Robert Sloan | fe7cd21 | 2017-08-07 09:03:39 -0700 | [diff] [blame] | 1 | #! /usr/bin/env perl |
| 2 | # Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved. |
| 3 | # |
| 4 | # Licensed under the OpenSSL license (the "License"). You may not use |
| 5 | # this file except in compliance with the License. You can obtain a copy |
| 6 | # in the file LICENSE in the source distribution or at |
| 7 | # https://www.openssl.org/source/license.html |
| 8 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 9 | # |
| 10 | # ==================================================================== |
| 11 | # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL |
| 12 | # project. The module is, however, dual licensed under OpenSSL and |
| 13 | # CRYPTOGAMS licenses depending on where you obtain it. For further |
| 14 | # details see http://www.openssl.org/~appro/cryptogams/. |
| 15 | # ==================================================================== |
| 16 | # |
| 17 | # |
| 18 | # AES-NI-CTR+GHASH stitch. |
| 19 | # |
| 20 | # February 2013 |
| 21 | # |
| 22 | # OpenSSL GCM implementation is organized in such way that its |
| 23 | # performance is rather close to the sum of its streamed components, |
| 24 | # in the context parallelized AES-NI CTR and modulo-scheduled |
| 25 | # PCLMULQDQ-enabled GHASH. Unfortunately, as no stitch implementation |
| 26 | # was observed to perform significantly better than the sum of the |
| 27 | # components on contemporary CPUs, the effort was deemed impossible to |
| 28 | # justify. This module is based on combination of Intel submissions, |
| 29 | # [1] and [2], with MOVBE twist suggested by Ilya Albrekht and Max |
| 30 | # Locktyukhin of Intel Corp. who verified that it reduces shuffles |
| 31 | # pressure with notable relative improvement, achieving 1.0 cycle per |
Robert Sloan | a94fe05 | 2017-02-21 08:49:28 -0800 | [diff] [blame] | 32 | # byte processed with 128-bit key on Haswell processor, 0.74 - on |
| 33 | # Broadwell, 0.63 - on Skylake... [Mentioned results are raw profiled |
| 34 | # measurements for favourable packet size, one divisible by 96. |
| 35 | # Applications using the EVP interface will observe a few percent |
| 36 | # worse performance.] |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 37 | # |
Robert Sloan | fe7cd21 | 2017-08-07 09:03:39 -0700 | [diff] [blame] | 38 | # Knights Landing processes 1 byte in 1.25 cycles (measured with EVP). |
| 39 | # |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 40 | # [1] http://rt.openssl.org/Ticket/Display.html?id=2900&user=guest&pass=guest |
| 41 | # [2] http://www.intel.com/content/dam/www/public/us/en/documents/software-support/enabling-high-performance-gcm.pdf |
| 42 | |
| 43 | $flavour = shift; |
| 44 | $output = shift; |
| 45 | if ($flavour =~ /\./) { $output = $flavour; undef $flavour; } |
| 46 | |
| 47 | $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/); |
| 48 | |
| 49 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 50 | ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or |
Robert Sloan | 9254e68 | 2017-04-24 09:42:06 -0700 | [diff] [blame] | 51 | ( $xlate="${dir}../../../perlasm/x86_64-xlate.pl" and -f $xlate) or |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 52 | die "can't locate x86_64-xlate.pl"; |
| 53 | |
Steven Valdez | b0b45c6 | 2017-01-17 16:23:54 -0500 | [diff] [blame] | 54 | # |$avx| in ghash-x86_64.pl must be set to at least 1; otherwise tags will |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 55 | # be computed incorrectly. |
| 56 | # |
Kenny Root | e99801b | 2015-11-06 15:31:15 -0800 | [diff] [blame] | 57 | # In upstream, this is controlled by shelling out to the compiler to check |
| 58 | # versions, but BoringSSL is intended to be used with pre-generated perlasm |
| 59 | # output, so this isn't useful anyway. |
Steven Valdez | b0b45c6 | 2017-01-17 16:23:54 -0500 | [diff] [blame] | 60 | # |
| 61 | # The upstream code uses the condition |$avx>1| even though no AVX2 |
| 62 | # instructions are used, because it assumes MOVBE is supported by the assembler |
| 63 | # if and only if AVX2 is also supported by the assembler; see |
| 64 | # https://marc.info/?l=openssl-dev&m=146567589526984&w=2. |
| 65 | $avx = 2; |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 66 | |
David Benjamin | c895d6b | 2016-08-11 13:26:41 -0400 | [diff] [blame] | 67 | open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\""; |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 68 | *STDOUT=*OUT; |
| 69 | |
Steven Valdez | b0b45c6 | 2017-01-17 16:23:54 -0500 | [diff] [blame] | 70 | # See the comment above regarding why the condition is ($avx>1) when there are |
| 71 | # no AVX2 instructions being used. |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 72 | if ($avx>1) {{{ |
| 73 | |
| 74 | ($inp,$out,$len,$key,$ivp,$Xip)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9"); |
| 75 | |
| 76 | ($Ii,$T1,$T2,$Hkey, |
| 77 | $Z0,$Z1,$Z2,$Z3,$Xi) = map("%xmm$_",(0..8)); |
| 78 | |
| 79 | ($inout0,$inout1,$inout2,$inout3,$inout4,$inout5,$rndkey) = map("%xmm$_",(9..15)); |
| 80 | |
| 81 | ($counter,$rounds,$ret,$const,$in0,$end0)=("%ebx","%ebp","%r10","%r11","%r14","%r15"); |
| 82 | |
| 83 | $code=<<___; |
| 84 | .text |
| 85 | |
| 86 | .type _aesni_ctr32_ghash_6x,\@abi-omnipotent |
| 87 | .align 32 |
| 88 | _aesni_ctr32_ghash_6x: |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 89 | .cfi_startproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 90 | vmovdqu 0x20($const),$T2 # borrow $T2, .Lone_msb |
| 91 | sub \$6,$len |
| 92 | vpxor $Z0,$Z0,$Z0 # $Z0 = 0 |
| 93 | vmovdqu 0x00-0x80($key),$rndkey |
| 94 | vpaddb $T2,$T1,$inout1 |
| 95 | vpaddb $T2,$inout1,$inout2 |
| 96 | vpaddb $T2,$inout2,$inout3 |
| 97 | vpaddb $T2,$inout3,$inout4 |
| 98 | vpaddb $T2,$inout4,$inout5 |
| 99 | vpxor $rndkey,$T1,$inout0 |
| 100 | vmovdqu $Z0,16+8(%rsp) # "$Z3" = 0 |
| 101 | jmp .Loop6x |
| 102 | |
| 103 | .align 32 |
| 104 | .Loop6x: |
| 105 | add \$`6<<24`,$counter |
| 106 | jc .Lhandle_ctr32 # discard $inout[1-5]? |
| 107 | vmovdqu 0x00-0x20($Xip),$Hkey # $Hkey^1 |
| 108 | vpaddb $T2,$inout5,$T1 # next counter value |
| 109 | vpxor $rndkey,$inout1,$inout1 |
| 110 | vpxor $rndkey,$inout2,$inout2 |
| 111 | |
| 112 | .Lresume_ctr32: |
| 113 | vmovdqu $T1,($ivp) # save next counter value |
| 114 | vpclmulqdq \$0x10,$Hkey,$Z3,$Z1 |
| 115 | vpxor $rndkey,$inout3,$inout3 |
| 116 | vmovups 0x10-0x80($key),$T2 # borrow $T2 for $rndkey |
| 117 | vpclmulqdq \$0x01,$Hkey,$Z3,$Z2 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 118 | |
| 119 | # At this point, the current block of 96 (0x60) bytes has already been |
| 120 | # loaded into registers. Concurrently with processing it, we want to |
| 121 | # load the next 96 bytes of input for the next round. Obviously, we can |
| 122 | # only do this if there are at least 96 more bytes of input beyond the |
| 123 | # input we're currently processing, or else we'd read past the end of |
| 124 | # the input buffer. Here, we set |%r12| to 96 if there are at least 96 |
| 125 | # bytes of input beyond the 96 bytes we're already processing, and we |
| 126 | # set |%r12| to 0 otherwise. In the case where we set |%r12| to 96, |
| 127 | # we'll read in the next block so that it is in registers for the next |
| 128 | # loop iteration. In the case where we set |%r12| to 0, we'll re-read |
| 129 | # the current block and then ignore what we re-read. |
| 130 | # |
| 131 | # At this point, |$in0| points to the current (already read into |
| 132 | # registers) block, and |$end0| points to 2*96 bytes before the end of |
| 133 | # the input. Thus, |$in0| > |$end0| means that we do not have the next |
| 134 | # 96-byte block to read in, and |$in0| <= |$end0| means we do. |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 135 | xor %r12,%r12 |
| 136 | cmp $in0,$end0 |
| 137 | |
| 138 | vaesenc $T2,$inout0,$inout0 |
| 139 | vmovdqu 0x30+8(%rsp),$Ii # I[4] |
| 140 | vpxor $rndkey,$inout4,$inout4 |
| 141 | vpclmulqdq \$0x00,$Hkey,$Z3,$T1 |
| 142 | vaesenc $T2,$inout1,$inout1 |
| 143 | vpxor $rndkey,$inout5,$inout5 |
| 144 | setnc %r12b |
| 145 | vpclmulqdq \$0x11,$Hkey,$Z3,$Z3 |
| 146 | vaesenc $T2,$inout2,$inout2 |
| 147 | vmovdqu 0x10-0x20($Xip),$Hkey # $Hkey^2 |
| 148 | neg %r12 |
| 149 | vaesenc $T2,$inout3,$inout3 |
| 150 | vpxor $Z1,$Z2,$Z2 |
| 151 | vpclmulqdq \$0x00,$Hkey,$Ii,$Z1 |
| 152 | vpxor $Z0,$Xi,$Xi # modulo-scheduled |
| 153 | vaesenc $T2,$inout4,$inout4 |
| 154 | vpxor $Z1,$T1,$Z0 |
| 155 | and \$0x60,%r12 |
| 156 | vmovups 0x20-0x80($key),$rndkey |
| 157 | vpclmulqdq \$0x10,$Hkey,$Ii,$T1 |
| 158 | vaesenc $T2,$inout5,$inout5 |
| 159 | |
| 160 | vpclmulqdq \$0x01,$Hkey,$Ii,$T2 |
| 161 | lea ($in0,%r12),$in0 |
| 162 | vaesenc $rndkey,$inout0,$inout0 |
| 163 | vpxor 16+8(%rsp),$Xi,$Xi # modulo-scheduled [vpxor $Z3,$Xi,$Xi] |
| 164 | vpclmulqdq \$0x11,$Hkey,$Ii,$Hkey |
| 165 | vmovdqu 0x40+8(%rsp),$Ii # I[3] |
| 166 | vaesenc $rndkey,$inout1,$inout1 |
| 167 | movbe 0x58($in0),%r13 |
| 168 | vaesenc $rndkey,$inout2,$inout2 |
| 169 | movbe 0x50($in0),%r12 |
| 170 | vaesenc $rndkey,$inout3,$inout3 |
| 171 | mov %r13,0x20+8(%rsp) |
| 172 | vaesenc $rndkey,$inout4,$inout4 |
| 173 | mov %r12,0x28+8(%rsp) |
| 174 | vmovdqu 0x30-0x20($Xip),$Z1 # borrow $Z1 for $Hkey^3 |
| 175 | vaesenc $rndkey,$inout5,$inout5 |
| 176 | |
| 177 | vmovups 0x30-0x80($key),$rndkey |
| 178 | vpxor $T1,$Z2,$Z2 |
| 179 | vpclmulqdq \$0x00,$Z1,$Ii,$T1 |
| 180 | vaesenc $rndkey,$inout0,$inout0 |
| 181 | vpxor $T2,$Z2,$Z2 |
| 182 | vpclmulqdq \$0x10,$Z1,$Ii,$T2 |
| 183 | vaesenc $rndkey,$inout1,$inout1 |
| 184 | vpxor $Hkey,$Z3,$Z3 |
| 185 | vpclmulqdq \$0x01,$Z1,$Ii,$Hkey |
| 186 | vaesenc $rndkey,$inout2,$inout2 |
| 187 | vpclmulqdq \$0x11,$Z1,$Ii,$Z1 |
| 188 | vmovdqu 0x50+8(%rsp),$Ii # I[2] |
| 189 | vaesenc $rndkey,$inout3,$inout3 |
| 190 | vaesenc $rndkey,$inout4,$inout4 |
| 191 | vpxor $T1,$Z0,$Z0 |
| 192 | vmovdqu 0x40-0x20($Xip),$T1 # borrow $T1 for $Hkey^4 |
| 193 | vaesenc $rndkey,$inout5,$inout5 |
| 194 | |
| 195 | vmovups 0x40-0x80($key),$rndkey |
| 196 | vpxor $T2,$Z2,$Z2 |
| 197 | vpclmulqdq \$0x00,$T1,$Ii,$T2 |
| 198 | vaesenc $rndkey,$inout0,$inout0 |
| 199 | vpxor $Hkey,$Z2,$Z2 |
| 200 | vpclmulqdq \$0x10,$T1,$Ii,$Hkey |
| 201 | vaesenc $rndkey,$inout1,$inout1 |
| 202 | movbe 0x48($in0),%r13 |
| 203 | vpxor $Z1,$Z3,$Z3 |
| 204 | vpclmulqdq \$0x01,$T1,$Ii,$Z1 |
| 205 | vaesenc $rndkey,$inout2,$inout2 |
| 206 | movbe 0x40($in0),%r12 |
| 207 | vpclmulqdq \$0x11,$T1,$Ii,$T1 |
| 208 | vmovdqu 0x60+8(%rsp),$Ii # I[1] |
| 209 | vaesenc $rndkey,$inout3,$inout3 |
| 210 | mov %r13,0x30+8(%rsp) |
| 211 | vaesenc $rndkey,$inout4,$inout4 |
| 212 | mov %r12,0x38+8(%rsp) |
| 213 | vpxor $T2,$Z0,$Z0 |
| 214 | vmovdqu 0x60-0x20($Xip),$T2 # borrow $T2 for $Hkey^5 |
| 215 | vaesenc $rndkey,$inout5,$inout5 |
| 216 | |
| 217 | vmovups 0x50-0x80($key),$rndkey |
| 218 | vpxor $Hkey,$Z2,$Z2 |
| 219 | vpclmulqdq \$0x00,$T2,$Ii,$Hkey |
| 220 | vaesenc $rndkey,$inout0,$inout0 |
| 221 | vpxor $Z1,$Z2,$Z2 |
| 222 | vpclmulqdq \$0x10,$T2,$Ii,$Z1 |
| 223 | vaesenc $rndkey,$inout1,$inout1 |
| 224 | movbe 0x38($in0),%r13 |
| 225 | vpxor $T1,$Z3,$Z3 |
| 226 | vpclmulqdq \$0x01,$T2,$Ii,$T1 |
| 227 | vpxor 0x70+8(%rsp),$Xi,$Xi # accumulate I[0] |
| 228 | vaesenc $rndkey,$inout2,$inout2 |
| 229 | movbe 0x30($in0),%r12 |
| 230 | vpclmulqdq \$0x11,$T2,$Ii,$T2 |
| 231 | vaesenc $rndkey,$inout3,$inout3 |
| 232 | mov %r13,0x40+8(%rsp) |
| 233 | vaesenc $rndkey,$inout4,$inout4 |
| 234 | mov %r12,0x48+8(%rsp) |
| 235 | vpxor $Hkey,$Z0,$Z0 |
| 236 | vmovdqu 0x70-0x20($Xip),$Hkey # $Hkey^6 |
| 237 | vaesenc $rndkey,$inout5,$inout5 |
| 238 | |
| 239 | vmovups 0x60-0x80($key),$rndkey |
| 240 | vpxor $Z1,$Z2,$Z2 |
| 241 | vpclmulqdq \$0x10,$Hkey,$Xi,$Z1 |
| 242 | vaesenc $rndkey,$inout0,$inout0 |
| 243 | vpxor $T1,$Z2,$Z2 |
| 244 | vpclmulqdq \$0x01,$Hkey,$Xi,$T1 |
| 245 | vaesenc $rndkey,$inout1,$inout1 |
| 246 | movbe 0x28($in0),%r13 |
| 247 | vpxor $T2,$Z3,$Z3 |
| 248 | vpclmulqdq \$0x00,$Hkey,$Xi,$T2 |
| 249 | vaesenc $rndkey,$inout2,$inout2 |
| 250 | movbe 0x20($in0),%r12 |
| 251 | vpclmulqdq \$0x11,$Hkey,$Xi,$Xi |
| 252 | vaesenc $rndkey,$inout3,$inout3 |
| 253 | mov %r13,0x50+8(%rsp) |
| 254 | vaesenc $rndkey,$inout4,$inout4 |
| 255 | mov %r12,0x58+8(%rsp) |
| 256 | vpxor $Z1,$Z2,$Z2 |
| 257 | vaesenc $rndkey,$inout5,$inout5 |
| 258 | vpxor $T1,$Z2,$Z2 |
| 259 | |
| 260 | vmovups 0x70-0x80($key),$rndkey |
| 261 | vpslldq \$8,$Z2,$Z1 |
| 262 | vpxor $T2,$Z0,$Z0 |
| 263 | vmovdqu 0x10($const),$Hkey # .Lpoly |
| 264 | |
| 265 | vaesenc $rndkey,$inout0,$inout0 |
| 266 | vpxor $Xi,$Z3,$Z3 |
| 267 | vaesenc $rndkey,$inout1,$inout1 |
| 268 | vpxor $Z1,$Z0,$Z0 |
| 269 | movbe 0x18($in0),%r13 |
| 270 | vaesenc $rndkey,$inout2,$inout2 |
| 271 | movbe 0x10($in0),%r12 |
| 272 | vpalignr \$8,$Z0,$Z0,$Ii # 1st phase |
| 273 | vpclmulqdq \$0x10,$Hkey,$Z0,$Z0 |
| 274 | mov %r13,0x60+8(%rsp) |
| 275 | vaesenc $rndkey,$inout3,$inout3 |
| 276 | mov %r12,0x68+8(%rsp) |
| 277 | vaesenc $rndkey,$inout4,$inout4 |
| 278 | vmovups 0x80-0x80($key),$T1 # borrow $T1 for $rndkey |
| 279 | vaesenc $rndkey,$inout5,$inout5 |
| 280 | |
| 281 | vaesenc $T1,$inout0,$inout0 |
| 282 | vmovups 0x90-0x80($key),$rndkey |
| 283 | vaesenc $T1,$inout1,$inout1 |
| 284 | vpsrldq \$8,$Z2,$Z2 |
| 285 | vaesenc $T1,$inout2,$inout2 |
| 286 | vpxor $Z2,$Z3,$Z3 |
| 287 | vaesenc $T1,$inout3,$inout3 |
| 288 | vpxor $Ii,$Z0,$Z0 |
| 289 | movbe 0x08($in0),%r13 |
| 290 | vaesenc $T1,$inout4,$inout4 |
| 291 | movbe 0x00($in0),%r12 |
| 292 | vaesenc $T1,$inout5,$inout5 |
| 293 | vmovups 0xa0-0x80($key),$T1 |
| 294 | cmp \$11,$rounds |
| 295 | jb .Lenc_tail # 128-bit key |
| 296 | |
| 297 | vaesenc $rndkey,$inout0,$inout0 |
| 298 | vaesenc $rndkey,$inout1,$inout1 |
| 299 | vaesenc $rndkey,$inout2,$inout2 |
| 300 | vaesenc $rndkey,$inout3,$inout3 |
| 301 | vaesenc $rndkey,$inout4,$inout4 |
| 302 | vaesenc $rndkey,$inout5,$inout5 |
| 303 | |
| 304 | vaesenc $T1,$inout0,$inout0 |
| 305 | vaesenc $T1,$inout1,$inout1 |
| 306 | vaesenc $T1,$inout2,$inout2 |
| 307 | vaesenc $T1,$inout3,$inout3 |
| 308 | vaesenc $T1,$inout4,$inout4 |
| 309 | vmovups 0xb0-0x80($key),$rndkey |
| 310 | vaesenc $T1,$inout5,$inout5 |
| 311 | vmovups 0xc0-0x80($key),$T1 |
| 312 | je .Lenc_tail # 192-bit key |
| 313 | |
| 314 | vaesenc $rndkey,$inout0,$inout0 |
| 315 | vaesenc $rndkey,$inout1,$inout1 |
| 316 | vaesenc $rndkey,$inout2,$inout2 |
| 317 | vaesenc $rndkey,$inout3,$inout3 |
| 318 | vaesenc $rndkey,$inout4,$inout4 |
| 319 | vaesenc $rndkey,$inout5,$inout5 |
| 320 | |
| 321 | vaesenc $T1,$inout0,$inout0 |
| 322 | vaesenc $T1,$inout1,$inout1 |
| 323 | vaesenc $T1,$inout2,$inout2 |
| 324 | vaesenc $T1,$inout3,$inout3 |
| 325 | vaesenc $T1,$inout4,$inout4 |
| 326 | vmovups 0xd0-0x80($key),$rndkey |
| 327 | vaesenc $T1,$inout5,$inout5 |
| 328 | vmovups 0xe0-0x80($key),$T1 |
| 329 | jmp .Lenc_tail # 256-bit key |
| 330 | |
| 331 | .align 32 |
| 332 | .Lhandle_ctr32: |
| 333 | vmovdqu ($const),$Ii # borrow $Ii for .Lbswap_mask |
| 334 | vpshufb $Ii,$T1,$Z2 # byte-swap counter |
| 335 | vmovdqu 0x30($const),$Z1 # borrow $Z1, .Ltwo_lsb |
| 336 | vpaddd 0x40($const),$Z2,$inout1 # .Lone_lsb |
| 337 | vpaddd $Z1,$Z2,$inout2 |
| 338 | vmovdqu 0x00-0x20($Xip),$Hkey # $Hkey^1 |
| 339 | vpaddd $Z1,$inout1,$inout3 |
| 340 | vpshufb $Ii,$inout1,$inout1 |
| 341 | vpaddd $Z1,$inout2,$inout4 |
| 342 | vpshufb $Ii,$inout2,$inout2 |
| 343 | vpxor $rndkey,$inout1,$inout1 |
| 344 | vpaddd $Z1,$inout3,$inout5 |
| 345 | vpshufb $Ii,$inout3,$inout3 |
| 346 | vpxor $rndkey,$inout2,$inout2 |
| 347 | vpaddd $Z1,$inout4,$T1 # byte-swapped next counter value |
| 348 | vpshufb $Ii,$inout4,$inout4 |
| 349 | vpshufb $Ii,$inout5,$inout5 |
| 350 | vpshufb $Ii,$T1,$T1 # next counter value |
| 351 | jmp .Lresume_ctr32 |
| 352 | |
| 353 | .align 32 |
| 354 | .Lenc_tail: |
| 355 | vaesenc $rndkey,$inout0,$inout0 |
| 356 | vmovdqu $Z3,16+8(%rsp) # postpone vpxor $Z3,$Xi,$Xi |
| 357 | vpalignr \$8,$Z0,$Z0,$Xi # 2nd phase |
| 358 | vaesenc $rndkey,$inout1,$inout1 |
| 359 | vpclmulqdq \$0x10,$Hkey,$Z0,$Z0 |
| 360 | vpxor 0x00($inp),$T1,$T2 |
| 361 | vaesenc $rndkey,$inout2,$inout2 |
| 362 | vpxor 0x10($inp),$T1,$Ii |
| 363 | vaesenc $rndkey,$inout3,$inout3 |
| 364 | vpxor 0x20($inp),$T1,$Z1 |
| 365 | vaesenc $rndkey,$inout4,$inout4 |
| 366 | vpxor 0x30($inp),$T1,$Z2 |
| 367 | vaesenc $rndkey,$inout5,$inout5 |
| 368 | vpxor 0x40($inp),$T1,$Z3 |
| 369 | vpxor 0x50($inp),$T1,$Hkey |
| 370 | vmovdqu ($ivp),$T1 # load next counter value |
| 371 | |
| 372 | vaesenclast $T2,$inout0,$inout0 |
| 373 | vmovdqu 0x20($const),$T2 # borrow $T2, .Lone_msb |
| 374 | vaesenclast $Ii,$inout1,$inout1 |
| 375 | vpaddb $T2,$T1,$Ii |
| 376 | mov %r13,0x70+8(%rsp) |
| 377 | lea 0x60($inp),$inp |
| 378 | vaesenclast $Z1,$inout2,$inout2 |
| 379 | vpaddb $T2,$Ii,$Z1 |
| 380 | mov %r12,0x78+8(%rsp) |
| 381 | lea 0x60($out),$out |
| 382 | vmovdqu 0x00-0x80($key),$rndkey |
| 383 | vaesenclast $Z2,$inout3,$inout3 |
| 384 | vpaddb $T2,$Z1,$Z2 |
| 385 | vaesenclast $Z3, $inout4,$inout4 |
| 386 | vpaddb $T2,$Z2,$Z3 |
| 387 | vaesenclast $Hkey,$inout5,$inout5 |
| 388 | vpaddb $T2,$Z3,$Hkey |
| 389 | |
| 390 | add \$0x60,$ret |
| 391 | sub \$0x6,$len |
| 392 | jc .L6x_done |
| 393 | |
| 394 | vmovups $inout0,-0x60($out) # save output |
| 395 | vpxor $rndkey,$T1,$inout0 |
| 396 | vmovups $inout1,-0x50($out) |
| 397 | vmovdqa $Ii,$inout1 # 0 latency |
| 398 | vmovups $inout2,-0x40($out) |
| 399 | vmovdqa $Z1,$inout2 # 0 latency |
| 400 | vmovups $inout3,-0x30($out) |
| 401 | vmovdqa $Z2,$inout3 # 0 latency |
| 402 | vmovups $inout4,-0x20($out) |
| 403 | vmovdqa $Z3,$inout4 # 0 latency |
| 404 | vmovups $inout5,-0x10($out) |
| 405 | vmovdqa $Hkey,$inout5 # 0 latency |
| 406 | vmovdqu 0x20+8(%rsp),$Z3 # I[5] |
| 407 | jmp .Loop6x |
| 408 | |
| 409 | .L6x_done: |
| 410 | vpxor 16+8(%rsp),$Xi,$Xi # modulo-scheduled |
| 411 | vpxor $Z0,$Xi,$Xi # modulo-scheduled |
| 412 | |
| 413 | ret |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 414 | .cfi_endproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 415 | .size _aesni_ctr32_ghash_6x,.-_aesni_ctr32_ghash_6x |
| 416 | ___ |
| 417 | ###################################################################### |
| 418 | # |
| 419 | # size_t aesni_gcm_[en|de]crypt(const void *inp, void *out, size_t len, |
| 420 | # const AES_KEY *key, unsigned char iv[16], |
| 421 | # struct { u128 Xi,H,Htbl[9]; } *Xip); |
| 422 | $code.=<<___; |
| 423 | .globl aesni_gcm_decrypt |
| 424 | .type aesni_gcm_decrypt,\@function,6 |
| 425 | .align 32 |
| 426 | aesni_gcm_decrypt: |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 427 | .cfi_startproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 428 | xor $ret,$ret |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 429 | |
| 430 | # We call |_aesni_ctr32_ghash_6x|, which requires at least 96 (0x60) |
| 431 | # bytes of input. |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 432 | cmp \$0x60,$len # minimal accepted length |
| 433 | jb .Lgcm_dec_abort |
| 434 | |
| 435 | lea (%rsp),%rax # save stack pointer |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 436 | .cfi_def_cfa_register %rax |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 437 | push %rbx |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 438 | .cfi_push %rbx |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 439 | push %rbp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 440 | .cfi_push %rbp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 441 | push %r12 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 442 | .cfi_push %r12 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 443 | push %r13 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 444 | .cfi_push %r13 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 445 | push %r14 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 446 | .cfi_push %r14 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 447 | push %r15 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 448 | .cfi_push %r15 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 449 | ___ |
| 450 | $code.=<<___ if ($win64); |
| 451 | lea -0xa8(%rsp),%rsp |
| 452 | movaps %xmm6,-0xd8(%rax) |
| 453 | movaps %xmm7,-0xc8(%rax) |
| 454 | movaps %xmm8,-0xb8(%rax) |
| 455 | movaps %xmm9,-0xa8(%rax) |
| 456 | movaps %xmm10,-0x98(%rax) |
| 457 | movaps %xmm11,-0x88(%rax) |
| 458 | movaps %xmm12,-0x78(%rax) |
| 459 | movaps %xmm13,-0x68(%rax) |
| 460 | movaps %xmm14,-0x58(%rax) |
| 461 | movaps %xmm15,-0x48(%rax) |
| 462 | .Lgcm_dec_body: |
| 463 | ___ |
| 464 | $code.=<<___; |
| 465 | vzeroupper |
| 466 | |
| 467 | vmovdqu ($ivp),$T1 # input counter value |
| 468 | add \$-128,%rsp |
| 469 | mov 12($ivp),$counter |
| 470 | lea .Lbswap_mask(%rip),$const |
| 471 | lea -0x80($key),$in0 # borrow $in0 |
| 472 | mov \$0xf80,$end0 # borrow $end0 |
| 473 | vmovdqu ($Xip),$Xi # load Xi |
| 474 | and \$-128,%rsp # ensure stack alignment |
| 475 | vmovdqu ($const),$Ii # borrow $Ii for .Lbswap_mask |
| 476 | lea 0x80($key),$key # size optimization |
| 477 | lea 0x20+0x20($Xip),$Xip # size optimization |
| 478 | mov 0xf0-0x80($key),$rounds |
| 479 | vpshufb $Ii,$Xi,$Xi |
| 480 | |
| 481 | and $end0,$in0 |
| 482 | and %rsp,$end0 |
| 483 | sub $in0,$end0 |
| 484 | jc .Ldec_no_key_aliasing |
| 485 | cmp \$768,$end0 |
| 486 | jnc .Ldec_no_key_aliasing |
| 487 | sub $end0,%rsp # avoid aliasing with key |
| 488 | .Ldec_no_key_aliasing: |
| 489 | |
| 490 | vmovdqu 0x50($inp),$Z3 # I[5] |
| 491 | lea ($inp),$in0 |
| 492 | vmovdqu 0x40($inp),$Z0 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 493 | |
| 494 | # |_aesni_ctr32_ghash_6x| requires |$end0| to point to 2*96 (0xc0) |
| 495 | # bytes before the end of the input. Note, in particular, that this is |
| 496 | # correct even if |$len| is not an even multiple of 96 or 16. XXX: This |
| 497 | # seems to require that |$inp| + |$len| >= 2*96 (0xc0); i.e. |$inp| must |
| 498 | # not be near the very beginning of the address space when |$len| < 2*96 |
| 499 | # (0xc0). |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 500 | lea -0xc0($inp,$len),$end0 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 501 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 502 | vmovdqu 0x30($inp),$Z1 |
| 503 | shr \$4,$len |
| 504 | xor $ret,$ret |
| 505 | vmovdqu 0x20($inp),$Z2 |
| 506 | vpshufb $Ii,$Z3,$Z3 # passed to _aesni_ctr32_ghash_6x |
| 507 | vmovdqu 0x10($inp),$T2 |
| 508 | vpshufb $Ii,$Z0,$Z0 |
| 509 | vmovdqu ($inp),$Hkey |
| 510 | vpshufb $Ii,$Z1,$Z1 |
| 511 | vmovdqu $Z0,0x30(%rsp) |
| 512 | vpshufb $Ii,$Z2,$Z2 |
| 513 | vmovdqu $Z1,0x40(%rsp) |
| 514 | vpshufb $Ii,$T2,$T2 |
| 515 | vmovdqu $Z2,0x50(%rsp) |
| 516 | vpshufb $Ii,$Hkey,$Hkey |
| 517 | vmovdqu $T2,0x60(%rsp) |
| 518 | vmovdqu $Hkey,0x70(%rsp) |
| 519 | |
| 520 | call _aesni_ctr32_ghash_6x |
| 521 | |
| 522 | vmovups $inout0,-0x60($out) # save output |
| 523 | vmovups $inout1,-0x50($out) |
| 524 | vmovups $inout2,-0x40($out) |
| 525 | vmovups $inout3,-0x30($out) |
| 526 | vmovups $inout4,-0x20($out) |
| 527 | vmovups $inout5,-0x10($out) |
| 528 | |
| 529 | vpshufb ($const),$Xi,$Xi # .Lbswap_mask |
| 530 | vmovdqu $Xi,-0x40($Xip) # output Xi |
| 531 | |
| 532 | vzeroupper |
| 533 | ___ |
| 534 | $code.=<<___ if ($win64); |
| 535 | movaps -0xd8(%rax),%xmm6 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 536 | movaps -0xc8(%rax),%xmm7 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 537 | movaps -0xb8(%rax),%xmm8 |
| 538 | movaps -0xa8(%rax),%xmm9 |
| 539 | movaps -0x98(%rax),%xmm10 |
| 540 | movaps -0x88(%rax),%xmm11 |
| 541 | movaps -0x78(%rax),%xmm12 |
| 542 | movaps -0x68(%rax),%xmm13 |
| 543 | movaps -0x58(%rax),%xmm14 |
| 544 | movaps -0x48(%rax),%xmm15 |
| 545 | ___ |
| 546 | $code.=<<___; |
| 547 | mov -48(%rax),%r15 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 548 | .cfi_restore %r15 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 549 | mov -40(%rax),%r14 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 550 | .cfi_restore %r14 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 551 | mov -32(%rax),%r13 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 552 | .cfi_restore %r13 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 553 | mov -24(%rax),%r12 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 554 | .cfi_restore %r12 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 555 | mov -16(%rax),%rbp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 556 | .cfi_restore %rbp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 557 | mov -8(%rax),%rbx |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 558 | .cfi_restore %rbx |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 559 | lea (%rax),%rsp # restore %rsp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 560 | .cfi_def_cfa_register %rsp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 561 | .Lgcm_dec_abort: |
| 562 | mov $ret,%rax # return value |
| 563 | ret |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 564 | .cfi_endproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 565 | .size aesni_gcm_decrypt,.-aesni_gcm_decrypt |
| 566 | ___ |
| 567 | |
| 568 | $code.=<<___; |
| 569 | .type _aesni_ctr32_6x,\@abi-omnipotent |
| 570 | .align 32 |
| 571 | _aesni_ctr32_6x: |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 572 | .cfi_startproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 573 | vmovdqu 0x00-0x80($key),$Z0 # borrow $Z0 for $rndkey |
| 574 | vmovdqu 0x20($const),$T2 # borrow $T2, .Lone_msb |
| 575 | lea -1($rounds),%r13 |
| 576 | vmovups 0x10-0x80($key),$rndkey |
| 577 | lea 0x20-0x80($key),%r12 |
| 578 | vpxor $Z0,$T1,$inout0 |
| 579 | add \$`6<<24`,$counter |
| 580 | jc .Lhandle_ctr32_2 |
| 581 | vpaddb $T2,$T1,$inout1 |
| 582 | vpaddb $T2,$inout1,$inout2 |
| 583 | vpxor $Z0,$inout1,$inout1 |
| 584 | vpaddb $T2,$inout2,$inout3 |
| 585 | vpxor $Z0,$inout2,$inout2 |
| 586 | vpaddb $T2,$inout3,$inout4 |
| 587 | vpxor $Z0,$inout3,$inout3 |
| 588 | vpaddb $T2,$inout4,$inout5 |
| 589 | vpxor $Z0,$inout4,$inout4 |
| 590 | vpaddb $T2,$inout5,$T1 |
| 591 | vpxor $Z0,$inout5,$inout5 |
| 592 | jmp .Loop_ctr32 |
| 593 | |
| 594 | .align 16 |
| 595 | .Loop_ctr32: |
| 596 | vaesenc $rndkey,$inout0,$inout0 |
| 597 | vaesenc $rndkey,$inout1,$inout1 |
| 598 | vaesenc $rndkey,$inout2,$inout2 |
| 599 | vaesenc $rndkey,$inout3,$inout3 |
| 600 | vaesenc $rndkey,$inout4,$inout4 |
| 601 | vaesenc $rndkey,$inout5,$inout5 |
| 602 | vmovups (%r12),$rndkey |
| 603 | lea 0x10(%r12),%r12 |
| 604 | dec %r13d |
| 605 | jnz .Loop_ctr32 |
| 606 | |
| 607 | vmovdqu (%r12),$Hkey # last round key |
| 608 | vaesenc $rndkey,$inout0,$inout0 |
| 609 | vpxor 0x00($inp),$Hkey,$Z0 |
| 610 | vaesenc $rndkey,$inout1,$inout1 |
| 611 | vpxor 0x10($inp),$Hkey,$Z1 |
| 612 | vaesenc $rndkey,$inout2,$inout2 |
| 613 | vpxor 0x20($inp),$Hkey,$Z2 |
| 614 | vaesenc $rndkey,$inout3,$inout3 |
| 615 | vpxor 0x30($inp),$Hkey,$Xi |
| 616 | vaesenc $rndkey,$inout4,$inout4 |
| 617 | vpxor 0x40($inp),$Hkey,$T2 |
| 618 | vaesenc $rndkey,$inout5,$inout5 |
| 619 | vpxor 0x50($inp),$Hkey,$Hkey |
| 620 | lea 0x60($inp),$inp |
| 621 | |
| 622 | vaesenclast $Z0,$inout0,$inout0 |
| 623 | vaesenclast $Z1,$inout1,$inout1 |
| 624 | vaesenclast $Z2,$inout2,$inout2 |
| 625 | vaesenclast $Xi,$inout3,$inout3 |
| 626 | vaesenclast $T2,$inout4,$inout4 |
| 627 | vaesenclast $Hkey,$inout5,$inout5 |
| 628 | vmovups $inout0,0x00($out) |
| 629 | vmovups $inout1,0x10($out) |
| 630 | vmovups $inout2,0x20($out) |
| 631 | vmovups $inout3,0x30($out) |
| 632 | vmovups $inout4,0x40($out) |
| 633 | vmovups $inout5,0x50($out) |
| 634 | lea 0x60($out),$out |
| 635 | |
| 636 | ret |
| 637 | .align 32 |
| 638 | .Lhandle_ctr32_2: |
| 639 | vpshufb $Ii,$T1,$Z2 # byte-swap counter |
| 640 | vmovdqu 0x30($const),$Z1 # borrow $Z1, .Ltwo_lsb |
| 641 | vpaddd 0x40($const),$Z2,$inout1 # .Lone_lsb |
| 642 | vpaddd $Z1,$Z2,$inout2 |
| 643 | vpaddd $Z1,$inout1,$inout3 |
| 644 | vpshufb $Ii,$inout1,$inout1 |
| 645 | vpaddd $Z1,$inout2,$inout4 |
| 646 | vpshufb $Ii,$inout2,$inout2 |
| 647 | vpxor $Z0,$inout1,$inout1 |
| 648 | vpaddd $Z1,$inout3,$inout5 |
| 649 | vpshufb $Ii,$inout3,$inout3 |
| 650 | vpxor $Z0,$inout2,$inout2 |
| 651 | vpaddd $Z1,$inout4,$T1 # byte-swapped next counter value |
| 652 | vpshufb $Ii,$inout4,$inout4 |
| 653 | vpxor $Z0,$inout3,$inout3 |
| 654 | vpshufb $Ii,$inout5,$inout5 |
| 655 | vpxor $Z0,$inout4,$inout4 |
| 656 | vpshufb $Ii,$T1,$T1 # next counter value |
| 657 | vpxor $Z0,$inout5,$inout5 |
| 658 | jmp .Loop_ctr32 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 659 | .cfi_endproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 660 | .size _aesni_ctr32_6x,.-_aesni_ctr32_6x |
| 661 | |
| 662 | .globl aesni_gcm_encrypt |
| 663 | .type aesni_gcm_encrypt,\@function,6 |
| 664 | .align 32 |
| 665 | aesni_gcm_encrypt: |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 666 | .cfi_startproc |
Robert Sloan | 4c22c5f | 2019-03-01 15:53:37 -0800 | [diff] [blame] | 667 | #ifndef NDEBUG |
| 668 | #ifndef BORINGSSL_FIPS |
| 669 | .extern BORINGSSL_function_hit |
| 670 | movb \$1,BORINGSSL_function_hit+2(%rip) |
| 671 | #endif |
| 672 | #endif |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 673 | xor $ret,$ret |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 674 | |
| 675 | # We call |_aesni_ctr32_6x| twice, each call consuming 96 bytes of |
| 676 | # input. Then we call |_aesni_ctr32_ghash_6x|, which requires at |
| 677 | # least 96 more bytes of input. |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 678 | cmp \$0x60*3,$len # minimal accepted length |
| 679 | jb .Lgcm_enc_abort |
| 680 | |
| 681 | lea (%rsp),%rax # save stack pointer |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 682 | .cfi_def_cfa_register %rax |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 683 | push %rbx |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 684 | .cfi_push %rbx |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 685 | push %rbp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 686 | .cfi_push %rbp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 687 | push %r12 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 688 | .cfi_push %r12 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 689 | push %r13 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 690 | .cfi_push %r13 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 691 | push %r14 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 692 | .cfi_push %r14 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 693 | push %r15 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 694 | .cfi_push %r15 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 695 | ___ |
| 696 | $code.=<<___ if ($win64); |
| 697 | lea -0xa8(%rsp),%rsp |
| 698 | movaps %xmm6,-0xd8(%rax) |
| 699 | movaps %xmm7,-0xc8(%rax) |
| 700 | movaps %xmm8,-0xb8(%rax) |
| 701 | movaps %xmm9,-0xa8(%rax) |
| 702 | movaps %xmm10,-0x98(%rax) |
| 703 | movaps %xmm11,-0x88(%rax) |
| 704 | movaps %xmm12,-0x78(%rax) |
| 705 | movaps %xmm13,-0x68(%rax) |
| 706 | movaps %xmm14,-0x58(%rax) |
| 707 | movaps %xmm15,-0x48(%rax) |
| 708 | .Lgcm_enc_body: |
| 709 | ___ |
| 710 | $code.=<<___; |
| 711 | vzeroupper |
| 712 | |
| 713 | vmovdqu ($ivp),$T1 # input counter value |
| 714 | add \$-128,%rsp |
| 715 | mov 12($ivp),$counter |
| 716 | lea .Lbswap_mask(%rip),$const |
| 717 | lea -0x80($key),$in0 # borrow $in0 |
| 718 | mov \$0xf80,$end0 # borrow $end0 |
| 719 | lea 0x80($key),$key # size optimization |
| 720 | vmovdqu ($const),$Ii # borrow $Ii for .Lbswap_mask |
| 721 | and \$-128,%rsp # ensure stack alignment |
| 722 | mov 0xf0-0x80($key),$rounds |
| 723 | |
| 724 | and $end0,$in0 |
| 725 | and %rsp,$end0 |
| 726 | sub $in0,$end0 |
| 727 | jc .Lenc_no_key_aliasing |
| 728 | cmp \$768,$end0 |
| 729 | jnc .Lenc_no_key_aliasing |
| 730 | sub $end0,%rsp # avoid aliasing with key |
| 731 | .Lenc_no_key_aliasing: |
| 732 | |
| 733 | lea ($out),$in0 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 734 | |
| 735 | # |_aesni_ctr32_ghash_6x| requires |$end0| to point to 2*96 (0xc0) |
| 736 | # bytes before the end of the input. Note, in particular, that this is |
| 737 | # correct even if |$len| is not an even multiple of 96 or 16. Unlike in |
| 738 | # the decryption case, there's no caveat that |$out| must not be near |
| 739 | # the very beginning of the address space, because we know that |
| 740 | # |$len| >= 3*96 from the check above, and so we know |
| 741 | # |$out| + |$len| >= 2*96 (0xc0). |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 742 | lea -0xc0($out,$len),$end0 |
David Benjamin | 4969cc9 | 2016-04-22 15:02:23 -0400 | [diff] [blame] | 743 | |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 744 | shr \$4,$len |
| 745 | |
| 746 | call _aesni_ctr32_6x |
| 747 | vpshufb $Ii,$inout0,$Xi # save bswapped output on stack |
| 748 | vpshufb $Ii,$inout1,$T2 |
| 749 | vmovdqu $Xi,0x70(%rsp) |
| 750 | vpshufb $Ii,$inout2,$Z0 |
| 751 | vmovdqu $T2,0x60(%rsp) |
| 752 | vpshufb $Ii,$inout3,$Z1 |
| 753 | vmovdqu $Z0,0x50(%rsp) |
| 754 | vpshufb $Ii,$inout4,$Z2 |
| 755 | vmovdqu $Z1,0x40(%rsp) |
| 756 | vpshufb $Ii,$inout5,$Z3 # passed to _aesni_ctr32_ghash_6x |
| 757 | vmovdqu $Z2,0x30(%rsp) |
| 758 | |
| 759 | call _aesni_ctr32_6x |
| 760 | |
| 761 | vmovdqu ($Xip),$Xi # load Xi |
| 762 | lea 0x20+0x20($Xip),$Xip # size optimization |
| 763 | sub \$12,$len |
| 764 | mov \$0x60*2,$ret |
| 765 | vpshufb $Ii,$Xi,$Xi |
| 766 | |
| 767 | call _aesni_ctr32_ghash_6x |
| 768 | vmovdqu 0x20(%rsp),$Z3 # I[5] |
| 769 | vmovdqu ($const),$Ii # borrow $Ii for .Lbswap_mask |
| 770 | vmovdqu 0x00-0x20($Xip),$Hkey # $Hkey^1 |
| 771 | vpunpckhqdq $Z3,$Z3,$T1 |
| 772 | vmovdqu 0x20-0x20($Xip),$rndkey # borrow $rndkey for $HK |
| 773 | vmovups $inout0,-0x60($out) # save output |
| 774 | vpshufb $Ii,$inout0,$inout0 # but keep bswapped copy |
| 775 | vpxor $Z3,$T1,$T1 |
| 776 | vmovups $inout1,-0x50($out) |
| 777 | vpshufb $Ii,$inout1,$inout1 |
| 778 | vmovups $inout2,-0x40($out) |
| 779 | vpshufb $Ii,$inout2,$inout2 |
| 780 | vmovups $inout3,-0x30($out) |
| 781 | vpshufb $Ii,$inout3,$inout3 |
| 782 | vmovups $inout4,-0x20($out) |
| 783 | vpshufb $Ii,$inout4,$inout4 |
| 784 | vmovups $inout5,-0x10($out) |
| 785 | vpshufb $Ii,$inout5,$inout5 |
| 786 | vmovdqu $inout0,0x10(%rsp) # free $inout0 |
| 787 | ___ |
| 788 | { my ($HK,$T3)=($rndkey,$inout0); |
| 789 | |
| 790 | $code.=<<___; |
| 791 | vmovdqu 0x30(%rsp),$Z2 # I[4] |
| 792 | vmovdqu 0x10-0x20($Xip),$Ii # borrow $Ii for $Hkey^2 |
| 793 | vpunpckhqdq $Z2,$Z2,$T2 |
| 794 | vpclmulqdq \$0x00,$Hkey,$Z3,$Z1 |
| 795 | vpxor $Z2,$T2,$T2 |
| 796 | vpclmulqdq \$0x11,$Hkey,$Z3,$Z3 |
| 797 | vpclmulqdq \$0x00,$HK,$T1,$T1 |
| 798 | |
| 799 | vmovdqu 0x40(%rsp),$T3 # I[3] |
| 800 | vpclmulqdq \$0x00,$Ii,$Z2,$Z0 |
| 801 | vmovdqu 0x30-0x20($Xip),$Hkey # $Hkey^3 |
| 802 | vpxor $Z1,$Z0,$Z0 |
| 803 | vpunpckhqdq $T3,$T3,$Z1 |
| 804 | vpclmulqdq \$0x11,$Ii,$Z2,$Z2 |
| 805 | vpxor $T3,$Z1,$Z1 |
| 806 | vpxor $Z3,$Z2,$Z2 |
| 807 | vpclmulqdq \$0x10,$HK,$T2,$T2 |
| 808 | vmovdqu 0x50-0x20($Xip),$HK |
| 809 | vpxor $T1,$T2,$T2 |
| 810 | |
| 811 | vmovdqu 0x50(%rsp),$T1 # I[2] |
| 812 | vpclmulqdq \$0x00,$Hkey,$T3,$Z3 |
| 813 | vmovdqu 0x40-0x20($Xip),$Ii # borrow $Ii for $Hkey^4 |
| 814 | vpxor $Z0,$Z3,$Z3 |
| 815 | vpunpckhqdq $T1,$T1,$Z0 |
| 816 | vpclmulqdq \$0x11,$Hkey,$T3,$T3 |
| 817 | vpxor $T1,$Z0,$Z0 |
| 818 | vpxor $Z2,$T3,$T3 |
| 819 | vpclmulqdq \$0x00,$HK,$Z1,$Z1 |
| 820 | vpxor $T2,$Z1,$Z1 |
| 821 | |
| 822 | vmovdqu 0x60(%rsp),$T2 # I[1] |
| 823 | vpclmulqdq \$0x00,$Ii,$T1,$Z2 |
| 824 | vmovdqu 0x60-0x20($Xip),$Hkey # $Hkey^5 |
| 825 | vpxor $Z3,$Z2,$Z2 |
| 826 | vpunpckhqdq $T2,$T2,$Z3 |
| 827 | vpclmulqdq \$0x11,$Ii,$T1,$T1 |
| 828 | vpxor $T2,$Z3,$Z3 |
| 829 | vpxor $T3,$T1,$T1 |
| 830 | vpclmulqdq \$0x10,$HK,$Z0,$Z0 |
| 831 | vmovdqu 0x80-0x20($Xip),$HK |
| 832 | vpxor $Z1,$Z0,$Z0 |
| 833 | |
| 834 | vpxor 0x70(%rsp),$Xi,$Xi # accumulate I[0] |
| 835 | vpclmulqdq \$0x00,$Hkey,$T2,$Z1 |
| 836 | vmovdqu 0x70-0x20($Xip),$Ii # borrow $Ii for $Hkey^6 |
| 837 | vpunpckhqdq $Xi,$Xi,$T3 |
| 838 | vpxor $Z2,$Z1,$Z1 |
| 839 | vpclmulqdq \$0x11,$Hkey,$T2,$T2 |
| 840 | vpxor $Xi,$T3,$T3 |
| 841 | vpxor $T1,$T2,$T2 |
| 842 | vpclmulqdq \$0x00,$HK,$Z3,$Z3 |
| 843 | vpxor $Z0,$Z3,$Z0 |
| 844 | |
| 845 | vpclmulqdq \$0x00,$Ii,$Xi,$Z2 |
| 846 | vmovdqu 0x00-0x20($Xip),$Hkey # $Hkey^1 |
| 847 | vpunpckhqdq $inout5,$inout5,$T1 |
| 848 | vpclmulqdq \$0x11,$Ii,$Xi,$Xi |
| 849 | vpxor $inout5,$T1,$T1 |
| 850 | vpxor $Z1,$Z2,$Z1 |
| 851 | vpclmulqdq \$0x10,$HK,$T3,$T3 |
| 852 | vmovdqu 0x20-0x20($Xip),$HK |
| 853 | vpxor $T2,$Xi,$Z3 |
| 854 | vpxor $Z0,$T3,$Z2 |
| 855 | |
| 856 | vmovdqu 0x10-0x20($Xip),$Ii # borrow $Ii for $Hkey^2 |
| 857 | vpxor $Z1,$Z3,$T3 # aggregated Karatsuba post-processing |
| 858 | vpclmulqdq \$0x00,$Hkey,$inout5,$Z0 |
| 859 | vpxor $T3,$Z2,$Z2 |
| 860 | vpunpckhqdq $inout4,$inout4,$T2 |
| 861 | vpclmulqdq \$0x11,$Hkey,$inout5,$inout5 |
| 862 | vpxor $inout4,$T2,$T2 |
| 863 | vpslldq \$8,$Z2,$T3 |
| 864 | vpclmulqdq \$0x00,$HK,$T1,$T1 |
| 865 | vpxor $T3,$Z1,$Xi |
| 866 | vpsrldq \$8,$Z2,$Z2 |
| 867 | vpxor $Z2,$Z3,$Z3 |
| 868 | |
| 869 | vpclmulqdq \$0x00,$Ii,$inout4,$Z1 |
| 870 | vmovdqu 0x30-0x20($Xip),$Hkey # $Hkey^3 |
| 871 | vpxor $Z0,$Z1,$Z1 |
| 872 | vpunpckhqdq $inout3,$inout3,$T3 |
| 873 | vpclmulqdq \$0x11,$Ii,$inout4,$inout4 |
| 874 | vpxor $inout3,$T3,$T3 |
| 875 | vpxor $inout5,$inout4,$inout4 |
| 876 | vpalignr \$8,$Xi,$Xi,$inout5 # 1st phase |
| 877 | vpclmulqdq \$0x10,$HK,$T2,$T2 |
| 878 | vmovdqu 0x50-0x20($Xip),$HK |
| 879 | vpxor $T1,$T2,$T2 |
| 880 | |
| 881 | vpclmulqdq \$0x00,$Hkey,$inout3,$Z0 |
| 882 | vmovdqu 0x40-0x20($Xip),$Ii # borrow $Ii for $Hkey^4 |
| 883 | vpxor $Z1,$Z0,$Z0 |
| 884 | vpunpckhqdq $inout2,$inout2,$T1 |
| 885 | vpclmulqdq \$0x11,$Hkey,$inout3,$inout3 |
| 886 | vpxor $inout2,$T1,$T1 |
| 887 | vpxor $inout4,$inout3,$inout3 |
| 888 | vxorps 0x10(%rsp),$Z3,$Z3 # accumulate $inout0 |
| 889 | vpclmulqdq \$0x00,$HK,$T3,$T3 |
| 890 | vpxor $T2,$T3,$T3 |
| 891 | |
| 892 | vpclmulqdq \$0x10,0x10($const),$Xi,$Xi |
| 893 | vxorps $inout5,$Xi,$Xi |
| 894 | |
| 895 | vpclmulqdq \$0x00,$Ii,$inout2,$Z1 |
| 896 | vmovdqu 0x60-0x20($Xip),$Hkey # $Hkey^5 |
| 897 | vpxor $Z0,$Z1,$Z1 |
| 898 | vpunpckhqdq $inout1,$inout1,$T2 |
| 899 | vpclmulqdq \$0x11,$Ii,$inout2,$inout2 |
| 900 | vpxor $inout1,$T2,$T2 |
| 901 | vpalignr \$8,$Xi,$Xi,$inout5 # 2nd phase |
| 902 | vpxor $inout3,$inout2,$inout2 |
| 903 | vpclmulqdq \$0x10,$HK,$T1,$T1 |
| 904 | vmovdqu 0x80-0x20($Xip),$HK |
| 905 | vpxor $T3,$T1,$T1 |
| 906 | |
| 907 | vxorps $Z3,$inout5,$inout5 |
| 908 | vpclmulqdq \$0x10,0x10($const),$Xi,$Xi |
| 909 | vxorps $inout5,$Xi,$Xi |
| 910 | |
| 911 | vpclmulqdq \$0x00,$Hkey,$inout1,$Z0 |
| 912 | vmovdqu 0x70-0x20($Xip),$Ii # borrow $Ii for $Hkey^6 |
| 913 | vpxor $Z1,$Z0,$Z0 |
| 914 | vpunpckhqdq $Xi,$Xi,$T3 |
| 915 | vpclmulqdq \$0x11,$Hkey,$inout1,$inout1 |
| 916 | vpxor $Xi,$T3,$T3 |
| 917 | vpxor $inout2,$inout1,$inout1 |
| 918 | vpclmulqdq \$0x00,$HK,$T2,$T2 |
| 919 | vpxor $T1,$T2,$T2 |
| 920 | |
| 921 | vpclmulqdq \$0x00,$Ii,$Xi,$Z1 |
| 922 | vpclmulqdq \$0x11,$Ii,$Xi,$Z3 |
| 923 | vpxor $Z0,$Z1,$Z1 |
| 924 | vpclmulqdq \$0x10,$HK,$T3,$Z2 |
| 925 | vpxor $inout1,$Z3,$Z3 |
| 926 | vpxor $T2,$Z2,$Z2 |
| 927 | |
| 928 | vpxor $Z1,$Z3,$Z0 # aggregated Karatsuba post-processing |
| 929 | vpxor $Z0,$Z2,$Z2 |
| 930 | vpslldq \$8,$Z2,$T1 |
| 931 | vmovdqu 0x10($const),$Hkey # .Lpoly |
| 932 | vpsrldq \$8,$Z2,$Z2 |
| 933 | vpxor $T1,$Z1,$Xi |
| 934 | vpxor $Z2,$Z3,$Z3 |
| 935 | |
| 936 | vpalignr \$8,$Xi,$Xi,$T2 # 1st phase |
| 937 | vpclmulqdq \$0x10,$Hkey,$Xi,$Xi |
| 938 | vpxor $T2,$Xi,$Xi |
| 939 | |
| 940 | vpalignr \$8,$Xi,$Xi,$T2 # 2nd phase |
| 941 | vpclmulqdq \$0x10,$Hkey,$Xi,$Xi |
| 942 | vpxor $Z3,$T2,$T2 |
| 943 | vpxor $T2,$Xi,$Xi |
| 944 | ___ |
| 945 | } |
| 946 | $code.=<<___; |
| 947 | vpshufb ($const),$Xi,$Xi # .Lbswap_mask |
| 948 | vmovdqu $Xi,-0x40($Xip) # output Xi |
| 949 | |
| 950 | vzeroupper |
| 951 | ___ |
| 952 | $code.=<<___ if ($win64); |
| 953 | movaps -0xd8(%rax),%xmm6 |
| 954 | movaps -0xc8(%rax),%xmm7 |
| 955 | movaps -0xb8(%rax),%xmm8 |
| 956 | movaps -0xa8(%rax),%xmm9 |
| 957 | movaps -0x98(%rax),%xmm10 |
| 958 | movaps -0x88(%rax),%xmm11 |
| 959 | movaps -0x78(%rax),%xmm12 |
| 960 | movaps -0x68(%rax),%xmm13 |
| 961 | movaps -0x58(%rax),%xmm14 |
| 962 | movaps -0x48(%rax),%xmm15 |
| 963 | ___ |
| 964 | $code.=<<___; |
| 965 | mov -48(%rax),%r15 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 966 | .cfi_restore %r15 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 967 | mov -40(%rax),%r14 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 968 | .cfi_restore %r14 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 969 | mov -32(%rax),%r13 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 970 | .cfi_restore %r13 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 971 | mov -24(%rax),%r12 |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 972 | .cfi_restore %r12 |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 973 | mov -16(%rax),%rbp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 974 | .cfi_restore %rbp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 975 | mov -8(%rax),%rbx |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 976 | .cfi_restore %rbx |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 977 | lea (%rax),%rsp # restore %rsp |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 978 | .cfi_def_cfa_register %rsp |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 979 | .Lgcm_enc_abort: |
| 980 | mov $ret,%rax # return value |
| 981 | ret |
Robert Sloan | d5c2215 | 2017-11-13 09:22:12 -0800 | [diff] [blame] | 982 | .cfi_endproc |
Adam Langley | d9e397b | 2015-01-22 14:27:53 -0800 | [diff] [blame] | 983 | .size aesni_gcm_encrypt,.-aesni_gcm_encrypt |
| 984 | ___ |
| 985 | |
| 986 | $code.=<<___; |
| 987 | .align 64 |
| 988 | .Lbswap_mask: |
| 989 | .byte 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 |
| 990 | .Lpoly: |
| 991 | .byte 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0xc2 |
| 992 | .Lone_msb: |
| 993 | .byte 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1 |
| 994 | .Ltwo_lsb: |
| 995 | .byte 2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 |
| 996 | .Lone_lsb: |
| 997 | .byte 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 |
| 998 | .asciz "AES-NI GCM module for x86_64, CRYPTOGAMS by <appro\@openssl.org>" |
| 999 | .align 64 |
| 1000 | ___ |
| 1001 | if ($win64) { |
| 1002 | $rec="%rcx"; |
| 1003 | $frame="%rdx"; |
| 1004 | $context="%r8"; |
| 1005 | $disp="%r9"; |
| 1006 | |
| 1007 | $code.=<<___ |
| 1008 | .extern __imp_RtlVirtualUnwind |
| 1009 | .type gcm_se_handler,\@abi-omnipotent |
| 1010 | .align 16 |
| 1011 | gcm_se_handler: |
| 1012 | push %rsi |
| 1013 | push %rdi |
| 1014 | push %rbx |
| 1015 | push %rbp |
| 1016 | push %r12 |
| 1017 | push %r13 |
| 1018 | push %r14 |
| 1019 | push %r15 |
| 1020 | pushfq |
| 1021 | sub \$64,%rsp |
| 1022 | |
| 1023 | mov 120($context),%rax # pull context->Rax |
| 1024 | mov 248($context),%rbx # pull context->Rip |
| 1025 | |
| 1026 | mov 8($disp),%rsi # disp->ImageBase |
| 1027 | mov 56($disp),%r11 # disp->HandlerData |
| 1028 | |
| 1029 | mov 0(%r11),%r10d # HandlerData[0] |
| 1030 | lea (%rsi,%r10),%r10 # prologue label |
| 1031 | cmp %r10,%rbx # context->Rip<prologue label |
| 1032 | jb .Lcommon_seh_tail |
| 1033 | |
| 1034 | mov 152($context),%rax # pull context->Rsp |
| 1035 | |
| 1036 | mov 4(%r11),%r10d # HandlerData[1] |
| 1037 | lea (%rsi,%r10),%r10 # epilogue label |
| 1038 | cmp %r10,%rbx # context->Rip>=epilogue label |
| 1039 | jae .Lcommon_seh_tail |
| 1040 | |
| 1041 | mov 120($context),%rax # pull context->Rax |
| 1042 | |
| 1043 | mov -48(%rax),%r15 |
| 1044 | mov -40(%rax),%r14 |
| 1045 | mov -32(%rax),%r13 |
| 1046 | mov -24(%rax),%r12 |
| 1047 | mov -16(%rax),%rbp |
| 1048 | mov -8(%rax),%rbx |
| 1049 | mov %r15,240($context) |
| 1050 | mov %r14,232($context) |
| 1051 | mov %r13,224($context) |
| 1052 | mov %r12,216($context) |
| 1053 | mov %rbp,160($context) |
| 1054 | mov %rbx,144($context) |
| 1055 | |
| 1056 | lea -0xd8(%rax),%rsi # %xmm save area |
| 1057 | lea 512($context),%rdi # & context.Xmm6 |
| 1058 | mov \$20,%ecx # 10*sizeof(%xmm0)/sizeof(%rax) |
| 1059 | .long 0xa548f3fc # cld; rep movsq |
| 1060 | |
| 1061 | .Lcommon_seh_tail: |
| 1062 | mov 8(%rax),%rdi |
| 1063 | mov 16(%rax),%rsi |
| 1064 | mov %rax,152($context) # restore context->Rsp |
| 1065 | mov %rsi,168($context) # restore context->Rsi |
| 1066 | mov %rdi,176($context) # restore context->Rdi |
| 1067 | |
| 1068 | mov 40($disp),%rdi # disp->ContextRecord |
| 1069 | mov $context,%rsi # context |
| 1070 | mov \$154,%ecx # sizeof(CONTEXT) |
| 1071 | .long 0xa548f3fc # cld; rep movsq |
| 1072 | |
| 1073 | mov $disp,%rsi |
| 1074 | xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER |
| 1075 | mov 8(%rsi),%rdx # arg2, disp->ImageBase |
| 1076 | mov 0(%rsi),%r8 # arg3, disp->ControlPc |
| 1077 | mov 16(%rsi),%r9 # arg4, disp->FunctionEntry |
| 1078 | mov 40(%rsi),%r10 # disp->ContextRecord |
| 1079 | lea 56(%rsi),%r11 # &disp->HandlerData |
| 1080 | lea 24(%rsi),%r12 # &disp->EstablisherFrame |
| 1081 | mov %r10,32(%rsp) # arg5 |
| 1082 | mov %r11,40(%rsp) # arg6 |
| 1083 | mov %r12,48(%rsp) # arg7 |
| 1084 | mov %rcx,56(%rsp) # arg8, (NULL) |
| 1085 | call *__imp_RtlVirtualUnwind(%rip) |
| 1086 | |
| 1087 | mov \$1,%eax # ExceptionContinueSearch |
| 1088 | add \$64,%rsp |
| 1089 | popfq |
| 1090 | pop %r15 |
| 1091 | pop %r14 |
| 1092 | pop %r13 |
| 1093 | pop %r12 |
| 1094 | pop %rbp |
| 1095 | pop %rbx |
| 1096 | pop %rdi |
| 1097 | pop %rsi |
| 1098 | ret |
| 1099 | .size gcm_se_handler,.-gcm_se_handler |
| 1100 | |
| 1101 | .section .pdata |
| 1102 | .align 4 |
| 1103 | .rva .LSEH_begin_aesni_gcm_decrypt |
| 1104 | .rva .LSEH_end_aesni_gcm_decrypt |
| 1105 | .rva .LSEH_gcm_dec_info |
| 1106 | |
| 1107 | .rva .LSEH_begin_aesni_gcm_encrypt |
| 1108 | .rva .LSEH_end_aesni_gcm_encrypt |
| 1109 | .rva .LSEH_gcm_enc_info |
| 1110 | .section .xdata |
| 1111 | .align 8 |
| 1112 | .LSEH_gcm_dec_info: |
| 1113 | .byte 9,0,0,0 |
| 1114 | .rva gcm_se_handler |
| 1115 | .rva .Lgcm_dec_body,.Lgcm_dec_abort |
| 1116 | .LSEH_gcm_enc_info: |
| 1117 | .byte 9,0,0,0 |
| 1118 | .rva gcm_se_handler |
| 1119 | .rva .Lgcm_enc_body,.Lgcm_enc_abort |
| 1120 | ___ |
| 1121 | } |
| 1122 | }}} else {{{ |
| 1123 | $code=<<___; # assembler is too old |
| 1124 | .text |
| 1125 | |
| 1126 | .globl aesni_gcm_encrypt |
| 1127 | .type aesni_gcm_encrypt,\@abi-omnipotent |
| 1128 | aesni_gcm_encrypt: |
| 1129 | xor %eax,%eax |
| 1130 | ret |
| 1131 | .size aesni_gcm_encrypt,.-aesni_gcm_encrypt |
| 1132 | |
| 1133 | .globl aesni_gcm_decrypt |
| 1134 | .type aesni_gcm_decrypt,\@abi-omnipotent |
| 1135 | aesni_gcm_decrypt: |
| 1136 | xor %eax,%eax |
| 1137 | ret |
| 1138 | .size aesni_gcm_decrypt,.-aesni_gcm_decrypt |
| 1139 | ___ |
| 1140 | }}} |
| 1141 | |
| 1142 | $code =~ s/\`([^\`]*)\`/eval($1)/gem; |
| 1143 | |
| 1144 | print $code; |
| 1145 | |
Srinivas Paladugu | dd42a61 | 2019-08-09 19:30:39 +0000 | [diff] [blame^] | 1146 | close STDOUT; |