x86[_64]cpuid.pl: harmonize usage of reserved bits #20 and #30.

[openssl.git] / crypto / x86_64cpuid.pl

diff --git a/crypto/x86_64cpuid.pl b/crypto/x86_64cpuid.pl
index 862118f285858439384cbed181c056bd6b58501d..7b76522bd881f8eac8a53939a07941d19f957185 100644 (file)
--- a/crypto/x86_64cpuid.pl
+++ b/crypto/x86_64cpuid.pl
@@ -9,13 +9,18 @@ $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 open STDOUT,"| $^X ${dir}perlasm/x86_64-xlate.pl $flavour $output";
 
-if ($win64)    { $arg1="%rcx"; $arg2="%rdx"; }
-else           { $arg1="%rdi"; $arg2="%rsi"; }
+($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") :        # Win64 order
+                                ("%rdi","%rsi","%rdx","%rcx"); # Unix order
+
 print<<___;
 .extern                OPENSSL_cpuid_setup
+.hidden                OPENSSL_cpuid_setup
 .section       .init
        call    OPENSSL_cpuid_setup
 
+.hidden        OPENSSL_ia32cap_P
+.comm  OPENSSL_ia32cap_P,8
+
 .text
 
 .globl OPENSSL_atomic_add
@@ -46,7 +51,7 @@ OPENSSL_rdtsc:
 .type  OPENSSL_ia32_cpuid,\@abi-omnipotent
 .align 16
 OPENSSL_ia32_cpuid:
-       mov     %rbx,%r8
+       mov     %rbx,%r8                # save %rbx
 
        xor     %eax,%eax
        cpuid
@@ -78,7 +83,15 @@ OPENSSL_ia32_cpuid:
        # AMD specific
        mov     \$0x80000000,%eax
        cpuid
-       cmp     \$0x80000008,%eax
+       cmp     \$0x80000001,%eax
+       jb      .Lintel
+       mov     %eax,%r10d
+       mov     \$0x80000001,%eax
+       cpuid
+       or      %ecx,%r9d
+       and     \$0x00000801,%r9d       # isolate AMD XOP bit, 1<<11
+
+       cmp     \$0x80000008,%r10d
        jb      .Lintel
 
        mov     \$0x80000008,%eax
@@ -89,12 +102,12 @@ OPENSSL_ia32_cpuid:
        mov     \$1,%eax
        cpuid
        bt      \$28,%edx               # test hyper-threading bit
-       jnc     .Ldone
+       jnc     .Lgeneric
        shr     \$16,%ebx               # number of logical processors
        cmp     %r10b,%bl
-       ja      .Ldone
+       ja      .Lgeneric
        and     \$0xefffffff,%edx       # ~(1<<28)
-       jmp     .Ldone
+       jmp     .Lgeneric
 
 .Lintel:
        cmp     \$4,%r11d
@@ -111,30 +124,47 @@ OPENSSL_ia32_cpuid:
 .Lnocacheinfo:
        mov     \$1,%eax
        cpuid
+       and     \$0xbfefffff,%edx       # force reserved bits to 0
        cmp     \$0,%r9d
        jne     .Lnotintel
-       or      \$0x00100000,%edx       # use reserved 20th bit to engage RC4_CHAR
+       or      \$0x40000000,%edx       # set reserved bit#30 on Intel CPUs
        and     \$15,%ah
        cmp     \$15,%ah                # examine Family ID
-       je      .Lnotintel
-       or      \$0x40000000,%edx       # use reserved bit to skip unrolled loop
+       jne     .Lnotintel
+       or      \$0x00100000,%edx       # set reserved bit#20 to engage RC4_CHAR
 .Lnotintel:
        bt      \$28,%edx               # test hyper-threading bit
-       jnc     .Ldone
+       jnc     .Lgeneric
        and     \$0xefffffff,%edx       # ~(1<<28)
        cmp     \$0,%r10d
-       je      .Ldone
+       je      .Lgeneric
 
        or      \$0x10000000,%edx       # 1<<28
        shr     \$16,%ebx
        cmp     \$1,%bl                 # see if cache is shared
-       ja      .Ldone
+       ja      .Lgeneric
        and     \$0xefffffff,%edx       # ~(1<<28)
+.Lgeneric:
+       and     \$0x00000800,%r9d       # isolate AMD XOP flag
+       and     \$0xfffff7ff,%ecx
+       or      %ecx,%r9d               # merge AMD XOP flag
+
+       mov     %edx,%r10d              # %r9d:%r10d is copy of %ecx:%edx
+       bt      \$27,%r9d               # check OSXSAVE bit
+       jnc     .Lclear_avx
+       xor     %ecx,%ecx               # XCR0
+       .byte   0x0f,0x01,0xd0          # xgetbv
+       and     \$6,%eax                # isolate XMM and YMM state support
+       cmp     \$6,%eax
+       je      .Ldone
+.Lclear_avx:
+       mov     \$0xefffe7ff,%eax       # ~(1<<28|1<<12|1<<11)
+       and     %eax,%r9d               # clear AVX, FMA and AMD XOP bits
 .Ldone:
-       shl     \$32,%rcx
-       mov     %edx,%eax
-       mov     %r8,%rbx
-       or      %rcx,%rax
+       shl     \$32,%r9
+       mov     %r10d,%eax
+       mov     %r8,%rbx                # restore %rbx
+       or      %r9,%rax
        ret
 .size  OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
 
@@ -145,11 +175,14 @@ OPENSSL_cleanse:
        xor     %rax,%rax
        cmp     \$15,$arg2
        jae     .Lot
+       cmp     \$0,$arg2
+       je      .Lret
 .Little:
        mov     %al,($arg1)
        sub     \$1,$arg2
        lea     1($arg1),$arg1
        jnz     .Little
+.Lret:
        ret
 .align 16
 .Lot:
@@ -225,5 +258,95 @@ OPENSSL_wipe_cpu:
        ret
 .size  OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
 ___
+{
+my $out="%r10";
+my $cnt="%rcx";
+my $max="%r11";
+my $lasttick="%r8d";
+my $lastdiff="%r9d";
+my $redzone=win64?8:-8;
+
+print<<___;
+.globl OPENSSL_instrument_bus
+.type  OPENSSL_instrument_bus,\@abi-omnipotent
+.align 16
+OPENSSL_instrument_bus:
+       mov     $arg1,$out      # tribute to Win64
+       mov     $arg2,$cnt
+       mov     $arg2,$max
+
+       rdtsc                   # collect 1st tick
+       mov     %eax,$lasttick  # lasttick = tick
+       mov     \$0,$lastdiff   # lastdiff = 0
+       clflush ($out)
+       .byte   0xf0            # lock
+       add     $lastdiff,($out)
+       jmp     .Loop
+.align 16
+.Loop: rdtsc
+       mov     %eax,%edx
+       sub     $lasttick,%eax
+       mov     %edx,$lasttick
+       mov     %eax,$lastdiff
+       clflush ($out)
+       .byte   0xf0            # lock
+       add     %eax,($out)
+       lea     4($out),$out
+       sub     \$1,$cnt
+       jnz     .Loop
+
+       mov     $max,%rax
+       ret
+.size  OPENSSL_instrument_bus,.-OPENSSL_instrument_bus
+
+.globl OPENSSL_instrument_bus2
+.type  OPENSSL_instrument_bus2,\@abi-omnipotent
+.align 16
+OPENSSL_instrument_bus2:
+       mov     $arg1,$out      # tribute to Win64
+       mov     $arg2,$cnt
+       mov     $arg3,$max
+       mov     $cnt,$redzone(%rsp)
+
+       rdtsc                   # collect 1st tick
+       mov     %eax,$lasttick  # lasttick = tick
+       mov     \$0,$lastdiff   # lastdiff = 0
+
+       clflush ($out)
+       .byte   0xf0            # lock
+       add     $lastdiff,($out)
+
+       rdtsc                   # collect 1st diff
+       mov     %eax,%edx
+       sub     $lasttick,%eax  # diff
+       mov     %edx,$lasttick  # lasttick = tick
+       mov     %eax,$lastdiff  # lastdiff = diff
+.Loop2:
+       clflush ($out)
+       .byte   0xf0            # lock
+       add     %eax,($out)     # accumulate diff
+
+       sub     \$1,$max
+       jz      .Ldone2
+
+       rdtsc
+       mov     %eax,%edx
+       sub     $lasttick,%eax  # diff
+       mov     %edx,$lasttick  # lasttick = tick
+       cmp     $lastdiff,%eax
+       mov     %eax,$lastdiff  # lastdiff = diff
+       mov     \$0,%edx
+       setne   %dl
+       sub     %rdx,$cnt       # conditional --$cnt
+       lea     ($out,%rdx,4),$out      # conditional ++$out
+       jnz     .Loop2
+
+.Ldone2:
+       mov     $redzone(%rsp),%rax
+       sub     $cnt,%rax
+       ret
+.size  OPENSSL_instrument_bus2,.-OPENSSL_instrument_bus2
+___
+}
 
 close STDOUT;  # flush