x86[_64]cpuid.pl: harmonize OPENSSL_ia32_cpuid [from HEAD].

diff --git a/crypto/x86_64cpuid.pl b/crypto/x86_64cpuid.pl
index c96821a3c811d55567f43caebce6f47a8308c790..503ef019ee3155bb602f2027906c4955f96abf39 100644 (file)
--- a/crypto/x86_64cpuid.pl
+++ b/crypto/x86_64cpuid.pl
@@ -7,15 +7,24 @@ if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
 $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";
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour $output";
+
+($arg1,$arg2,$arg3,$arg4)=$win64?("%rcx","%rdx","%r8", "%r9") :        # Win64 order
+                                ("%rdi","%rsi","%rdx","%rcx"); # Unix order
 
-if ($win64)    { $arg1="%rcx"; $arg2="%rdx"; }
-else           { $arg1="%rdi"; $arg2="%rsi"; }
 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 +55,7 @@ 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 +87,15 @@ 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 +106,12 @@ 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 +128,47 @@ 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
 

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index a7464af19b771d48aca8b3669e827ff843d29c82..b8961ba4ad6412fce5c669fbbc2b35eddd98e15b 100644 (file)
--- a/crypto/x86cpuid.pl
+++ b/crypto/x86cpuid.pl
@@ -20,7 +20,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &pop    ("eax");
        &xor    ("ecx","eax");
        &bt     ("ecx",21);
-       &jnc    (&label("done"));
+       &jnc    (&label("generic"));
        &xor    ("eax","eax");
        &cpuid  ();
        &mov    ("edi","eax");          # max value for standard query level
@@ -51,7 +51,14 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        # AMD specific
        &mov    ("eax",0x80000000);
        &cpuid  ();
-       &cmp    ("eax",0x80000008);
+       &cmp    ("eax",0x80000001);
+       &jb     (&label("intel"));
+       &mov    ("esi","eax");
+       &mov    ("eax",0x80000001);
+       &cpuid  ();
+       &or     ("ebp","ecx");
+       &and    ("ebp",1<<11|1);        # isolate XOP bit
+       &cmp    ("esi",0x80000008);
        &jb     (&label("intel"));
 
        &mov    ("eax",0x80000008);
@@ -62,13 +69,13 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &mov    ("eax",1);
        &cpuid  ();
        &bt     ("edx",28);
-       &jnc    (&label("done"));
+       &jnc    (&label("generic"));
        &shr    ("ebx",16);
        &and    ("ebx",0xff);
        &cmp    ("ebx","esi");
-       &ja     (&label("done"));
+       &ja     (&label("generic"));
        &and    ("edx",0xefffffff);     # clear hyper-threading bit
-       &jmp    (&label("done"));
+       &jmp    (&label("generic"));
        
 &set_label("intel");
        &cmp    ("edi",4);
@@ -85,27 +92,52 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 &set_label("nocacheinfo");
        &mov    ("eax",1);
        &cpuid  ();
+       &and    ("edx",0xbfefffff);     # force reserved bits #20, #30 to 0
        &cmp    ("ebp",0);
-       &jne    (&label("notP4"));
+       &jne    (&label("notintel"));
+       &or     ("edx",1<<30);          # set reserved bit#30 on Intel CPUs
        &and    (&HB("eax"),15);        # familiy ID
        &cmp    (&HB("eax"),15);        # P4?
-       &jne    (&label("notP4"));
-       &or     ("edx",1<<20);          # use reserved bit to engage RC4_CHAR
-&set_label("notP4");
+       &jne    (&label("notintel"));
+       &or     ("edx",1<<20);          # set reserved bit#20 to engage RC4_CHAR
+&set_label("notintel");
        &bt     ("edx",28);             # test hyper-threading bit
-       &jnc    (&label("done"));
+       &jnc    (&label("generic"));
        &and    ("edx",0xefffffff);
        &cmp    ("edi",0);
-       &je     (&label("done"));
+       &je     (&label("generic"));
 
        &or     ("edx",0x10000000);
        &shr    ("ebx",16);
        &cmp    (&LB("ebx"),1);
-       &ja     (&label("done"));
+       &ja     (&label("generic"));
        &and    ("edx",0xefffffff);     # clear hyper-threading bit if not
+
+&set_label("generic");
+       &and    ("ebp",1<<11);          # isolate AMD XOP flag
+       &and    ("ecx",0xfffff7ff);     # force 11th bit to 0
+       &mov    ("esi","edx");
+       &or     ("ebp","ecx");          # merge AMD XOP flag
+
+       &bt     ("ecx",26);             # check XSAVE bit
+       &jnc    (&label("done"));
+       &bt     ("ecx",27);             # check OSXSAVE bit
+       &jnc    (&label("clear_xmm"));
+       &xor    ("ecx","ecx");
+       &data_byte(0x0f,0x01,0xd0);     # xgetbv
+       &and    ("eax",6);
+       &cmp    ("eax",6);
+       &je     (&label("done"));
+       &cmp    ("eax",2);
+       &je     (&label("clear_avx"));
+&set_label("clear_xmm");
+       &and    ("ebp",0xfdfffffd);     # clear AESNI and PCLMULQDQ bits
+       &and    ("esi",0xfeffffff);     # clear FXSR
+&set_label("clear_avx");
+       &and    ("ebp",0xefffe7ff);     # clear AVX, FMA and AMD XOP bits
 &set_label("done");
-       &mov    ("eax","edx");
-       &mov    ("edx","ecx");
+       &mov    ("eax","esi");
+       &mov    ("edx","ebp");
 &function_end("OPENSSL_ia32_cpuid");
 
 &external_label("OPENSSL_ia32cap_P");
@@ -199,8 +231,9 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &bt     (&DWP(0,"ecx"),1);
        &jnc    (&label("no_x87"));
        if ($sse2) {
-               &bt     (&DWP(0,"ecx"),26);
-               &jnc    (&label("no_sse2"));
+               &and    ("ecx",1<<26|1<<24);    # check SSE2 and FXSR bits
+               &cmp    ("ecx",1<<26|1<<24);
+               &jne    (&label("no_sse2"));
                &pxor   ("xmm0","xmm0");
                &pxor   ("xmm1","xmm1");
                &pxor   ("xmm2","xmm2");