X-Git-Url: https://git.openssl.org/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fx86cpuid.pl;h=dca37e7cd265878d04a77d6dea0235add06e4c74;hp=70c8a2d671a50ee3347672e2278eb273507c94b8;hb=a90081576c94f9f54de1755188a00ccc1760549a;hpb=18f5603c5333744ddbe55c963ffd88bb3f59b0d7

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index 70c8a2d671..dca37e7cd2 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");
+	&bt	("ecx",21);
+	&jnc	(&label("nocpuid"));
+	&mov	("esi",&wparam(0));
+	&mov	(&DWP(8,"esi"),"eax");	# clear 3rd word
 	&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"));
@@ -78,6 +81,16 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&jmp	(&label("generic"));
 	
 &set_label("intel");
+	&cmp	("edi",7);
+	&jb	(&label("cacheinfo"));
+
+	&mov	("esi",&wparam(0));
+	&mov	("eax",7);
+	&xor	("ecx","ecx");
+	&cpuid	();
+	&mov	(&DWP(8,"esi"),"ebx");
+
+&set_label("cacheinfo");
 	&cmp	("edi",4);
 	&mov	("edi",-1);
 	&jb	(&label("nocacheinfo"));
@@ -91,6 +104,7 @@ 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);
@@ -119,10 +133,8 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&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"));
+	&jnc	(&label("clear_avx"));
 	&xor	("ecx","ecx");
 	&data_byte(0x0f,0x01,0xd0);	# xgetbv
 	&and	("eax",6);
@@ -135,9 +147,12 @@ 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
+	&mov	("edi",&wparam(0));
+	&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");
@@ -153,7 +168,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&ret	();
 &function_end_B("OPENSSL_rdtsc");
 
-# This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host],
+# This works in Ring 0 only [read MS-DOS+privileged DPMI host],
 # but it's safe to call it on any [supported] 32-bit platform...
 # Just check for [non-]zero return value...
 &function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
@@ -166,7 +181,7 @@ 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
 
@@ -187,41 +202,6 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 	&ret	();
 &function_end_B("OPENSSL_instrument_halt");
 
-# Essentially there is only one use for this function. Under DJGPP:
-#
-#	#include <go32.h>
-#	...
-#	i=OPENSSL_far_spin(_dos_ds,0x46c);
-#	...
-# to obtain the number of spins till closest timer interrupt.
-
-&function_begin_B("OPENSSL_far_spin");
-	&pushf	();
-	&pop	("eax")
-	&bt	("eax",9);
-	&jnc	(&label("nospin"));	# interrupts are disabled
-
-	&mov	("eax",&DWP(4,"esp"));
-	&mov	("ecx",&DWP(8,"esp"));
-	&data_word (0x90d88e1e);	# push %ds, mov %eax,%ds
-	&xor	("eax","eax");
-	&mov	("edx",&DWP(0,"ecx"));
-	&jmp	(&label("spin"));
-
-	&align	(16);
-&set_label("spin");
-	&inc	("eax");
-	&cmp	("edx",&DWP(0,"ecx"));
-	&je	(&label("spin"));
-
-	&data_word (0x1f909090);	# pop	%ds
-	&ret	();
-
-&set_label("nospin");
-	&xor	("eax","eax");
-	&xor	("edx","edx");
-	&ret	();
-&function_end_B("OPENSSL_far_spin");
 
 &function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
 	&xor	("eax","eax");
@@ -281,7 +261,7 @@ 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");
@@ -442,6 +422,33 @@ my $max = "ebp";
 &function_end("OPENSSL_instrument_bus2");
 }
 
+&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");
+
+&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();