| 1 |
#!/usr/bin/env perl |
| 2 |
|
| 3 |
###################################################################### |
| 4 |
## Constant-time SSSE3 AES core implementation. |
| 5 |
## version 0.1 |
| 6 |
## |
| 7 |
## By Mike Hamburg (Stanford University), 2009 |
| 8 |
## Public domain. |
| 9 |
## |
| 10 |
## For details see http://shiftleft.org/papers/vector_aes/ and |
| 11 |
## http://crypto.stanford.edu/vpaes/. |
| 12 |
|
| 13 |
###################################################################### |
| 14 |
# September 2011. |
| 15 |
# |
| 16 |
# Interface to OpenSSL as "almost" drop-in replacement for |
| 17 |
# aes-x86_64.pl. "Almost" refers to the fact that AES_cbc_encrypt |
| 18 |
# doesn't handle partial vectors (doesn't have to if called from |
| 19 |
# EVP only). "Drop-in" implies that this module doesn't share key |
| 20 |
# schedule structure with the original nor does it make assumption |
| 21 |
# about its alignment... |
| 22 |
# |
| 23 |
# Performance summary. aes-x86_64.pl column lists large-block CBC |
| 24 |
# encrypt/decrypt/with-hyper-threading-off(*) results in cycles per |
| 25 |
# byte processed with 128-bit key, and vpaes-x86_64.pl column - |
| 26 |
# [also large-block CBC] encrypt/decrypt. |
| 27 |
# |
| 28 |
# aes-x86_64.pl vpaes-x86_64.pl |
| 29 |
# |
| 30 |
# Core 2(**) 29.6/41.1/14.3 21.9/25.2(***) |
| 31 |
# Nehalem 29.6/40.3/14.6 10.0/11.8 |
| 32 |
# Atom 57.3/74.2/32.1 60.9/77.2(***) |
| 33 |
# Silvermont 52.7/64.0/19.5 48.8/60.8(***) |
| 34 |
# |
| 35 |
# (*) "Hyper-threading" in the context refers rather to cache shared |
| 36 |
# among multiple cores, than to specifically Intel HTT. As vast |
| 37 |
# majority of contemporary cores share cache, slower code path |
| 38 |
# is common place. In other words "with-hyper-threading-off" |
| 39 |
# results are presented mostly for reference purposes. |
| 40 |
# |
| 41 |
# (**) "Core 2" refers to initial 65nm design, a.k.a. Conroe. |
| 42 |
# |
| 43 |
# (***) Less impressive improvement on Core 2 and Atom is due to slow |
| 44 |
# pshufb, yet it's respectable +36%/62% improvement on Core 2 |
| 45 |
# (as implied, over "hyper-threading-safe" code path). |
| 46 |
# |
| 47 |
# <[email protected]> |
| 48 |
|
| 49 |
$flavour = shift; |
| 50 |
$output = shift; |
| 51 |
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } |
| 52 |
|
| 53 |
$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); |
| 54 |
|
| 55 |
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 56 |
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or |
| 57 |
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or |
| 58 |
die "can't locate x86_64-xlate.pl"; |
| 59 |
|
| 60 |
open OUT,"| \"$^X\" $xlate $flavour $output"; |
| 61 |
*STDOUT=*OUT; |
| 62 |
|
| 63 |
$PREFIX="vpaes"; |
| 64 |
|
| 65 |
$code.=<<___; |
| 66 |
.text |
| 67 |
|
| 68 |
## |
| 69 |
## _aes_encrypt_core |
| 70 |
## |
| 71 |
## AES-encrypt %xmm0. |
| 72 |
## |
| 73 |
## Inputs: |
| 74 |
## %xmm0 = input |
| 75 |
## %xmm9-%xmm15 as in _vpaes_preheat |
| 76 |
## (%rdx) = scheduled keys |
| 77 |
## |
| 78 |
## Output in %xmm0 |
| 79 |
## Clobbers %xmm1-%xmm5, %r9, %r10, %r11, %rax |
| 80 |
## Preserves %xmm6 - %xmm8 so you get some local vectors |
| 81 |
## |
| 82 |
## |
| 83 |
.type _vpaes_encrypt_core,\@abi-omnipotent |
| 84 |
.align 16 |
| 85 |
_vpaes_encrypt_core: |
| 86 |
mov %rdx, %r9 |
| 87 |
mov \$16, %r11 |
| 88 |
mov 240(%rdx),%eax |
| 89 |
movdqa %xmm9, %xmm1 |
| 90 |
movdqa .Lk_ipt(%rip), %xmm2 # iptlo |
| 91 |
pandn %xmm0, %xmm1 |
| 92 |
movdqu (%r9), %xmm5 # round0 key |
| 93 |
psrld \$4, %xmm1 |
| 94 |
pand %xmm9, %xmm0 |
| 95 |
pshufb %xmm0, %xmm2 |
| 96 |
movdqa .Lk_ipt+16(%rip), %xmm0 # ipthi |
| 97 |
pshufb %xmm1, %xmm0 |
| 98 |
pxor %xmm5, %xmm2 |
| 99 |
add \$16, %r9 |
| 100 |
pxor %xmm2, %xmm0 |
| 101 |
lea .Lk_mc_backward(%rip),%r10 |
| 102 |
jmp .Lenc_entry |
| 103 |
|
| 104 |
.align 16 |
| 105 |
.Lenc_loop: |
| 106 |
# middle of middle round |
| 107 |
movdqa %xmm13, %xmm4 # 4 : sb1u |
| 108 |
movdqa %xmm12, %xmm0 # 0 : sb1t |
| 109 |
pshufb %xmm2, %xmm4 # 4 = sb1u |
| 110 |
pshufb %xmm3, %xmm0 # 0 = sb1t |
| 111 |
pxor %xmm5, %xmm4 # 4 = sb1u + k |
| 112 |
movdqa %xmm15, %xmm5 # 4 : sb2u |
| 113 |
pxor %xmm4, %xmm0 # 0 = A |
| 114 |
movdqa -0x40(%r11,%r10), %xmm1 # .Lk_mc_forward[] |
| 115 |
pshufb %xmm2, %xmm5 # 4 = sb2u |
| 116 |
movdqa (%r11,%r10), %xmm4 # .Lk_mc_backward[] |
| 117 |
movdqa %xmm14, %xmm2 # 2 : sb2t |
| 118 |
pshufb %xmm3, %xmm2 # 2 = sb2t |
| 119 |
movdqa %xmm0, %xmm3 # 3 = A |
| 120 |
pxor %xmm5, %xmm2 # 2 = 2A |
| 121 |
pshufb %xmm1, %xmm0 # 0 = B |
| 122 |
add \$16, %r9 # next key |
| 123 |
pxor %xmm2, %xmm0 # 0 = 2A+B |
| 124 |
pshufb %xmm4, %xmm3 # 3 = D |
| 125 |
add \$16, %r11 # next mc |
| 126 |
pxor %xmm0, %xmm3 # 3 = 2A+B+D |
| 127 |
pshufb %xmm1, %xmm0 # 0 = 2B+C |
| 128 |
and \$0x30, %r11 # ... mod 4 |
| 129 |
sub \$1,%rax # nr-- |
| 130 |
pxor %xmm3, %xmm0 # 0 = 2A+3B+C+D |
| 131 |
|
| 132 |
.Lenc_entry: |
| 133 |
# top of round |
| 134 |
movdqa %xmm9, %xmm1 # 1 : i |
| 135 |
movdqa %xmm11, %xmm5 # 2 : a/k |
| 136 |
pandn %xmm0, %xmm1 # 1 = i<<4 |
| 137 |
psrld \$4, %xmm1 # 1 = i |
| 138 |
pand %xmm9, %xmm0 # 0 = k |
| 139 |
pshufb %xmm0, %xmm5 # 2 = a/k |
| 140 |
movdqa %xmm10, %xmm3 # 3 : 1/i |
| 141 |
pxor %xmm1, %xmm0 # 0 = j |
| 142 |
pshufb %xmm1, %xmm3 # 3 = 1/i |
| 143 |
movdqa %xmm10, %xmm4 # 4 : 1/j |
| 144 |
pxor %xmm5, %xmm3 # 3 = iak = 1/i + a/k |
| 145 |
pshufb %xmm0, %xmm4 # 4 = 1/j |
| 146 |
movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 147 |
pxor %xmm5, %xmm4 # 4 = jak = 1/j + a/k |
| 148 |
pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 149 |
movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 150 |
pxor %xmm0, %xmm2 # 2 = io |
| 151 |
pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 152 |
movdqu (%r9), %xmm5 |
| 153 |
pxor %xmm1, %xmm3 # 3 = jo |
| 154 |
jnz .Lenc_loop |
| 155 |
|
| 156 |
# middle of last round |
| 157 |
movdqa -0x60(%r10), %xmm4 # 3 : sbou .Lk_sbo |
| 158 |
movdqa -0x50(%r10), %xmm0 # 0 : sbot .Lk_sbo+16 |
| 159 |
pshufb %xmm2, %xmm4 # 4 = sbou |
| 160 |
pxor %xmm5, %xmm4 # 4 = sb1u + k |
| 161 |
pshufb %xmm3, %xmm0 # 0 = sb1t |
| 162 |
movdqa 0x40(%r11,%r10), %xmm1 # .Lk_sr[] |
| 163 |
pxor %xmm4, %xmm0 # 0 = A |
| 164 |
pshufb %xmm1, %xmm0 |
| 165 |
ret |
| 166 |
.size _vpaes_encrypt_core,.-_vpaes_encrypt_core |
| 167 |
|
| 168 |
## |
| 169 |
## Decryption core |
| 170 |
## |
| 171 |
## Same API as encryption core. |
| 172 |
## |
| 173 |
.type _vpaes_decrypt_core,\@abi-omnipotent |
| 174 |
.align 16 |
| 175 |
_vpaes_decrypt_core: |
| 176 |
mov %rdx, %r9 # load key |
| 177 |
mov 240(%rdx),%eax |
| 178 |
movdqa %xmm9, %xmm1 |
| 179 |
movdqa .Lk_dipt(%rip), %xmm2 # iptlo |
| 180 |
pandn %xmm0, %xmm1 |
| 181 |
mov %rax, %r11 |
| 182 |
psrld \$4, %xmm1 |
| 183 |
movdqu (%r9), %xmm5 # round0 key |
| 184 |
shl \$4, %r11 |
| 185 |
pand %xmm9, %xmm0 |
| 186 |
pshufb %xmm0, %xmm2 |
| 187 |
movdqa .Lk_dipt+16(%rip), %xmm0 # ipthi |
| 188 |
xor \$0x30, %r11 |
| 189 |
lea .Lk_dsbd(%rip),%r10 |
| 190 |
pshufb %xmm1, %xmm0 |
| 191 |
and \$0x30, %r11 |
| 192 |
pxor %xmm5, %xmm2 |
| 193 |
movdqa .Lk_mc_forward+48(%rip), %xmm5 |
| 194 |
pxor %xmm2, %xmm0 |
| 195 |
add \$16, %r9 |
| 196 |
add %r10, %r11 |
| 197 |
jmp .Ldec_entry |
| 198 |
|
| 199 |
.align 16 |
| 200 |
.Ldec_loop: |
| 201 |
## |
| 202 |
## Inverse mix columns |
| 203 |
## |
| 204 |
movdqa -0x20(%r10),%xmm4 # 4 : sb9u |
| 205 |
movdqa -0x10(%r10),%xmm1 # 0 : sb9t |
| 206 |
pshufb %xmm2, %xmm4 # 4 = sb9u |
| 207 |
pshufb %xmm3, %xmm1 # 0 = sb9t |
| 208 |
pxor %xmm4, %xmm0 |
| 209 |
movdqa 0x00(%r10),%xmm4 # 4 : sbdu |
| 210 |
pxor %xmm1, %xmm0 # 0 = ch |
| 211 |
movdqa 0x10(%r10),%xmm1 # 0 : sbdt |
| 212 |
|
| 213 |
pshufb %xmm2, %xmm4 # 4 = sbdu |
| 214 |
pshufb %xmm5, %xmm0 # MC ch |
| 215 |
pshufb %xmm3, %xmm1 # 0 = sbdt |
| 216 |
pxor %xmm4, %xmm0 # 4 = ch |
| 217 |
movdqa 0x20(%r10),%xmm4 # 4 : sbbu |
| 218 |
pxor %xmm1, %xmm0 # 0 = ch |
| 219 |
movdqa 0x30(%r10),%xmm1 # 0 : sbbt |
| 220 |
|
| 221 |
pshufb %xmm2, %xmm4 # 4 = sbbu |
| 222 |
pshufb %xmm5, %xmm0 # MC ch |
| 223 |
pshufb %xmm3, %xmm1 # 0 = sbbt |
| 224 |
pxor %xmm4, %xmm0 # 4 = ch |
| 225 |
movdqa 0x40(%r10),%xmm4 # 4 : sbeu |
| 226 |
pxor %xmm1, %xmm0 # 0 = ch |
| 227 |
movdqa 0x50(%r10),%xmm1 # 0 : sbet |
| 228 |
|
| 229 |
pshufb %xmm2, %xmm4 # 4 = sbeu |
| 230 |
pshufb %xmm5, %xmm0 # MC ch |
| 231 |
pshufb %xmm3, %xmm1 # 0 = sbet |
| 232 |
pxor %xmm4, %xmm0 # 4 = ch |
| 233 |
add \$16, %r9 # next round key |
| 234 |
palignr \$12, %xmm5, %xmm5 |
| 235 |
pxor %xmm1, %xmm0 # 0 = ch |
| 236 |
sub \$1,%rax # nr-- |
| 237 |
|
| 238 |
.Ldec_entry: |
| 239 |
# top of round |
| 240 |
movdqa %xmm9, %xmm1 # 1 : i |
| 241 |
pandn %xmm0, %xmm1 # 1 = i<<4 |
| 242 |
movdqa %xmm11, %xmm2 # 2 : a/k |
| 243 |
psrld \$4, %xmm1 # 1 = i |
| 244 |
pand %xmm9, %xmm0 # 0 = k |
| 245 |
pshufb %xmm0, %xmm2 # 2 = a/k |
| 246 |
movdqa %xmm10, %xmm3 # 3 : 1/i |
| 247 |
pxor %xmm1, %xmm0 # 0 = j |
| 248 |
pshufb %xmm1, %xmm3 # 3 = 1/i |
| 249 |
movdqa %xmm10, %xmm4 # 4 : 1/j |
| 250 |
pxor %xmm2, %xmm3 # 3 = iak = 1/i + a/k |
| 251 |
pshufb %xmm0, %xmm4 # 4 = 1/j |
| 252 |
pxor %xmm2, %xmm4 # 4 = jak = 1/j + a/k |
| 253 |
movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 254 |
pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 255 |
movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 256 |
pxor %xmm0, %xmm2 # 2 = io |
| 257 |
pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 258 |
movdqu (%r9), %xmm0 |
| 259 |
pxor %xmm1, %xmm3 # 3 = jo |
| 260 |
jnz .Ldec_loop |
| 261 |
|
| 262 |
# middle of last round |
| 263 |
movdqa 0x60(%r10), %xmm4 # 3 : sbou |
| 264 |
pshufb %xmm2, %xmm4 # 4 = sbou |
| 265 |
pxor %xmm0, %xmm4 # 4 = sb1u + k |
| 266 |
movdqa 0x70(%r10), %xmm0 # 0 : sbot |
| 267 |
movdqa -0x160(%r11), %xmm2 # .Lk_sr-.Lk_dsbd=-0x160 |
| 268 |
pshufb %xmm3, %xmm0 # 0 = sb1t |
| 269 |
pxor %xmm4, %xmm0 # 0 = A |
| 270 |
pshufb %xmm2, %xmm0 |
| 271 |
ret |
| 272 |
.size _vpaes_decrypt_core,.-_vpaes_decrypt_core |
| 273 |
|
| 274 |
######################################################## |
| 275 |
## ## |
| 276 |
## AES key schedule ## |
| 277 |
## ## |
| 278 |
######################################################## |
| 279 |
.type _vpaes_schedule_core,\@abi-omnipotent |
| 280 |
.align 16 |
| 281 |
_vpaes_schedule_core: |
| 282 |
# rdi = key |
| 283 |
# rsi = size in bits |
| 284 |
# rdx = buffer |
| 285 |
# rcx = direction. 0=encrypt, 1=decrypt |
| 286 |
|
| 287 |
call _vpaes_preheat # load the tables |
| 288 |
movdqa .Lk_rcon(%rip), %xmm8 # load rcon |
| 289 |
movdqu (%rdi), %xmm0 # load key (unaligned) |
| 290 |
|
| 291 |
# input transform |
| 292 |
movdqa %xmm0, %xmm3 |
| 293 |
lea .Lk_ipt(%rip), %r11 |
| 294 |
call _vpaes_schedule_transform |
| 295 |
movdqa %xmm0, %xmm7 |
| 296 |
|
| 297 |
lea .Lk_sr(%rip),%r10 |
| 298 |
test %rcx, %rcx |
| 299 |
jnz .Lschedule_am_decrypting |
| 300 |
|
| 301 |
# encrypting, output zeroth round key after transform |
| 302 |
movdqu %xmm0, (%rdx) |
| 303 |
jmp .Lschedule_go |
| 304 |
|
| 305 |
.Lschedule_am_decrypting: |
| 306 |
# decrypting, output zeroth round key after shiftrows |
| 307 |
movdqa (%r8,%r10),%xmm1 |
| 308 |
pshufb %xmm1, %xmm3 |
| 309 |
movdqu %xmm3, (%rdx) |
| 310 |
xor \$0x30, %r8 |
| 311 |
|
| 312 |
.Lschedule_go: |
| 313 |
cmp \$192, %esi |
| 314 |
ja .Lschedule_256 |
| 315 |
je .Lschedule_192 |
| 316 |
# 128: fall though |
| 317 |
|
| 318 |
## |
| 319 |
## .schedule_128 |
| 320 |
## |
| 321 |
## 128-bit specific part of key schedule. |
| 322 |
## |
| 323 |
## This schedule is really simple, because all its parts |
| 324 |
## are accomplished by the subroutines. |
| 325 |
## |
| 326 |
.Lschedule_128: |
| 327 |
mov \$10, %esi |
| 328 |
|
| 329 |
.Loop_schedule_128: |
| 330 |
call _vpaes_schedule_round |
| 331 |
dec %rsi |
| 332 |
jz .Lschedule_mangle_last |
| 333 |
call _vpaes_schedule_mangle # write output |
| 334 |
jmp .Loop_schedule_128 |
| 335 |
|
| 336 |
## |
| 337 |
## .aes_schedule_192 |
| 338 |
## |
| 339 |
## 192-bit specific part of key schedule. |
| 340 |
## |
| 341 |
## The main body of this schedule is the same as the 128-bit |
| 342 |
## schedule, but with more smearing. The long, high side is |
| 343 |
## stored in %xmm7 as before, and the short, low side is in |
| 344 |
## the high bits of %xmm6. |
| 345 |
## |
| 346 |
## This schedule is somewhat nastier, however, because each |
| 347 |
## round produces 192 bits of key material, or 1.5 round keys. |
| 348 |
## Therefore, on each cycle we do 2 rounds and produce 3 round |
| 349 |
## keys. |
| 350 |
## |
| 351 |
.align 16 |
| 352 |
.Lschedule_192: |
| 353 |
movdqu 8(%rdi),%xmm0 # load key part 2 (very unaligned) |
| 354 |
call _vpaes_schedule_transform # input transform |
| 355 |
movdqa %xmm0, %xmm6 # save short part |
| 356 |
pxor %xmm4, %xmm4 # clear 4 |
| 357 |
movhlps %xmm4, %xmm6 # clobber low side with zeros |
| 358 |
mov \$4, %esi |
| 359 |
|
| 360 |
.Loop_schedule_192: |
| 361 |
call _vpaes_schedule_round |
| 362 |
palignr \$8,%xmm6,%xmm0 |
| 363 |
call _vpaes_schedule_mangle # save key n |
| 364 |
call _vpaes_schedule_192_smear |
| 365 |
call _vpaes_schedule_mangle # save key n+1 |
| 366 |
call _vpaes_schedule_round |
| 367 |
dec %rsi |
| 368 |
jz .Lschedule_mangle_last |
| 369 |
call _vpaes_schedule_mangle # save key n+2 |
| 370 |
call _vpaes_schedule_192_smear |
| 371 |
jmp .Loop_schedule_192 |
| 372 |
|
| 373 |
## |
| 374 |
## .aes_schedule_256 |
| 375 |
## |
| 376 |
## 256-bit specific part of key schedule. |
| 377 |
## |
| 378 |
## The structure here is very similar to the 128-bit |
| 379 |
## schedule, but with an additional "low side" in |
| 380 |
## %xmm6. The low side's rounds are the same as the |
| 381 |
## high side's, except no rcon and no rotation. |
| 382 |
## |
| 383 |
.align 16 |
| 384 |
.Lschedule_256: |
| 385 |
movdqu 16(%rdi),%xmm0 # load key part 2 (unaligned) |
| 386 |
call _vpaes_schedule_transform # input transform |
| 387 |
mov \$7, %esi |
| 388 |
|
| 389 |
.Loop_schedule_256: |
| 390 |
call _vpaes_schedule_mangle # output low result |
| 391 |
movdqa %xmm0, %xmm6 # save cur_lo in xmm6 |
| 392 |
|
| 393 |
# high round |
| 394 |
call _vpaes_schedule_round |
| 395 |
dec %rsi |
| 396 |
jz .Lschedule_mangle_last |
| 397 |
call _vpaes_schedule_mangle |
| 398 |
|
| 399 |
# low round. swap xmm7 and xmm6 |
| 400 |
pshufd \$0xFF, %xmm0, %xmm0 |
| 401 |
movdqa %xmm7, %xmm5 |
| 402 |
movdqa %xmm6, %xmm7 |
| 403 |
call _vpaes_schedule_low_round |
| 404 |
movdqa %xmm5, %xmm7 |
| 405 |
|
| 406 |
jmp .Loop_schedule_256 |
| 407 |
|
| 408 |
|
| 409 |
## |
| 410 |
## .aes_schedule_mangle_last |
| 411 |
## |
| 412 |
## Mangler for last round of key schedule |
| 413 |
## Mangles %xmm0 |
| 414 |
## when encrypting, outputs out(%xmm0) ^ 63 |
| 415 |
## when decrypting, outputs unskew(%xmm0) |
| 416 |
## |
| 417 |
## Always called right before return... jumps to cleanup and exits |
| 418 |
## |
| 419 |
.align 16 |
| 420 |
.Lschedule_mangle_last: |
| 421 |
# schedule last round key from xmm0 |
| 422 |
lea .Lk_deskew(%rip),%r11 # prepare to deskew |
| 423 |
test %rcx, %rcx |
| 424 |
jnz .Lschedule_mangle_last_dec |
| 425 |
|
| 426 |
# encrypting |
| 427 |
movdqa (%r8,%r10),%xmm1 |
| 428 |
pshufb %xmm1, %xmm0 # output permute |
| 429 |
lea .Lk_opt(%rip), %r11 # prepare to output transform |
| 430 |
add \$32, %rdx |
| 431 |
|
| 432 |
.Lschedule_mangle_last_dec: |
| 433 |
add \$-16, %rdx |
| 434 |
pxor .Lk_s63(%rip), %xmm0 |
| 435 |
call _vpaes_schedule_transform # output transform |
| 436 |
movdqu %xmm0, (%rdx) # save last key |
| 437 |
|
| 438 |
# cleanup |
| 439 |
pxor %xmm0, %xmm0 |
| 440 |
pxor %xmm1, %xmm1 |
| 441 |
pxor %xmm2, %xmm2 |
| 442 |
pxor %xmm3, %xmm3 |
| 443 |
pxor %xmm4, %xmm4 |
| 444 |
pxor %xmm5, %xmm5 |
| 445 |
pxor %xmm6, %xmm6 |
| 446 |
pxor %xmm7, %xmm7 |
| 447 |
ret |
| 448 |
.size _vpaes_schedule_core,.-_vpaes_schedule_core |
| 449 |
|
| 450 |
## |
| 451 |
## .aes_schedule_192_smear |
| 452 |
## |
| 453 |
## Smear the short, low side in the 192-bit key schedule. |
| 454 |
## |
| 455 |
## Inputs: |
| 456 |
## %xmm7: high side, b a x y |
| 457 |
## %xmm6: low side, d c 0 0 |
| 458 |
## %xmm13: 0 |
| 459 |
## |
| 460 |
## Outputs: |
| 461 |
## %xmm6: b+c+d b+c 0 0 |
| 462 |
## %xmm0: b+c+d b+c b a |
| 463 |
## |
| 464 |
.type _vpaes_schedule_192_smear,\@abi-omnipotent |
| 465 |
.align 16 |
| 466 |
_vpaes_schedule_192_smear: |
| 467 |
pshufd \$0x80, %xmm6, %xmm1 # d c 0 0 -> c 0 0 0 |
| 468 |
pshufd \$0xFE, %xmm7, %xmm0 # b a _ _ -> b b b a |
| 469 |
pxor %xmm1, %xmm6 # -> c+d c 0 0 |
| 470 |
pxor %xmm1, %xmm1 |
| 471 |
pxor %xmm0, %xmm6 # -> b+c+d b+c b a |
| 472 |
movdqa %xmm6, %xmm0 |
| 473 |
movhlps %xmm1, %xmm6 # clobber low side with zeros |
| 474 |
ret |
| 475 |
.size _vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear |
| 476 |
|
| 477 |
## |
| 478 |
## .aes_schedule_round |
| 479 |
## |
| 480 |
## Runs one main round of the key schedule on %xmm0, %xmm7 |
| 481 |
## |
| 482 |
## Specifically, runs subbytes on the high dword of %xmm0 |
| 483 |
## then rotates it by one byte and xors into the low dword of |
| 484 |
## %xmm7. |
| 485 |
## |
| 486 |
## Adds rcon from low byte of %xmm8, then rotates %xmm8 for |
| 487 |
## next rcon. |
| 488 |
## |
| 489 |
## Smears the dwords of %xmm7 by xoring the low into the |
| 490 |
## second low, result into third, result into highest. |
| 491 |
## |
| 492 |
## Returns results in %xmm7 = %xmm0. |
| 493 |
## Clobbers %xmm1-%xmm4, %r11. |
| 494 |
## |
| 495 |
.type _vpaes_schedule_round,\@abi-omnipotent |
| 496 |
.align 16 |
| 497 |
_vpaes_schedule_round: |
| 498 |
# extract rcon from xmm8 |
| 499 |
pxor %xmm1, %xmm1 |
| 500 |
palignr \$15, %xmm8, %xmm1 |
| 501 |
palignr \$15, %xmm8, %xmm8 |
| 502 |
pxor %xmm1, %xmm7 |
| 503 |
|
| 504 |
# rotate |
| 505 |
pshufd \$0xFF, %xmm0, %xmm0 |
| 506 |
palignr \$1, %xmm0, %xmm0 |
| 507 |
|
| 508 |
# fall through... |
| 509 |
|
| 510 |
# low round: same as high round, but no rotation and no rcon. |
| 511 |
_vpaes_schedule_low_round: |
| 512 |
# smear xmm7 |
| 513 |
movdqa %xmm7, %xmm1 |
| 514 |
pslldq \$4, %xmm7 |
| 515 |
pxor %xmm1, %xmm7 |
| 516 |
movdqa %xmm7, %xmm1 |
| 517 |
pslldq \$8, %xmm7 |
| 518 |
pxor %xmm1, %xmm7 |
| 519 |
pxor .Lk_s63(%rip), %xmm7 |
| 520 |
|
| 521 |
# subbytes |
| 522 |
movdqa %xmm9, %xmm1 |
| 523 |
pandn %xmm0, %xmm1 |
| 524 |
psrld \$4, %xmm1 # 1 = i |
| 525 |
pand %xmm9, %xmm0 # 0 = k |
| 526 |
movdqa %xmm11, %xmm2 # 2 : a/k |
| 527 |
pshufb %xmm0, %xmm2 # 2 = a/k |
| 528 |
pxor %xmm1, %xmm0 # 0 = j |
| 529 |
movdqa %xmm10, %xmm3 # 3 : 1/i |
| 530 |
pshufb %xmm1, %xmm3 # 3 = 1/i |
| 531 |
pxor %xmm2, %xmm3 # 3 = iak = 1/i + a/k |
| 532 |
movdqa %xmm10, %xmm4 # 4 : 1/j |
| 533 |
pshufb %xmm0, %xmm4 # 4 = 1/j |
| 534 |
pxor %xmm2, %xmm4 # 4 = jak = 1/j + a/k |
| 535 |
movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 536 |
pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 537 |
pxor %xmm0, %xmm2 # 2 = io |
| 538 |
movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 539 |
pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 540 |
pxor %xmm1, %xmm3 # 3 = jo |
| 541 |
movdqa %xmm13, %xmm4 # 4 : sbou |
| 542 |
pshufb %xmm2, %xmm4 # 4 = sbou |
| 543 |
movdqa %xmm12, %xmm0 # 0 : sbot |
| 544 |
pshufb %xmm3, %xmm0 # 0 = sb1t |
| 545 |
pxor %xmm4, %xmm0 # 0 = sbox output |
| 546 |
|
| 547 |
# add in smeared stuff |
| 548 |
pxor %xmm7, %xmm0 |
| 549 |
movdqa %xmm0, %xmm7 |
| 550 |
ret |
| 551 |
.size _vpaes_schedule_round,.-_vpaes_schedule_round |
| 552 |
|
| 553 |
## |
| 554 |
## .aes_schedule_transform |
| 555 |
## |
| 556 |
## Linear-transform %xmm0 according to tables at (%r11) |
| 557 |
## |
| 558 |
## Requires that %xmm9 = 0x0F0F... as in preheat |
| 559 |
## Output in %xmm0 |
| 560 |
## Clobbers %xmm1, %xmm2 |
| 561 |
## |
| 562 |
.type _vpaes_schedule_transform,\@abi-omnipotent |
| 563 |
.align 16 |
| 564 |
_vpaes_schedule_transform: |
| 565 |
movdqa %xmm9, %xmm1 |
| 566 |
pandn %xmm0, %xmm1 |
| 567 |
psrld \$4, %xmm1 |
| 568 |
pand %xmm9, %xmm0 |
| 569 |
movdqa (%r11), %xmm2 # lo |
| 570 |
pshufb %xmm0, %xmm2 |
| 571 |
movdqa 16(%r11), %xmm0 # hi |
| 572 |
pshufb %xmm1, %xmm0 |
| 573 |
pxor %xmm2, %xmm0 |
| 574 |
ret |
| 575 |
.size _vpaes_schedule_transform,.-_vpaes_schedule_transform |
| 576 |
|
| 577 |
## |
| 578 |
## .aes_schedule_mangle |
| 579 |
## |
| 580 |
## Mangle xmm0 from (basis-transformed) standard version |
| 581 |
## to our version. |
| 582 |
## |
| 583 |
## On encrypt, |
| 584 |
## xor with 0x63 |
| 585 |
## multiply by circulant 0,1,1,1 |
| 586 |
## apply shiftrows transform |
| 587 |
## |
| 588 |
## On decrypt, |
| 589 |
## xor with 0x63 |
| 590 |
## multiply by "inverse mixcolumns" circulant E,B,D,9 |
| 591 |
## deskew |
| 592 |
## apply shiftrows transform |
| 593 |
## |
| 594 |
## |
| 595 |
## Writes out to (%rdx), and increments or decrements it |
| 596 |
## Keeps track of round number mod 4 in %r8 |
| 597 |
## Preserves xmm0 |
| 598 |
## Clobbers xmm1-xmm5 |
| 599 |
## |
| 600 |
.type _vpaes_schedule_mangle,\@abi-omnipotent |
| 601 |
.align 16 |
| 602 |
_vpaes_schedule_mangle: |
| 603 |
movdqa %xmm0, %xmm4 # save xmm0 for later |
| 604 |
movdqa .Lk_mc_forward(%rip),%xmm5 |
| 605 |
test %rcx, %rcx |
| 606 |
jnz .Lschedule_mangle_dec |
| 607 |
|
| 608 |
# encrypting |
| 609 |
add \$16, %rdx |
| 610 |
pxor .Lk_s63(%rip),%xmm4 |
| 611 |
pshufb %xmm5, %xmm4 |
| 612 |
movdqa %xmm4, %xmm3 |
| 613 |
pshufb %xmm5, %xmm4 |
| 614 |
pxor %xmm4, %xmm3 |
| 615 |
pshufb %xmm5, %xmm4 |
| 616 |
pxor %xmm4, %xmm3 |
| 617 |
|
| 618 |
jmp .Lschedule_mangle_both |
| 619 |
.align 16 |
| 620 |
.Lschedule_mangle_dec: |
| 621 |
# inverse mix columns |
| 622 |
lea .Lk_dksd(%rip),%r11 |
| 623 |
movdqa %xmm9, %xmm1 |
| 624 |
pandn %xmm4, %xmm1 |
| 625 |
psrld \$4, %xmm1 # 1 = hi |
| 626 |
pand %xmm9, %xmm4 # 4 = lo |
| 627 |
|
| 628 |
movdqa 0x00(%r11), %xmm2 |
| 629 |
pshufb %xmm4, %xmm2 |
| 630 |
movdqa 0x10(%r11), %xmm3 |
| 631 |
pshufb %xmm1, %xmm3 |
| 632 |
pxor %xmm2, %xmm3 |
| 633 |
pshufb %xmm5, %xmm3 |
| 634 |
|
| 635 |
movdqa 0x20(%r11), %xmm2 |
| 636 |
pshufb %xmm4, %xmm2 |
| 637 |
pxor %xmm3, %xmm2 |
| 638 |
movdqa 0x30(%r11), %xmm3 |
| 639 |
pshufb %xmm1, %xmm3 |
| 640 |
pxor %xmm2, %xmm3 |
| 641 |
pshufb %xmm5, %xmm3 |
| 642 |
|
| 643 |
movdqa 0x40(%r11), %xmm2 |
| 644 |
pshufb %xmm4, %xmm2 |
| 645 |
pxor %xmm3, %xmm2 |
| 646 |
movdqa 0x50(%r11), %xmm3 |
| 647 |
pshufb %xmm1, %xmm3 |
| 648 |
pxor %xmm2, %xmm3 |
| 649 |
pshufb %xmm5, %xmm3 |
| 650 |
|
| 651 |
movdqa 0x60(%r11), %xmm2 |
| 652 |
pshufb %xmm4, %xmm2 |
| 653 |
pxor %xmm3, %xmm2 |
| 654 |
movdqa 0x70(%r11), %xmm3 |
| 655 |
pshufb %xmm1, %xmm3 |
| 656 |
pxor %xmm2, %xmm3 |
| 657 |
|
| 658 |
add \$-16, %rdx |
| 659 |
|
| 660 |
.Lschedule_mangle_both: |
| 661 |
movdqa (%r8,%r10),%xmm1 |
| 662 |
pshufb %xmm1,%xmm3 |
| 663 |
add \$-16, %r8 |
| 664 |
and \$0x30, %r8 |
| 665 |
movdqu %xmm3, (%rdx) |
| 666 |
ret |
| 667 |
.size _vpaes_schedule_mangle,.-_vpaes_schedule_mangle |
| 668 |
|
| 669 |
# |
| 670 |
# Interface to OpenSSL |
| 671 |
# |
| 672 |
.globl ${PREFIX}_set_encrypt_key |
| 673 |
.type ${PREFIX}_set_encrypt_key,\@function,3 |
| 674 |
.align 16 |
| 675 |
${PREFIX}_set_encrypt_key: |
| 676 |
___ |
| 677 |
$code.=<<___ if ($win64); |
| 678 |
lea -0xb8(%rsp),%rsp |
| 679 |
movaps %xmm6,0x10(%rsp) |
| 680 |
movaps %xmm7,0x20(%rsp) |
| 681 |
movaps %xmm8,0x30(%rsp) |
| 682 |
movaps %xmm9,0x40(%rsp) |
| 683 |
movaps %xmm10,0x50(%rsp) |
| 684 |
movaps %xmm11,0x60(%rsp) |
| 685 |
movaps %xmm12,0x70(%rsp) |
| 686 |
movaps %xmm13,0x80(%rsp) |
| 687 |
movaps %xmm14,0x90(%rsp) |
| 688 |
movaps %xmm15,0xa0(%rsp) |
| 689 |
.Lenc_key_body: |
| 690 |
___ |
| 691 |
$code.=<<___; |
| 692 |
mov %esi,%eax |
| 693 |
shr \$5,%eax |
| 694 |
add \$5,%eax |
| 695 |
mov %eax,240(%rdx) # AES_KEY->rounds = nbits/32+5; |
| 696 |
|
| 697 |
mov \$0,%ecx |
| 698 |
mov \$0x30,%r8d |
| 699 |
call _vpaes_schedule_core |
| 700 |
___ |
| 701 |
$code.=<<___ if ($win64); |
| 702 |
movaps 0x10(%rsp),%xmm6 |
| 703 |
movaps 0x20(%rsp),%xmm7 |
| 704 |
movaps 0x30(%rsp),%xmm8 |
| 705 |
movaps 0x40(%rsp),%xmm9 |
| 706 |
movaps 0x50(%rsp),%xmm10 |
| 707 |
movaps 0x60(%rsp),%xmm11 |
| 708 |
movaps 0x70(%rsp),%xmm12 |
| 709 |
movaps 0x80(%rsp),%xmm13 |
| 710 |
movaps 0x90(%rsp),%xmm14 |
| 711 |
movaps 0xa0(%rsp),%xmm15 |
| 712 |
lea 0xb8(%rsp),%rsp |
| 713 |
.Lenc_key_epilogue: |
| 714 |
___ |
| 715 |
$code.=<<___; |
| 716 |
xor %eax,%eax |
| 717 |
ret |
| 718 |
.size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key |
| 719 |
|
| 720 |
.globl ${PREFIX}_set_decrypt_key |
| 721 |
.type ${PREFIX}_set_decrypt_key,\@function,3 |
| 722 |
.align 16 |
| 723 |
${PREFIX}_set_decrypt_key: |
| 724 |
___ |
| 725 |
$code.=<<___ if ($win64); |
| 726 |
lea -0xb8(%rsp),%rsp |
| 727 |
movaps %xmm6,0x10(%rsp) |
| 728 |
movaps %xmm7,0x20(%rsp) |
| 729 |
movaps %xmm8,0x30(%rsp) |
| 730 |
movaps %xmm9,0x40(%rsp) |
| 731 |
movaps %xmm10,0x50(%rsp) |
| 732 |
movaps %xmm11,0x60(%rsp) |
| 733 |
movaps %xmm12,0x70(%rsp) |
| 734 |
movaps %xmm13,0x80(%rsp) |
| 735 |
movaps %xmm14,0x90(%rsp) |
| 736 |
movaps %xmm15,0xa0(%rsp) |
| 737 |
.Ldec_key_body: |
| 738 |
___ |
| 739 |
$code.=<<___; |
| 740 |
mov %esi,%eax |
| 741 |
shr \$5,%eax |
| 742 |
add \$5,%eax |
| 743 |
mov %eax,240(%rdx) # AES_KEY->rounds = nbits/32+5; |
| 744 |
shl \$4,%eax |
| 745 |
lea 16(%rdx,%rax),%rdx |
| 746 |
|
| 747 |
mov \$1,%ecx |
| 748 |
mov %esi,%r8d |
| 749 |
shr \$1,%r8d |
| 750 |
and \$32,%r8d |
| 751 |
xor \$32,%r8d # nbits==192?0:32 |
| 752 |
call _vpaes_schedule_core |
| 753 |
___ |
| 754 |
$code.=<<___ if ($win64); |
| 755 |
movaps 0x10(%rsp),%xmm6 |
| 756 |
movaps 0x20(%rsp),%xmm7 |
| 757 |
movaps 0x30(%rsp),%xmm8 |
| 758 |
movaps 0x40(%rsp),%xmm9 |
| 759 |
movaps 0x50(%rsp),%xmm10 |
| 760 |
movaps 0x60(%rsp),%xmm11 |
| 761 |
movaps 0x70(%rsp),%xmm12 |
| 762 |
movaps 0x80(%rsp),%xmm13 |
| 763 |
movaps 0x90(%rsp),%xmm14 |
| 764 |
movaps 0xa0(%rsp),%xmm15 |
| 765 |
lea 0xb8(%rsp),%rsp |
| 766 |
.Ldec_key_epilogue: |
| 767 |
___ |
| 768 |
$code.=<<___; |
| 769 |
xor %eax,%eax |
| 770 |
ret |
| 771 |
.size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key |
| 772 |
|
| 773 |
.globl ${PREFIX}_encrypt |
| 774 |
.type ${PREFIX}_encrypt,\@function,3 |
| 775 |
.align 16 |
| 776 |
${PREFIX}_encrypt: |
| 777 |
___ |
| 778 |
$code.=<<___ if ($win64); |
| 779 |
lea -0xb8(%rsp),%rsp |
| 780 |
movaps %xmm6,0x10(%rsp) |
| 781 |
movaps %xmm7,0x20(%rsp) |
| 782 |
movaps %xmm8,0x30(%rsp) |
| 783 |
movaps %xmm9,0x40(%rsp) |
| 784 |
movaps %xmm10,0x50(%rsp) |
| 785 |
movaps %xmm11,0x60(%rsp) |
| 786 |
movaps %xmm12,0x70(%rsp) |
| 787 |
movaps %xmm13,0x80(%rsp) |
| 788 |
movaps %xmm14,0x90(%rsp) |
| 789 |
movaps %xmm15,0xa0(%rsp) |
| 790 |
.Lenc_body: |
| 791 |
___ |
| 792 |
$code.=<<___; |
| 793 |
movdqu (%rdi),%xmm0 |
| 794 |
call _vpaes_preheat |
| 795 |
call _vpaes_encrypt_core |
| 796 |
movdqu %xmm0,(%rsi) |
| 797 |
___ |
| 798 |
$code.=<<___ if ($win64); |
| 799 |
movaps 0x10(%rsp),%xmm6 |
| 800 |
movaps 0x20(%rsp),%xmm7 |
| 801 |
movaps 0x30(%rsp),%xmm8 |
| 802 |
movaps 0x40(%rsp),%xmm9 |
| 803 |
movaps 0x50(%rsp),%xmm10 |
| 804 |
movaps 0x60(%rsp),%xmm11 |
| 805 |
movaps 0x70(%rsp),%xmm12 |
| 806 |
movaps 0x80(%rsp),%xmm13 |
| 807 |
movaps 0x90(%rsp),%xmm14 |
| 808 |
movaps 0xa0(%rsp),%xmm15 |
| 809 |
lea 0xb8(%rsp),%rsp |
| 810 |
.Lenc_epilogue: |
| 811 |
___ |
| 812 |
$code.=<<___; |
| 813 |
ret |
| 814 |
.size ${PREFIX}_encrypt,.-${PREFIX}_encrypt |
| 815 |
|
| 816 |
.globl ${PREFIX}_decrypt |
| 817 |
.type ${PREFIX}_decrypt,\@function,3 |
| 818 |
.align 16 |
| 819 |
${PREFIX}_decrypt: |
| 820 |
___ |
| 821 |
$code.=<<___ if ($win64); |
| 822 |
lea -0xb8(%rsp),%rsp |
| 823 |
movaps %xmm6,0x10(%rsp) |
| 824 |
movaps %xmm7,0x20(%rsp) |
| 825 |
movaps %xmm8,0x30(%rsp) |
| 826 |
movaps %xmm9,0x40(%rsp) |
| 827 |
movaps %xmm10,0x50(%rsp) |
| 828 |
movaps %xmm11,0x60(%rsp) |
| 829 |
movaps %xmm12,0x70(%rsp) |
| 830 |
movaps %xmm13,0x80(%rsp) |
| 831 |
movaps %xmm14,0x90(%rsp) |
| 832 |
movaps %xmm15,0xa0(%rsp) |
| 833 |
.Ldec_body: |
| 834 |
___ |
| 835 |
$code.=<<___; |
| 836 |
movdqu (%rdi),%xmm0 |
| 837 |
call _vpaes_preheat |
| 838 |
call _vpaes_decrypt_core |
| 839 |
movdqu %xmm0,(%rsi) |
| 840 |
___ |
| 841 |
$code.=<<___ if ($win64); |
| 842 |
movaps 0x10(%rsp),%xmm6 |
| 843 |
movaps 0x20(%rsp),%xmm7 |
| 844 |
movaps 0x30(%rsp),%xmm8 |
| 845 |
movaps 0x40(%rsp),%xmm9 |
| 846 |
movaps 0x50(%rsp),%xmm10 |
| 847 |
movaps 0x60(%rsp),%xmm11 |
| 848 |
movaps 0x70(%rsp),%xmm12 |
| 849 |
movaps 0x80(%rsp),%xmm13 |
| 850 |
movaps 0x90(%rsp),%xmm14 |
| 851 |
movaps 0xa0(%rsp),%xmm15 |
| 852 |
lea 0xb8(%rsp),%rsp |
| 853 |
.Ldec_epilogue: |
| 854 |
___ |
| 855 |
$code.=<<___; |
| 856 |
ret |
| 857 |
.size ${PREFIX}_decrypt,.-${PREFIX}_decrypt |
| 858 |
___ |
| 859 |
{ |
| 860 |
my ($inp,$out,$len,$key,$ivp,$enc)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9"); |
| 861 |
# void AES_cbc_encrypt (const void char *inp, unsigned char *out, |
| 862 |
# size_t length, const AES_KEY *key, |
| 863 |
# unsigned char *ivp,const int enc); |
| 864 |
$code.=<<___; |
| 865 |
.globl ${PREFIX}_cbc_encrypt |
| 866 |
.type ${PREFIX}_cbc_encrypt,\@function,6 |
| 867 |
.align 16 |
| 868 |
${PREFIX}_cbc_encrypt: |
| 869 |
xchg $key,$len |
| 870 |
___ |
| 871 |
($len,$key)=($key,$len); |
| 872 |
$code.=<<___; |
| 873 |
sub \$16,$len |
| 874 |
jc .Lcbc_abort |
| 875 |
___ |
| 876 |
$code.=<<___ if ($win64); |
| 877 |
lea -0xb8(%rsp),%rsp |
| 878 |
movaps %xmm6,0x10(%rsp) |
| 879 |
movaps %xmm7,0x20(%rsp) |
| 880 |
movaps %xmm8,0x30(%rsp) |
| 881 |
movaps %xmm9,0x40(%rsp) |
| 882 |
movaps %xmm10,0x50(%rsp) |
| 883 |
movaps %xmm11,0x60(%rsp) |
| 884 |
movaps %xmm12,0x70(%rsp) |
| 885 |
movaps %xmm13,0x80(%rsp) |
| 886 |
movaps %xmm14,0x90(%rsp) |
| 887 |
movaps %xmm15,0xa0(%rsp) |
| 888 |
.Lcbc_body: |
| 889 |
___ |
| 890 |
$code.=<<___; |
| 891 |
movdqu ($ivp),%xmm6 # load IV |
| 892 |
sub $inp,$out |
| 893 |
call _vpaes_preheat |
| 894 |
cmp \$0,${enc}d |
| 895 |
je .Lcbc_dec_loop |
| 896 |
jmp .Lcbc_enc_loop |
| 897 |
.align 16 |
| 898 |
.Lcbc_enc_loop: |
| 899 |
movdqu ($inp),%xmm0 |
| 900 |
pxor %xmm6,%xmm0 |
| 901 |
call _vpaes_encrypt_core |
| 902 |
movdqa %xmm0,%xmm6 |
| 903 |
movdqu %xmm0,($out,$inp) |
| 904 |
lea 16($inp),$inp |
| 905 |
sub \$16,$len |
| 906 |
jnc .Lcbc_enc_loop |
| 907 |
jmp .Lcbc_done |
| 908 |
.align 16 |
| 909 |
.Lcbc_dec_loop: |
| 910 |
movdqu ($inp),%xmm0 |
| 911 |
movdqa %xmm0,%xmm7 |
| 912 |
call _vpaes_decrypt_core |
| 913 |
pxor %xmm6,%xmm0 |
| 914 |
movdqa %xmm7,%xmm6 |
| 915 |
movdqu %xmm0,($out,$inp) |
| 916 |
lea 16($inp),$inp |
| 917 |
sub \$16,$len |
| 918 |
jnc .Lcbc_dec_loop |
| 919 |
.Lcbc_done: |
| 920 |
movdqu %xmm6,($ivp) # save IV |
| 921 |
___ |
| 922 |
$code.=<<___ if ($win64); |
| 923 |
movaps 0x10(%rsp),%xmm6 |
| 924 |
movaps 0x20(%rsp),%xmm7 |
| 925 |
movaps 0x30(%rsp),%xmm8 |
| 926 |
movaps 0x40(%rsp),%xmm9 |
| 927 |
movaps 0x50(%rsp),%xmm10 |
| 928 |
movaps 0x60(%rsp),%xmm11 |
| 929 |
movaps 0x70(%rsp),%xmm12 |
| 930 |
movaps 0x80(%rsp),%xmm13 |
| 931 |
movaps 0x90(%rsp),%xmm14 |
| 932 |
movaps 0xa0(%rsp),%xmm15 |
| 933 |
lea 0xb8(%rsp),%rsp |
| 934 |
.Lcbc_epilogue: |
| 935 |
___ |
| 936 |
$code.=<<___; |
| 937 |
.Lcbc_abort: |
| 938 |
ret |
| 939 |
.size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt |
| 940 |
___ |
| 941 |
} |
| 942 |
$code.=<<___; |
| 943 |
## |
| 944 |
## _aes_preheat |
| 945 |
## |
| 946 |
## Fills register %r10 -> .aes_consts (so you can -fPIC) |
| 947 |
## and %xmm9-%xmm15 as specified below. |
| 948 |
## |
| 949 |
.type _vpaes_preheat,\@abi-omnipotent |
| 950 |
.align 16 |
| 951 |
_vpaes_preheat: |
| 952 |
lea .Lk_s0F(%rip), %r10 |
| 953 |
movdqa -0x20(%r10), %xmm10 # .Lk_inv |
| 954 |
movdqa -0x10(%r10), %xmm11 # .Lk_inv+16 |
| 955 |
movdqa 0x00(%r10), %xmm9 # .Lk_s0F |
| 956 |
movdqa 0x30(%r10), %xmm13 # .Lk_sb1 |
| 957 |
movdqa 0x40(%r10), %xmm12 # .Lk_sb1+16 |
| 958 |
movdqa 0x50(%r10), %xmm15 # .Lk_sb2 |
| 959 |
movdqa 0x60(%r10), %xmm14 # .Lk_sb2+16 |
| 960 |
ret |
| 961 |
.size _vpaes_preheat,.-_vpaes_preheat |
| 962 |
######################################################## |
| 963 |
## ## |
| 964 |
## Constants ## |
| 965 |
## ## |
| 966 |
######################################################## |
| 967 |
.type _vpaes_consts,\@object |
| 968 |
.align 64 |
| 969 |
_vpaes_consts: |
| 970 |
.Lk_inv: # inv, inva |
| 971 |
.quad 0x0E05060F0D080180, 0x040703090A0B0C02 |
| 972 |
.quad 0x01040A060F0B0780, 0x030D0E0C02050809 |
| 973 |
|
| 974 |
.Lk_s0F: # s0F |
| 975 |
.quad 0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F |
| 976 |
|
| 977 |
.Lk_ipt: # input transform (lo, hi) |
| 978 |
.quad 0xC2B2E8985A2A7000, 0xCABAE09052227808 |
| 979 |
.quad 0x4C01307D317C4D00, 0xCD80B1FCB0FDCC81 |
| 980 |
|
| 981 |
.Lk_sb1: # sb1u, sb1t |
| 982 |
.quad 0xB19BE18FCB503E00, 0xA5DF7A6E142AF544 |
| 983 |
.quad 0x3618D415FAE22300, 0x3BF7CCC10D2ED9EF |
| 984 |
.Lk_sb2: # sb2u, sb2t |
| 985 |
.quad 0xE27A93C60B712400, 0x5EB7E955BC982FCD |
| 986 |
.quad 0x69EB88400AE12900, 0xC2A163C8AB82234A |
| 987 |
.Lk_sbo: # sbou, sbot |
| 988 |
.quad 0xD0D26D176FBDC700, 0x15AABF7AC502A878 |
| 989 |
.quad 0xCFE474A55FBB6A00, 0x8E1E90D1412B35FA |
| 990 |
|
| 991 |
.Lk_mc_forward: # mc_forward |
| 992 |
.quad 0x0407060500030201, 0x0C0F0E0D080B0A09 |
| 993 |
.quad 0x080B0A0904070605, 0x000302010C0F0E0D |
| 994 |
.quad 0x0C0F0E0D080B0A09, 0x0407060500030201 |
| 995 |
.quad 0x000302010C0F0E0D, 0x080B0A0904070605 |
| 996 |
|
| 997 |
.Lk_mc_backward:# mc_backward |
| 998 |
.quad 0x0605040702010003, 0x0E0D0C0F0A09080B |
| 999 |
.quad 0x020100030E0D0C0F, 0x0A09080B06050407 |
| 1000 |
.quad 0x0E0D0C0F0A09080B, 0x0605040702010003 |
| 1001 |
.quad 0x0A09080B06050407, 0x020100030E0D0C0F |
| 1002 |
|
| 1003 |
.Lk_sr: # sr |
| 1004 |
.quad 0x0706050403020100, 0x0F0E0D0C0B0A0908 |
| 1005 |
.quad 0x030E09040F0A0500, 0x0B06010C07020D08 |
| 1006 |
.quad 0x0F060D040B020900, 0x070E050C030A0108 |
| 1007 |
.quad 0x0B0E0104070A0D00, 0x0306090C0F020508 |
| 1008 |
|
| 1009 |
.Lk_rcon: # rcon |
| 1010 |
.quad 0x1F8391B9AF9DEEB6, 0x702A98084D7C7D81 |
| 1011 |
|
| 1012 |
.Lk_s63: # s63: all equal to 0x63 transformed |
| 1013 |
.quad 0x5B5B5B5B5B5B5B5B, 0x5B5B5B5B5B5B5B5B |
| 1014 |
|
| 1015 |
.Lk_opt: # output transform |
| 1016 |
.quad 0xFF9F4929D6B66000, 0xF7974121DEBE6808 |
| 1017 |
.quad 0x01EDBD5150BCEC00, 0xE10D5DB1B05C0CE0 |
| 1018 |
|
| 1019 |
.Lk_deskew: # deskew tables: inverts the sbox's "skew" |
| 1020 |
.quad 0x07E4A34047A4E300, 0x1DFEB95A5DBEF91A |
| 1021 |
.quad 0x5F36B5DC83EA6900, 0x2841C2ABF49D1E77 |
| 1022 |
|
| 1023 |
## |
| 1024 |
## Decryption stuff |
| 1025 |
## Key schedule constants |
| 1026 |
## |
| 1027 |
.Lk_dksd: # decryption key schedule: invskew x*D |
| 1028 |
.quad 0xFEB91A5DA3E44700, 0x0740E3A45A1DBEF9 |
| 1029 |
.quad 0x41C277F4B5368300, 0x5FDC69EAAB289D1E |
| 1030 |
.Lk_dksb: # decryption key schedule: invskew x*B |
| 1031 |
.quad 0x9A4FCA1F8550D500, 0x03D653861CC94C99 |
| 1032 |
.quad 0x115BEDA7B6FC4A00, 0xD993256F7E3482C8 |
| 1033 |
.Lk_dkse: # decryption key schedule: invskew x*E + 0x63 |
| 1034 |
.quad 0xD5031CCA1FC9D600, 0x53859A4C994F5086 |
| 1035 |
.quad 0xA23196054FDC7BE8, 0xCD5EF96A20B31487 |
| 1036 |
.Lk_dks9: # decryption key schedule: invskew x*9 |
| 1037 |
.quad 0xB6116FC87ED9A700, 0x4AED933482255BFC |
| 1038 |
.quad 0x4576516227143300, 0x8BB89FACE9DAFDCE |
| 1039 |
|
| 1040 |
## |
| 1041 |
## Decryption stuff |
| 1042 |
## Round function constants |
| 1043 |
## |
| 1044 |
.Lk_dipt: # decryption input transform |
| 1045 |
.quad 0x0F505B040B545F00, 0x154A411E114E451A |
| 1046 |
.quad 0x86E383E660056500, 0x12771772F491F194 |
| 1047 |
|
| 1048 |
.Lk_dsb9: # decryption sbox output *9*u, *9*t |
| 1049 |
.quad 0x851C03539A86D600, 0xCAD51F504F994CC9 |
| 1050 |
.quad 0xC03B1789ECD74900, 0x725E2C9EB2FBA565 |
| 1051 |
.Lk_dsbd: # decryption sbox output *D*u, *D*t |
| 1052 |
.quad 0x7D57CCDFE6B1A200, 0xF56E9B13882A4439 |
| 1053 |
.quad 0x3CE2FAF724C6CB00, 0x2931180D15DEEFD3 |
| 1054 |
.Lk_dsbb: # decryption sbox output *B*u, *B*t |
| 1055 |
.quad 0xD022649296B44200, 0x602646F6B0F2D404 |
| 1056 |
.quad 0xC19498A6CD596700, 0xF3FF0C3E3255AA6B |
| 1057 |
.Lk_dsbe: # decryption sbox output *E*u, *E*t |
| 1058 |
.quad 0x46F2929626D4D000, 0x2242600464B4F6B0 |
| 1059 |
.quad 0x0C55A6CDFFAAC100, 0x9467F36B98593E32 |
| 1060 |
.Lk_dsbo: # decryption sbox final output |
| 1061 |
.quad 0x1387EA537EF94000, 0xC7AA6DB9D4943E2D |
| 1062 |
.quad 0x12D7560F93441D00, 0xCA4B8159D8C58E9C |
| 1063 |
.asciz "Vector Permutation AES for x86_64/SSSE3, Mike Hamburg (Stanford University)" |
| 1064 |
.align 64 |
| 1065 |
.size _vpaes_consts,.-_vpaes_consts |
| 1066 |
___ |
| 1067 |
|
| 1068 |
if ($win64) { |
| 1069 |
# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, |
| 1070 |
# CONTEXT *context,DISPATCHER_CONTEXT *disp) |
| 1071 |
$rec="%rcx"; |
| 1072 |
$frame="%rdx"; |
| 1073 |
$context="%r8"; |
| 1074 |
$disp="%r9"; |
| 1075 |
|
| 1076 |
$code.=<<___; |
| 1077 |
.extern __imp_RtlVirtualUnwind |
| 1078 |
.type se_handler,\@abi-omnipotent |
| 1079 |
.align 16 |
| 1080 |
se_handler: |
| 1081 |
push %rsi |
| 1082 |
push %rdi |
| 1083 |
push %rbx |
| 1084 |
push %rbp |
| 1085 |
push %r12 |
| 1086 |
push %r13 |
| 1087 |
push %r14 |
| 1088 |
push %r15 |
| 1089 |
pushfq |
| 1090 |
sub \$64,%rsp |
| 1091 |
|
| 1092 |
mov 120($context),%rax # pull context->Rax |
| 1093 |
mov 248($context),%rbx # pull context->Rip |
| 1094 |
|
| 1095 |
mov 8($disp),%rsi # disp->ImageBase |
| 1096 |
mov 56($disp),%r11 # disp->HandlerData |
| 1097 |
|
| 1098 |
mov 0(%r11),%r10d # HandlerData[0] |
| 1099 |
lea (%rsi,%r10),%r10 # prologue label |
| 1100 |
cmp %r10,%rbx # context->Rip<prologue label |
| 1101 |
jb .Lin_prologue |
| 1102 |
|
| 1103 |
mov 152($context),%rax # pull context->Rsp |
| 1104 |
|
| 1105 |
mov 4(%r11),%r10d # HandlerData[1] |
| 1106 |
lea (%rsi,%r10),%r10 # epilogue label |
| 1107 |
cmp %r10,%rbx # context->Rip>=epilogue label |
| 1108 |
jae .Lin_prologue |
| 1109 |
|
| 1110 |
lea 16(%rax),%rsi # %xmm save area |
| 1111 |
lea 512($context),%rdi # &context.Xmm6 |
| 1112 |
mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) |
| 1113 |
.long 0xa548f3fc # cld; rep movsq |
| 1114 |
lea 0xb8(%rax),%rax # adjust stack pointer |
| 1115 |
|
| 1116 |
.Lin_prologue: |
| 1117 |
mov 8(%rax),%rdi |
| 1118 |
mov 16(%rax),%rsi |
| 1119 |
mov %rax,152($context) # restore context->Rsp |
| 1120 |
mov %rsi,168($context) # restore context->Rsi |
| 1121 |
mov %rdi,176($context) # restore context->Rdi |
| 1122 |
|
| 1123 |
mov 40($disp),%rdi # disp->ContextRecord |
| 1124 |
mov $context,%rsi # context |
| 1125 |
mov \$`1232/8`,%ecx # sizeof(CONTEXT) |
| 1126 |
.long 0xa548f3fc # cld; rep movsq |
| 1127 |
|
| 1128 |
mov $disp,%rsi |
| 1129 |
xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER |
| 1130 |
mov 8(%rsi),%rdx # arg2, disp->ImageBase |
| 1131 |
mov 0(%rsi),%r8 # arg3, disp->ControlPc |
| 1132 |
mov 16(%rsi),%r9 # arg4, disp->FunctionEntry |
| 1133 |
mov 40(%rsi),%r10 # disp->ContextRecord |
| 1134 |
lea 56(%rsi),%r11 # &disp->HandlerData |
| 1135 |
lea 24(%rsi),%r12 # &disp->EstablisherFrame |
| 1136 |
mov %r10,32(%rsp) # arg5 |
| 1137 |
mov %r11,40(%rsp) # arg6 |
| 1138 |
mov %r12,48(%rsp) # arg7 |
| 1139 |
mov %rcx,56(%rsp) # arg8, (NULL) |
| 1140 |
call *__imp_RtlVirtualUnwind(%rip) |
| 1141 |
|
| 1142 |
mov \$1,%eax # ExceptionContinueSearch |
| 1143 |
add \$64,%rsp |
| 1144 |
popfq |
| 1145 |
pop %r15 |
| 1146 |
pop %r14 |
| 1147 |
pop %r13 |
| 1148 |
pop %r12 |
| 1149 |
pop %rbp |
| 1150 |
pop %rbx |
| 1151 |
pop %rdi |
| 1152 |
pop %rsi |
| 1153 |
ret |
| 1154 |
.size se_handler,.-se_handler |
| 1155 |
|
| 1156 |
.section .pdata |
| 1157 |
.align 4 |
| 1158 |
.rva .LSEH_begin_${PREFIX}_set_encrypt_key |
| 1159 |
.rva .LSEH_end_${PREFIX}_set_encrypt_key |
| 1160 |
.rva .LSEH_info_${PREFIX}_set_encrypt_key |
| 1161 |
|
| 1162 |
.rva .LSEH_begin_${PREFIX}_set_decrypt_key |
| 1163 |
.rva .LSEH_end_${PREFIX}_set_decrypt_key |
| 1164 |
.rva .LSEH_info_${PREFIX}_set_decrypt_key |
| 1165 |
|
| 1166 |
.rva .LSEH_begin_${PREFIX}_encrypt |
| 1167 |
.rva .LSEH_end_${PREFIX}_encrypt |
| 1168 |
.rva .LSEH_info_${PREFIX}_encrypt |
| 1169 |
|
| 1170 |
.rva .LSEH_begin_${PREFIX}_decrypt |
| 1171 |
.rva .LSEH_end_${PREFIX}_decrypt |
| 1172 |
.rva .LSEH_info_${PREFIX}_decrypt |
| 1173 |
|
| 1174 |
.rva .LSEH_begin_${PREFIX}_cbc_encrypt |
| 1175 |
.rva .LSEH_end_${PREFIX}_cbc_encrypt |
| 1176 |
.rva .LSEH_info_${PREFIX}_cbc_encrypt |
| 1177 |
|
| 1178 |
.section .xdata |
| 1179 |
.align 8 |
| 1180 |
.LSEH_info_${PREFIX}_set_encrypt_key: |
| 1181 |
.byte 9,0,0,0 |
| 1182 |
.rva se_handler |
| 1183 |
.rva .Lenc_key_body,.Lenc_key_epilogue # HandlerData[] |
| 1184 |
.LSEH_info_${PREFIX}_set_decrypt_key: |
| 1185 |
.byte 9,0,0,0 |
| 1186 |
.rva se_handler |
| 1187 |
.rva .Ldec_key_body,.Ldec_key_epilogue # HandlerData[] |
| 1188 |
.LSEH_info_${PREFIX}_encrypt: |
| 1189 |
.byte 9,0,0,0 |
| 1190 |
.rva se_handler |
| 1191 |
.rva .Lenc_body,.Lenc_epilogue # HandlerData[] |
| 1192 |
.LSEH_info_${PREFIX}_decrypt: |
| 1193 |
.byte 9,0,0,0 |
| 1194 |
.rva se_handler |
| 1195 |
.rva .Ldec_body,.Ldec_epilogue # HandlerData[] |
| 1196 |
.LSEH_info_${PREFIX}_cbc_encrypt: |
| 1197 |
.byte 9,0,0,0 |
| 1198 |
.rva se_handler |
| 1199 |
.rva .Lcbc_body,.Lcbc_epilogue # HandlerData[] |
| 1200 |
___ |
| 1201 |
} |
| 1202 |
|
| 1203 |
$code =~ s/\`([^\`]*)\`/eval($1)/gem; |
| 1204 |
|
| 1205 |
print $code; |
| 1206 |
|
| 1207 |
close STDOUT; |