | movups 16*1($inp),$inout1 | |
1671 | movups 16*2($inp),$inout2 | |
1672 | xorps @tweak[0],$inout0 | |
1673 | movups 16*3($inp),$inout3 | |
1674 | lea 16*4($inp),$inp | |
1675 | xorps @tweak[1],$inout1 | |
1676 | xorps @tweak[2],$inout2 | |
1677 | xorps @tweak[3],$inout3 | |
1678 | ||
1679 | call _aesni_encrypt4 | |
1680 | ||
1681 | xorps @tweak[0],$inout0 | |
1682 | movdqa @tweak[5],@tweak[0] | |
1683 | xorps @tweak[1],$inout1 | |
1684 | xorps @tweak[2],$inout2 | |
1685 | movups $inout0,($out) | |
1686 | xorps @tweak[3],$inout3 | |
1687 | movups $inout1,16*1($out) | |
1688 | movups $inout2,16*2($out) | |
1689 | movups $inout3,16*3($out) | |
1690 | lea 16*4($out),$out | |
1691 | jmp .Lxts_enc_done | |
1692 | ||
1693 | .align 16 | |
1694 | .Lxts_enc_done: | |
1695 | and \$15,$len_ | |
1696 | jz .Lxts_enc_ret | |
1697 | mov $len_,$len | |
1698 | ||
1699 | .Lxts_enc_steal: | |
1700 | movzb ($inp),%eax # borrow $rounds ... | |
1701 | movzb -16($out),%ecx # ... and $key | |
1702 | lea 1($inp),$inp | |
1703 | mov %al,-16($out) | |
1704 | mov %cl,0($out) | |
1705 | lea 1($out),$out | |
1706 | sub \$1,$len | |
1707 | jnz .Lxts_enc_steal | |
1708 | ||
1709 | sub $len_,$out # rewind $out | |
1710 | mov $key_,$key # restore $key | |
1711 | mov $rnds_,$rounds # restore $rounds | |
1712 | ||
1713 | movups -16($out),$inout0 | |
1714 | xorps @tweak[0],$inout0 | |
1715 | ___ | |
1716 | &aesni_generate1("enc",$key,$rounds); | |
1717 | $code.=<<___; | |
1718 | xorps @tweak[0],$inout0 | |
1719 | movups $inout0,-16($out) | |
1720 | ||
1721 | .Lxts_enc_ret: | |
1722 | ___ | |
1723 | $code.=<<___ if ($win64); | |
1724 | movaps 0x60(%rsp),%xmm6 | |
1725 | movaps 0x70(%rsp),%xmm7 | |
1726 | movaps 0x80(%rsp),%xmm8 | |
1727 | movaps 0x90(%rsp),%xmm9 | |
1728 | movaps 0xa0(%rsp),%xmm10 | |
1729 | movaps 0xb0(%rsp),%xmm11 | |
1730 | movaps 0xc0(%rsp),%xmm12 | |
1731 | movaps 0xd0(%rsp),%xmm13 | |
1732 | movaps 0xe0(%rsp),%xmm14 | |
1733 | movaps 0xf0(%rsp),%xmm15 | |
1734 | ___ | |
1735 | $code.=<<___; | |
6a40ebe8 AP |
1736 | lea (%rbp),%rsp |
1737 | pop %rbp | |
f8501464 AP |
1738 | .Lxts_enc_epilogue: |
1739 | ret | |
1740 | .size aesni_xts_encrypt,.-aesni_xts_encrypt | |
d7d119a3 | 1741 | ___ |
6c83629b AP |
1742 | |
1743 | $code.=<<___; | |
f8501464 AP |
1744 | .globl aesni_xts_decrypt |
1745 | .type aesni_xts_decrypt,\@function,6 | |
6c83629b | 1746 | .align 16 |
f8501464 | 1747 | aesni_xts_decrypt: |
6a40ebe8 AP |
1748 | lea (%rsp),%rax |
1749 | push %rbp | |
1750 | sub \$$frame_size,%rsp | |
1751 | and \$-16,%rsp # Linux kernel stack can be incorrectly seeded | |
6c83629b AP |
1752 | ___ |
1753 | $code.=<<___ if ($win64); | |
f8501464 AP |
1754 | movaps %xmm6,0x60(%rsp) |
1755 | movaps %xmm7,0x70(%rsp) | |
1756 | movaps %xmm8,0x80(%rsp) | |
1757 | movaps %xmm9,0x90(%rsp) | |
1758 | movaps %xmm10,0xa0(%rsp) | |
1759 | movaps %xmm11,0xb0(%rsp) | |
1760 | movaps %xmm12,0xc0(%rsp) | |
1761 | movaps %xmm13,0xd0(%rsp) | |
1762 | movaps %xmm14,0xe0(%rsp) | |
1763 | movaps %xmm15,0xf0(%rsp) | |
1764 | .Lxts_dec_body: | |
6c83629b AP |
1765 | ___ |
1766 | $code.=<<___; | |
6a40ebe8 | 1767 | lea -8(%rax),%rbp |
f8501464 AP |
1768 | movups ($ivp),@tweak[5] # load clear-text tweak |
1769 | mov 240($key2),$rounds # key2->rounds | |
1770 | mov 240($key),$rnds_ # key1->rounds | |
1771 | ___ | |
1772 | # generate the tweak | |
1773 | &aesni_generate1("enc",$key2,$rounds,@tweak[5]); | |
1774 | $code.=<<___; | |
1775 | xor %eax,%eax # if ($len%16) len-=16; | |
1776 | test \$15,$len | |
1777 | setnz %al | |
1778 | shl \$4,%rax | |
1779 | sub %rax,$len | |
1780 | ||
1781 | mov $key,$key_ # backup $key | |
1782 | mov $rnds_,$rounds # backup $rounds | |
1783 | mov $len,$len_ # backup $len | |
1784 | and \$-16,$len | |
6c83629b | 1785 | |
f8501464 AP |
1786 | movdqa .Lxts_magic(%rip),$twmask |
1787 | pxor $twtmp,$twtmp | |
1788 | pcmpgtd @tweak[5],$twtmp # broadcast upper bits | |
1789 | ___ | |
1790 | for ($i=0;$i<4;$i++) { | |
1791 | $code.=<<___; | |
1792 | pshufd \$0x13,$twtmp,$twres | |
1793 | pxor $twtmp,$twtmp | |
1794 | movdqa @tweak[5],@tweak[$i] | |
1795 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1796 | pand $twmask,$twres # isolate carry and residue | |
1797 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1798 | pxor $twres,@tweak[5] | |
1799 | ___ | |
1800 | } | |
1801 | $code.=<<___; | |
1802 | sub \$16*6,$len | |
1803 | jc .Lxts_dec_short | |
6c83629b | 1804 | |
f8501464 AP |
1805 | shr \$1,$rounds |
1806 | sub \$1,$rounds | |
6c83629b | 1807 | mov $rounds,$rnds_ |
f8501464 | 1808 | jmp .Lxts_dec_grandloop |
6c83629b | 1809 | |
f8501464 AP |
1810 | .align 16 |
1811 | .Lxts_dec_grandloop: | |
1812 | pshufd \$0x13,$twtmp,$twres | |
1813 | movdqa @tweak[5],@tweak[4] | |
1814 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1815 | movdqu `16*0`($inp),$inout0 # load input | |
1816 | pand $twmask,$twres # isolate carry and residue | |
1817 | movdqu `16*1`($inp),$inout1 | |
1818 | pxor $twres,@tweak[5] | |
1819 | ||
1820 | movdqu `16*2`($inp),$inout2 | |
1821 | pxor @tweak[0],$inout0 # input^=tweak | |
1822 | movdqu `16*3`($inp),$inout3 | |
1823 | pxor @tweak[1],$inout1 | |
1824 | movdqu `16*4`($inp),$inout4 | |
1825 | pxor @tweak[2],$inout2 | |
1826 | movdqu `16*5`($inp),$inout5 | |
1827 | lea `16*6`($inp),$inp | |
1828 | pxor @tweak[3],$inout3 | |
1829 | $movkey ($key_),$rndkey0 | |
1830 | pxor @tweak[4],$inout4 | |
1831 | pxor @tweak[5],$inout5 | |
1832 | ||
1833 | # inline _aesni_decrypt6 and interleave first and last rounds | |
d7d119a3 | 1834 | # with own code... |
f8501464 AP |
1835 | $movkey 16($key_),$rndkey1 |
1836 | pxor $rndkey0,$inout0 | |
1837 | pxor $rndkey0,$inout1 | |
1838 | movdqa @tweak[0],`16*0`(%rsp) # put aside tweaks | |
1839 | aesdec $rndkey1,$inout0 | |
1840 | lea 32($key_),$key | |
1841 | pxor $rndkey0,$inout2 | |
1842 | movdqa @tweak[1],`16*1`(%rsp) | |
1843 | aesdec $rndkey1,$inout1 | |
1844 | pxor $rndkey0,$inout3 | |
1845 | movdqa @tweak[2],`16*2`(%rsp) | |
1846 | aesdec $rndkey1,$inout2 | |
1847 | pxor $rndkey0,$inout4 | |
1848 | movdqa @tweak[3],`16*3`(%rsp) | |
1849 | aesdec $rndkey1,$inout3 | |
1850 | pxor $rndkey0,$inout5 | |
1851 | $movkey ($key),$rndkey0 | |
1852 | dec $rounds | |
1853 | movdqa @tweak[4],`16*4`(%rsp) | |
1854 | aesdec $rndkey1,$inout4 | |
1855 | movdqa @tweak[5],`16*5`(%rsp) | |
1856 | aesdec $rndkey1,$inout5 | |
1857 | pxor $twtmp,$twtmp | |
1858 | pcmpgtd @tweak[5],$twtmp | |
1859 | jmp .Lxts_dec_loop6_enter | |
6c83629b | 1860 | |
d7d119a3 | 1861 | .align 16 |
f8501464 AP |
1862 | .Lxts_dec_loop6: |
1863 | aesdec $rndkey1,$inout0 | |
1864 | aesdec $rndkey1,$inout1 | |
d7d119a3 | 1865 | dec $rounds |
f8501464 AP |
1866 | aesdec $rndkey1,$inout2 |
1867 | aesdec $rndkey1,$inout3 | |
1868 | aesdec $rndkey1,$inout4 | |
1869 | aesdec $rndkey1,$inout5 | |
1870 | .Lxts_dec_loop6_enter: | |
d7d119a3 | 1871 | $movkey 16($key),$rndkey1 |
f8501464 AP |
1872 | aesdec $rndkey0,$inout0 |
1873 | aesdec $rndkey0,$inout1 | |
d7d119a3 | 1874 | lea 32($key),$key |
f8501464 AP |
1875 | aesdec $rndkey0,$inout2 |
1876 | aesdec $rndkey0,$inout3 | |
1877 | aesdec $rndkey0,$inout4 | |
1878 | aesdec $rndkey0,$inout5 | |
d7d119a3 | 1879 | $movkey ($key),$rndkey0 |
f8501464 AP |
1880 | jnz .Lxts_dec_loop6 |
1881 | ||
1882 | pshufd \$0x13,$twtmp,$twres | |
1883 | pxor $twtmp,$twtmp | |
1884 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1885 | aesdec $rndkey1,$inout0 | |
1886 | pand $twmask,$twres # isolate carry and residue | |
1887 | aesdec $rndkey1,$inout1 | |
1888 | pcmpgtd @tweak[5],$twtmp # broadcast upper bits | |
1889 | aesdec $rndkey1,$inout2 | |
1890 | pxor $twres,@tweak[5] | |
1891 | aesdec $rndkey1,$inout3 | |
1892 | aesdec $rndkey1,$inout4 | |
1893 | aesdec $rndkey1,$inout5 | |
1894 | $movkey 16($key),$rndkey1 | |
1895 | ||
1896 | pshufd \$0x13,$twtmp,$twres | |
1897 | pxor $twtmp,$twtmp | |
1898 | movdqa @tweak[5],@tweak[0] | |
1899 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1900 | aesdec $rndkey0,$inout0 | |
1901 | pand $twmask,$twres # isolate carry and residue | |
1902 | aesdec $rndkey0,$inout1 | |
1903 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1904 | aesdec $rndkey0,$inout2 | |
1905 | pxor $twres,@tweak[5] | |
1906 | aesdec $rndkey0,$inout3 | |
1907 | aesdec $rndkey0,$inout4 | |
1908 | aesdec $rndkey0,$inout5 | |
1909 | $movkey 32($key),$rndkey0 | |
1910 | ||
1911 | pshufd \$0x13,$twtmp,$twres | |
1912 | pxor $twtmp,$twtmp | |
1913 | movdqa @tweak[5],@tweak[1] | |
1914 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1915 | aesdec $rndkey1,$inout0 | |
1916 | pand $twmask,$twres # isolate carry and residue | |
1917 | aesdec $rndkey1,$inout1 | |
1918 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1919 | aesdec $rndkey1,$inout2 | |
1920 | pxor $twres,@tweak[5] | |
1921 | aesdec $rndkey1,$inout3 | |
1922 | aesdec $rndkey1,$inout4 | |
1923 | aesdec $rndkey1,$inout5 | |
1924 | ||
1925 | pshufd \$0x13,$twtmp,$twres | |
1926 | pxor $twtmp,$twtmp | |
1927 | movdqa @tweak[5],@tweak[2] | |
1928 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1929 | aesdeclast $rndkey0,$inout0 | |
1930 | pand $twmask,$twres # isolate carry and residue | |
1931 | aesdeclast $rndkey0,$inout1 | |
1932 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1933 | aesdeclast $rndkey0,$inout2 | |
1934 | pxor $twres,@tweak[5] | |
1935 | aesdeclast $rndkey0,$inout3 | |
1936 | aesdeclast $rndkey0,$inout4 | |
1937 | aesdeclast $rndkey0,$inout5 | |
1938 | ||
1939 | pshufd \$0x13,$twtmp,$twres | |
1940 | pxor $twtmp,$twtmp | |
1941 | movdqa @tweak[5],@tweak[3] | |
1942 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1943 | xorps `16*0`(%rsp),$inout0 # output^=tweak | |
1944 | pand $twmask,$twres # isolate carry and residue | |
1945 | xorps `16*1`(%rsp),$inout1 | |
1946 | pcmpgtd @tweak[5],$twtmp # broadcat upper bits | |
1947 | pxor $twres,@tweak[5] | |
1948 | ||
1949 | xorps `16*2`(%rsp),$inout2 | |
1950 | movups $inout0,`16*0`($out) # write output | |
1951 | xorps `16*3`(%rsp),$inout3 | |
1952 | movups $inout1,`16*1`($out) | |
1953 | xorps `16*4`(%rsp),$inout4 | |
1954 | movups $inout2,`16*2`($out) | |
1955 | xorps `16*5`(%rsp),$inout5 | |
1956 | movups $inout3,`16*3`($out) | |
1957 | mov $rnds_,$rounds # restore $rounds | |
1958 | movups $inout4,`16*4`($out) | |
1959 | movups $inout5,`16*5`($out) | |
1960 | lea `16*6`($out),$out | |
1961 | sub \$16*6,$len | |
1962 | jnc .Lxts_dec_grandloop | |
1963 | ||
1964 | lea 3($rounds,$rounds),$rounds # restore original value | |
1965 | mov $key_,$key # restore $key | |
1966 | mov $rounds,$rnds_ # backup $rounds | |
1967 | ||
1968 | .Lxts_dec_short: | |
1969 | add \$16*6,$len | |
1970 | jz .Lxts_dec_done | |
d7d119a3 | 1971 | |
f8501464 AP |
1972 | cmp \$0x20,$len |
1973 | jb .Lxts_dec_one | |
1974 | je .Lxts_dec_two | |
d7d119a3 | 1975 | |
f8501464 AP |
1976 | cmp \$0x40,$len |
1977 | jb .Lxts_dec_three | |
1978 | je .Lxts_dec_four | |
1979 | ||
1980 | pshufd \$0x13,$twtmp,$twres | |
1981 | movdqa @tweak[5],@tweak[4] | |
1982 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
1983 | movdqu ($inp),$inout0 | |
1984 | pand $twmask,$twres # isolate carry and residue | |
1985 | movdqu 16*1($inp),$inout1 | |
1986 | pxor $twres,@tweak[5] | |
1987 | ||
1988 | movdqu 16*2($inp),$inout2 | |
1989 | pxor @tweak[0],$inout0 | |
1990 | movdqu 16*3($inp),$inout3 | |
1991 | pxor @tweak[1],$inout1 | |
1992 | movdqu 16*4($inp),$inout4 | |
1993 | lea 16*5($inp),$inp | |
1994 | pxor @tweak[2],$inout2 | |
1995 | pxor @tweak[3],$inout3 | |
1996 | pxor @tweak[4],$inout4 | |
1997 | ||
1998 | call _aesni_decrypt6 | |
1999 | ||
2000 | xorps @tweak[0],$inout0 | |
2001 | xorps @tweak[1],$inout1 | |
2002 | xorps @tweak[2],$inout2 | |
2003 | movdqu $inout0,($out) | |
2004 | xorps @tweak[3],$inout3 | |
2005 | movdqu $inout1,16*1($out) | |
2006 | xorps @tweak[4],$inout4 | |
2007 | movdqu $inout2,16*2($out) | |
2008 | pxor $twtmp,$twtmp | |
2009 | movdqu $inout3,16*3($out) | |
2010 | pcmpgtd @tweak[5],$twtmp | |
2011 | movdqu $inout4,16*4($out) | |
2012 | lea 16*5($out),$out | |
2013 | pshufd \$0x13,$twtmp,@tweak[1] # $twres | |
2014 | and \$15,$len_ | |
2015 | jz .Lxts_dec_ret | |
2016 | ||
2017 | movdqa @tweak[5],@tweak[0] | |
2018 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
2019 | pand $twmask,@tweak[1] # isolate carry and residue | |
2020 | pxor @tweak[5],@tweak[1] | |
2021 | jmp .Lxts_dec_done2 | |
d7d119a3 | 2022 | |
f8501464 AP |
2023 | .align 16 |
2024 | .Lxts_dec_one: | |
2025 | movups ($inp),$inout0 | |
2026 | lea 16*1($inp),$inp | |
2027 | xorps @tweak[0],$inout0 | |
2028 | ___ | |
2029 | &aesni_generate1("dec",$key,$rounds); | |
2030 | $code.=<<___; | |
2031 | xorps @tweak[0],$inout0 | |
2032 | movdqa @tweak[1],@tweak[0] | |
2033 | movups $inout0,($out) | |
2034 | movdqa @tweak[2],@tweak[1] | |
2035 | lea 16*1($out),$out | |
2036 | jmp .Lxts_dec_done | |
6c83629b | 2037 | |
f8501464 AP |
2038 | .align 16 |
2039 | .Lxts_dec_two: | |
2040 | movups ($inp),$inout0 | |
2041 | movups 16($inp),$inout1 | |
2042 | lea 32($inp),$inp | |
2043 | xorps @tweak[0],$inout0 | |
2044 | xorps @tweak[1],$inout1 | |
6c83629b | 2045 | |
f8501464 | 2046 | call _aesni_decrypt3 |
6c83629b | 2047 | |
f8501464 AP |
2048 | xorps @tweak[0],$inout0 |
2049 | movdqa @tweak[2],@tweak[0] | |
2050 | xorps @tweak[1],$inout1 | |
2051 | movdqa @tweak[3],@tweak[1] | |
2052 | movups $inout0,($out) | |
2053 | movups $inout1,16*1($out) | |
2054 | lea 16*2($out),$out | |
2055 | jmp .Lxts_dec_done | |
6c83629b | 2056 | |
f8501464 AP |
2057 | .align 16 |
2058 | .Lxts_dec_three: | |
2059 | movups ($inp),$inout0 | |
2060 | movups 16*1($inp),$inout1 | |
2061 | movups 16*2($inp),$inout2 | |
2062 | lea 16*3($inp),$inp | |
2063 | xorps @tweak[0],$inout0 | |
2064 | xorps @tweak[1],$inout1 | |
2065 | xorps @tweak[2],$inout2 | |
6c83629b | 2066 | |
f8501464 | 2067 | call _aesni_decrypt3 |
6c83629b | 2068 | |
f8501464 AP |
2069 | xorps @tweak[0],$inout0 |
2070 | movdqa @tweak[3],@tweak[0] | |
2071 | xorps @tweak[1],$inout1 | |
2072 | movdqa @tweak[5],@tweak[1] | |
2073 | xorps @tweak[2],$inout2 | |
2074 | movups $inout0,($out) | |
2075 | movups $inout1,16*1($out) | |
2076 | movups $inout2,16*2($out) | |
2077 | lea 16*3($out),$out | |
2078 | jmp .Lxts_dec_done | |
6c83629b AP |
2079 | |
2080 | .align 16 | |
f8501464 AP |
2081 | .Lxts_dec_four: |
2082 | pshufd \$0x13,$twtmp,$twres | |
2083 | movdqa @tweak[5],@tweak[4] | |
2084 | paddq @tweak[5],@tweak[5] # psllq 1,$tweak | |
2085 | movups ($inp),$inout0 | |
2086 | pand $twmask,$twres # isolate carry and residue | |
2087 | movups 16*1($inp),$inout1 | |
2088 | pxor $twres,@tweak[5] | |
2089 | ||
2090 | movups 16*2($inp),$inout2 | |
2091 | xorps @tweak[0],$inout0 | |
2092 | movups 16*3($inp),$inout3 | |
2093 | lea 16*4($inp),$inp | |
2094 | xorps @tweak[1],$inout1 | |
2095 | xorps @tweak[2],$inout2 | |
2096 | xorps @tweak[3],$inout3 | |
2097 | ||
2098 | call _aesni_decrypt4 | |
2099 | ||
2100 | xorps @tweak[0],$inout0 | |
2101 | movdqa @tweak[4],@tweak[0] | |
2102 | xorps @tweak[1],$inout1 | |
2103 | movdqa @tweak[5],@tweak[1] | |
2104 | xorps @tweak[2],$inout2 | |
2105 | movups $inout0,($out) | |
2106 | xorps @tweak[3],$inout3 | |
2107 | movups $inout1,16*1($out) | |
2108 | movups $inout2,16*2($out) | |
2109 | movups $inout3,16*3($out) | |
2110 | lea 16*4($out),$out | |
2111 | jmp .Lxts_dec_done | |
6c83629b AP |
2112 | |
2113 | .align 16 | |
f8501464 AP |
2114 | .Lxts_dec_done: |
2115 | and \$15,$len_ | |
2116 | jz .Lxts_dec_ret | |
2117 | .Lxts_dec_done2: | |
2118 | mov $len_,$len | |
2119 | mov $key_,$key # restore $key | |
2120 | mov $rnds_,$rounds # restore $rounds | |
6c83629b | 2121 | |
f8501464 AP |
2122 | movups ($inp),$inout0 |
2123 | xorps @tweak[1],$inout0 | |
2124 | ___ | |
2125 | &aesni_generate1("dec",$key,$rounds); | |
2126 | $code.=<<___; | |
2127 | xorps @tweak[1],$inout0 | |
2128 | movups $inout0,($out) | |
2129 | ||
2130 | .Lxts_dec_steal: | |
2131 | movzb 16($inp),%eax # borrow $rounds ... | |
2132 | movzb ($out),%ecx # ... and $key | |
2133 | lea 1($inp),$inp | |
2134 | mov %al,($out) | |
2135 | mov %cl,16($out) | |
2136 | lea 1($out),$out | |
2137 | sub \$1,$len | |
2138 | jnz .Lxts_dec_steal | |
2139 | ||
2140 | sub $len_,$out # rewind $out | |
2141 | mov $key_,$key # restore $key | |
2142 | mov $rnds_,$rounds # restore $rounds | |
2143 | ||
2144 | movups ($out),$inout0 | |
2145 | xorps @tweak[0],$inout0 | |
6c83629b | 2146 | ___ |
f8501464 AP |
2147 | &aesni_generate1("dec",$key,$rounds); |
2148 | $code.=<<___; | |
2149 | xorps @tweak[0],$inout0 | |
2150 | movups $inout0,($out) | |
6c83629b | 2151 | |
f8501464 AP |
2152 | .Lxts_dec_ret: |
2153 | ___ | |
6c83629b | 2154 | $code.=<<___ if ($win64); |
f8501464 AP |
2155 | movaps 0x60(%rsp),%xmm6 |
2156 | movaps 0x70(%rsp),%xmm7 | |
2157 | movaps 0x80(%rsp),%xmm8 | |
2158 | movaps 0x90(%rsp),%xmm9 | |
2159 | movaps 0xa0(%rsp),%xmm10 | |
2160 | movaps 0xb0(%rsp),%xmm11 | |
2161 | movaps 0xc0(%rsp),%xmm12 | |
2162 | movaps 0xd0(%rsp),%xmm13 | |
2163 | movaps 0xe0(%rsp),%xmm14 | |
2164 | movaps 0xf0(%rsp),%xmm15 | |
6c83629b AP |
2165 | ___ |
2166 | $code.=<<___; | |
6a40ebe8 AP |
2167 | lea (%rbp),%rsp |
2168 | pop %rbp | |
f8501464 | 2169 | .Lxts_dec_epilogue: |
6c83629b | 2170 | ret |
f8501464 | 2171 | .size aesni_xts_decrypt,.-aesni_xts_decrypt |
6c83629b | 2172 | ___ |
f8501464 | 2173 | } }} |
d64a7232 | 2174 | \f |
6c83629b | 2175 | ######################################################################## |
d64a7232 AP |
2176 | # void $PREFIX_cbc_encrypt (const void *inp, void *out, |
2177 | # size_t length, const AES_KEY *key, | |
2178 | # unsigned char *ivp,const int enc); | |
f8501464 | 2179 | { |
6a40ebe8 | 2180 | my $frame_size = 0x10 + ($win64?0x40:0); # used in decrypt |
d64a7232 AP |
2181 | $code.=<<___; |
2182 | .globl ${PREFIX}_cbc_encrypt | |
2183 | .type ${PREFIX}_cbc_encrypt,\@function,6 | |
2184 | .align 16 | |
2185 | ${PREFIX}_cbc_encrypt: | |
2186 | test $len,$len # check length | |
2187 | jz .Lcbc_ret | |
d608b4d6 | 2188 | |
f8501464 | 2189 | mov 240($key),$rnds_ # key->rounds |
d64a7232 | 2190 | mov $key,$key_ # backup $key |
d608b4d6 | 2191 | test %r9d,%r9d # 6th argument |
d64a7232 AP |
2192 | jz .Lcbc_decrypt |
2193 | #--------------------------- CBC ENCRYPT ------------------------------# | |
f8501464 | 2194 | movups ($ivp),$inout0 # load iv as initial state |
d608b4d6 | 2195 | mov $rnds_,$rounds |
d7d119a3 | 2196 | cmp \$16,$len |
d64a7232 AP |
2197 | jb .Lcbc_enc_tail |
2198 | sub \$16,$len | |
2199 | jmp .Lcbc_enc_loop | |
d7d119a3 | 2200 | .align 16 |
d64a7232 | 2201 | .Lcbc_enc_loop: |
f8501464 | 2202 | movups ($inp),$inout1 # load input |
d64a7232 | 2203 | lea 16($inp),$inp |
f8501464 | 2204 | #xorps $inout1,$inout0 |
d64a7232 | 2205 | ___ |
f8501464 | 2206 | &aesni_generate1("enc",$key,$rounds,$inout0,$inout1); |
d64a7232 | 2207 | $code.=<<___; |
d608b4d6 AP |
2208 | mov $rnds_,$rounds # restore $rounds |
2209 | mov $key_,$key # restore $key | |
d7d119a3 AP |
2210 | movups $inout0,0($out) # store output |
2211 | lea 16($out),$out | |
2212 | sub \$16,$len | |
d64a7232 AP |
2213 | jnc .Lcbc_enc_loop |
2214 | add \$16,$len | |
2215 | jnz .Lcbc_enc_tail | |
d608b4d6 | 2216 | movups $inout0,($ivp) |
d64a7232 AP |
2217 | jmp .Lcbc_ret |
2218 | ||
2219 | .Lcbc_enc_tail: | |
2220 | mov $len,%rcx # zaps $key | |
2221 | xchg $inp,$out # $inp is %rsi and $out is %rdi now | |
2222 | .long 0x9066A4F3 # rep movsb | |
2223 | mov \$16,%ecx # zero tail | |
2224 | sub $len,%rcx | |
2225 | xor %eax,%eax | |
2226 | .long 0x9066AAF3 # rep stosb | |
2227 | lea -16(%rdi),%rdi # rewind $out by 1 block | |
2228 | mov $rnds_,$rounds # restore $rounds | |
2229 | mov %rdi,%rsi # $inp and $out are the same | |
2230 | mov $key_,$key # restore $key | |
2231 | xor $len,$len # len=16 | |
2232 | jmp .Lcbc_enc_loop # one more spin | |
2233 | \f#--------------------------- CBC DECRYPT ------------------------------# | |
2234 | .align 16 | |
2235 | .Lcbc_decrypt: | |
6a40ebe8 AP |
2236 | lea (%rsp),%rax |
2237 | push %rbp | |
2238 | sub \$$frame_size,%rsp | |
2239 | and \$-16,%rsp # Linux kernel stack can be incorrectly seeded | |
d64a7232 AP |
2240 | ___ |
2241 | $code.=<<___ if ($win64); | |
6a40ebe8 AP |
2242 | movaps %xmm6,0x10(%rsp) |
2243 | movaps %xmm7,0x20(%rsp) | |
2244 | movaps %xmm8,0x30(%rsp) | |
2245 | movaps %xmm9,0x40(%rsp) | |
d608b4d6 | 2246 | .Lcbc_decrypt_body: |
d64a7232 AP |
2247 | ___ |
2248 | $code.=<<___; | |
6a40ebe8 | 2249 | lea -8(%rax),%rbp |
d64a7232 | 2250 | movups ($ivp),$iv |
d608b4d6 | 2251 | mov $rnds_,$rounds |
f8501464 | 2252 | cmp \$0x70,$len |
d608b4d6 | 2253 | jbe .Lcbc_dec_tail |
f8501464 AP |
2254 | shr \$1,$rnds_ |
2255 | sub \$0x70,$len | |
2256 | mov $rnds_,$rounds | |
6a40ebe8 | 2257 | movaps $iv,(%rsp) |
f8501464 | 2258 | jmp .Lcbc_dec_loop8_enter |
d7d119a3 | 2259 | .align 16 |
f8501464 | 2260 | .Lcbc_dec_loop8: |
6a40ebe8 | 2261 | movaps $rndkey0,(%rsp) # save IV |
f8501464 AP |
2262 | movups $inout7,($out) |
2263 | lea 0x10($out),$out | |
2264 | .Lcbc_dec_loop8_enter: | |
2265 | $movkey ($key),$rndkey0 | |
2266 | movups ($inp),$inout0 # load input | |
d64a7232 | 2267 | movups 0x10($inp),$inout1 |
f8501464 | 2268 | $movkey 16($key),$rndkey1 |
d7d119a3 | 2269 | |
f8501464 AP |
2270 | lea 32($key),$key |
2271 | movdqu 0x20($inp),$inout2 | |
2272 | xorps $rndkey0,$inout0 | |
2273 | movdqu 0x30($inp),$inout3 | |
2274 | xorps $rndkey0,$inout1 | |
2275 | movdqu 0x40($inp),$inout4 | |
2276 | aesdec $rndkey1,$inout0 | |
2277 | pxor $rndkey0,$inout2 | |
2278 | movdqu 0x50($inp),$inout5 | |
2279 | aesdec $rndkey1,$inout1 | |
2280 | pxor $rndkey0,$inout3 | |
2281 | movdqu 0x60($inp),$inout6 | |
2282 | aesdec $rndkey1,$inout2 | |
2283 | pxor $rndkey0,$inout4 | |
2284 | movdqu 0x70($inp),$inout7 | |
2285 | aesdec $rndkey1,$inout3 | |
2286 | pxor $rndkey0,$inout5 | |
2287 | dec $rounds | |
2288 | aesdec $rndkey1,$inout4 | |
2289 | pxor $rndkey0,$inout6 | |
2290 | aesdec $rndkey1,$inout5 | |
2291 | pxor $rndkey0,$inout7 | |
2292 | $movkey ($key),$rndkey0 | |
2293 | aesdec $rndkey1,$inout6 | |
2294 | aesdec $rndkey1,$inout7 | |
2295 | $movkey 16($key),$rndkey1 | |
d64a7232 | 2296 | |
f8501464 AP |
2297 | call .Ldec_loop8_enter |
2298 | ||
2299 | movups ($inp),$rndkey1 # re-load input | |
2300 | movups 0x10($inp),$rndkey0 | |
6a40ebe8 | 2301 | xorps (%rsp),$inout0 # ^= IV |
f8501464 AP |
2302 | xorps $rndkey1,$inout1 |
2303 | movups 0x20($inp),$rndkey1 | |
2304 | xorps $rndkey0,$inout2 | |
2305 | movups 0x30($inp),$rndkey0 | |
2306 | xorps $rndkey1,$inout3 | |
2307 | movups 0x40($inp),$rndkey1 | |
2308 | xorps $rndkey0,$inout4 | |
2309 | movups 0x50($inp),$rndkey0 | |
2310 | xorps $rndkey1,$inout5 | |
2311 | movups 0x60($inp),$rndkey1 | |
2312 | xorps $rndkey0,$inout6 | |
2313 | movups 0x70($inp),$rndkey0 # IV | |
2314 | xorps $rndkey1,$inout7 | |
2315 | movups $inout0,($out) | |
2316 | movups $inout1,0x10($out) | |
2317 | movups $inout2,0x20($out) | |
2318 | movups $inout3,0x30($out) | |
2319 | mov $rnds_,$rounds # restore $rounds | |
2320 | movups $inout4,0x40($out) | |
2321 | mov $key_,$key # restore $key | |
2322 | movups $inout5,0x50($out) | |
2323 | lea 0x80($inp),$inp | |
2324 | movups $inout6,0x60($out) | |
2325 | lea 0x70($out),$out | |
2326 | sub \$0x80,$len | |
2327 | ja .Lcbc_dec_loop8 | |
2328 | ||
2329 | movaps $inout7,$inout0 | |
2330 | movaps $rndkey0,$iv | |
2331 | add \$0x70,$len | |
2332 | jle .Lcbc_dec_tail_collected | |
2333 | movups $inout0,($out) | |
2334 | lea 1($rnds_,$rnds_),$rounds | |
2335 | lea 0x10($out),$out | |
6c83629b | 2336 | .Lcbc_dec_tail: |
d64a7232 | 2337 | movups ($inp),$inout0 |
d64a7232 | 2338 | movaps $inout0,$in0 |
d7d119a3 | 2339 | cmp \$0x10,$len |
d64a7232 | 2340 | jbe .Lcbc_dec_one |
f8501464 | 2341 | |
d64a7232 | 2342 | movups 0x10($inp),$inout1 |
d64a7232 | 2343 | movaps $inout1,$in1 |
d7d119a3 | 2344 | cmp \$0x20,$len |
d64a7232 | 2345 | jbe .Lcbc_dec_two |
f8501464 | 2346 | |
d64a7232 | 2347 | movups 0x20($inp),$inout2 |
d64a7232 | 2348 | movaps $inout2,$in2 |
d7d119a3 | 2349 | cmp \$0x30,$len |
d64a7232 | 2350 | jbe .Lcbc_dec_three |
f8501464 | 2351 | |
d64a7232 | 2352 | movups 0x30($inp),$inout3 |
f8501464 AP |
2353 | cmp \$0x40,$len |
2354 | jbe .Lcbc_dec_four | |
2355 | ||
2356 | movups 0x40($inp),$inout4 | |
2357 | cmp \$0x50,$len | |
2358 | jbe .Lcbc_dec_five | |
2359 | ||
2360 | movups 0x50($inp),$inout5 | |
2361 | cmp \$0x60,$len | |
2362 | jbe .Lcbc_dec_six | |
2363 | ||
2364 | movups 0x60($inp),$inout6 | |
6a40ebe8 | 2365 | movaps $iv,(%rsp) # save IV |
f8501464 AP |
2366 | call _aesni_decrypt8 |
2367 | movups ($inp),$rndkey1 | |
2368 | movups 0x10($inp),$rndkey0 | |
6a40ebe8 | 2369 | xorps (%rsp),$inout0 # ^= IV |
f8501464 AP |
2370 | xorps $rndkey1,$inout1 |
2371 | movups 0x20($inp),$rndkey1 | |
2372 | xorps $rndkey0,$inout2 | |
2373 | movups 0x30($inp),$rndkey0 | |
2374 | xorps $rndkey1,$inout3 | |
2375 | movups 0x40($inp),$rndkey1 | |
2376 | xorps $rndkey0,$inout4 | |
2377 | movups 0x50($inp),$rndkey0 | |
2378 | xorps $rndkey1,$inout5 | |
2379 | movups 0x60($inp),$iv # IV | |
2380 | xorps $rndkey0,$inout6 | |
2381 | movups $inout0,($out) | |
2382 | movups $inout1,0x10($out) | |
2383 | movups $inout2,0x20($out) | |
2384 | movups $inout3,0x30($out) | |
2385 | movups $inout4,0x40($out) | |
2386 | movups $inout5,0x50($out) | |
2387 | lea 0x60($out),$out | |
2388 | movaps $inout6,$inout0 | |
2389 | sub \$0x70,$len | |
d64a7232 AP |
2390 | jmp .Lcbc_dec_tail_collected |
2391 | .align 16 | |
2392 | .Lcbc_dec_one: | |
2393 | ___ | |
d608b4d6 | 2394 | &aesni_generate1("dec",$key,$rounds); |
d64a7232 | 2395 | $code.=<<___; |
f8501464 | 2396 | xorps $iv,$inout0 |
d64a7232 | 2397 | movaps $in0,$iv |
f8501464 | 2398 | sub \$0x10,$len |
d64a7232 AP |
2399 | jmp .Lcbc_dec_tail_collected |
2400 | .align 16 | |
2401 | .Lcbc_dec_two: | |
f8501464 | 2402 | xorps $inout2,$inout2 |
d608b4d6 | 2403 | call _aesni_decrypt3 |
f8501464 AP |
2404 | xorps $iv,$inout0 |
2405 | xorps $in0,$inout1 | |
2406 | movups $inout0,($out) | |
d64a7232 | 2407 | movaps $in1,$iv |
f8501464 | 2408 | movaps $inout1,$inout0 |
d64a7232 | 2409 | lea 0x10($out),$out |
f8501464 | 2410 | sub \$0x20,$len |
d64a7232 AP |
2411 | jmp .Lcbc_dec_tail_collected |
2412 | .align 16 | |
2413 | .Lcbc_dec_three: | |
d608b4d6 | 2414 | call _aesni_decrypt3 |
f8501464 AP |
2415 | xorps $iv,$inout0 |
2416 | xorps $in0,$inout1 | |
2417 | movups $inout0,($out) | |
2418 | xorps $in1,$inout2 | |
2419 | movups $inout1,0x10($out) | |
d64a7232 | 2420 | movaps $in2,$iv |
f8501464 | 2421 | movaps $inout2,$inout0 |
d64a7232 | 2422 | lea 0x20($out),$out |
f8501464 AP |
2423 | sub \$0x30,$len |
2424 | jmp .Lcbc_dec_tail_collected | |
2425 | .align 16 | |
2426 | .Lcbc_dec_four: | |
2427 | call _aesni_decrypt4 | |
2428 | xorps $iv,$inout0 | |
2429 | movups 0x30($inp),$iv | |
2430 | xorps $in0,$inout1 | |
2431 | movups $inout0,($out) | |
2432 | xorps $in1,$inout2 | |
2433 | movups $inout1,0x10($out) | |
2434 | xorps $in2,$inout3 | |
2435 | movups $inout2,0x20($out) | |
2436 | movaps $inout3,$inout0 | |
2437 | lea 0x30($out),$out | |
2438 | sub \$0x40,$len | |
2439 | jmp .Lcbc_dec_tail_collected | |
2440 | .align 16 | |
2441 | .Lcbc_dec_five: | |
2442 | xorps $inout5,$inout5 | |
2443 | call _aesni_decrypt6 | |
2444 | movups 0x10($inp),$rndkey1 | |
2445 | movups 0x20($inp),$rndkey0 | |
2446 | xorps $iv,$inout0 | |
2447 | xorps $in0,$inout1 | |
2448 | xorps $rndkey1,$inout2 | |
2449 | movups 0x30($inp),$rndkey1 | |
2450 | xorps $rndkey0,$inout3 | |
2451 | movups 0x40($inp),$iv | |
2452 | xorps $rndkey1,$inout4 | |
2453 | movups $inout0,($out) | |
2454 | movups $inout1,0x10($out) | |
2455 | movups $inout2,0x20($out) | |
2456 | movups $inout3,0x30($out) | |
2457 | lea 0x40($out),$out | |
2458 | movaps $inout4,$inout0 | |
2459 | sub \$0x50,$len | |
2460 | jmp .Lcbc_dec_tail_collected | |
2461 | .align 16 | |
2462 | .Lcbc_dec_six: | |
2463 | call _aesni_decrypt6 | |
2464 | movups 0x10($inp),$rndkey1 | |
2465 | movups 0x20($inp),$rndkey0 | |
2466 | xorps $iv,$inout0 | |
2467 | xorps $in0,$inout1 | |
2468 | xorps $rndkey1,$inout2 | |
2469 | movups 0x30($inp),$rndkey1 | |
2470 | xorps $rndkey0,$inout3 | |
2471 | movups 0x40($inp),$rndkey0 | |
2472 | xorps $rndkey1,$inout4 | |
2473 | movups 0x50($inp),$iv | |
2474 | xorps $rndkey0,$inout5 | |
2475 | movups $inout0,($out) | |
2476 | movups $inout1,0x10($out) | |
2477 | movups $inout2,0x20($out) | |
2478 | movups $inout3,0x30($out) | |
2479 | movups $inout4,0x40($out) | |
2480 | lea 0x50($out),$out | |
2481 | movaps $inout5,$inout0 | |
2482 | sub \$0x60,$len | |
d64a7232 AP |
2483 | jmp .Lcbc_dec_tail_collected |
2484 | .align 16 | |
d64a7232 AP |
2485 | .Lcbc_dec_tail_collected: |
2486 | and \$15,$len | |
2487 | movups $iv,($ivp) | |
2488 | jnz .Lcbc_dec_tail_partial | |
f8501464 | 2489 | movups $inout0,($out) |
d64a7232 | 2490 | jmp .Lcbc_dec_ret |
d7d119a3 | 2491 | .align 16 |
d64a7232 | 2492 | .Lcbc_dec_tail_partial: |
6a40ebe8 | 2493 | movaps $inout0,(%rsp) |
f8501464 | 2494 | mov \$16,%rcx |
d64a7232 | 2495 | mov $out,%rdi |
f8501464 | 2496 | sub $len,%rcx |
6a40ebe8 | 2497 | lea (%rsp),%rsi |
d64a7232 AP |
2498 | .long 0x9066A4F3 # rep movsb |
2499 | ||
2500 | .Lcbc_dec_ret: | |
2501 | ___ | |
2502 | $code.=<<___ if ($win64); | |
6a40ebe8 AP |
2503 | movaps 0x10(%rsp),%xmm6 |
2504 | movaps 0x20(%rsp),%xmm7 | |
2505 | movaps 0x30(%rsp),%xmm8 | |
2506 | movaps 0x40(%rsp),%xmm9 | |
d64a7232 AP |
2507 | ___ |
2508 | $code.=<<___; | |
6a40ebe8 AP |
2509 | lea (%rbp),%rsp |
2510 | pop %rbp | |
d64a7232 AP |
2511 | .Lcbc_ret: |
2512 | ret | |
2513 | .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt | |
2514 | ___ | |
f8501464 | 2515 | } \f |
d608b4d6 AP |
2516 | # int $PREFIX_set_[en|de]crypt_key (const unsigned char *userKey, |
2517 | # int bits, AES_KEY *key) | |
2518 | { my ($inp,$bits,$key) = @_4args; | |
2519 | $bits =~ s/%r/%e/; | |
2520 | ||
d64a7232 AP |
2521 | $code.=<<___; |
2522 | .globl ${PREFIX}_set_decrypt_key | |
d608b4d6 | 2523 | .type ${PREFIX}_set_decrypt_key,\@abi-omnipotent |
d64a7232 AP |
2524 | .align 16 |
2525 | ${PREFIX}_set_decrypt_key: | |
d608b4d6 | 2526 | .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 |
fb2f3411 | 2527 | call __aesni_set_encrypt_key |
d608b4d6 | 2528 | shl \$4,$bits # rounds-1 after _aesni_set_encrypt_key |
d64a7232 AP |
2529 | test %eax,%eax |
2530 | jnz .Ldec_key_ret | |
d608b4d6 AP |
2531 | lea 16($key,$bits),$inp # points at the end of key schedule |
2532 | ||
2533 | $movkey ($key),%xmm0 # just swap | |
2534 | $movkey ($inp),%xmm1 | |
2535 | $movkey %xmm0,($inp) | |
2536 | $movkey %xmm1,($key) | |
2537 | lea 16($key),$key | |
2538 | lea -16($inp),$inp | |
2539 | ||
d64a7232 | 2540 | .Ldec_key_inverse: |
d608b4d6 AP |
2541 | $movkey ($key),%xmm0 # swap and inverse |
2542 | $movkey ($inp),%xmm1 | |
d64a7232 AP |
2543 | aesimc %xmm0,%xmm0 |
2544 | aesimc %xmm1,%xmm1 | |
d608b4d6 AP |
2545 | lea 16($key),$key |
2546 | lea -16($inp),$inp | |
d608b4d6 AP |
2547 | $movkey %xmm0,16($inp) |
2548 | $movkey %xmm1,-16($key) | |
d7d119a3 | 2549 | cmp $key,$inp |
d64a7232 AP |
2550 | ja .Ldec_key_inverse |
2551 | ||
d608b4d6 | 2552 | $movkey ($key),%xmm0 # inverse middle |
d64a7232 | 2553 | aesimc %xmm0,%xmm0 |
d608b4d6 | 2554 | $movkey %xmm0,($inp) |
d64a7232 | 2555 | .Ldec_key_ret: |
d608b4d6 | 2556 | add \$8,%rsp |
d64a7232 | 2557 | ret |
d608b4d6 | 2558 | .LSEH_end_set_decrypt_key: |
d64a7232 AP |
2559 | .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key |
2560 | ___ | |
2561 | \f | |
2562 | # This is based on submission by | |
2563 | # | |
2564 | # Huang Ying <ying.huang@intel.com> | |
2565 | # Vinodh Gopal <vinodh.gopal@intel.com> | |
2566 | # Kahraman Akdemir | |
2567 | # | |
2568 | # Agressively optimized in respect to aeskeygenassist's critical path | |
2569 | # and is contained in %xmm0-5 to meet Win64 ABI requirement. | |
2570 | # | |
2571 | $code.=<<___; | |
d608b4d6 AP |
2572 | .globl ${PREFIX}_set_encrypt_key |
2573 | .type ${PREFIX}_set_encrypt_key,\@abi-omnipotent | |
d64a7232 | 2574 | .align 16 |
d608b4d6 | 2575 | ${PREFIX}_set_encrypt_key: |
fb2f3411 | 2576 | __aesni_set_encrypt_key: |
d608b4d6 | 2577 | .byte 0x48,0x83,0xEC,0x08 # sub rsp,8 |
d608b4d6 | 2578 | mov \$-1,%rax |
d7d119a3 | 2579 | test $inp,$inp |
d608b4d6 AP |
2580 | jz .Lenc_key_ret |
2581 | test $key,$key | |
2582 | jz .Lenc_key_ret | |
2583 | ||
2584 | movups ($inp),%xmm0 # pull first 128 bits of *userKey | |
f8501464 | 2585 | xorps %xmm4,%xmm4 # low dword of xmm4 is assumed 0 |
d608b4d6 AP |
2586 | lea 16($key),%rax |
2587 | cmp \$256,$bits | |
d64a7232 | 2588 | je .L14rounds |
d608b4d6 | 2589 | cmp \$192,$bits |
d64a7232 | 2590 | je .L12rounds |
d608b4d6 | 2591 | cmp \$128,$bits |
d64a7232 | 2592 | jne .Lbad_keybits |
d608b4d6 | 2593 | |
d64a7232 | 2594 | .L10rounds: |
d608b4d6 AP |
2595 | mov \$9,$bits # 10 rounds for 128-bit key |
2596 | $movkey %xmm0,($key) # round 0 | |
d64a7232 AP |
2597 | aeskeygenassist \$0x1,%xmm0,%xmm1 # round 1 |
2598 | call .Lkey_expansion_128_cold | |
2599 | aeskeygenassist \$0x2,%xmm0,%xmm1 # round 2 | |
2600 | call .Lkey_expansion_128 | |
2601 | aeskeygenassist \$0x4,%xmm0,%xmm1 # round 3 | |
2602 | call .Lkey_expansion_128 | |
2603 | aeskeygenassist \$0x8,%xmm0,%xmm1 # round 4 | |
2604 | call .Lkey_expansion_128 | |
2605 | aeskeygenassist \$0x10,%xmm0,%xmm1 # round 5 | |
2606 | call .Lkey_expansion_128 | |
2607 | aeskeygenassist \$0x20,%xmm0,%xmm1 # round 6 | |
2608 | call .Lkey_expansion_128 | |
2609 | aeskeygenassist \$0x40,%xmm0,%xmm1 # round 7 | |
2610 | call .Lkey_expansion_128 | |
2611 | aeskeygenassist \$0x80,%xmm0,%xmm1 # round 8 | |
2612 | call .Lkey_expansion_128 | |
2613 | aeskeygenassist \$0x1b,%xmm0,%xmm1 # round 9 | |
2614 | call .Lkey_expansion_128 | |
2615 | aeskeygenassist \$0x36,%xmm0,%xmm1 # round 10 | |
2616 | call .Lkey_expansion_128 | |
d608b4d6 AP |
2617 | $movkey %xmm0,(%rax) |
2618 | mov $bits,80(%rax) # 240(%rdx) | |
d64a7232 | 2619 | xor %eax,%eax |
d608b4d6 | 2620 | jmp .Lenc_key_ret |
d64a7232 | 2621 | |
d64a7232 AP |
2622 | .align 16 |
2623 | .L12rounds: | |
d608b4d6 AP |
2624 | movq 16($inp),%xmm2 # remaining 1/3 of *userKey |
2625 | mov \$11,$bits # 12 rounds for 192 | |
2626 | $movkey %xmm0,($key) # round 0 | |
d64a7232 AP |
2627 | aeskeygenassist \$0x1,%xmm2,%xmm1 # round 1,2 |
2628 | call .Lkey_expansion_192a_cold | |
2629 | aeskeygenassist \$0x2,%xmm2,%xmm1 # round 2,3 | |
2630 | call .Lkey_expansion_192b | |
2631 | aeskeygenassist \$0x4,%xmm2,%xmm1 # round 4,5 | |
2632 | call .Lkey_expansion_192a | |
2633 | aeskeygenassist \$0x8,%xmm2,%xmm1 # round 5,6 | |
2634 | call .Lkey_expansion_192b | |
2635 | aeskeygenassist \$0x10,%xmm2,%xmm1 # round 7,8 | |
2636 | call .Lkey_expansion_192a | |
2637 | aeskeygenassist \$0x20,%xmm2,%xmm1 # round 8,9 | |
2638 | call .Lkey_expansion_192b | |
2639 | aeskeygenassist \$0x40,%xmm2,%xmm1 # round 10,11 | |
2640 | call .Lkey_expansion_192a | |
2641 | aeskeygenassist \$0x80,%xmm2,%xmm1 # round 11,12 | |
2642 | call .Lkey_expansion_192b | |
d608b4d6 AP |
2643 | $movkey %xmm0,(%rax) |
2644 | mov $bits,48(%rax) # 240(%rdx) | |
d64a7232 | 2645 | xor %rax, %rax |
d608b4d6 | 2646 | jmp .Lenc_key_ret |
d64a7232 | 2647 | |
d64a7232 AP |
2648 | .align 16 |
2649 | .L14rounds: | |
d608b4d6 AP |
2650 | movups 16($inp),%xmm2 # remaning half of *userKey |
2651 | mov \$13,$bits # 14 rounds for 256 | |
2652 | lea 16(%rax),%rax | |
2653 | $movkey %xmm0,($key) # round 0 | |
2654 | $movkey %xmm2,16($key) # round 1 | |
d64a7232 AP |
2655 | aeskeygenassist \$0x1,%xmm2,%xmm1 # round 2 |
2656 | call .Lkey_expansion_256a_cold | |
2657 | aeskeygenassist \$0x1,%xmm0,%xmm1 # round 3 | |
2658 | call .Lkey_expansion_256b | |
2659 | aeskeygenassist \$0x2,%xmm2,%xmm1 # round 4 | |
2660 | call .Lkey_expansion_256a | |
2661 | aeskeygenassist \$0x2,%xmm0,%xmm1 # round 5 | |
2662 | call .Lkey_expansion_256b | |
2663 | aeskeygenassist \$0x4,%xmm2,%xmm1 # round 6 | |
2664 | call .Lkey_expansion_256a | |
2665 | aeskeygenassist \$0x4,%xmm0,%xmm1 # round 7 | |
2666 | call .Lkey_expansion_256b | |
2667 | aeskeygenassist \$0x8,%xmm2,%xmm1 # round 8 | |
2668 | call .Lkey_expansion_256a | |
2669 | aeskeygenassist \$0x8,%xmm0,%xmm1 # round 9 | |
2670 | call .Lkey_expansion_256b | |
2671 | aeskeygenassist \$0x10,%xmm2,%xmm1 # round 10 | |
2672 | call .Lkey_expansion_256a | |
2673 | aeskeygenassist \$0x10,%xmm0,%xmm1 # round 11 | |
2674 | call .Lkey_expansion_256b | |
2675 | aeskeygenassist \$0x20,%xmm2,%xmm1 # round 12 | |
2676 | call .Lkey_expansion_256a | |
2677 | aeskeygenassist \$0x20,%xmm0,%xmm1 # round 13 | |
2678 | call .Lkey_expansion_256b | |
2679 | aeskeygenassist \$0x40,%xmm2,%xmm1 # round 14 | |
2680 | call .Lkey_expansion_256a | |
d608b4d6 AP |
2681 | $movkey %xmm0,(%rax) |
2682 | mov $bits,16(%rax) # 240(%rdx) | |
d64a7232 | 2683 | xor %rax,%rax |
d608b4d6 AP |
2684 | jmp .Lenc_key_ret |
2685 | ||
2686 | .align 16 | |
2687 | .Lbad_keybits: | |
2688 | mov \$-2,%rax | |
2689 | .Lenc_key_ret: | |
2690 | add \$8,%rsp | |
2691 | ret | |
2692 | .LSEH_end_set_encrypt_key: | |
2693 | \f | |
2694 | .align 16 | |
2695 | .Lkey_expansion_128: | |
2696 | $movkey %xmm0,(%rax) | |
2697 | lea 16(%rax),%rax | |
2698 | .Lkey_expansion_128_cold: | |
2699 | shufps \$0b00010000,%xmm0,%xmm4 | |
f8501464 | 2700 | xorps %xmm4, %xmm0 |
d608b4d6 | 2701 | shufps \$0b10001100,%xmm0,%xmm4 |
f8501464 AP |
2702 | xorps %xmm4, %xmm0 |
2703 | shufps \$0b11111111,%xmm1,%xmm1 # critical path | |
2704 | xorps %xmm1,%xmm0 | |
d608b4d6 AP |
2705 | ret |
2706 | ||
2707 | .align 16 | |
2708 | .Lkey_expansion_192a: | |
2709 | $movkey %xmm0,(%rax) | |
2710 | lea 16(%rax),%rax | |
2711 | .Lkey_expansion_192a_cold: | |
2712 | movaps %xmm2, %xmm5 | |
2713 | .Lkey_expansion_192b_warm: | |
2714 | shufps \$0b00010000,%xmm0,%xmm4 | |
f8501464 AP |
2715 | movdqa %xmm2,%xmm3 |
2716 | xorps %xmm4,%xmm0 | |
d608b4d6 AP |
2717 | shufps \$0b10001100,%xmm0,%xmm4 |
2718 | pslldq \$4,%xmm3 | |
f8501464 | 2719 | xorps %xmm4,%xmm0 |
d608b4d6 AP |
2720 | pshufd \$0b01010101,%xmm1,%xmm1 # critical path |
2721 | pxor %xmm3,%xmm2 | |
2722 | pxor %xmm1,%xmm0 | |
2723 | pshufd \$0b11111111,%xmm0,%xmm3 | |
2724 | pxor %xmm3,%xmm2 | |
d64a7232 AP |
2725 | ret |
2726 | ||
d608b4d6 AP |
2727 | .align 16 |
2728 | .Lkey_expansion_192b: | |
2729 | movaps %xmm0,%xmm3 | |
2730 | shufps \$0b01000100,%xmm0,%xmm5 | |
2731 | $movkey %xmm5,(%rax) | |
2732 | shufps \$0b01001110,%xmm2,%xmm3 | |
2733 | $movkey %xmm3,16(%rax) | |
2734 | lea 32(%rax),%rax | |
2735 | jmp .Lkey_expansion_192b_warm | |
2736 | ||
d64a7232 AP |
2737 | .align 16 |
2738 | .Lkey_expansion_256a: | |
d608b4d6 AP |
2739 | $movkey %xmm2,(%rax) |
2740 | lea 16(%rax),%rax | |
d64a7232 AP |
2741 | .Lkey_expansion_256a_cold: |
2742 | shufps \$0b00010000,%xmm0,%xmm4 | |
f8501464 | 2743 | xorps %xmm4,%xmm0 |
d64a7232 | 2744 | shufps \$0b10001100,%xmm0,%xmm4 |
f8501464 AP |
2745 | xorps %xmm4,%xmm0 |
2746 | shufps \$0b11111111,%xmm1,%xmm1 # critical path | |
2747 | xorps %xmm1,%xmm0 | |
d64a7232 AP |
2748 | ret |
2749 | ||
2750 | .align 16 | |
2751 | .Lkey_expansion_256b: | |
d608b4d6 AP |
2752 | $movkey %xmm0,(%rax) |
2753 | lea 16(%rax),%rax | |
d64a7232 AP |
2754 | |
2755 | shufps \$0b00010000,%xmm2,%xmm4 | |
f8501464 | 2756 | xorps %xmm4,%xmm2 |
d64a7232 | 2757 | shufps \$0b10001100,%xmm2,%xmm4 |
f8501464 AP |
2758 | xorps %xmm4,%xmm2 |
2759 | shufps \$0b10101010,%xmm1,%xmm1 # critical path | |
2760 | xorps %xmm1,%xmm2 | |
d64a7232 | 2761 | ret |
d608b4d6 | 2762 | .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key |
f8501464 | 2763 | .size __aesni_set_encrypt_key,.-__aesni_set_encrypt_key |
d64a7232 AP |
2764 | ___ |
2765 | } | |
2766 | \f | |
2767 | $code.=<<___; | |
6c83629b AP |
2768 | .align 64 |
2769 | .Lbswap_mask: | |
2770 | .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 | |
d7d119a3 | 2771 | .Lincrement32: |
f8501464 | 2772 | .long 6,6,6,0 |
d7d119a3 AP |
2773 | .Lincrement64: |
2774 | .long 1,0,0,0 | |
f8501464 AP |
2775 | .Lxts_magic: |
2776 | .long 0x87,0,1,0 | |
9282c335 AP |
2777 | .Lincrement1: |
2778 | .byte 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1 | |
f8501464 | 2779 | |
d64a7232 AP |
2780 | .asciz "AES for Intel AES-NI, CRYPTOGAMS by <appro\@openssl.org>" |
2781 | .align 64 | |
2782 | ___ | |
2783 | ||
d608b4d6 AP |
2784 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, |
2785 | # CONTEXT *context,DISPATCHER_CONTEXT *disp) | |
2786 | if ($win64) { | |
2787 | $rec="%rcx"; | |
2788 | $frame="%rdx"; | |
2789 | $context="%r8"; | |
2790 | $disp="%r9"; | |
2791 | ||
2792 | $code.=<<___; | |
2793 | .extern __imp_RtlVirtualUnwind | |
6c83629b AP |
2794 | ___ |
2795 | $code.=<<___ if ($PREFIX eq "aesni"); | |
2796 | .type ecb_se_handler,\@abi-omnipotent | |
d608b4d6 | 2797 | .align 16 |
6c83629b | 2798 | ecb_se_handler: |
d608b4d6 AP |
2799 | push %rsi |
2800 | push %rdi | |
2801 | push %rbx | |
2802 | push %rbp | |
2803 | push %r12 | |
2804 | push %r13 | |
2805 | push %r14 | |
2806 | push %r15 | |
2807 | pushfq | |
2808 | sub \$64,%rsp | |
2809 | ||
2810 | mov 152($context),%rax # pull context->Rsp | |
6c83629b | 2811 | |
f8501464 | 2812 | jmp .Lcommon_seh_tail |
6c83629b AP |
2813 | .size ecb_se_handler,.-ecb_se_handler |
2814 | ||
d7d119a3 AP |
2815 | .type ccm64_se_handler,\@abi-omnipotent |
2816 | .align 16 | |
2817 | ccm64_se_handler: | |
2818 | push %rsi | |
2819 | push %rdi | |
2820 | push %rbx | |
2821 | push %rbp | |
2822 | push %r12 | |
2823 | push %r13 | |
2824 | push %r14 | |
2825 | push %r15 | |
2826 | pushfq | |
2827 | sub \$64,%rsp | |
2828 | ||
2829 | mov 120($context),%rax # pull context->Rax | |
2830 | mov 248($context),%rbx # pull context->Rip | |
2831 | ||
2832 | mov 8($disp),%rsi # disp->ImageBase | |
02f358da | 2833 | mov 56($disp),%r11 # disp->HandlerData |
d7d119a3 AP |
2834 | |
2835 | mov 0(%r11),%r10d # HandlerData[0] | |
2836 | lea (%rsi,%r10),%r10 # prologue label | |
2837 | cmp %r10,%rbx # context->Rip<prologue label | |
f8501464 | 2838 | jb .Lcommon_seh_tail |
d7d119a3 AP |
2839 | |
2840 | mov 152($context),%rax # pull context->Rsp | |
2841 | ||
2842 | mov 4(%r11),%r10d # HandlerData[1] | |
2843 | lea (%rsi,%r10),%r10 # epilogue label | |
2844 | cmp %r10,%rbx # context->Rip>=epilogue label | |
f8501464 | 2845 | jae .Lcommon_seh_tail |
d7d119a3 | 2846 | |
f8501464 | 2847 | lea 0(%rax),%rsi # %xmm save area |
d7d119a3 AP |
2848 | lea 512($context),%rdi # &context.Xmm6 |
2849 | mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) | |
2850 | .long 0xa548f3fc # cld; rep movsq | |
2851 | lea 0x58(%rax),%rax # adjust stack pointer | |
2852 | ||
f8501464 | 2853 | jmp .Lcommon_seh_tail |
d7d119a3 AP |
2854 | .size ccm64_se_handler,.-ccm64_se_handler |
2855 | ||
6c83629b AP |
2856 | .type ctr32_se_handler,\@abi-omnipotent |
2857 | .align 16 | |
2858 | ctr32_se_handler: | |
2859 | push %rsi | |
2860 | push %rdi | |
2861 | push %rbx | |
2862 | push %rbp | |
2863 | push %r12 | |
2864 | push %r13 | |
2865 | push %r14 | |
2866 | push %r15 | |
2867 | pushfq | |
2868 | sub \$64,%rsp | |
2869 | ||
2870 | mov 120($context),%rax # pull context->Rax | |
d608b4d6 AP |
2871 | mov 248($context),%rbx # pull context->Rip |
2872 | ||
6c83629b | 2873 | lea .Lctr32_body(%rip),%r10 |
d608b4d6 | 2874 | cmp %r10,%rbx # context->Rip<"prologue" label |
f8501464 | 2875 | jb .Lcommon_seh_tail |
d608b4d6 | 2876 | |
6c83629b | 2877 | mov 152($context),%rax # pull context->Rsp |
d608b4d6 | 2878 | |
6c83629b AP |
2879 | lea .Lctr32_ret(%rip),%r10 |
2880 | cmp %r10,%rbx | |
f8501464 | 2881 | jae .Lcommon_seh_tail |
d608b4d6 | 2882 | |
9282c335 | 2883 | lea (%rax),%rsi # %xmm save area |
d608b4d6 | 2884 | lea 512($context),%rdi # &context.Xmm6 |
f8501464 | 2885 | mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) |
d608b4d6 | 2886 | .long 0xa548f3fc # cld; rep movsq |
9282c335 | 2887 | lea 0xa8(%rax),%rax # adjust stack pointer |
d608b4d6 | 2888 | |
9282c335 | 2889 | jmp .Lcommon_seh_tail |
6c83629b | 2890 | .size ctr32_se_handler,.-ctr32_se_handler |
f8501464 AP |
2891 | |
2892 | .type xts_se_handler,\@abi-omnipotent | |
2893 | .align 16 | |
2894 | xts_se_handler: | |
2895 | push %rsi | |
2896 | push %rdi | |
2897 | push %rbx | |
2898 | push %rbp | |
2899 | push %r12 | |
2900 | push %r13 | |
2901 | push %r14 | |
2902 | push %r15 | |
2903 | pushfq | |
2904 | sub \$64,%rsp | |
2905 | ||
2906 | mov 120($context),%rax # pull context->Rax | |
2907 | mov 248($context),%rbx # pull context->Rip | |
2908 | ||
2909 | mov 8($disp),%rsi # disp->ImageBase | |
2910 | mov 56($disp),%r11 # disp->HandlerData | |
2911 | ||
2912 | mov 0(%r11),%r10d # HandlerData[0] | |
2913 | lea (%rsi,%r10),%r10 # prologue lable | |
2914 | cmp %r10,%rbx # context->Rip<prologue label | |
2915 | jb .Lcommon_seh_tail | |
2916 | ||
2917 | mov 152($context),%rax # pull context->Rsp | |
2918 | ||
2919 | mov 4(%r11),%r10d # HandlerData[1] | |
2920 | lea (%rsi,%r10),%r10 # epilogue label | |
2921 | cmp %r10,%rbx # context->Rip>=epilogue label | |
2922 | jae .Lcommon_seh_tail | |
2923 | ||
2924 | lea 0x60(%rax),%rsi # %xmm save area | |
2925 | lea 512($context),%rdi # & context.Xmm6 | |
2926 | mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) | |
2927 | .long 0xa548f3fc # cld; rep movsq | |
f8501464 | 2928 | |
6a40ebe8 | 2929 | jmp .Lcommon_rbp_tail |
f8501464 | 2930 | .size xts_se_handler,.-xts_se_handler |
6c83629b AP |
2931 | ___ |
2932 | $code.=<<___; | |
2933 | .type cbc_se_handler,\@abi-omnipotent | |
d608b4d6 | 2934 | .align 16 |
6c83629b | 2935 | cbc_se_handler: |
d608b4d6 AP |
2936 | push %rsi |
2937 | push %rdi | |
2938 | push %rbx | |
2939 | push %rbp | |
2940 | push %r12 | |
2941 | push %r13 | |
2942 | push %r14 | |
2943 | push %r15 | |
2944 | pushfq | |
2945 | sub \$64,%rsp | |
2946 | ||
2947 | mov 152($context),%rax # pull context->Rsp | |
6c83629b AP |
2948 | mov 248($context),%rbx # pull context->Rip |
2949 | ||
2950 | lea .Lcbc_decrypt(%rip),%r10 | |
2951 | cmp %r10,%rbx # context->Rip<"prologue" label | |
f8501464 | 2952 | jb .Lcommon_seh_tail |
6c83629b AP |
2953 | |
2954 | lea .Lcbc_decrypt_body(%rip),%r10 | |
2955 | cmp %r10,%rbx # context->Rip<cbc_decrypt_body | |
2956 | jb .Lrestore_cbc_rax | |
2957 | ||
2958 | lea .Lcbc_ret(%rip),%r10 | |
2959 | cmp %r10,%rbx # context->Rip>="epilogue" label | |
f8501464 | 2960 | jae .Lcommon_seh_tail |
6c83629b | 2961 | |
6a40ebe8 | 2962 | lea 16(%rax),%rsi # %xmm save area |
6c83629b AP |
2963 | lea 512($context),%rdi # &context.Xmm6 |
2964 | mov \$8,%ecx # 4*sizeof(%xmm0)/sizeof(%rax) | |
2965 | .long 0xa548f3fc # cld; rep movsq | |
6a40ebe8 AP |
2966 | |
2967 | .Lcommon_rbp_tail: | |
2968 | mov 160($context),%rax # pull context->Rbp | |
2969 | mov (%rax),%rbp # restore saved %rbp | |
2970 | lea 8(%rax),%rax # adjust stack pointer | |
2971 | mov %rbp,160($context) # restore context->Rbp | |
f8501464 | 2972 | jmp .Lcommon_seh_tail |
6c83629b AP |
2973 | |
2974 | .Lrestore_cbc_rax: | |
2975 | mov 120($context),%rax | |
f8501464 AP |
2976 | |
2977 | .Lcommon_seh_tail: | |
d608b4d6 AP |
2978 | mov 8(%rax),%rdi |
2979 | mov 16(%rax),%rsi | |
6c83629b | 2980 | mov %rax,152($context) # restore context->Rsp |
d608b4d6 AP |
2981 | mov %rsi,168($context) # restore context->Rsi |
2982 | mov %rdi,176($context) # restore context->Rdi | |
2983 | ||
d608b4d6 AP |
2984 | mov 40($disp),%rdi # disp->ContextRecord |
2985 | mov $context,%rsi # context | |
2986 | mov \$154,%ecx # sizeof(CONTEXT) | |
2987 | .long 0xa548f3fc # cld; rep movsq | |
2988 | ||
2989 | mov $disp,%rsi | |
2990 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER | |
2991 | mov 8(%rsi),%rdx # arg2, disp->ImageBase | |
2992 | mov 0(%rsi),%r8 # arg3, disp->ControlPc | |
2993 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry | |
2994 | mov 40(%rsi),%r10 # disp->ContextRecord | |
2995 | lea 56(%rsi),%r11 # &disp->HandlerData | |
2996 | lea 24(%rsi),%r12 # &disp->EstablisherFrame | |
2997 | mov %r10,32(%rsp) # arg5 | |
2998 | mov %r11,40(%rsp) # arg6 | |
2999 | mov %r12,48(%rsp) # arg7 | |
3000 | mov %rcx,56(%rsp) # arg8, (NULL) | |
3001 | call *__imp_RtlVirtualUnwind(%rip) | |
3002 | ||
3003 | mov \$1,%eax # ExceptionContinueSearch | |
3004 | add \$64,%rsp | |
3005 | popfq | |
3006 | pop %r15 | |
3007 | pop %r14 | |
3008 | pop %r13 | |
3009 | pop %r12 | |
3010 | pop %rbp | |
3011 | pop %rbx | |
3012 | pop %rdi | |
3013 | pop %rsi | |
3014 | ret | |
3015 | .size cbc_se_handler,.-cbc_se_handler | |
3016 | ||
3017 | .section .pdata | |
3018 | .align 4 | |
6c83629b AP |
3019 | ___ |
3020 | $code.=<<___ if ($PREFIX eq "aesni"); | |
3021 | .rva .LSEH_begin_aesni_ecb_encrypt | |
3022 | .rva .LSEH_end_aesni_ecb_encrypt | |
d608b4d6 AP |
3023 | .rva .LSEH_info_ecb |
3024 | ||
d7d119a3 AP |
3025 | .rva .LSEH_begin_aesni_ccm64_encrypt_blocks |
3026 | .rva .LSEH_end_aesni_ccm64_encrypt_blocks | |
02f358da | 3027 | .rva .LSEH_info_ccm64_enc |
d7d119a3 AP |
3028 | |
3029 | .rva .LSEH_begin_aesni_ccm64_decrypt_blocks | |
3030 | .rva .LSEH_end_aesni_ccm64_decrypt_blocks | |
02f358da | 3031 | .rva .LSEH_info_ccm64_dec |
d7d119a3 | 3032 | |
6c83629b AP |
3033 | .rva .LSEH_begin_aesni_ctr32_encrypt_blocks |
3034 | .rva .LSEH_end_aesni_ctr32_encrypt_blocks | |
3035 | .rva .LSEH_info_ctr32 | |
f8501464 AP |
3036 | |
3037 | .rva .LSEH_begin_aesni_xts_encrypt | |
3038 | .rva .LSEH_end_aesni_xts_encrypt | |
3039 | .rva .LSEH_info_xts_enc | |
3040 | ||
3041 | .rva .LSEH_begin_aesni_xts_decrypt | |
3042 | .rva .LSEH_end_aesni_xts_decrypt | |
3043 | .rva .LSEH_info_xts_dec | |
6c83629b AP |
3044 | ___ |
3045 | $code.=<<___; | |
d608b4d6 AP |
3046 | .rva .LSEH_begin_${PREFIX}_cbc_encrypt |
3047 | .rva .LSEH_end_${PREFIX}_cbc_encrypt | |
3048 | .rva .LSEH_info_cbc | |
3049 | ||
d608b4d6 AP |
3050 | .rva ${PREFIX}_set_decrypt_key |
3051 | .rva .LSEH_end_set_decrypt_key | |
3052 | .rva .LSEH_info_key | |
c5036d78 AP |
3053 | |
3054 | .rva ${PREFIX}_set_encrypt_key | |
3055 | .rva .LSEH_end_set_encrypt_key | |
3056 | .rva .LSEH_info_key | |
d608b4d6 AP |
3057 | .section .xdata |
3058 | .align 8 | |
6c83629b AP |
3059 | ___ |
3060 | $code.=<<___ if ($PREFIX eq "aesni"); | |
d608b4d6 AP |
3061 | .LSEH_info_ecb: |
3062 | .byte 9,0,0,0 | |
3063 | .rva ecb_se_handler | |
02f358da | 3064 | .LSEH_info_ccm64_enc: |
d7d119a3 AP |
3065 | .byte 9,0,0,0 |
3066 | .rva ccm64_se_handler | |
02f358da AP |
3067 | .rva .Lccm64_enc_body,.Lccm64_enc_ret # HandlerData[] |
3068 | .LSEH_info_ccm64_dec: | |
3069 | .byte 9,0,0,0 | |
3070 | .rva ccm64_se_handler | |
3071 | .rva .Lccm64_dec_body,.Lccm64_dec_ret # HandlerData[] | |
6c83629b AP |
3072 | .LSEH_info_ctr32: |
3073 | .byte 9,0,0,0 | |
3074 | .rva ctr32_se_handler | |
f8501464 AP |
3075 | .LSEH_info_xts_enc: |
3076 | .byte 9,0,0,0 | |
3077 | .rva xts_se_handler | |
3078 | .rva .Lxts_enc_body,.Lxts_enc_epilogue # HandlerData[] | |
3079 | .LSEH_info_xts_dec: | |
3080 | .byte 9,0,0,0 | |
3081 | .rva xts_se_handler | |
3082 | .rva .Lxts_dec_body,.Lxts_dec_epilogue # HandlerData[] | |
6c83629b AP |
3083 | ___ |
3084 | $code.=<<___; | |
d608b4d6 AP |
3085 | .LSEH_info_cbc: |
3086 | .byte 9,0,0,0 | |
3087 | .rva cbc_se_handler | |
3088 | .LSEH_info_key: | |
3089 | .byte 0x01,0x04,0x01,0x00 | |
d7d119a3 | 3090 | .byte 0x04,0x02,0x00,0x00 # sub rsp,8 |
d608b4d6 AP |
3091 | ___ |
3092 | } | |
3093 | ||
d64a7232 | 3094 | sub rex { |
0a9a692e AP |
3095 | local *opcode=shift; |
3096 | my ($dst,$src)=@_; | |
3097 | my $rex=0; | |
3098 | ||
3099 | $rex|=0x04 if($dst>=8); | |
3100 | $rex|=0x01 if($src>=8); | |
3101 | push @opcode,$rex|0x40 if($rex); | |
d64a7232 AP |
3102 | } |
3103 | ||
3104 | sub aesni { | |
3105 | my $line=shift; | |
3106 | my @opcode=(0x66); | |
3107 | ||
3108 | if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) { | |
3109 | rex(\@opcode,$4,$3); | |
3110 | push @opcode,0x0f,0x3a,0xdf; | |
3111 | push @opcode,0xc0|($3&7)|(($4&7)<<3); # ModR/M | |
3112 | my $c=$2; | |
3113 | push @opcode,$c=~/^0/?oct($c):$c; | |
3114 | return ".byte\t".join(',',@opcode); | |
3115 | } | |
3116 | elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) { | |
3117 | my %opcodelet = ( | |
3118 | "aesimc" => 0xdb, | |
3119 | "aesenc" => 0xdc, "aesenclast" => 0xdd, | |
3120 | "aesdec" => 0xde, "aesdeclast" => 0xdf | |
3121 | ); | |
3122 | return undef if (!defined($opcodelet{$1})); | |
3123 | rex(\@opcode,$3,$2); | |
3124 | push @opcode,0x0f,0x38,$opcodelet{$1}; | |
3125 | push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M | |
3126 | return ".byte\t".join(',',@opcode); | |
3127 | } | |
3128 | return $line; | |
3129 | } | |
3130 | ||
3131 | $code =~ s/\`([^\`]*)\`/eval($1)/gem; | |
3132 | $code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem; | |
3133 | ||
3134 | print $code; | |
3135 | ||
3136 | close STDOUT; |