crypto/x86_64cpuid.pl

   1 #! /usr/bin/env perl
   2 # Copyright 2005-2016 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
   9
  10 $flavour = shift;
  11 $output  = shift;
  12 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
  13
  14 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
  15
  16 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  17 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
  18 ( $xlate="${dir}perlasm/x86_64-xlate.pl" and -f $xlate) or
  19 die "can't locate x86_64-xlate.pl";
  20
  21 open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
  22 *STDOUT=*OUT;
  23
  24 ($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") : # Win64 order
  25                                  ("%rdi","%rsi","%rdx","%rcx"); # Unix order
  26
  27 print<<___;
  28 .extern         OPENSSL_cpuid_setup
  29 .hidden         OPENSSL_cpuid_setup
  30 .section        .init
  31         call    OPENSSL_cpuid_setup
  32
  33 .hidden OPENSSL_ia32cap_P
  34 .comm   OPENSSL_ia32cap_P,16,4
  35
  36 .text
  37
  38 .globl  OPENSSL_atomic_add
  39 .type   OPENSSL_atomic_add,\@abi-omnipotent
  40 .align  16
  41 OPENSSL_atomic_add:
  42         movl    ($arg1),%eax
  43 .Lspin: leaq    ($arg2,%rax),%r8
  44         .byte   0xf0            # lock
  45         cmpxchgl        %r8d,($arg1)
  46         jne     .Lspin
  47         movl    %r8d,%eax
  48         .byte   0x48,0x98       # cltq/cdqe
  49         ret
  50 .size   OPENSSL_atomic_add,.-OPENSSL_atomic_add
  51
  52 .globl  OPENSSL_rdtsc
  53 .type   OPENSSL_rdtsc,\@abi-omnipotent
  54 .align  16
  55 OPENSSL_rdtsc:
  56         rdtsc
  57         shl     \$32,%rdx
  58         or      %rdx,%rax
  59         ret
  60 .size   OPENSSL_rdtsc,.-OPENSSL_rdtsc
  61
  62 .globl  OPENSSL_ia32_cpuid
  63 .type   OPENSSL_ia32_cpuid,\@function,1
  64 .align  16
  65 OPENSSL_ia32_cpuid:
  66 .cfi_startproc
  67         mov     %rbx,%r8                # save %rbx
  68 .cfi_register   %rbx,%r8
  69
  70         xor     %eax,%eax
  71         mov     %rax,8(%rdi)            # clear extended feature flags
  72         cpuid
  73         mov     %eax,%r11d              # max value for standard query level
  74
  75         xor     %eax,%eax
  76         cmp     \$0x756e6547,%ebx       # "Genu"
  77         setne   %al
  78         mov     %eax,%r9d
  79         cmp     \$0x49656e69,%edx       # "ineI"
  80         setne   %al
  81         or      %eax,%r9d
  82         cmp     \$0x6c65746e,%ecx       # "ntel"
  83         setne   %al
  84         or      %eax,%r9d               # 0 indicates Intel CPU
  85         jz      .Lintel
  86
  87         cmp     \$0x68747541,%ebx       # "Auth"
  88         setne   %al
  89         mov     %eax,%r10d
  90         cmp     \$0x69746E65,%edx       # "enti"
  91         setne   %al
  92         or      %eax,%r10d
  93         cmp     \$0x444D4163,%ecx       # "cAMD"
  94         setne   %al
  95         or      %eax,%r10d              # 0 indicates AMD CPU
  96         jnz     .Lintel
  97
  98         # AMD specific
  99         mov     \$0x80000000,%eax
 100         cpuid
 101         cmp     \$0x80000001,%eax
 102         jb      .Lintel
 103         mov     %eax,%r10d
 104         mov     \$0x80000001,%eax
 105         cpuid
 106         or      %ecx,%r9d
 107         and     \$0x00000801,%r9d       # isolate AMD XOP bit, 1<<11
 108
 109         cmp     \$0x80000008,%r10d
 110         jb      .Lintel
 111
 112         mov     \$0x80000008,%eax
 113         cpuid
 114         movzb   %cl,%r10                # number of cores - 1
 115         inc     %r10                    # number of cores
 116
 117         mov     \$1,%eax
 118         cpuid
 119         bt      \$28,%edx               # test hyper-threading bit
 120         jnc     .Lgeneric
 121         shr     \$16,%ebx               # number of logical processors
 122         cmp     %r10b,%bl
 123         ja      .Lgeneric
 124         and     \$0xefffffff,%edx       # ~(1<<28)
 125         jmp     .Lgeneric
 126
 127 .Lintel:
 128         cmp     \$4,%r11d
 129         mov     \$-1,%r10d
 130         jb      .Lnocacheinfo
 131
 132         mov     \$4,%eax
 133         mov     \$0,%ecx                # query L1D
 134         cpuid
 135         mov     %eax,%r10d
 136         shr     \$14,%r10d
 137         and     \$0xfff,%r10d           # number of cores -1 per L1D
 138
 139 .Lnocacheinfo:
 140         mov     \$1,%eax
 141         cpuid
 142         movd    %eax,%xmm0              # put aside processor id
 143         and     \$0xbfefffff,%edx       # force reserved bits to 0
 144         cmp     \$0,%r9d
 145         jne     .Lnotintel
 146         or      \$0x40000000,%edx       # set reserved bit#30 on Intel CPUs
 147         and     \$15,%ah
 148         cmp     \$15,%ah                # examine Family ID
 149         jne     .LnotP4
 150         or      \$0x00100000,%edx       # set reserved bit#20 to engage RC4_CHAR
 151 .LnotP4:
 152         cmp     \$6,%ah
 153         jne     .Lnotintel
 154         and     \$0x0fff0ff0,%eax
 155         cmp     \$0x00050670,%eax       # Knights Landing
 156         je      .Lknights
 157         cmp     \$0x00080650,%eax       # Knights Mill (according to sde)
 158         jne     .Lnotintel
 159 .Lknights:
 160         and     \$0xfbffffff,%ecx       # clear XSAVE flag to mimic Silvermont
 161
 162 .Lnotintel:
 163         bt      \$28,%edx               # test hyper-threading bit
 164         jnc     .Lgeneric
 165         and     \$0xefffffff,%edx       # ~(1<<28)
 166         cmp     \$0,%r10d
 167         je      .Lgeneric
 168
 169         or      \$0x10000000,%edx       # 1<<28
 170         shr     \$16,%ebx
 171         cmp     \$1,%bl                 # see if cache is shared
 172         ja      .Lgeneric
 173         and     \$0xefffffff,%edx       # ~(1<<28)
 174 .Lgeneric:
 175         and     \$0x00000800,%r9d       # isolate AMD XOP flag
 176         and     \$0xfffff7ff,%ecx
 177         or      %ecx,%r9d               # merge AMD XOP flag
 178
 179         mov     %edx,%r10d              # %r9d:%r10d is copy of %ecx:%edx
 180
 181         cmp     \$7,%r11d
 182         jb      .Lno_extended_info
 183         mov     \$7,%eax
 184         xor     %ecx,%ecx
 185         cpuid
 186         bt      \$26,%r9d               # check XSAVE bit, cleared on Knights
 187         jc      .Lnotknights
 188         and     \$0xfff7ffff,%ebx       # clear ADCX/ADOX flag
 189 .Lnotknights:
 190         movd    %xmm0,%eax              # restore processor id
 191         and     \$0x0fff0ff0,%eax
 192         cmp     \$0x00050650,%eax       # Skylake-X
 193         jne     .Lnotskylakex
 194         and     \$0xfffeffff,%ebx       # ~(1<<16)
 195                                         # suppress AVX512F flag on Skylake-X
 196 .Lnotskylakex:
 197         mov     %ebx,8(%rdi)            # save extended feature flags
 198         mov     %ecx,12(%rdi)
 199 .Lno_extended_info:
 200
 201         bt      \$27,%r9d               # check OSXSAVE bit
 202         jnc     .Lclear_avx
 203         xor     %ecx,%ecx               # XCR0
 204         .byte   0x0f,0x01,0xd0          # xgetbv
 205         and     \$0xe6,%eax             # isolate XMM, YMM and ZMM state support
 206         cmp     \$0xe6,%eax
 207         je      .Ldone
 208         andl    \$0x3fdeffff,8(%rdi)    # ~(1<<31|1<<30|1<<21|1<<16)
 209                                         # clear AVX512F+BW+VL+FIMA, all of
 210                                         # them are EVEX-encoded, which requires
 211                                         # ZMM state support even if one uses
 212                                         # only XMM and YMM :-(
 213         and     \$6,%eax                # isolate XMM and YMM state support
 214         cmp     \$6,%eax
 215         je      .Ldone
 216 .Lclear_avx:
 217         mov     \$0xefffe7ff,%eax       # ~(1<<28|1<<12|1<<11)
 218         and     %eax,%r9d               # clear AVX, FMA and AMD XOP bits
 219         mov     \$0x3fdeffdf,%eax       # ~(1<<31|1<<30|1<<21|1<<16|1<<5)
 220         and     %eax,8(%rdi)            # clear AVX2 and AVX512* bits
 221 .Ldone:
 222         shl     \$32,%r9
 223         mov     %r10d,%eax
 224         mov     %r8,%rbx                # restore %rbx
 225 .cfi_restore    %rbx
 226         or      %r9,%rax
 227         ret
 228 .cfi_endproc
 229 .size   OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
 230
 231 .globl  OPENSSL_cleanse
 232 .type   OPENSSL_cleanse,\@abi-omnipotent
 233 .align  16
 234 OPENSSL_cleanse:
 235         xor     %rax,%rax
 236         cmp     \$15,$arg2
 237         jae     .Lot
 238         cmp     \$0,$arg2
 239         je      .Lret
 240 .Little:
 241         mov     %al,($arg1)
 242         sub     \$1,$arg2
 243         lea     1($arg1),$arg1
 244         jnz     .Little
 245 .Lret:
 246         ret
 247 .align  16
 248 .Lot:
 249         test    \$7,$arg1
 250         jz      .Laligned
 251         mov     %al,($arg1)
 252         lea     -1($arg2),$arg2
 253         lea     1($arg1),$arg1
 254         jmp     .Lot
 255 .Laligned:
 256         mov     %rax,($arg1)
 257         lea     -8($arg2),$arg2
 258         test    \$-8,$arg2
 259         lea     8($arg1),$arg1
 260         jnz     .Laligned
 261         cmp     \$0,$arg2
 262         jne     .Little
 263         ret
 264 .size   OPENSSL_cleanse,.-OPENSSL_cleanse
 265
 266 .globl  CRYPTO_memcmp
 267 .type   CRYPTO_memcmp,\@abi-omnipotent
 268 .align  16
 269 CRYPTO_memcmp:
 270         xor     %rax,%rax
 271         xor     %r10,%r10
 272         cmp     \$0,$arg3
 273         je      .Lno_data
 274 .Loop_cmp:
 275         mov     ($arg1),%r10b
 276         lea     1($arg1),$arg1
 277         xor     ($arg2),%r10b
 278         lea     1($arg2),$arg2
 279         or      %r10b,%al
 280         dec     $arg3
 281         jnz     .Loop_cmp
 282         neg     %rax
 283         shr     \$63,%rax
 284 .Lno_data:
 285         ret
 286 .size   CRYPTO_memcmp,.-CRYPTO_memcmp
 287 ___
 288
 289 print<<___ if (!$win64);
 290 .globl  OPENSSL_wipe_cpu
 291 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 292 .align  16
 293 OPENSSL_wipe_cpu:
 294         pxor    %xmm0,%xmm0
 295         pxor    %xmm1,%xmm1
 296         pxor    %xmm2,%xmm2
 297         pxor    %xmm3,%xmm3
 298         pxor    %xmm4,%xmm4
 299         pxor    %xmm5,%xmm5
 300         pxor    %xmm6,%xmm6
 301         pxor    %xmm7,%xmm7
 302         pxor    %xmm8,%xmm8
 303         pxor    %xmm9,%xmm9
 304         pxor    %xmm10,%xmm10
 305         pxor    %xmm11,%xmm11
 306         pxor    %xmm12,%xmm12
 307         pxor    %xmm13,%xmm13
 308         pxor    %xmm14,%xmm14
 309         pxor    %xmm15,%xmm15
 310         xorq    %rcx,%rcx
 311         xorq    %rdx,%rdx
 312         xorq    %rsi,%rsi
 313         xorq    %rdi,%rdi
 314         xorq    %r8,%r8
 315         xorq    %r9,%r9
 316         xorq    %r10,%r10
 317         xorq    %r11,%r11
 318         leaq    8(%rsp),%rax
 319         ret
 320 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 321 ___
 322 print<<___ if ($win64);
 323 .globl  OPENSSL_wipe_cpu
 324 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 325 .align  16
 326 OPENSSL_wipe_cpu:
 327         pxor    %xmm0,%xmm0
 328         pxor    %xmm1,%xmm1
 329         pxor    %xmm2,%xmm2
 330         pxor    %xmm3,%xmm3
 331         pxor    %xmm4,%xmm4
 332         pxor    %xmm5,%xmm5
 333         xorq    %rcx,%rcx
 334         xorq    %rdx,%rdx
 335         xorq    %r8,%r8
 336         xorq    %r9,%r9
 337         xorq    %r10,%r10
 338         xorq    %r11,%r11
 339         leaq    8(%rsp),%rax
 340         ret
 341 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 342 ___
 343 {
 344 my $out="%r10";
 345 my $cnt="%rcx";
 346 my $max="%r11";
 347 my $lasttick="%r8d";
 348 my $lastdiff="%r9d";
 349 my $redzone=win64?8:-8;
 350
 351 print<<___;
 352 .globl  OPENSSL_instrument_bus
 353 .type   OPENSSL_instrument_bus,\@abi-omnipotent
 354 .align  16
 355 OPENSSL_instrument_bus:
 356         mov     $arg1,$out      # tribute to Win64
 357         mov     $arg2,$cnt
 358         mov     $arg2,$max
 359
 360         rdtsc                   # collect 1st tick
 361         mov     %eax,$lasttick  # lasttick = tick
 362         mov     \$0,$lastdiff   # lastdiff = 0
 363         clflush ($out)
 364         .byte   0xf0            # lock
 365         add     $lastdiff,($out)
 366         jmp     .Loop
 367 .align  16
 368 .Loop:  rdtsc
 369         mov     %eax,%edx
 370         sub     $lasttick,%eax
 371         mov     %edx,$lasttick
 372         mov     %eax,$lastdiff
 373         clflush ($out)
 374         .byte   0xf0            # lock
 375         add     %eax,($out)
 376         lea     4($out),$out
 377         sub     \$1,$cnt
 378         jnz     .Loop
 379
 380         mov     $max,%rax
 381         ret
 382 .size   OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
 383
 384 .globl  OPENSSL_instrument_bus2
 385 .type   OPENSSL_instrument_bus2,\@abi-omnipotent
 386 .align  16
 387 OPENSSL_instrument_bus2:
 388         mov     $arg1,$out      # tribute to Win64
 389         mov     $arg2,$cnt
 390         mov     $arg3,$max
 391         mov     $cnt,$redzone(%rsp)
 392
 393         rdtsc                   # collect 1st tick
 394         mov     %eax,$lasttick  # lasttick = tick
 395         mov     \$0,$lastdiff   # lastdiff = 0
 396
 397         clflush ($out)
 398         .byte   0xf0            # lock
 399         add     $lastdiff,($out)
 400
 401         rdtsc                   # collect 1st diff
 402         mov     %eax,%edx
 403         sub     $lasttick,%eax  # diff
 404         mov     %edx,$lasttick  # lasttick = tick
 405         mov     %eax,$lastdiff  # lastdiff = diff
 406 .Loop2:
 407         clflush ($out)
 408         .byte   0xf0            # lock
 409         add     %eax,($out)     # accumulate diff
 410
 411         sub     \$1,$max
 412         jz      .Ldone2
 413
 414         rdtsc
 415         mov     %eax,%edx
 416         sub     $lasttick,%eax  # diff
 417         mov     %edx,$lasttick  # lasttick = tick
 418         cmp     $lastdiff,%eax
 419         mov     %eax,$lastdiff  # lastdiff = diff
 420         mov     \$0,%edx
 421         setne   %dl
 422         sub     %rdx,$cnt       # conditional --$cnt
 423         lea     ($out,%rdx,4),$out      # conditional ++$out
 424         jnz     .Loop2
 425
 426 .Ldone2:
 427         mov     $redzone(%rsp),%rax
 428         sub     $cnt,%rax
 429         ret
 430 .size   OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
 431 ___
 432 }
 433
 434 sub gen_random {
 435 my $rdop = shift;
 436 print<<___;
 437 .globl  OPENSSL_ia32_${rdop}
 438 .type   OPENSSL_ia32_${rdop},\@abi-omnipotent
 439 .align  16
 440 OPENSSL_ia32_${rdop}:
 441         mov     \$8,%ecx
 442 .Loop_${rdop}:
 443         ${rdop} %rax
 444         jc      .Lbreak_${rdop}
 445         loop    .Loop_${rdop}
 446 .Lbreak_${rdop}:
 447         cmp     \$0,%rax
 448         cmove   %rcx,%rax
 449         ret
 450 .size   OPENSSL_ia32_${rdop},.-OPENSSL_ia32_${rdop}
 451
 452 .globl  OPENSSL_ia32_${rdop}_bytes
 453 .type   OPENSSL_ia32_${rdop}_bytes,\@abi-omnipotent
 454 .align  16
 455 OPENSSL_ia32_${rdop}_bytes:
 456         xor     %rax, %rax      # return value
 457         cmp     \$0,$arg2
 458         je      .Ldone_${rdop}_bytes
 459
 460         mov     \$8,%r11
 461 .Loop_${rdop}_bytes:
 462         ${rdop} %r10
 463         jc      .Lbreak_${rdop}_bytes
 464         dec     %r11
 465         jnz     .Loop_${rdop}_bytes
 466         jmp     .Ldone_${rdop}_bytes
 467
 468 .align  16
 469 .Lbreak_${rdop}_bytes:
 470         cmp     \$8,$arg2
 471         jb      .Ltail_${rdop}_bytes
 472         mov     %r10,($arg1)
 473         lea     8($arg1),$arg1
 474         add     \$8,%rax
 475         sub     \$8,$arg2
 476         jz      .Ldone_${rdop}_bytes
 477         mov     \$8,%r11
 478         jmp     .Loop_${rdop}_bytes
 479
 480 .align  16
 481 .Ltail_${rdop}_bytes:
 482         mov     %r10b,($arg1)
 483         lea     1($arg1),$arg1
 484         inc     %rax
 485         shr     \$8,%r8
 486         dec     $arg2
 487         jnz     .Ltail_${rdop}_bytes
 488
 489 .Ldone_${rdop}_bytes:
 490         ret
 491 .size   OPENSSL_ia32_${rdop}_bytes,.-OPENSSL_ia32_${rdop}_bytes
 492 ___
 493 }
 494 gen_random("rdrand");
 495 gen_random("rdseed");
 496
 497 close STDOUT;   # flush