X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fx86cpuid.pl;h=90ed196c09cd1580fd4d3a4e7f2b69f98e99311c;hp=4073d4d77303491cd1795624543213dc10eb8cae;hb=c14a3be5adf88edc002199835d8fa40f1296e381;hpb=cb45708061511fe50b0a7de1778d77757208bb49

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index 4073d4d773..90ed196c09 100644
--- a/crypto/x86cpuid.pl
+++ b/crypto/x86cpuid.pl
@@ -19,9 +19,11 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&pushf	();
 	&pop	("eax");
 	&xor	("ecx","eax");
-	&bt	("ecx",21);
-	&jnc	(&label("generic"));
 	&xor	("eax","eax");
+	&mov	("esi",&wparam(0));
+	&mov	(&DWP(8,"esi"),"eax");	# clear extended feature flags
+	&bt	("ecx",21);
+	&jnc	(&label("nocpuid"));
 	&cpuid	();
 	&mov	("edi","eax");		# max value for standard query level
 
@@ -67,6 +69,7 @@ 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("generic"));
@@ -79,18 +82,19 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	
 &set_label("intel");
 	&cmp	("edi",4);
-	&mov	("edi",-1);
+	&mov	("esi",-1);
 	&jb	(&label("nocacheinfo"));
 
 	&mov	("eax",4);
 	&mov	("ecx",0);		# query L1D
 	&cpuid	();
-	&mov	("edi","eax");
-	&shr	("edi",14);
-	&and	("edi",0xfff);		# number of cores -1 per L1D
+	&mov	("esi","eax");
+	&shr	("esi",14);
+	&and	("esi",0xfff);		# number of cores -1 per L1D
 
 &set_label("nocacheinfo");
 	&mov	("eax",1);
+	&xor	("ecx","ecx");
 	&cpuid	();
 	&and	("edx",0xbfefffff);	# force reserved bits #20, #30 to 0
 	&cmp	("ebp",0);
@@ -104,7 +108,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&bt	("edx",28);		# test hyper-threading bit
 	&jnc	(&label("generic"));
 	&and	("edx",0xefffffff);
-	&cmp	("edi",0);
+	&cmp	("esi",0);
 	&je	(&label("generic"));
 
 	&or	("edx",0x10000000);
@@ -116,12 +120,19 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 &set_label("generic");
 	&and	("ebp",1<<11);		# isolate AMD XOP flag
 	&and	("ecx",0xfffff7ff);	# force 11th bit to 0
-	&mov	("esi","edx");
+	&mov	("esi","edx");		# %ebp:%esi is copy of %ecx:%edx
 	&or	("ebp","ecx");		# merge AMD XOP flag
 
-	&bt	("ecx",26);		# check XSAVE bit
-	&jnc	(&label("done"));
-	&bt	("ecx",27);		# check OSXSAVE bit
+	&cmp	("edi",7);
+	&mov	("edi",&wparam(0));
+	&jb	(&label("no_extended_info"));
+	&mov	("eax",7);
+	&xor	("ecx","ecx");
+	&cpuid	();
+	&mov	(&DWP(8,"edi"),"ebx");	# save extended feature flag
+&set_label("no_extended_info");
+
+	&bt	("ebp",27);		# check OSXSAVE bit
 	&jnc	(&label("clear_avx"));
 	&xor	("ecx","ecx");
 	&data_byte(0x0f,0x01,0xd0);	# xgetbv
@@ -135,9 +146,11 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&and	("esi",0xfeffffff);	# clear FXSR
 &set_label("clear_avx");
 	&and	("ebp",0xefffe7ff);	# clear AVX, FMA and AMD XOP bits
+	&and	(&DWP(8,"edi"),0xffffffdf);	# clear AVX2
 &set_label("done");
 	&mov	("eax","esi");
 	&mov	("edx","ebp");
+&set_label("nocpuid");
 &function_end("OPENSSL_ia32_cpuid");
 
 &external_label("OPENSSL_ia32cap_P");
@@ -166,7 +179,7 @@ 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
 
@@ -197,7 +210,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 
 &function_begin_B("OPENSSL_far_spin");
 	&pushf	();
-	&pop	("eax")
+	&pop	("eax");
 	&bt	("eax",9);
 	&jnc	(&label("nospin"));	# interrupts are disabled
 
@@ -281,7 +294,7 @@ 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");
@@ -352,6 +365,21 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&ret	();
 &function_end_B("OPENSSL_ia32_rdrand");
 
+&function_begin_B("OPENSSL_ia32_rdseed");
+	&mov	("ecx",8);
+&set_label("loop");
+	&rdseed	("eax");
+	&jc	(&label("break"));
+	&loop	(&label("loop"));
+&set_label("break");
+	&cmp	("eax",0);
+	&cmove	("eax","ecx");
+	&ret	();
+&function_end_B("OPENSSL_ia32_rdseed");
+
 &initseg("OPENSSL_cpuid_setup");
 
+&hidden("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_ia32cap_P");
+
 &asm_finish();