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         mov     %rbx,%r8                # save %rbx
  67
  68         xor     %eax,%eax
  69         mov     %eax,8(%rdi)            # clear 3rd word
  70         cpuid
  71         mov     %eax,%r11d              # max value for standard query level
  72
  73         xor     %eax,%eax
  74         cmp     \$0x756e6547,%ebx       # "Genu"
  75         setne   %al
  76         mov     %eax,%r9d
  77         cmp     \$0x49656e69,%edx       # "ineI"
  78         setne   %al
  79         or      %eax,%r9d
  80         cmp     \$0x6c65746e,%ecx       # "ntel"
  81         setne   %al
  82         or      %eax,%r9d               # 0 indicates Intel CPU
  83         jz      .Lintel
  84
  85         cmp     \$0x68747541,%ebx       # "Auth"
  86         setne   %al
  87         mov     %eax,%r10d
  88         cmp     \$0x69746E65,%edx       # "enti"
  89         setne   %al
  90         or      %eax,%r10d
  91         cmp     \$0x444D4163,%ecx       # "cAMD"
  92         setne   %al
  93         or      %eax,%r10d              # 0 indicates AMD CPU
  94         jnz     .Lintel
  95
  96         # AMD specific
  97         mov     \$0x80000000,%eax
  98         cpuid
  99         cmp     \$0x80000001,%eax
 100         jb      .Lintel
 101         mov     %eax,%r10d
 102         mov     \$0x80000001,%eax
 103         cpuid
 104         or      %ecx,%r9d
 105         and     \$0x00000801,%r9d       # isolate AMD XOP bit, 1<<11
 106
 107         cmp     \$0x80000008,%r10d
 108         jb      .Lintel
 109
 110         mov     \$0x80000008,%eax
 111         cpuid
 112         movzb   %cl,%r10                # number of cores - 1
 113         inc     %r10                    # number of cores
 114
 115         mov     \$1,%eax
 116         cpuid
 117         bt      \$28,%edx               # test hyper-threading bit
 118         jnc     .Lgeneric
 119         shr     \$16,%ebx               # number of logical processors
 120         cmp     %r10b,%bl
 121         ja      .Lgeneric
 122         and     \$0xefffffff,%edx       # ~(1<<28)
 123         jmp     .Lgeneric
 124
 125 .Lintel:
 126         cmp     \$4,%r11d
 127         mov     \$-1,%r10d
 128         jb      .Lnocacheinfo
 129
 130         mov     \$4,%eax
 131         mov     \$0,%ecx                # query L1D
 132         cpuid
 133         mov     %eax,%r10d
 134         shr     \$14,%r10d
 135         and     \$0xfff,%r10d           # number of cores -1 per L1D
 136
 137         cmp     \$7,%r11d
 138         jb      .Lnocacheinfo
 139
 140         mov     \$7,%eax
 141         xor     %ecx,%ecx
 142         cpuid
 143         mov     %ebx,8(%rdi)
 144
 145 .Lnocacheinfo:
 146         mov     \$1,%eax
 147         cpuid
 148         and     \$0xbfefffff,%edx       # force reserved bits to 0
 149         cmp     \$0,%r9d
 150         jne     .Lnotintel
 151         or      \$0x40000000,%edx       # set reserved bit#30 on Intel CPUs
 152         and     \$15,%ah
 153         cmp     \$15,%ah                # examine Family ID
 154         jne     .Lnotintel
 155         or      \$0x00100000,%edx       # set reserved bit#20 to engage RC4_CHAR
 156 .Lnotintel:
 157         bt      \$28,%edx               # test hyper-threading bit
 158         jnc     .Lgeneric
 159         and     \$0xefffffff,%edx       # ~(1<<28)
 160         cmp     \$0,%r10d
 161         je      .Lgeneric
 162
 163         or      \$0x10000000,%edx       # 1<<28
 164         shr     \$16,%ebx
 165         cmp     \$1,%bl                 # see if cache is shared
 166         ja      .Lgeneric
 167         and     \$0xefffffff,%edx       # ~(1<<28)
 168 .Lgeneric:
 169         and     \$0x00000800,%r9d       # isolate AMD XOP flag
 170         and     \$0xfffff7ff,%ecx
 171         or      %ecx,%r9d               # merge AMD XOP flag
 172
 173         mov     %edx,%r10d              # %r9d:%r10d is copy of %ecx:%edx
 174         bt      \$27,%r9d               # check OSXSAVE bit
 175         jnc     .Lclear_avx
 176         xor     %ecx,%ecx               # XCR0
 177         .byte   0x0f,0x01,0xd0          # xgetbv
 178         and     \$6,%eax                # isolate XMM and YMM state support
 179         cmp     \$6,%eax
 180         je      .Ldone
 181 .Lclear_avx:
 182         mov     \$0xefffe7ff,%eax       # ~(1<<28|1<<12|1<<11)
 183         and     %eax,%r9d               # clear AVX, FMA and AMD XOP bits
 184         andl    \$0xffffffdf,8(%rdi)    # cleax AVX2, ~(1<<5)
 185 .Ldone:
 186         shl     \$32,%r9
 187         mov     %r10d,%eax
 188         mov     %r8,%rbx                # restore %rbx
 189         or      %r9,%rax
 190         ret
 191 .size   OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
 192
 193 .globl  OPENSSL_cleanse
 194 .type   OPENSSL_cleanse,\@abi-omnipotent
 195 .align  16
 196 OPENSSL_cleanse:
 197         xor     %rax,%rax
 198         cmp     \$15,$arg2
 199         jae     .Lot
 200         cmp     \$0,$arg2
 201         je      .Lret
 202 .Little:
 203         mov     %al,($arg1)
 204         sub     \$1,$arg2
 205         lea     1($arg1),$arg1
 206         jnz     .Little
 207 .Lret:
 208         ret
 209 .align  16
 210 .Lot:
 211         test    \$7,$arg1
 212         jz      .Laligned
 213         mov     %al,($arg1)
 214         lea     -1($arg2),$arg2
 215         lea     1($arg1),$arg1
 216         jmp     .Lot
 217 .Laligned:
 218         mov     %rax,($arg1)
 219         lea     -8($arg2),$arg2
 220         test    \$-8,$arg2
 221         lea     8($arg1),$arg1
 222         jnz     .Laligned
 223         cmp     \$0,$arg2
 224         jne     .Little
 225         ret
 226 .size   OPENSSL_cleanse,.-OPENSSL_cleanse
 227
 228 .globl  CRYPTO_memcmp
 229 .type   CRYPTO_memcmp,\@abi-omnipotent
 230 .align  16
 231 CRYPTO_memcmp:
 232         xor     %rax,%rax
 233         xor     %r10,%r10
 234         cmp     \$0,$arg3
 235         je      .Lno_data
 236 .Loop_cmp:
 237         mov     ($arg1),%r10b
 238         lea     1($arg1),$arg1
 239         xor     ($arg2),%r10b
 240         lea     1($arg2),$arg2
 241         or      %r10b,%al
 242         dec     $arg3
 243         jnz     .Loop_cmp
 244         neg     %rax
 245         shr     \$63,%rax
 246 .Lno_data:
 247         ret
 248 .size   CRYPTO_memcmp,.-CRYPTO_memcmp
 249 ___
 250
 251 print<<___ if (!$win64);
 252 .globl  OPENSSL_wipe_cpu
 253 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 254 .align  16
 255 OPENSSL_wipe_cpu:
 256         pxor    %xmm0,%xmm0
 257         pxor    %xmm1,%xmm1
 258         pxor    %xmm2,%xmm2
 259         pxor    %xmm3,%xmm3
 260         pxor    %xmm4,%xmm4
 261         pxor    %xmm5,%xmm5
 262         pxor    %xmm6,%xmm6
 263         pxor    %xmm7,%xmm7
 264         pxor    %xmm8,%xmm8
 265         pxor    %xmm9,%xmm9
 266         pxor    %xmm10,%xmm10
 267         pxor    %xmm11,%xmm11
 268         pxor    %xmm12,%xmm12
 269         pxor    %xmm13,%xmm13
 270         pxor    %xmm14,%xmm14
 271         pxor    %xmm15,%xmm15
 272         xorq    %rcx,%rcx
 273         xorq    %rdx,%rdx
 274         xorq    %rsi,%rsi
 275         xorq    %rdi,%rdi
 276         xorq    %r8,%r8
 277         xorq    %r9,%r9
 278         xorq    %r10,%r10
 279         xorq    %r11,%r11
 280         leaq    8(%rsp),%rax
 281         ret
 282 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 283 ___
 284 print<<___ if ($win64);
 285 .globl  OPENSSL_wipe_cpu
 286 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 287 .align  16
 288 OPENSSL_wipe_cpu:
 289         pxor    %xmm0,%xmm0
 290         pxor    %xmm1,%xmm1
 291         pxor    %xmm2,%xmm2
 292         pxor    %xmm3,%xmm3
 293         pxor    %xmm4,%xmm4
 294         pxor    %xmm5,%xmm5
 295         xorq    %rcx,%rcx
 296         xorq    %rdx,%rdx
 297         xorq    %r8,%r8
 298         xorq    %r9,%r9
 299         xorq    %r10,%r10
 300         xorq    %r11,%r11
 301         leaq    8(%rsp),%rax
 302         ret
 303 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 304 ___
 305 {
 306 my $out="%r10";
 307 my $cnt="%rcx";
 308 my $max="%r11";
 309 my $lasttick="%r8d";
 310 my $lastdiff="%r9d";
 311 my $redzone=win64?8:-8;
 312
 313 print<<___;
 314 .globl  OPENSSL_instrument_bus
 315 .type   OPENSSL_instrument_bus,\@abi-omnipotent
 316 .align  16
 317 OPENSSL_instrument_bus:
 318         mov     $arg1,$out      # tribute to Win64
 319         mov     $arg2,$cnt
 320         mov     $arg2,$max
 321
 322         rdtsc                   # collect 1st tick
 323         mov     %eax,$lasttick  # lasttick = tick
 324         mov     \$0,$lastdiff   # lastdiff = 0
 325         clflush ($out)
 326         .byte   0xf0            # lock
 327         add     $lastdiff,($out)
 328         jmp     .Loop
 329 .align  16
 330 .Loop:  rdtsc
 331         mov     %eax,%edx
 332         sub     $lasttick,%eax
 333         mov     %edx,$lasttick
 334         mov     %eax,$lastdiff
 335         clflush ($out)
 336         .byte   0xf0            # lock
 337         add     %eax,($out)
 338         lea     4($out),$out
 339         sub     \$1,$cnt
 340         jnz     .Loop
 341
 342         mov     $max,%rax
 343         ret
 344 .size   OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
 345
 346 .globl  OPENSSL_instrument_bus2
 347 .type   OPENSSL_instrument_bus2,\@abi-omnipotent
 348 .align  16
 349 OPENSSL_instrument_bus2:
 350         mov     $arg1,$out      # tribute to Win64
 351         mov     $arg2,$cnt
 352         mov     $arg3,$max
 353         mov     $cnt,$redzone(%rsp)
 354
 355         rdtsc                   # collect 1st tick
 356         mov     %eax,$lasttick  # lasttick = tick
 357         mov     \$0,$lastdiff   # lastdiff = 0
 358
 359         clflush ($out)
 360         .byte   0xf0            # lock
 361         add     $lastdiff,($out)
 362
 363         rdtsc                   # collect 1st diff
 364         mov     %eax,%edx
 365         sub     $lasttick,%eax  # diff
 366         mov     %edx,$lasttick  # lasttick = tick
 367         mov     %eax,$lastdiff  # lastdiff = diff
 368 .Loop2:
 369         clflush ($out)
 370         .byte   0xf0            # lock
 371         add     %eax,($out)     # accumulate diff
 372
 373         sub     \$1,$max
 374         jz      .Ldone2
 375
 376         rdtsc
 377         mov     %eax,%edx
 378         sub     $lasttick,%eax  # diff
 379         mov     %edx,$lasttick  # lasttick = tick
 380         cmp     $lastdiff,%eax
 381         mov     %eax,$lastdiff  # lastdiff = diff
 382         mov     \$0,%edx
 383         setne   %dl
 384         sub     %rdx,$cnt       # conditional --$cnt
 385         lea     ($out,%rdx,4),$out      # conditional ++$out
 386         jnz     .Loop2
 387
 388 .Ldone2:
 389         mov     $redzone(%rsp),%rax
 390         sub     $cnt,%rax
 391         ret
 392 .size   OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
 393 ___
 394 }
 395
 396 sub gen_random {
 397 my $rdop = shift;
 398 print<<___;
 399 .globl  OPENSSL_ia32_${rdop}
 400 .type   OPENSSL_ia32_${rdop},\@abi-omnipotent
 401 .align  16
 402 OPENSSL_ia32_${rdop}:
 403         mov     \$8,%ecx
 404 .Loop_${rdop}:
 405         ${rdop} %rax
 406         jc      .Lbreak_${rdop}
 407         loop    .Loop_${rdop}
 408 .Lbreak_${rdop}:
 409         cmp     \$0,%rax
 410         cmove   %rcx,%rax
 411         ret
 412 .size   OPENSSL_ia32_${rdop},.-OPENSSL_ia32_${rdop}
 413
 414 .globl  OPENSSL_ia32_${rdop}_bytes
 415 .type   OPENSSL_ia32_${rdop}_bytes,\@abi-omnipotent
 416 .align  16
 417 OPENSSL_ia32_${rdop}_bytes:
 418         xor     %rax, %rax      # return value
 419         cmp     \$0,$arg2
 420         je      .Ldone_${rdop}_bytes
 421
 422         mov     \$8,%r11
 423 .Loop_${rdop}_bytes:
 424         ${rdop} %r10
 425         jc      .Lbreak_${rdop}_bytes
 426         dec     %r11
 427         jnz     .Loop_${rdop}_bytes
 428         jmp     .Ldone_${rdop}_bytes
 429
 430 .align  16
 431 .Lbreak_${rdop}_bytes:
 432         cmp     \$8,$arg2
 433         jb      .Ltail_${rdop}_bytes
 434         mov     %r10,($arg1)
 435         lea     8($arg1),$arg1
 436         add     \$8,%rax
 437         sub     \$8,$arg2
 438         jz      .Ldone_${rdop}_bytes
 439         mov     \$8,%r11
 440         jmp     .Loop_${rdop}_bytes
 441
 442 .align  16
 443 .Ltail_${rdop}_bytes:
 444         mov     %r10b,($arg1)
 445         lea     1($arg1),$arg1
 446         inc     %rax
 447         shr     \$8,%r8
 448         dec     $arg2
 449         jnz     .Ltail_${rdop}_bytes
 450
 451 .Ldone_${rdop}_bytes:
 452         ret
 453 .size   OPENSSL_ia32_${rdop}_bytes,.-OPENSSL_ia32_${rdop}_bytes
 454 ___
 455 }
 456 gen_random("rdrand");
 457 gen_random("rdseed");
 458
 459 close STDOUT;   # flush