summaryrefslogtreecommitdiff
path: root/app/openssl/crypto/x86cpuid.pl
diff options
context:
space:
mode:
authorParménides GV <parmegv@sdf.org>2014-06-13 12:21:38 +0200
committerParménides GV <parmegv@sdf.org>2014-06-13 12:21:38 +0200
commitd09435a9d2adb0d0cafa0b1f9f6cfe326346cbc6 (patch)
tree946278cd836aa6af639675f4eb5ec7378d783dd8 /app/openssl/crypto/x86cpuid.pl
parent5d47fde84a03acb1201f8650c55dc0cd981df679 (diff)
parent3a71bc9e4aa4296f460e2e3c55de74c9852477ad (diff)
Merge branch 'develop' into release-0.5.2
Diffstat (limited to 'app/openssl/crypto/x86cpuid.pl')
-rw-r--r--app/openssl/crypto/x86cpuid.pl84
1 files changed, 65 insertions, 19 deletions
diff --git a/app/openssl/crypto/x86cpuid.pl b/app/openssl/crypto/x86cpuid.pl
index a7464af1..b270b443 100644
--- a/app/openssl/crypto/x86cpuid.pl
+++ b/app/openssl/crypto/x86cpuid.pl
@@ -19,9 +19,9 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&pushf ();
&pop ("eax");
&xor ("ecx","eax");
- &bt ("ecx",21);
- &jnc (&label("done"));
&xor ("eax","eax");
+ &bt ("ecx",21);
+ &jnc (&label("nocpuid"));
&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);
@@ -60,15 +67,16 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&inc ("esi"); # number of cores
&mov ("eax",1);
+ &xor ("ecx","ecx");
&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);
@@ -84,28 +92,53 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&set_label("nocacheinfo");
&mov ("eax",1);
+ &xor ("ecx","ecx");
&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",27); # check OSXSAVE bit
+ &jnc (&label("clear_avx"));
+ &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");
+&set_label("nocpuid");
&function_end("OPENSSL_ia32_cpuid");
&external_label("OPENSSL_ia32cap_P");
@@ -134,7 +167,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&jnz (&label("nohalt")); # not enough privileges
&pushf ();
- &pop ("eax")
+ &pop ("eax");
&bt ("eax",9);
&jnc (&label("nohalt")); # interrupts are disabled
@@ -199,8 +232,9 @@ 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");
@@ -248,7 +282,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
# arguments is 1 or 2!
&function_begin_B("OPENSSL_indirect_call");
{
- my $i,$max=7; # $max has to be chosen as 4*n-1
+ my ($max,$i)=(7,); # $max has to be chosen as 4*n-1
# in order to preserve eventual
# stack alignment
&push ("ebp");
@@ -307,6 +341,18 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
&ret ();
&function_end_B("OPENSSL_cleanse");
+&function_begin_B("OPENSSL_ia32_rdrand");
+ &mov ("ecx",8);
+&set_label("loop");
+ &rdrand ("eax");
+ &jc (&label("break"));
+ &loop (&label("loop"));
+&set_label("break");
+ &cmp ("eax",0);
+ &cmove ("eax","ecx");
+ &ret ();
+&function_end_B("OPENSSL_ia32_rdrand");
+
&initseg("OPENSSL_cpuid_setup");
&asm_finish();