Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 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(***) |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 34 | # Goldmont 38.9/49.0/17.8 10.6/12.6 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 35 | # |
| 36 | # (*) "Hyper-threading" in the context refers rather to cache shared |
| 37 | # among multiple cores, than to specifically Intel HTT. As vast |
| 38 | # majority of contemporary cores share cache, slower code path |
| 39 | # is common place. In other words "with-hyper-threading-off" |
| 40 | # results are presented mostly for reference purposes. |
| 41 | # |
| 42 | # (**) "Core 2" refers to initial 65nm design, a.k.a. Conroe. |
| 43 | # |
| 44 | # (***) Less impressive improvement on Core 2 and Atom is due to slow |
| 45 | # pshufb, yet it's respectable +36%/62% improvement on Core 2 |
| 46 | # (as implied, over "hyper-threading-safe" code path). |
| 47 | # |
| 48 | # <appro@openssl.org> |
| 49 | |
| 50 | $flavour = shift; |
| 51 | $output = shift; |
| 52 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } |
| 53 | |
| 54 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); |
| 55 | |
| 56 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 57 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or |
Robert Sloan | 572a4e2 | 2017-04-17 10:52:19 -0700 | [diff] [blame] | 58 | ( $xlate="${dir}../../../perlasm/x86_64-xlate.pl" and -f $xlate) or |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 59 | die "can't locate x86_64-xlate.pl"; |
| 60 | |
David Benjamin | c895d6b | 2016-08-11 13:26:41 -0400 | [diff] [blame] | 61 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 62 | *STDOUT=*OUT; |
| 63 | |
| 64 | $PREFIX="vpaes"; |
| 65 | |
| 66 | $code.=<<___; |
| 67 | .text |
| 68 | |
| 69 | ## |
| 70 | ## _aes_encrypt_core |
| 71 | ## |
| 72 | ## AES-encrypt %xmm0. |
| 73 | ## |
| 74 | ## Inputs: |
| 75 | ## %xmm0 = input |
| 76 | ## %xmm9-%xmm15 as in _vpaes_preheat |
| 77 | ## (%rdx) = scheduled keys |
| 78 | ## |
| 79 | ## Output in %xmm0 |
| 80 | ## Clobbers %xmm1-%xmm5, %r9, %r10, %r11, %rax |
| 81 | ## Preserves %xmm6 - %xmm8 so you get some local vectors |
| 82 | ## |
| 83 | ## |
| 84 | .type _vpaes_encrypt_core,\@abi-omnipotent |
| 85 | .align 16 |
| 86 | _vpaes_encrypt_core: |
| 87 | mov %rdx, %r9 |
| 88 | mov \$16, %r11 |
| 89 | mov 240(%rdx),%eax |
| 90 | movdqa %xmm9, %xmm1 |
| 91 | movdqa .Lk_ipt(%rip), %xmm2 # iptlo |
| 92 | pandn %xmm0, %xmm1 |
| 93 | movdqu (%r9), %xmm5 # round0 key |
| 94 | psrld \$4, %xmm1 |
| 95 | pand %xmm9, %xmm0 |
| 96 | pshufb %xmm0, %xmm2 |
| 97 | movdqa .Lk_ipt+16(%rip), %xmm0 # ipthi |
| 98 | pshufb %xmm1, %xmm0 |
| 99 | pxor %xmm5, %xmm2 |
| 100 | add \$16, %r9 |
| 101 | pxor %xmm2, %xmm0 |
| 102 | lea .Lk_mc_backward(%rip),%r10 |
| 103 | jmp .Lenc_entry |
| 104 | |
| 105 | .align 16 |
| 106 | .Lenc_loop: |
| 107 | # middle of middle round |
| 108 | movdqa %xmm13, %xmm4 # 4 : sb1u |
| 109 | movdqa %xmm12, %xmm0 # 0 : sb1t |
| 110 | pshufb %xmm2, %xmm4 # 4 = sb1u |
| 111 | pshufb %xmm3, %xmm0 # 0 = sb1t |
| 112 | pxor %xmm5, %xmm4 # 4 = sb1u + k |
| 113 | movdqa %xmm15, %xmm5 # 4 : sb2u |
| 114 | pxor %xmm4, %xmm0 # 0 = A |
| 115 | movdqa -0x40(%r11,%r10), %xmm1 # .Lk_mc_forward[] |
| 116 | pshufb %xmm2, %xmm5 # 4 = sb2u |
| 117 | movdqa (%r11,%r10), %xmm4 # .Lk_mc_backward[] |
| 118 | movdqa %xmm14, %xmm2 # 2 : sb2t |
| 119 | pshufb %xmm3, %xmm2 # 2 = sb2t |
| 120 | movdqa %xmm0, %xmm3 # 3 = A |
| 121 | pxor %xmm5, %xmm2 # 2 = 2A |
| 122 | pshufb %xmm1, %xmm0 # 0 = B |
| 123 | add \$16, %r9 # next key |
| 124 | pxor %xmm2, %xmm0 # 0 = 2A+B |
| 125 | pshufb %xmm4, %xmm3 # 3 = D |
| 126 | add \$16, %r11 # next mc |
| 127 | pxor %xmm0, %xmm3 # 3 = 2A+B+D |
| 128 | pshufb %xmm1, %xmm0 # 0 = 2B+C |
| 129 | and \$0x30, %r11 # ... mod 4 |
| 130 | sub \$1,%rax # nr-- |
| 131 | pxor %xmm3, %xmm0 # 0 = 2A+3B+C+D |
| 132 | |
| 133 | .Lenc_entry: |
| 134 | # top of round |
| 135 | movdqa %xmm9, %xmm1 # 1 : i |
| 136 | movdqa %xmm11, %xmm5 # 2 : a/k |
| 137 | pandn %xmm0, %xmm1 # 1 = i<<4 |
| 138 | psrld \$4, %xmm1 # 1 = i |
| 139 | pand %xmm9, %xmm0 # 0 = k |
| 140 | pshufb %xmm0, %xmm5 # 2 = a/k |
| 141 | movdqa %xmm10, %xmm3 # 3 : 1/i |
| 142 | pxor %xmm1, %xmm0 # 0 = j |
| 143 | pshufb %xmm1, %xmm3 # 3 = 1/i |
| 144 | movdqa %xmm10, %xmm4 # 4 : 1/j |
| 145 | pxor %xmm5, %xmm3 # 3 = iak = 1/i + a/k |
| 146 | pshufb %xmm0, %xmm4 # 4 = 1/j |
| 147 | movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 148 | pxor %xmm5, %xmm4 # 4 = jak = 1/j + a/k |
| 149 | pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 150 | movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 151 | pxor %xmm0, %xmm2 # 2 = io |
| 152 | pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 153 | movdqu (%r9), %xmm5 |
| 154 | pxor %xmm1, %xmm3 # 3 = jo |
| 155 | jnz .Lenc_loop |
| 156 | |
| 157 | # middle of last round |
| 158 | movdqa -0x60(%r10), %xmm4 # 3 : sbou .Lk_sbo |
| 159 | movdqa -0x50(%r10), %xmm0 # 0 : sbot .Lk_sbo+16 |
| 160 | pshufb %xmm2, %xmm4 # 4 = sbou |
| 161 | pxor %xmm5, %xmm4 # 4 = sb1u + k |
| 162 | pshufb %xmm3, %xmm0 # 0 = sb1t |
| 163 | movdqa 0x40(%r11,%r10), %xmm1 # .Lk_sr[] |
| 164 | pxor %xmm4, %xmm0 # 0 = A |
| 165 | pshufb %xmm1, %xmm0 |
| 166 | ret |
| 167 | .size _vpaes_encrypt_core,.-_vpaes_encrypt_core |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 168 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 169 | ## |
| 170 | ## Decryption core |
| 171 | ## |
| 172 | ## Same API as encryption core. |
| 173 | ## |
| 174 | .type _vpaes_decrypt_core,\@abi-omnipotent |
| 175 | .align 16 |
| 176 | _vpaes_decrypt_core: |
| 177 | mov %rdx, %r9 # load key |
| 178 | mov 240(%rdx),%eax |
| 179 | movdqa %xmm9, %xmm1 |
| 180 | movdqa .Lk_dipt(%rip), %xmm2 # iptlo |
| 181 | pandn %xmm0, %xmm1 |
| 182 | mov %rax, %r11 |
| 183 | psrld \$4, %xmm1 |
| 184 | movdqu (%r9), %xmm5 # round0 key |
| 185 | shl \$4, %r11 |
| 186 | pand %xmm9, %xmm0 |
| 187 | pshufb %xmm0, %xmm2 |
| 188 | movdqa .Lk_dipt+16(%rip), %xmm0 # ipthi |
| 189 | xor \$0x30, %r11 |
| 190 | lea .Lk_dsbd(%rip),%r10 |
| 191 | pshufb %xmm1, %xmm0 |
| 192 | and \$0x30, %r11 |
| 193 | pxor %xmm5, %xmm2 |
| 194 | movdqa .Lk_mc_forward+48(%rip), %xmm5 |
| 195 | pxor %xmm2, %xmm0 |
| 196 | add \$16, %r9 |
| 197 | add %r10, %r11 |
| 198 | jmp .Ldec_entry |
| 199 | |
| 200 | .align 16 |
| 201 | .Ldec_loop: |
| 202 | ## |
| 203 | ## Inverse mix columns |
| 204 | ## |
| 205 | movdqa -0x20(%r10),%xmm4 # 4 : sb9u |
| 206 | movdqa -0x10(%r10),%xmm1 # 0 : sb9t |
| 207 | pshufb %xmm2, %xmm4 # 4 = sb9u |
| 208 | pshufb %xmm3, %xmm1 # 0 = sb9t |
| 209 | pxor %xmm4, %xmm0 |
| 210 | movdqa 0x00(%r10),%xmm4 # 4 : sbdu |
| 211 | pxor %xmm1, %xmm0 # 0 = ch |
| 212 | movdqa 0x10(%r10),%xmm1 # 0 : sbdt |
| 213 | |
| 214 | pshufb %xmm2, %xmm4 # 4 = sbdu |
| 215 | pshufb %xmm5, %xmm0 # MC ch |
| 216 | pshufb %xmm3, %xmm1 # 0 = sbdt |
| 217 | pxor %xmm4, %xmm0 # 4 = ch |
| 218 | movdqa 0x20(%r10),%xmm4 # 4 : sbbu |
| 219 | pxor %xmm1, %xmm0 # 0 = ch |
| 220 | movdqa 0x30(%r10),%xmm1 # 0 : sbbt |
| 221 | |
| 222 | pshufb %xmm2, %xmm4 # 4 = sbbu |
| 223 | pshufb %xmm5, %xmm0 # MC ch |
| 224 | pshufb %xmm3, %xmm1 # 0 = sbbt |
| 225 | pxor %xmm4, %xmm0 # 4 = ch |
| 226 | movdqa 0x40(%r10),%xmm4 # 4 : sbeu |
| 227 | pxor %xmm1, %xmm0 # 0 = ch |
| 228 | movdqa 0x50(%r10),%xmm1 # 0 : sbet |
| 229 | |
| 230 | pshufb %xmm2, %xmm4 # 4 = sbeu |
| 231 | pshufb %xmm5, %xmm0 # MC ch |
| 232 | pshufb %xmm3, %xmm1 # 0 = sbet |
| 233 | pxor %xmm4, %xmm0 # 4 = ch |
| 234 | add \$16, %r9 # next round key |
| 235 | palignr \$12, %xmm5, %xmm5 |
| 236 | pxor %xmm1, %xmm0 # 0 = ch |
| 237 | sub \$1,%rax # nr-- |
| 238 | |
| 239 | .Ldec_entry: |
| 240 | # top of round |
| 241 | movdqa %xmm9, %xmm1 # 1 : i |
| 242 | pandn %xmm0, %xmm1 # 1 = i<<4 |
| 243 | movdqa %xmm11, %xmm2 # 2 : a/k |
| 244 | psrld \$4, %xmm1 # 1 = i |
| 245 | pand %xmm9, %xmm0 # 0 = k |
| 246 | pshufb %xmm0, %xmm2 # 2 = a/k |
| 247 | movdqa %xmm10, %xmm3 # 3 : 1/i |
| 248 | pxor %xmm1, %xmm0 # 0 = j |
| 249 | pshufb %xmm1, %xmm3 # 3 = 1/i |
| 250 | movdqa %xmm10, %xmm4 # 4 : 1/j |
| 251 | pxor %xmm2, %xmm3 # 3 = iak = 1/i + a/k |
| 252 | pshufb %xmm0, %xmm4 # 4 = 1/j |
| 253 | pxor %xmm2, %xmm4 # 4 = jak = 1/j + a/k |
| 254 | movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 255 | pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 256 | movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 257 | pxor %xmm0, %xmm2 # 2 = io |
| 258 | pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 259 | movdqu (%r9), %xmm0 |
| 260 | pxor %xmm1, %xmm3 # 3 = jo |
| 261 | jnz .Ldec_loop |
| 262 | |
| 263 | # middle of last round |
| 264 | movdqa 0x60(%r10), %xmm4 # 3 : sbou |
| 265 | pshufb %xmm2, %xmm4 # 4 = sbou |
| 266 | pxor %xmm0, %xmm4 # 4 = sb1u + k |
| 267 | movdqa 0x70(%r10), %xmm0 # 0 : sbot |
| 268 | movdqa -0x160(%r11), %xmm2 # .Lk_sr-.Lk_dsbd=-0x160 |
| 269 | pshufb %xmm3, %xmm0 # 0 = sb1t |
| 270 | pxor %xmm4, %xmm0 # 0 = A |
| 271 | pshufb %xmm2, %xmm0 |
| 272 | ret |
| 273 | .size _vpaes_decrypt_core,.-_vpaes_decrypt_core |
| 274 | |
| 275 | ######################################################## |
| 276 | ## ## |
| 277 | ## AES key schedule ## |
| 278 | ## ## |
| 279 | ######################################################## |
| 280 | .type _vpaes_schedule_core,\@abi-omnipotent |
| 281 | .align 16 |
| 282 | _vpaes_schedule_core: |
| 283 | # rdi = key |
| 284 | # rsi = size in bits |
| 285 | # rdx = buffer |
| 286 | # rcx = direction. 0=encrypt, 1=decrypt |
| 287 | |
| 288 | call _vpaes_preheat # load the tables |
| 289 | movdqa .Lk_rcon(%rip), %xmm8 # load rcon |
| 290 | movdqu (%rdi), %xmm0 # load key (unaligned) |
| 291 | |
| 292 | # input transform |
| 293 | movdqa %xmm0, %xmm3 |
| 294 | lea .Lk_ipt(%rip), %r11 |
| 295 | call _vpaes_schedule_transform |
| 296 | movdqa %xmm0, %xmm7 |
| 297 | |
| 298 | lea .Lk_sr(%rip),%r10 |
| 299 | test %rcx, %rcx |
| 300 | jnz .Lschedule_am_decrypting |
| 301 | |
| 302 | # encrypting, output zeroth round key after transform |
| 303 | movdqu %xmm0, (%rdx) |
| 304 | jmp .Lschedule_go |
| 305 | |
| 306 | .Lschedule_am_decrypting: |
| 307 | # decrypting, output zeroth round key after shiftrows |
| 308 | movdqa (%r8,%r10),%xmm1 |
| 309 | pshufb %xmm1, %xmm3 |
| 310 | movdqu %xmm3, (%rdx) |
| 311 | xor \$0x30, %r8 |
| 312 | |
| 313 | .Lschedule_go: |
| 314 | cmp \$192, %esi |
| 315 | ja .Lschedule_256 |
| 316 | je .Lschedule_192 |
| 317 | # 128: fall though |
| 318 | |
| 319 | ## |
| 320 | ## .schedule_128 |
| 321 | ## |
| 322 | ## 128-bit specific part of key schedule. |
| 323 | ## |
| 324 | ## This schedule is really simple, because all its parts |
| 325 | ## are accomplished by the subroutines. |
| 326 | ## |
| 327 | .Lschedule_128: |
| 328 | mov \$10, %esi |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 329 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 330 | .Loop_schedule_128: |
| 331 | call _vpaes_schedule_round |
| 332 | dec %rsi |
| 333 | jz .Lschedule_mangle_last |
| 334 | call _vpaes_schedule_mangle # write output |
| 335 | jmp .Loop_schedule_128 |
| 336 | |
| 337 | ## |
| 338 | ## .aes_schedule_192 |
| 339 | ## |
| 340 | ## 192-bit specific part of key schedule. |
| 341 | ## |
| 342 | ## The main body of this schedule is the same as the 128-bit |
| 343 | ## schedule, but with more smearing. The long, high side is |
| 344 | ## stored in %xmm7 as before, and the short, low side is in |
| 345 | ## the high bits of %xmm6. |
| 346 | ## |
| 347 | ## This schedule is somewhat nastier, however, because each |
| 348 | ## round produces 192 bits of key material, or 1.5 round keys. |
| 349 | ## Therefore, on each cycle we do 2 rounds and produce 3 round |
| 350 | ## keys. |
| 351 | ## |
| 352 | .align 16 |
| 353 | .Lschedule_192: |
| 354 | movdqu 8(%rdi),%xmm0 # load key part 2 (very unaligned) |
| 355 | call _vpaes_schedule_transform # input transform |
| 356 | movdqa %xmm0, %xmm6 # save short part |
| 357 | pxor %xmm4, %xmm4 # clear 4 |
| 358 | movhlps %xmm4, %xmm6 # clobber low side with zeros |
| 359 | mov \$4, %esi |
| 360 | |
| 361 | .Loop_schedule_192: |
| 362 | call _vpaes_schedule_round |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 363 | palignr \$8,%xmm6,%xmm0 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 364 | call _vpaes_schedule_mangle # save key n |
| 365 | call _vpaes_schedule_192_smear |
| 366 | call _vpaes_schedule_mangle # save key n+1 |
| 367 | call _vpaes_schedule_round |
| 368 | dec %rsi |
| 369 | jz .Lschedule_mangle_last |
| 370 | call _vpaes_schedule_mangle # save key n+2 |
| 371 | call _vpaes_schedule_192_smear |
| 372 | jmp .Loop_schedule_192 |
| 373 | |
| 374 | ## |
| 375 | ## .aes_schedule_256 |
| 376 | ## |
| 377 | ## 256-bit specific part of key schedule. |
| 378 | ## |
| 379 | ## The structure here is very similar to the 128-bit |
| 380 | ## schedule, but with an additional "low side" in |
| 381 | ## %xmm6. The low side's rounds are the same as the |
| 382 | ## high side's, except no rcon and no rotation. |
| 383 | ## |
| 384 | .align 16 |
| 385 | .Lschedule_256: |
| 386 | movdqu 16(%rdi),%xmm0 # load key part 2 (unaligned) |
| 387 | call _vpaes_schedule_transform # input transform |
| 388 | mov \$7, %esi |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 389 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 390 | .Loop_schedule_256: |
| 391 | call _vpaes_schedule_mangle # output low result |
| 392 | movdqa %xmm0, %xmm6 # save cur_lo in xmm6 |
| 393 | |
| 394 | # high round |
| 395 | call _vpaes_schedule_round |
| 396 | dec %rsi |
| 397 | jz .Lschedule_mangle_last |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 398 | call _vpaes_schedule_mangle |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 399 | |
| 400 | # low round. swap xmm7 and xmm6 |
| 401 | pshufd \$0xFF, %xmm0, %xmm0 |
| 402 | movdqa %xmm7, %xmm5 |
| 403 | movdqa %xmm6, %xmm7 |
| 404 | call _vpaes_schedule_low_round |
| 405 | movdqa %xmm5, %xmm7 |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 406 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 407 | jmp .Loop_schedule_256 |
| 408 | |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 409 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 410 | ## |
| 411 | ## .aes_schedule_mangle_last |
| 412 | ## |
| 413 | ## Mangler for last round of key schedule |
| 414 | ## Mangles %xmm0 |
| 415 | ## when encrypting, outputs out(%xmm0) ^ 63 |
| 416 | ## when decrypting, outputs unskew(%xmm0) |
| 417 | ## |
| 418 | ## Always called right before return... jumps to cleanup and exits |
| 419 | ## |
| 420 | .align 16 |
| 421 | .Lschedule_mangle_last: |
| 422 | # schedule last round key from xmm0 |
| 423 | lea .Lk_deskew(%rip),%r11 # prepare to deskew |
| 424 | test %rcx, %rcx |
| 425 | jnz .Lschedule_mangle_last_dec |
| 426 | |
| 427 | # encrypting |
| 428 | movdqa (%r8,%r10),%xmm1 |
| 429 | pshufb %xmm1, %xmm0 # output permute |
| 430 | lea .Lk_opt(%rip), %r11 # prepare to output transform |
| 431 | add \$32, %rdx |
| 432 | |
| 433 | .Lschedule_mangle_last_dec: |
| 434 | add \$-16, %rdx |
| 435 | pxor .Lk_s63(%rip), %xmm0 |
| 436 | call _vpaes_schedule_transform # output transform |
| 437 | movdqu %xmm0, (%rdx) # save last key |
| 438 | |
| 439 | # cleanup |
| 440 | pxor %xmm0, %xmm0 |
| 441 | pxor %xmm1, %xmm1 |
| 442 | pxor %xmm2, %xmm2 |
| 443 | pxor %xmm3, %xmm3 |
| 444 | pxor %xmm4, %xmm4 |
| 445 | pxor %xmm5, %xmm5 |
| 446 | pxor %xmm6, %xmm6 |
| 447 | pxor %xmm7, %xmm7 |
| 448 | ret |
| 449 | .size _vpaes_schedule_core,.-_vpaes_schedule_core |
| 450 | |
| 451 | ## |
| 452 | ## .aes_schedule_192_smear |
| 453 | ## |
| 454 | ## Smear the short, low side in the 192-bit key schedule. |
| 455 | ## |
| 456 | ## Inputs: |
| 457 | ## %xmm7: high side, b a x y |
| 458 | ## %xmm6: low side, d c 0 0 |
| 459 | ## %xmm13: 0 |
| 460 | ## |
| 461 | ## Outputs: |
| 462 | ## %xmm6: b+c+d b+c 0 0 |
| 463 | ## %xmm0: b+c+d b+c b a |
| 464 | ## |
| 465 | .type _vpaes_schedule_192_smear,\@abi-omnipotent |
| 466 | .align 16 |
| 467 | _vpaes_schedule_192_smear: |
| 468 | pshufd \$0x80, %xmm6, %xmm1 # d c 0 0 -> c 0 0 0 |
| 469 | pshufd \$0xFE, %xmm7, %xmm0 # b a _ _ -> b b b a |
| 470 | pxor %xmm1, %xmm6 # -> c+d c 0 0 |
| 471 | pxor %xmm1, %xmm1 |
| 472 | pxor %xmm0, %xmm6 # -> b+c+d b+c b a |
| 473 | movdqa %xmm6, %xmm0 |
| 474 | movhlps %xmm1, %xmm6 # clobber low side with zeros |
| 475 | ret |
| 476 | .size _vpaes_schedule_192_smear,.-_vpaes_schedule_192_smear |
| 477 | |
| 478 | ## |
| 479 | ## .aes_schedule_round |
| 480 | ## |
| 481 | ## Runs one main round of the key schedule on %xmm0, %xmm7 |
| 482 | ## |
| 483 | ## Specifically, runs subbytes on the high dword of %xmm0 |
| 484 | ## then rotates it by one byte and xors into the low dword of |
| 485 | ## %xmm7. |
| 486 | ## |
| 487 | ## Adds rcon from low byte of %xmm8, then rotates %xmm8 for |
| 488 | ## next rcon. |
| 489 | ## |
| 490 | ## Smears the dwords of %xmm7 by xoring the low into the |
| 491 | ## second low, result into third, result into highest. |
| 492 | ## |
| 493 | ## Returns results in %xmm7 = %xmm0. |
| 494 | ## Clobbers %xmm1-%xmm4, %r11. |
| 495 | ## |
| 496 | .type _vpaes_schedule_round,\@abi-omnipotent |
| 497 | .align 16 |
| 498 | _vpaes_schedule_round: |
| 499 | # extract rcon from xmm8 |
| 500 | pxor %xmm1, %xmm1 |
| 501 | palignr \$15, %xmm8, %xmm1 |
| 502 | palignr \$15, %xmm8, %xmm8 |
| 503 | pxor %xmm1, %xmm7 |
| 504 | |
| 505 | # rotate |
| 506 | pshufd \$0xFF, %xmm0, %xmm0 |
| 507 | palignr \$1, %xmm0, %xmm0 |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 508 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 509 | # fall through... |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 510 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 511 | # low round: same as high round, but no rotation and no rcon. |
| 512 | _vpaes_schedule_low_round: |
| 513 | # smear xmm7 |
| 514 | movdqa %xmm7, %xmm1 |
| 515 | pslldq \$4, %xmm7 |
| 516 | pxor %xmm1, %xmm7 |
| 517 | movdqa %xmm7, %xmm1 |
| 518 | pslldq \$8, %xmm7 |
| 519 | pxor %xmm1, %xmm7 |
| 520 | pxor .Lk_s63(%rip), %xmm7 |
| 521 | |
| 522 | # subbytes |
| 523 | movdqa %xmm9, %xmm1 |
| 524 | pandn %xmm0, %xmm1 |
| 525 | psrld \$4, %xmm1 # 1 = i |
| 526 | pand %xmm9, %xmm0 # 0 = k |
| 527 | movdqa %xmm11, %xmm2 # 2 : a/k |
| 528 | pshufb %xmm0, %xmm2 # 2 = a/k |
| 529 | pxor %xmm1, %xmm0 # 0 = j |
| 530 | movdqa %xmm10, %xmm3 # 3 : 1/i |
| 531 | pshufb %xmm1, %xmm3 # 3 = 1/i |
| 532 | pxor %xmm2, %xmm3 # 3 = iak = 1/i + a/k |
| 533 | movdqa %xmm10, %xmm4 # 4 : 1/j |
| 534 | pshufb %xmm0, %xmm4 # 4 = 1/j |
| 535 | pxor %xmm2, %xmm4 # 4 = jak = 1/j + a/k |
| 536 | movdqa %xmm10, %xmm2 # 2 : 1/iak |
| 537 | pshufb %xmm3, %xmm2 # 2 = 1/iak |
| 538 | pxor %xmm0, %xmm2 # 2 = io |
| 539 | movdqa %xmm10, %xmm3 # 3 : 1/jak |
| 540 | pshufb %xmm4, %xmm3 # 3 = 1/jak |
| 541 | pxor %xmm1, %xmm3 # 3 = jo |
| 542 | movdqa %xmm13, %xmm4 # 4 : sbou |
| 543 | pshufb %xmm2, %xmm4 # 4 = sbou |
| 544 | movdqa %xmm12, %xmm0 # 0 : sbot |
| 545 | pshufb %xmm3, %xmm0 # 0 = sb1t |
| 546 | pxor %xmm4, %xmm0 # 0 = sbox output |
| 547 | |
| 548 | # add in smeared stuff |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 549 | pxor %xmm7, %xmm0 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 550 | movdqa %xmm0, %xmm7 |
| 551 | ret |
| 552 | .size _vpaes_schedule_round,.-_vpaes_schedule_round |
| 553 | |
| 554 | ## |
| 555 | ## .aes_schedule_transform |
| 556 | ## |
| 557 | ## Linear-transform %xmm0 according to tables at (%r11) |
| 558 | ## |
| 559 | ## Requires that %xmm9 = 0x0F0F... as in preheat |
| 560 | ## Output in %xmm0 |
| 561 | ## Clobbers %xmm1, %xmm2 |
| 562 | ## |
| 563 | .type _vpaes_schedule_transform,\@abi-omnipotent |
| 564 | .align 16 |
| 565 | _vpaes_schedule_transform: |
| 566 | movdqa %xmm9, %xmm1 |
| 567 | pandn %xmm0, %xmm1 |
| 568 | psrld \$4, %xmm1 |
| 569 | pand %xmm9, %xmm0 |
| 570 | movdqa (%r11), %xmm2 # lo |
| 571 | pshufb %xmm0, %xmm2 |
| 572 | movdqa 16(%r11), %xmm0 # hi |
| 573 | pshufb %xmm1, %xmm0 |
| 574 | pxor %xmm2, %xmm0 |
| 575 | ret |
| 576 | .size _vpaes_schedule_transform,.-_vpaes_schedule_transform |
| 577 | |
| 578 | ## |
| 579 | ## .aes_schedule_mangle |
| 580 | ## |
| 581 | ## Mangle xmm0 from (basis-transformed) standard version |
| 582 | ## to our version. |
| 583 | ## |
| 584 | ## On encrypt, |
| 585 | ## xor with 0x63 |
| 586 | ## multiply by circulant 0,1,1,1 |
| 587 | ## apply shiftrows transform |
| 588 | ## |
| 589 | ## On decrypt, |
| 590 | ## xor with 0x63 |
| 591 | ## multiply by "inverse mixcolumns" circulant E,B,D,9 |
| 592 | ## deskew |
| 593 | ## apply shiftrows transform |
| 594 | ## |
| 595 | ## |
| 596 | ## Writes out to (%rdx), and increments or decrements it |
| 597 | ## Keeps track of round number mod 4 in %r8 |
| 598 | ## Preserves xmm0 |
| 599 | ## Clobbers xmm1-xmm5 |
| 600 | ## |
| 601 | .type _vpaes_schedule_mangle,\@abi-omnipotent |
| 602 | .align 16 |
| 603 | _vpaes_schedule_mangle: |
| 604 | movdqa %xmm0, %xmm4 # save xmm0 for later |
| 605 | movdqa .Lk_mc_forward(%rip),%xmm5 |
| 606 | test %rcx, %rcx |
| 607 | jnz .Lschedule_mangle_dec |
| 608 | |
| 609 | # encrypting |
| 610 | add \$16, %rdx |
| 611 | pxor .Lk_s63(%rip),%xmm4 |
| 612 | pshufb %xmm5, %xmm4 |
| 613 | movdqa %xmm4, %xmm3 |
| 614 | pshufb %xmm5, %xmm4 |
| 615 | pxor %xmm4, %xmm3 |
| 616 | pshufb %xmm5, %xmm4 |
| 617 | pxor %xmm4, %xmm3 |
| 618 | |
| 619 | jmp .Lschedule_mangle_both |
| 620 | .align 16 |
| 621 | .Lschedule_mangle_dec: |
| 622 | # inverse mix columns |
| 623 | lea .Lk_dksd(%rip),%r11 |
| 624 | movdqa %xmm9, %xmm1 |
| 625 | pandn %xmm4, %xmm1 |
| 626 | psrld \$4, %xmm1 # 1 = hi |
| 627 | pand %xmm9, %xmm4 # 4 = lo |
| 628 | |
| 629 | movdqa 0x00(%r11), %xmm2 |
| 630 | pshufb %xmm4, %xmm2 |
| 631 | movdqa 0x10(%r11), %xmm3 |
| 632 | pshufb %xmm1, %xmm3 |
| 633 | pxor %xmm2, %xmm3 |
| 634 | pshufb %xmm5, %xmm3 |
| 635 | |
| 636 | movdqa 0x20(%r11), %xmm2 |
| 637 | pshufb %xmm4, %xmm2 |
| 638 | pxor %xmm3, %xmm2 |
| 639 | movdqa 0x30(%r11), %xmm3 |
| 640 | pshufb %xmm1, %xmm3 |
| 641 | pxor %xmm2, %xmm3 |
| 642 | pshufb %xmm5, %xmm3 |
| 643 | |
| 644 | movdqa 0x40(%r11), %xmm2 |
| 645 | pshufb %xmm4, %xmm2 |
| 646 | pxor %xmm3, %xmm2 |
| 647 | movdqa 0x50(%r11), %xmm3 |
| 648 | pshufb %xmm1, %xmm3 |
| 649 | pxor %xmm2, %xmm3 |
| 650 | pshufb %xmm5, %xmm3 |
| 651 | |
| 652 | movdqa 0x60(%r11), %xmm2 |
| 653 | pshufb %xmm4, %xmm2 |
| 654 | pxor %xmm3, %xmm2 |
| 655 | movdqa 0x70(%r11), %xmm3 |
| 656 | pshufb %xmm1, %xmm3 |
| 657 | pxor %xmm2, %xmm3 |
| 658 | |
| 659 | add \$-16, %rdx |
| 660 | |
| 661 | .Lschedule_mangle_both: |
| 662 | movdqa (%r8,%r10),%xmm1 |
| 663 | pshufb %xmm1,%xmm3 |
| 664 | add \$-16, %r8 |
| 665 | and \$0x30, %r8 |
| 666 | movdqu %xmm3, (%rdx) |
| 667 | ret |
| 668 | .size _vpaes_schedule_mangle,.-_vpaes_schedule_mangle |
| 669 | |
| 670 | # |
| 671 | # Interface to OpenSSL |
| 672 | # |
| 673 | .globl ${PREFIX}_set_encrypt_key |
| 674 | .type ${PREFIX}_set_encrypt_key,\@function,3 |
| 675 | .align 16 |
| 676 | ${PREFIX}_set_encrypt_key: |
| 677 | ___ |
| 678 | $code.=<<___ if ($win64); |
| 679 | lea -0xb8(%rsp),%rsp |
| 680 | movaps %xmm6,0x10(%rsp) |
| 681 | movaps %xmm7,0x20(%rsp) |
| 682 | movaps %xmm8,0x30(%rsp) |
| 683 | movaps %xmm9,0x40(%rsp) |
| 684 | movaps %xmm10,0x50(%rsp) |
| 685 | movaps %xmm11,0x60(%rsp) |
| 686 | movaps %xmm12,0x70(%rsp) |
| 687 | movaps %xmm13,0x80(%rsp) |
| 688 | movaps %xmm14,0x90(%rsp) |
| 689 | movaps %xmm15,0xa0(%rsp) |
| 690 | .Lenc_key_body: |
| 691 | ___ |
| 692 | $code.=<<___; |
| 693 | mov %esi,%eax |
| 694 | shr \$5,%eax |
| 695 | add \$5,%eax |
| 696 | mov %eax,240(%rdx) # AES_KEY->rounds = nbits/32+5; |
| 697 | |
| 698 | mov \$0,%ecx |
| 699 | mov \$0x30,%r8d |
| 700 | call _vpaes_schedule_core |
| 701 | ___ |
| 702 | $code.=<<___ if ($win64); |
| 703 | movaps 0x10(%rsp),%xmm6 |
| 704 | movaps 0x20(%rsp),%xmm7 |
| 705 | movaps 0x30(%rsp),%xmm8 |
| 706 | movaps 0x40(%rsp),%xmm9 |
| 707 | movaps 0x50(%rsp),%xmm10 |
| 708 | movaps 0x60(%rsp),%xmm11 |
| 709 | movaps 0x70(%rsp),%xmm12 |
| 710 | movaps 0x80(%rsp),%xmm13 |
| 711 | movaps 0x90(%rsp),%xmm14 |
| 712 | movaps 0xa0(%rsp),%xmm15 |
| 713 | lea 0xb8(%rsp),%rsp |
| 714 | .Lenc_key_epilogue: |
| 715 | ___ |
| 716 | $code.=<<___; |
| 717 | xor %eax,%eax |
| 718 | ret |
| 719 | .size ${PREFIX}_set_encrypt_key,.-${PREFIX}_set_encrypt_key |
| 720 | |
| 721 | .globl ${PREFIX}_set_decrypt_key |
| 722 | .type ${PREFIX}_set_decrypt_key,\@function,3 |
| 723 | .align 16 |
| 724 | ${PREFIX}_set_decrypt_key: |
| 725 | ___ |
| 726 | $code.=<<___ if ($win64); |
| 727 | lea -0xb8(%rsp),%rsp |
| 728 | movaps %xmm6,0x10(%rsp) |
| 729 | movaps %xmm7,0x20(%rsp) |
| 730 | movaps %xmm8,0x30(%rsp) |
| 731 | movaps %xmm9,0x40(%rsp) |
| 732 | movaps %xmm10,0x50(%rsp) |
| 733 | movaps %xmm11,0x60(%rsp) |
| 734 | movaps %xmm12,0x70(%rsp) |
| 735 | movaps %xmm13,0x80(%rsp) |
| 736 | movaps %xmm14,0x90(%rsp) |
| 737 | movaps %xmm15,0xa0(%rsp) |
| 738 | .Ldec_key_body: |
| 739 | ___ |
| 740 | $code.=<<___; |
| 741 | mov %esi,%eax |
| 742 | shr \$5,%eax |
| 743 | add \$5,%eax |
| 744 | mov %eax,240(%rdx) # AES_KEY->rounds = nbits/32+5; |
| 745 | shl \$4,%eax |
| 746 | lea 16(%rdx,%rax),%rdx |
| 747 | |
| 748 | mov \$1,%ecx |
| 749 | mov %esi,%r8d |
| 750 | shr \$1,%r8d |
| 751 | and \$32,%r8d |
| 752 | xor \$32,%r8d # nbits==192?0:32 |
| 753 | call _vpaes_schedule_core |
| 754 | ___ |
| 755 | $code.=<<___ if ($win64); |
| 756 | movaps 0x10(%rsp),%xmm6 |
| 757 | movaps 0x20(%rsp),%xmm7 |
| 758 | movaps 0x30(%rsp),%xmm8 |
| 759 | movaps 0x40(%rsp),%xmm9 |
| 760 | movaps 0x50(%rsp),%xmm10 |
| 761 | movaps 0x60(%rsp),%xmm11 |
| 762 | movaps 0x70(%rsp),%xmm12 |
| 763 | movaps 0x80(%rsp),%xmm13 |
| 764 | movaps 0x90(%rsp),%xmm14 |
| 765 | movaps 0xa0(%rsp),%xmm15 |
| 766 | lea 0xb8(%rsp),%rsp |
| 767 | .Ldec_key_epilogue: |
| 768 | ___ |
| 769 | $code.=<<___; |
| 770 | xor %eax,%eax |
| 771 | ret |
| 772 | .size ${PREFIX}_set_decrypt_key,.-${PREFIX}_set_decrypt_key |
| 773 | |
| 774 | .globl ${PREFIX}_encrypt |
| 775 | .type ${PREFIX}_encrypt,\@function,3 |
| 776 | .align 16 |
| 777 | ${PREFIX}_encrypt: |
| 778 | ___ |
| 779 | $code.=<<___ if ($win64); |
| 780 | lea -0xb8(%rsp),%rsp |
| 781 | movaps %xmm6,0x10(%rsp) |
| 782 | movaps %xmm7,0x20(%rsp) |
| 783 | movaps %xmm8,0x30(%rsp) |
| 784 | movaps %xmm9,0x40(%rsp) |
| 785 | movaps %xmm10,0x50(%rsp) |
| 786 | movaps %xmm11,0x60(%rsp) |
| 787 | movaps %xmm12,0x70(%rsp) |
| 788 | movaps %xmm13,0x80(%rsp) |
| 789 | movaps %xmm14,0x90(%rsp) |
| 790 | movaps %xmm15,0xa0(%rsp) |
| 791 | .Lenc_body: |
| 792 | ___ |
| 793 | $code.=<<___; |
| 794 | movdqu (%rdi),%xmm0 |
| 795 | call _vpaes_preheat |
| 796 | call _vpaes_encrypt_core |
| 797 | movdqu %xmm0,(%rsi) |
| 798 | ___ |
| 799 | $code.=<<___ if ($win64); |
| 800 | movaps 0x10(%rsp),%xmm6 |
| 801 | movaps 0x20(%rsp),%xmm7 |
| 802 | movaps 0x30(%rsp),%xmm8 |
| 803 | movaps 0x40(%rsp),%xmm9 |
| 804 | movaps 0x50(%rsp),%xmm10 |
| 805 | movaps 0x60(%rsp),%xmm11 |
| 806 | movaps 0x70(%rsp),%xmm12 |
| 807 | movaps 0x80(%rsp),%xmm13 |
| 808 | movaps 0x90(%rsp),%xmm14 |
| 809 | movaps 0xa0(%rsp),%xmm15 |
| 810 | lea 0xb8(%rsp),%rsp |
| 811 | .Lenc_epilogue: |
| 812 | ___ |
| 813 | $code.=<<___; |
| 814 | ret |
| 815 | .size ${PREFIX}_encrypt,.-${PREFIX}_encrypt |
| 816 | |
| 817 | .globl ${PREFIX}_decrypt |
| 818 | .type ${PREFIX}_decrypt,\@function,3 |
| 819 | .align 16 |
| 820 | ${PREFIX}_decrypt: |
| 821 | ___ |
| 822 | $code.=<<___ if ($win64); |
| 823 | lea -0xb8(%rsp),%rsp |
| 824 | movaps %xmm6,0x10(%rsp) |
| 825 | movaps %xmm7,0x20(%rsp) |
| 826 | movaps %xmm8,0x30(%rsp) |
| 827 | movaps %xmm9,0x40(%rsp) |
| 828 | movaps %xmm10,0x50(%rsp) |
| 829 | movaps %xmm11,0x60(%rsp) |
| 830 | movaps %xmm12,0x70(%rsp) |
| 831 | movaps %xmm13,0x80(%rsp) |
| 832 | movaps %xmm14,0x90(%rsp) |
| 833 | movaps %xmm15,0xa0(%rsp) |
| 834 | .Ldec_body: |
| 835 | ___ |
| 836 | $code.=<<___; |
| 837 | movdqu (%rdi),%xmm0 |
| 838 | call _vpaes_preheat |
| 839 | call _vpaes_decrypt_core |
| 840 | movdqu %xmm0,(%rsi) |
| 841 | ___ |
| 842 | $code.=<<___ if ($win64); |
| 843 | movaps 0x10(%rsp),%xmm6 |
| 844 | movaps 0x20(%rsp),%xmm7 |
| 845 | movaps 0x30(%rsp),%xmm8 |
| 846 | movaps 0x40(%rsp),%xmm9 |
| 847 | movaps 0x50(%rsp),%xmm10 |
| 848 | movaps 0x60(%rsp),%xmm11 |
| 849 | movaps 0x70(%rsp),%xmm12 |
| 850 | movaps 0x80(%rsp),%xmm13 |
| 851 | movaps 0x90(%rsp),%xmm14 |
| 852 | movaps 0xa0(%rsp),%xmm15 |
| 853 | lea 0xb8(%rsp),%rsp |
| 854 | .Ldec_epilogue: |
| 855 | ___ |
| 856 | $code.=<<___; |
| 857 | ret |
| 858 | .size ${PREFIX}_decrypt,.-${PREFIX}_decrypt |
| 859 | ___ |
| 860 | { |
| 861 | my ($inp,$out,$len,$key,$ivp,$enc)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9"); |
| 862 | # void AES_cbc_encrypt (const void char *inp, unsigned char *out, |
| 863 | # size_t length, const AES_KEY *key, |
| 864 | # unsigned char *ivp,const int enc); |
| 865 | $code.=<<___; |
| 866 | .globl ${PREFIX}_cbc_encrypt |
| 867 | .type ${PREFIX}_cbc_encrypt,\@function,6 |
| 868 | .align 16 |
| 869 | ${PREFIX}_cbc_encrypt: |
| 870 | xchg $key,$len |
| 871 | ___ |
| 872 | ($len,$key)=($key,$len); |
| 873 | $code.=<<___; |
| 874 | sub \$16,$len |
| 875 | jc .Lcbc_abort |
| 876 | ___ |
| 877 | $code.=<<___ if ($win64); |
| 878 | lea -0xb8(%rsp),%rsp |
| 879 | movaps %xmm6,0x10(%rsp) |
| 880 | movaps %xmm7,0x20(%rsp) |
| 881 | movaps %xmm8,0x30(%rsp) |
| 882 | movaps %xmm9,0x40(%rsp) |
| 883 | movaps %xmm10,0x50(%rsp) |
| 884 | movaps %xmm11,0x60(%rsp) |
| 885 | movaps %xmm12,0x70(%rsp) |
| 886 | movaps %xmm13,0x80(%rsp) |
| 887 | movaps %xmm14,0x90(%rsp) |
| 888 | movaps %xmm15,0xa0(%rsp) |
| 889 | .Lcbc_body: |
| 890 | ___ |
| 891 | $code.=<<___; |
| 892 | movdqu ($ivp),%xmm6 # load IV |
| 893 | sub $inp,$out |
| 894 | call _vpaes_preheat |
| 895 | cmp \$0,${enc}d |
| 896 | je .Lcbc_dec_loop |
| 897 | jmp .Lcbc_enc_loop |
| 898 | .align 16 |
| 899 | .Lcbc_enc_loop: |
| 900 | movdqu ($inp),%xmm0 |
| 901 | pxor %xmm6,%xmm0 |
| 902 | call _vpaes_encrypt_core |
| 903 | movdqa %xmm0,%xmm6 |
| 904 | movdqu %xmm0,($out,$inp) |
| 905 | lea 16($inp),$inp |
| 906 | sub \$16,$len |
| 907 | jnc .Lcbc_enc_loop |
| 908 | jmp .Lcbc_done |
| 909 | .align 16 |
| 910 | .Lcbc_dec_loop: |
| 911 | movdqu ($inp),%xmm0 |
| 912 | movdqa %xmm0,%xmm7 |
| 913 | call _vpaes_decrypt_core |
| 914 | pxor %xmm6,%xmm0 |
| 915 | movdqa %xmm7,%xmm6 |
| 916 | movdqu %xmm0,($out,$inp) |
| 917 | lea 16($inp),$inp |
| 918 | sub \$16,$len |
| 919 | jnc .Lcbc_dec_loop |
| 920 | .Lcbc_done: |
| 921 | movdqu %xmm6,($ivp) # save IV |
| 922 | ___ |
| 923 | $code.=<<___ if ($win64); |
| 924 | movaps 0x10(%rsp),%xmm6 |
| 925 | movaps 0x20(%rsp),%xmm7 |
| 926 | movaps 0x30(%rsp),%xmm8 |
| 927 | movaps 0x40(%rsp),%xmm9 |
| 928 | movaps 0x50(%rsp),%xmm10 |
| 929 | movaps 0x60(%rsp),%xmm11 |
| 930 | movaps 0x70(%rsp),%xmm12 |
| 931 | movaps 0x80(%rsp),%xmm13 |
| 932 | movaps 0x90(%rsp),%xmm14 |
| 933 | movaps 0xa0(%rsp),%xmm15 |
| 934 | lea 0xb8(%rsp),%rsp |
| 935 | .Lcbc_epilogue: |
| 936 | ___ |
| 937 | $code.=<<___; |
| 938 | .Lcbc_abort: |
| 939 | ret |
| 940 | .size ${PREFIX}_cbc_encrypt,.-${PREFIX}_cbc_encrypt |
| 941 | ___ |
| 942 | } |
| 943 | $code.=<<___; |
| 944 | ## |
| 945 | ## _aes_preheat |
| 946 | ## |
| 947 | ## Fills register %r10 -> .aes_consts (so you can -fPIC) |
| 948 | ## and %xmm9-%xmm15 as specified below. |
| 949 | ## |
| 950 | .type _vpaes_preheat,\@abi-omnipotent |
| 951 | .align 16 |
| 952 | _vpaes_preheat: |
| 953 | lea .Lk_s0F(%rip), %r10 |
| 954 | movdqa -0x20(%r10), %xmm10 # .Lk_inv |
| 955 | movdqa -0x10(%r10), %xmm11 # .Lk_inv+16 |
| 956 | movdqa 0x00(%r10), %xmm9 # .Lk_s0F |
| 957 | movdqa 0x30(%r10), %xmm13 # .Lk_sb1 |
| 958 | movdqa 0x40(%r10), %xmm12 # .Lk_sb1+16 |
| 959 | movdqa 0x50(%r10), %xmm15 # .Lk_sb2 |
| 960 | movdqa 0x60(%r10), %xmm14 # .Lk_sb2+16 |
| 961 | ret |
| 962 | .size _vpaes_preheat,.-_vpaes_preheat |
| 963 | ######################################################## |
| 964 | ## ## |
| 965 | ## Constants ## |
| 966 | ## ## |
| 967 | ######################################################## |
| 968 | .type _vpaes_consts,\@object |
| 969 | .align 64 |
| 970 | _vpaes_consts: |
| 971 | .Lk_inv: # inv, inva |
| 972 | .quad 0x0E05060F0D080180, 0x040703090A0B0C02 |
| 973 | .quad 0x01040A060F0B0780, 0x030D0E0C02050809 |
| 974 | |
| 975 | .Lk_s0F: # s0F |
| 976 | .quad 0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F |
| 977 | |
| 978 | .Lk_ipt: # input transform (lo, hi) |
| 979 | .quad 0xC2B2E8985A2A7000, 0xCABAE09052227808 |
| 980 | .quad 0x4C01307D317C4D00, 0xCD80B1FCB0FDCC81 |
| 981 | |
| 982 | .Lk_sb1: # sb1u, sb1t |
| 983 | .quad 0xB19BE18FCB503E00, 0xA5DF7A6E142AF544 |
| 984 | .quad 0x3618D415FAE22300, 0x3BF7CCC10D2ED9EF |
| 985 | .Lk_sb2: # sb2u, sb2t |
| 986 | .quad 0xE27A93C60B712400, 0x5EB7E955BC982FCD |
| 987 | .quad 0x69EB88400AE12900, 0xC2A163C8AB82234A |
| 988 | .Lk_sbo: # sbou, sbot |
| 989 | .quad 0xD0D26D176FBDC700, 0x15AABF7AC502A878 |
| 990 | .quad 0xCFE474A55FBB6A00, 0x8E1E90D1412B35FA |
| 991 | |
| 992 | .Lk_mc_forward: # mc_forward |
| 993 | .quad 0x0407060500030201, 0x0C0F0E0D080B0A09 |
| 994 | .quad 0x080B0A0904070605, 0x000302010C0F0E0D |
| 995 | .quad 0x0C0F0E0D080B0A09, 0x0407060500030201 |
| 996 | .quad 0x000302010C0F0E0D, 0x080B0A0904070605 |
| 997 | |
| 998 | .Lk_mc_backward:# mc_backward |
| 999 | .quad 0x0605040702010003, 0x0E0D0C0F0A09080B |
| 1000 | .quad 0x020100030E0D0C0F, 0x0A09080B06050407 |
| 1001 | .quad 0x0E0D0C0F0A09080B, 0x0605040702010003 |
| 1002 | .quad 0x0A09080B06050407, 0x020100030E0D0C0F |
| 1003 | |
| 1004 | .Lk_sr: # sr |
| 1005 | .quad 0x0706050403020100, 0x0F0E0D0C0B0A0908 |
| 1006 | .quad 0x030E09040F0A0500, 0x0B06010C07020D08 |
| 1007 | .quad 0x0F060D040B020900, 0x070E050C030A0108 |
| 1008 | .quad 0x0B0E0104070A0D00, 0x0306090C0F020508 |
| 1009 | |
| 1010 | .Lk_rcon: # rcon |
| 1011 | .quad 0x1F8391B9AF9DEEB6, 0x702A98084D7C7D81 |
| 1012 | |
| 1013 | .Lk_s63: # s63: all equal to 0x63 transformed |
| 1014 | .quad 0x5B5B5B5B5B5B5B5B, 0x5B5B5B5B5B5B5B5B |
| 1015 | |
| 1016 | .Lk_opt: # output transform |
| 1017 | .quad 0xFF9F4929D6B66000, 0xF7974121DEBE6808 |
| 1018 | .quad 0x01EDBD5150BCEC00, 0xE10D5DB1B05C0CE0 |
| 1019 | |
| 1020 | .Lk_deskew: # deskew tables: inverts the sbox's "skew" |
| 1021 | .quad 0x07E4A34047A4E300, 0x1DFEB95A5DBEF91A |
| 1022 | .quad 0x5F36B5DC83EA6900, 0x2841C2ABF49D1E77 |
| 1023 | |
| 1024 | ## |
| 1025 | ## Decryption stuff |
| 1026 | ## Key schedule constants |
| 1027 | ## |
| 1028 | .Lk_dksd: # decryption key schedule: invskew x*D |
| 1029 | .quad 0xFEB91A5DA3E44700, 0x0740E3A45A1DBEF9 |
| 1030 | .quad 0x41C277F4B5368300, 0x5FDC69EAAB289D1E |
| 1031 | .Lk_dksb: # decryption key schedule: invskew x*B |
| 1032 | .quad 0x9A4FCA1F8550D500, 0x03D653861CC94C99 |
| 1033 | .quad 0x115BEDA7B6FC4A00, 0xD993256F7E3482C8 |
| 1034 | .Lk_dkse: # decryption key schedule: invskew x*E + 0x63 |
| 1035 | .quad 0xD5031CCA1FC9D600, 0x53859A4C994F5086 |
| 1036 | .quad 0xA23196054FDC7BE8, 0xCD5EF96A20B31487 |
| 1037 | .Lk_dks9: # decryption key schedule: invskew x*9 |
| 1038 | .quad 0xB6116FC87ED9A700, 0x4AED933482255BFC |
| 1039 | .quad 0x4576516227143300, 0x8BB89FACE9DAFDCE |
| 1040 | |
| 1041 | ## |
| 1042 | ## Decryption stuff |
| 1043 | ## Round function constants |
| 1044 | ## |
| 1045 | .Lk_dipt: # decryption input transform |
| 1046 | .quad 0x0F505B040B545F00, 0x154A411E114E451A |
| 1047 | .quad 0x86E383E660056500, 0x12771772F491F194 |
| 1048 | |
| 1049 | .Lk_dsb9: # decryption sbox output *9*u, *9*t |
| 1050 | .quad 0x851C03539A86D600, 0xCAD51F504F994CC9 |
| 1051 | .quad 0xC03B1789ECD74900, 0x725E2C9EB2FBA565 |
| 1052 | .Lk_dsbd: # decryption sbox output *D*u, *D*t |
| 1053 | .quad 0x7D57CCDFE6B1A200, 0xF56E9B13882A4439 |
| 1054 | .quad 0x3CE2FAF724C6CB00, 0x2931180D15DEEFD3 |
| 1055 | .Lk_dsbb: # decryption sbox output *B*u, *B*t |
| 1056 | .quad 0xD022649296B44200, 0x602646F6B0F2D404 |
| 1057 | .quad 0xC19498A6CD596700, 0xF3FF0C3E3255AA6B |
| 1058 | .Lk_dsbe: # decryption sbox output *E*u, *E*t |
| 1059 | .quad 0x46F2929626D4D000, 0x2242600464B4F6B0 |
| 1060 | .quad 0x0C55A6CDFFAAC100, 0x9467F36B98593E32 |
| 1061 | .Lk_dsbo: # decryption sbox final output |
| 1062 | .quad 0x1387EA537EF94000, 0xC7AA6DB9D4943E2D |
| 1063 | .quad 0x12D7560F93441D00, 0xCA4B8159D8C58E9C |
| 1064 | .asciz "Vector Permutation AES for x86_64/SSSE3, Mike Hamburg (Stanford University)" |
| 1065 | .align 64 |
| 1066 | .size _vpaes_consts,.-_vpaes_consts |
| 1067 | ___ |
| 1068 | |
| 1069 | if ($win64) { |
| 1070 | # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, |
| 1071 | # CONTEXT *context,DISPATCHER_CONTEXT *disp) |
| 1072 | $rec="%rcx"; |
| 1073 | $frame="%rdx"; |
| 1074 | $context="%r8"; |
| 1075 | $disp="%r9"; |
| 1076 | |
| 1077 | $code.=<<___; |
| 1078 | .extern __imp_RtlVirtualUnwind |
| 1079 | .type se_handler,\@abi-omnipotent |
| 1080 | .align 16 |
| 1081 | se_handler: |
| 1082 | push %rsi |
| 1083 | push %rdi |
| 1084 | push %rbx |
| 1085 | push %rbp |
| 1086 | push %r12 |
| 1087 | push %r13 |
| 1088 | push %r14 |
| 1089 | push %r15 |
| 1090 | pushfq |
| 1091 | sub \$64,%rsp |
| 1092 | |
| 1093 | mov 120($context),%rax # pull context->Rax |
| 1094 | mov 248($context),%rbx # pull context->Rip |
| 1095 | |
| 1096 | mov 8($disp),%rsi # disp->ImageBase |
| 1097 | mov 56($disp),%r11 # disp->HandlerData |
| 1098 | |
| 1099 | mov 0(%r11),%r10d # HandlerData[0] |
| 1100 | lea (%rsi,%r10),%r10 # prologue label |
| 1101 | cmp %r10,%rbx # context->Rip<prologue label |
| 1102 | jb .Lin_prologue |
| 1103 | |
| 1104 | mov 152($context),%rax # pull context->Rsp |
| 1105 | |
| 1106 | mov 4(%r11),%r10d # HandlerData[1] |
| 1107 | lea (%rsi,%r10),%r10 # epilogue label |
| 1108 | cmp %r10,%rbx # context->Rip>=epilogue label |
| 1109 | jae .Lin_prologue |
| 1110 | |
| 1111 | lea 16(%rax),%rsi # %xmm save area |
| 1112 | lea 512($context),%rdi # &context.Xmm6 |
| 1113 | mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) |
| 1114 | .long 0xa548f3fc # cld; rep movsq |
| 1115 | lea 0xb8(%rax),%rax # adjust stack pointer |
| 1116 | |
| 1117 | .Lin_prologue: |
| 1118 | mov 8(%rax),%rdi |
| 1119 | mov 16(%rax),%rsi |
| 1120 | mov %rax,152($context) # restore context->Rsp |
| 1121 | mov %rsi,168($context) # restore context->Rsi |
| 1122 | mov %rdi,176($context) # restore context->Rdi |
| 1123 | |
| 1124 | mov 40($disp),%rdi # disp->ContextRecord |
| 1125 | mov $context,%rsi # context |
| 1126 | mov \$`1232/8`,%ecx # sizeof(CONTEXT) |
| 1127 | .long 0xa548f3fc # cld; rep movsq |
| 1128 | |
| 1129 | mov $disp,%rsi |
| 1130 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER |
| 1131 | mov 8(%rsi),%rdx # arg2, disp->ImageBase |
| 1132 | mov 0(%rsi),%r8 # arg3, disp->ControlPc |
| 1133 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry |
| 1134 | mov 40(%rsi),%r10 # disp->ContextRecord |
| 1135 | lea 56(%rsi),%r11 # &disp->HandlerData |
| 1136 | lea 24(%rsi),%r12 # &disp->EstablisherFrame |
| 1137 | mov %r10,32(%rsp) # arg5 |
| 1138 | mov %r11,40(%rsp) # arg6 |
| 1139 | mov %r12,48(%rsp) # arg7 |
| 1140 | mov %rcx,56(%rsp) # arg8, (NULL) |
| 1141 | call *__imp_RtlVirtualUnwind(%rip) |
| 1142 | |
| 1143 | mov \$1,%eax # ExceptionContinueSearch |
| 1144 | add \$64,%rsp |
| 1145 | popfq |
| 1146 | pop %r15 |
| 1147 | pop %r14 |
| 1148 | pop %r13 |
| 1149 | pop %r12 |
| 1150 | pop %rbp |
| 1151 | pop %rbx |
| 1152 | pop %rdi |
| 1153 | pop %rsi |
| 1154 | ret |
| 1155 | .size se_handler,.-se_handler |
| 1156 | |
| 1157 | .section .pdata |
| 1158 | .align 4 |
| 1159 | .rva .LSEH_begin_${PREFIX}_set_encrypt_key |
| 1160 | .rva .LSEH_end_${PREFIX}_set_encrypt_key |
| 1161 | .rva .LSEH_info_${PREFIX}_set_encrypt_key |
| 1162 | |
| 1163 | .rva .LSEH_begin_${PREFIX}_set_decrypt_key |
| 1164 | .rva .LSEH_end_${PREFIX}_set_decrypt_key |
| 1165 | .rva .LSEH_info_${PREFIX}_set_decrypt_key |
| 1166 | |
| 1167 | .rva .LSEH_begin_${PREFIX}_encrypt |
| 1168 | .rva .LSEH_end_${PREFIX}_encrypt |
| 1169 | .rva .LSEH_info_${PREFIX}_encrypt |
| 1170 | |
| 1171 | .rva .LSEH_begin_${PREFIX}_decrypt |
| 1172 | .rva .LSEH_end_${PREFIX}_decrypt |
| 1173 | .rva .LSEH_info_${PREFIX}_decrypt |
| 1174 | |
| 1175 | .rva .LSEH_begin_${PREFIX}_cbc_encrypt |
| 1176 | .rva .LSEH_end_${PREFIX}_cbc_encrypt |
| 1177 | .rva .LSEH_info_${PREFIX}_cbc_encrypt |
| 1178 | |
| 1179 | .section .xdata |
| 1180 | .align 8 |
| 1181 | .LSEH_info_${PREFIX}_set_encrypt_key: |
| 1182 | .byte 9,0,0,0 |
| 1183 | .rva se_handler |
| 1184 | .rva .Lenc_key_body,.Lenc_key_epilogue # HandlerData[] |
| 1185 | .LSEH_info_${PREFIX}_set_decrypt_key: |
| 1186 | .byte 9,0,0,0 |
| 1187 | .rva se_handler |
| 1188 | .rva .Ldec_key_body,.Ldec_key_epilogue # HandlerData[] |
| 1189 | .LSEH_info_${PREFIX}_encrypt: |
| 1190 | .byte 9,0,0,0 |
| 1191 | .rva se_handler |
| 1192 | .rva .Lenc_body,.Lenc_epilogue # HandlerData[] |
| 1193 | .LSEH_info_${PREFIX}_decrypt: |
| 1194 | .byte 9,0,0,0 |
| 1195 | .rva se_handler |
| 1196 | .rva .Ldec_body,.Ldec_epilogue # HandlerData[] |
| 1197 | .LSEH_info_${PREFIX}_cbc_encrypt: |
| 1198 | .byte 9,0,0,0 |
| 1199 | .rva se_handler |
| 1200 | .rva .Lcbc_body,.Lcbc_epilogue # HandlerData[] |
| 1201 | ___ |
| 1202 | } |
| 1203 | |
| 1204 | $code =~ s/\`([^\`]*)\`/eval($1)/gem; |
| 1205 | |
| 1206 | print $code; |
| 1207 | |
| 1208 | close STDOUT; |