crypto/x86_64cpuid.pl

   1 #!/usr/bin/env perl
   2
   3 $flavour = shift;
   4 $output  = shift;
   5 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
   6
   7 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
   8
   9 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  10 open STDOUT,"| $^X ${dir}perlasm/x86_64-xlate.pl $flavour $output";
  11
  12 ($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") : # Win64 order
  13                                  ("%rdi","%rsi","%rdx","%rcx"); # Unix order
  14
  15 print<<___;
  16 .extern         OPENSSL_cpuid_setup
  17 .hidden         OPENSSL_cpuid_setup
  18 .section        .init
  19         call    OPENSSL_cpuid_setup
  20
  21 .hidden OPENSSL_ia32cap_P
  22 .comm   OPENSSL_ia32cap_P,8
  23
  24 .text
  25
  26 .globl  OPENSSL_atomic_add
  27 .type   OPENSSL_atomic_add,\@abi-omnipotent
  28 .align  16
  29 OPENSSL_atomic_add:
  30         movl    ($arg1),%eax
  31 .Lspin: leaq    ($arg2,%rax),%r8
  32         .byte   0xf0            # lock
  33         cmpxchgl        %r8d,($arg1)
  34         jne     .Lspin
  35         movl    %r8d,%eax
  36         .byte   0x48,0x98       # cltq/cdqe
  37         ret
  38 .size   OPENSSL_atomic_add,.-OPENSSL_atomic_add
  39
  40 .globl  OPENSSL_rdtsc
  41 .type   OPENSSL_rdtsc,\@abi-omnipotent
  42 .align  16
  43 OPENSSL_rdtsc:
  44         rdtsc
  45         shl     \$32,%rdx
  46         or      %rdx,%rax
  47         ret
  48 .size   OPENSSL_rdtsc,.-OPENSSL_rdtsc
  49
  50 .globl  OPENSSL_ia32_cpuid
  51 .type   OPENSSL_ia32_cpuid,\@abi-omnipotent
  52 .align  16
  53 OPENSSL_ia32_cpuid:
  54         mov     %rbx,%r8                # save %rbx
  55
  56         xor     %eax,%eax
  57         cpuid
  58         mov     %eax,%r11d              # max value for standard query level
  59
  60         xor     %eax,%eax
  61         cmp     \$0x756e6547,%ebx       # "Genu"
  62         setne   %al
  63         mov     %eax,%r9d
  64         cmp     \$0x49656e69,%edx       # "ineI"
  65         setne   %al
  66         or      %eax,%r9d
  67         cmp     \$0x6c65746e,%ecx       # "ntel"
  68         setne   %al
  69         or      %eax,%r9d               # 0 indicates Intel CPU
  70         jz      .Lintel
  71
  72         cmp     \$0x68747541,%ebx       # "Auth"
  73         setne   %al
  74         mov     %eax,%r10d
  75         cmp     \$0x69746E65,%edx       # "enti"
  76         setne   %al
  77         or      %eax,%r10d
  78         cmp     \$0x444D4163,%ecx       # "cAMD"
  79         setne   %al
  80         or      %eax,%r10d              # 0 indicates AMD CPU
  81         jnz     .Lintel
  82
  83         # AMD specific
  84         mov     \$0x80000000,%eax
  85         cpuid
  86         cmp     \$0x80000001,%eax
  87         jb      .Lintel
  88         mov     %eax,%r10d
  89         mov     \$0x80000001,%eax
  90         cpuid
  91         or      %ecx,%r9d
  92         and     \$0x00000801,%r9d       # isolate AMD XOP bit, 1<<11
  93
  94         cmp     \$0x80000008,%r10d
  95         jb      .Lintel
  96
  97         mov     \$0x80000008,%eax
  98         cpuid
  99         movzb   %cl,%r10                # number of cores - 1
 100         inc     %r10                    # number of cores
 101
 102         mov     \$1,%eax
 103         cpuid
 104         bt      \$28,%edx               # test hyper-threading bit
 105         jnc     .Lgeneric
 106         shr     \$16,%ebx               # number of logical processors
 107         cmp     %r10b,%bl
 108         ja      .Lgeneric
 109         and     \$0xefffffff,%edx       # ~(1<<28)
 110         jmp     .Lgeneric
 111
 112 .Lintel:
 113         cmp     \$4,%r11d
 114         mov     \$-1,%r10d
 115         jb      .Lnocacheinfo
 116
 117         mov     \$4,%eax
 118         mov     \$0,%ecx                # query L1D
 119         cpuid
 120         mov     %eax,%r10d
 121         shr     \$14,%r10d
 122         and     \$0xfff,%r10d           # number of cores -1 per L1D
 123
 124 .Lnocacheinfo:
 125         mov     \$1,%eax
 126         cpuid
 127         cmp     \$0,%r9d
 128         jne     .Lnotintel
 129         or      \$0x00100000,%edx       # use reserved 20th bit to engage RC4_CHAR
 130         and     \$15,%ah
 131         cmp     \$15,%ah                # examine Family ID
 132         je      .Lnotintel
 133         or      \$0x40000000,%edx       # use reserved bit to skip unrolled loop
 134 .Lnotintel:
 135         bt      \$28,%edx               # test hyper-threading bit
 136         jnc     .Lgeneric
 137         and     \$0xefffffff,%edx       # ~(1<<28)
 138         cmp     \$0,%r10d
 139         je      .Lgeneric
 140
 141         or      \$0x10000000,%edx       # 1<<28
 142         shr     \$16,%ebx
 143         cmp     \$1,%bl                 # see if cache is shared
 144         ja      .Lgeneric
 145         and     \$0xefffffff,%edx       # ~(1<<28)
 146 .Lgeneric:
 147         and     \$0x00000800,%r9d       # isolate AMD XOP flag
 148         and     \$0xfffff7ff,%ecx
 149         or      %r9d,%ecx               # merge AMD XOP flag
 150
 151         shl     \$32,%rcx
 152         mov     %edx,%ebx
 153         or      %rcx,%rbx               # compose capability vector in %rbx
 154         bt      \$27+32,%rcx            # check OSXSAVE bit
 155         jnc     .Lclear_avx
 156         xor     %ecx,%ecx               # XCR0
 157         .byte   0x0f,0x01,0xd0          # xgetbv
 158         and     \$6,%eax                # isolate XMM and YMM state support
 159         cmp     \$6,%eax
 160         je      .Ldone
 161 .Lclear_avx:
 162         mov     \$0xefffe7ff,%eax       # ~(1<<28|1<<12|1<<11)
 163         shl     \$32,%rax
 164         and     %rax,%rbx               # clear AVX, FMA and AMD XOP bits
 165 .Ldone:
 166         mov     %rbx,%rax
 167         mov     %r8,%rbx                # restore %rbx
 168         ret
 169 .size   OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
 170
 171 .globl  OPENSSL_cleanse
 172 .type   OPENSSL_cleanse,\@abi-omnipotent
 173 .align  16
 174 OPENSSL_cleanse:
 175         xor     %rax,%rax
 176         cmp     \$15,$arg2
 177         jae     .Lot
 178         cmp     \$0,$arg2
 179         je      .Lret
 180 .Little:
 181         mov     %al,($arg1)
 182         sub     \$1,$arg2
 183         lea     1($arg1),$arg1
 184         jnz     .Little
 185 .Lret:
 186         ret
 187 .align  16
 188 .Lot:
 189         test    \$7,$arg1
 190         jz      .Laligned
 191         mov     %al,($arg1)
 192         lea     -1($arg2),$arg2
 193         lea     1($arg1),$arg1
 194         jmp     .Lot
 195 .Laligned:
 196         mov     %rax,($arg1)
 197         lea     -8($arg2),$arg2
 198         test    \$-8,$arg2
 199         lea     8($arg1),$arg1
 200         jnz     .Laligned
 201         cmp     \$0,$arg2
 202         jne     .Little
 203         ret
 204 .size   OPENSSL_cleanse,.-OPENSSL_cleanse
 205 ___
 206
 207 print<<___ if (!$win64);
 208 .globl  OPENSSL_wipe_cpu
 209 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 210 .align  16
 211 OPENSSL_wipe_cpu:
 212         pxor    %xmm0,%xmm0
 213         pxor    %xmm1,%xmm1
 214         pxor    %xmm2,%xmm2
 215         pxor    %xmm3,%xmm3
 216         pxor    %xmm4,%xmm4
 217         pxor    %xmm5,%xmm5
 218         pxor    %xmm6,%xmm6
 219         pxor    %xmm7,%xmm7
 220         pxor    %xmm8,%xmm8
 221         pxor    %xmm9,%xmm9
 222         pxor    %xmm10,%xmm10
 223         pxor    %xmm11,%xmm11
 224         pxor    %xmm12,%xmm12
 225         pxor    %xmm13,%xmm13
 226         pxor    %xmm14,%xmm14
 227         pxor    %xmm15,%xmm15
 228         xorq    %rcx,%rcx
 229         xorq    %rdx,%rdx
 230         xorq    %rsi,%rsi
 231         xorq    %rdi,%rdi
 232         xorq    %r8,%r8
 233         xorq    %r9,%r9
 234         xorq    %r10,%r10
 235         xorq    %r11,%r11
 236         leaq    8(%rsp),%rax
 237         ret
 238 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 239 ___
 240 print<<___ if ($win64);
 241 .globl  OPENSSL_wipe_cpu
 242 .type   OPENSSL_wipe_cpu,\@abi-omnipotent
 243 .align  16
 244 OPENSSL_wipe_cpu:
 245         pxor    %xmm0,%xmm0
 246         pxor    %xmm1,%xmm1
 247         pxor    %xmm2,%xmm2
 248         pxor    %xmm3,%xmm3
 249         pxor    %xmm4,%xmm4
 250         pxor    %xmm5,%xmm5
 251         xorq    %rcx,%rcx
 252         xorq    %rdx,%rdx
 253         xorq    %r8,%r8
 254         xorq    %r9,%r9
 255         xorq    %r10,%r10
 256         xorq    %r11,%r11
 257         leaq    8(%rsp),%rax
 258         ret
 259 .size   OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 260 ___
 261 {
 262 my $out="%r10";
 263 my $cnt="%rcx";
 264 my $max="%r11";
 265 my $lasttick="%r8d";
 266 my $lastdiff="%r9d";
 267 my $redzone=win64?8:-8;
 268
 269 print<<___;
 270 .globl  OPENSSL_instrument_bus
 271 .type   OPENSSL_instrument_bus,\@abi-omnipotent
 272 .align  16
 273 OPENSSL_instrument_bus:
 274         mov     $arg1,$out      # tribute to Win64
 275         mov     $arg2,$cnt
 276         mov     $arg2,$max
 277
 278         rdtsc                   # collect 1st tick
 279         mov     %eax,$lasttick  # lasttick = tick
 280         mov     \$0,$lastdiff   # lastdiff = 0
 281         clflush ($out)
 282         .byte   0xf0            # lock
 283         add     $lastdiff,($out)
 284         jmp     .Loop
 285 .align  16
 286 .Loop:  rdtsc
 287         mov     %eax,%edx
 288         sub     $lasttick,%eax
 289         mov     %edx,$lasttick
 290         mov     %eax,$lastdiff
 291         clflush ($out)
 292         .byte   0xf0            # lock
 293         add     %eax,($out)
 294         lea     4($out),$out
 295         sub     \$1,$cnt
 296         jnz     .Loop
 297
 298         mov     $max,%rax
 299         ret
 300 .size   OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
 301
 302 .globl  OPENSSL_instrument_bus2
 303 .type   OPENSSL_instrument_bus2,\@abi-omnipotent
 304 .align  16
 305 OPENSSL_instrument_bus2:
 306         mov     $arg1,$out      # tribute to Win64
 307         mov     $arg2,$cnt
 308         mov     $arg3,$max
 309         mov     $cnt,$redzone(%rsp)
 310
 311         rdtsc                   # collect 1st tick
 312         mov     %eax,$lasttick  # lasttick = tick
 313         mov     \$0,$lastdiff   # lastdiff = 0
 314
 315         clflush ($out)
 316         .byte   0xf0            # lock
 317         add     $lastdiff,($out)
 318
 319         rdtsc                   # collect 1st diff
 320         mov     %eax,%edx
 321         sub     $lasttick,%eax  # diff
 322         mov     %edx,$lasttick  # lasttick = tick
 323         mov     %eax,$lastdiff  # lastdiff = diff
 324 .Loop2:
 325         clflush ($out)
 326         .byte   0xf0            # lock
 327         add     %eax,($out)     # accumulate diff
 328
 329         sub     \$1,$max
 330         jz      .Ldone2
 331
 332         rdtsc
 333         mov     %eax,%edx
 334         sub     $lasttick,%eax  # diff
 335         mov     %edx,$lasttick  # lasttick = tick
 336         cmp     $lastdiff,%eax
 337         mov     %eax,$lastdiff  # lastdiff = diff
 338         mov     \$0,%edx
 339         setne   %dl
 340         sub     %rdx,$cnt       # conditional --$cnt
 341         lea     ($out,%rdx,4),$out      # conditional ++$out
 342         jnz     .Loop2
 343
 344 .Ldone2:
 345         mov     $redzone(%rsp),%rax
 346         sub     $cnt,%rax
 347         ret
 348 .size   OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
 349 ___
 350 }
 351
 352 close STDOUT;   # flush