X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=blobdiff_plain;f=crypto%2Fx86cpuid.pl;h=c3291112032b3ff7b365830ecf123f3a4efc4df8;hp=85fbef7417b5e64b02bb989b697ed821a00e47d5;hb=4d1f3f7a6c4dd0162c903c8fb96265db1ba0b91c;hpb=14e21f863a3e3278bb8660ea9844e92e52e1f2f7

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index 85fbef7417..c329111203 100644
--- a/crypto/x86cpuid.pl
+++ b/crypto/x86cpuid.pl
@@ -1,10 +1,13 @@
 #!/usr/bin/env perl
 
-push(@INC,"perlasm");
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC, "${dir}perlasm", "perlasm");
 require "x86asm.pl";
 
 &asm_init($ARGV[0],"x86cpuid");
 
+for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
+
 &function_begin("OPENSSL_ia32_cpuid");
 	&xor	("edx","edx");
 	&pushf	();
@@ -17,20 +20,47 @@ require "x86asm.pl";
 	&pop	("eax");
 	&xor	("ecx","eax");
 	&bt	("ecx",21);
-	&jnc	(&label("nocpuid"));
+	&jnc	(&label("done"));
+	&xor	("eax","eax");
+	&cpuid	();
+	&xor	("eax","eax");
+	&cmp	("ebx",0x756e6547);	# "Genu"
+	&setne	(&LB("eax"));
+	&mov	("ebp","eax");
+	&cmp	("edx",0x49656e69);	# "ineI"
+	&setne	(&LB("eax"));
+	&or	("ebp","eax");
+	&cmp	("ecx",0x6c65746e);	# "ntel"
+	&setne	(&LB("eax"));
+	&or	("ebp","eax");
 	&mov	("eax",1);
 	&cpuid	();
-&set_label("nocpuid");
+	&cmp	("ebp",0);
+	&jne	(&label("notP4"));
+	&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");
+	&bt	("edx",28);		# test hyper-threading bit
+	&jnc	(&label("done"));
+	&shr	("ebx",16);
+	&cmp	(&LB("ebx"),1);		# see if cache is shared(*)
+	&ja	(&label("done"));
+	&and	("edx",0xefffffff);	# clear hyper-threading bit if not
+&set_label("done");
 	&mov	("eax","edx");
 	&mov	("edx","ecx");
 &function_end("OPENSSL_ia32_cpuid");
+# (*)	on Core2 this value is set to 2 denoting the fact that L2
+#	cache is shared between cores.
 
-&external_label("OPENSSL_ia32cap");
+&external_label("OPENSSL_ia32cap_P");
 
-&function_begin_B("OPENSSL_rdtsc");
+&function_begin_B("OPENSSL_rdtsc","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
 	&xor	("eax","eax");
 	&xor	("edx","edx");
-	&picmeup("ecx","OPENSSL_ia32cap");
+	&picmeup("ecx","OPENSSL_ia32cap_P");
 	&bt	(&DWP(0,"ecx"),4);
 	&jnc	(&label("notsc"));
 	&rdtsc	();
@@ -38,6 +68,189 @@ require "x86asm.pl";
 	&ret	();
 &function_end_B("OPENSSL_rdtsc");
 
-&initseg("OPENSSL_cpuid_setup") if ($main'elf);
+# This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host],
+# but it's safe to call it on any [supported] 32-bit platform...
+# Just check for [non-]zero return value...
+&function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
+	&picmeup("ecx","OPENSSL_ia32cap_P");
+	&bt	(&DWP(0,"ecx"),4);
+	&jnc	(&label("nohalt"));	# no TSC
+
+	&data_word(0x9058900e);		# push %cs; pop %eax
+	&and	("eax",3);
+	&jnz	(&label("nohalt"));	# not enough privileges
+
+	&pushf	();
+	&pop	("eax")
+	&bt	("eax",9);
+	&jnc	(&label("nohalt"));	# interrupts are disabled
+
+	&rdtsc	();
+	&push	("edx");
+	&push	("eax");
+	&halt	();
+	&rdtsc	();
+
+	&sub	("eax",&DWP(0,"esp"));
+	&sbb	("edx",&DWP(4,"esp"));
+	&add	("esp",8);
+	&ret	();
+
+&set_label("nohalt");
+	&xor	("eax","eax");
+	&xor	("edx","edx");
+	&ret	();
+&function_end_B("OPENSSL_instrument_halt");
+
+# Essentially there is only one use for this function. Under DJGPP:
+#
+#	#include <go32.h>
+#	...
+#	i=OPENSSL_far_spin(_dos_ds,0x46c);
+#	...
+# to obtain the number of spins till closest timer interrupt.
+
+&function_begin_B("OPENSSL_far_spin");
+	&pushf	();
+	&pop	("eax")
+	&bt	("eax",9);
+	&jnc	(&label("nospin"));	# interrupts are disabled
+
+	&mov	("eax",&DWP(4,"esp"));
+	&mov	("ecx",&DWP(8,"esp"));
+	&data_word (0x90d88e1e);	# push %ds, mov %eax,%ds
+	&xor	("eax","eax");
+	&mov	("edx",&DWP(0,"ecx"));
+	&jmp	(&label("spin"));
+
+	&align	(16);
+&set_label("spin");
+	&inc	("eax");
+	&cmp	("edx",&DWP(0,"ecx"));
+	&je	(&label("spin"));
+
+	&data_word (0x1f909090);	# pop	%ds
+	&ret	();
+
+&set_label("nospin");
+	&xor	("eax","eax");
+	&xor	("edx","edx");
+	&ret	();
+&function_end_B("OPENSSL_far_spin");
+
+&function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
+	&xor	("eax","eax");
+	&xor	("edx","edx");
+	&picmeup("ecx","OPENSSL_ia32cap_P");
+	&mov	("ecx",&DWP(0,"ecx"));
+	&bt	(&DWP(0,"ecx"),1);
+	&jnc	(&label("no_x87"));
+	if ($sse2) {
+		&bt	(&DWP(0,"ecx"),26);
+		&jnc	(&label("no_sse2"));
+		&pxor	("xmm0","xmm0");
+		&pxor	("xmm1","xmm1");
+		&pxor	("xmm2","xmm2");
+		&pxor	("xmm3","xmm3");
+		&pxor	("xmm4","xmm4");
+		&pxor	("xmm5","xmm5");
+		&pxor	("xmm6","xmm6");
+		&pxor	("xmm7","xmm7");
+	&set_label("no_sse2");
+	}
+	# just a bunch of fldz to zap the fp/mm bank followed by finit...
+	&data_word(0xeed9eed9,0xeed9eed9,0xeed9eed9,0xeed9eed9,0x90e3db9b);
+&set_label("no_x87");
+	&lea	("eax",&DWP(4,"esp"));
+	&ret	();
+&function_end_B("OPENSSL_wipe_cpu");
+
+&function_begin_B("OPENSSL_atomic_add");
+	&mov	("edx",&DWP(4,"esp"));	# fetch the pointer, 1st arg
+	&mov	("ecx",&DWP(8,"esp"));	# fetch the increment, 2nd arg
+	&push	("ebx");
+	&nop	();
+	&mov	("eax",&DWP(0,"edx"));
+&set_label("spin");
+	&lea	("ebx",&DWP(0,"eax","ecx"));
+	&nop	();
+	&data_word(0x1ab10ff0);	# lock;	cmpxchg	%ebx,(%edx)	# %eax is envolved and is always reloaded
+	&jne	(&label("spin"));
+	&mov	("eax","ebx");	# OpenSSL expects the new value
+	&pop	("ebx");
+	&ret	();
+&function_end_B("OPENSSL_atomic_add");
+
+# This function can become handy under Win32 in situations when
+# we don't know which calling convention, __stdcall or __cdecl(*),
+# indirect callee is using. In C it can be deployed as
+#
+#ifdef OPENSSL_CPUID_OBJ
+#	type OPENSSL_indirect_call(void *f,...);
+#	...
+#	OPENSSL_indirect_call(func,[up to $max arguments]);
+#endif
+#
+# (*)	it's designed to work even for __fastcall if number of
+#	arguments is 1 or 2!
+&function_begin_B("OPENSSL_indirect_call");
+	{
+	my $i,$max=7;		# $max has to be chosen as 4*n-1
+				# in order to preserve eventual
+				# stack alignment
+	&push	("ebp");
+	&mov	("ebp","esp");
+	&sub	("esp",$max*4);
+	&mov	("ecx",&DWP(12,"ebp"));
+	&mov	(&DWP(0,"esp"),"ecx");
+	&mov	("edx",&DWP(16,"ebp"));
+	&mov	(&DWP(4,"esp"),"edx");
+	for($i=2;$i<$max;$i++)
+		{
+		# Some copies will be redundant/bogus...
+		&mov	("eax",&DWP(12+$i*4,"ebp"));
+		&mov	(&DWP(0+$i*4,"esp"),"eax");
+		}
+	&call_ptr	(&DWP(8,"ebp"));# make the call...
+	&mov	("esp","ebp");	# ... and just restore the stack pointer
+				# without paying attention to what we called,
+				# (__cdecl *func) or (__stdcall *one).
+	&pop	("ebp");
+	&ret	();
+	}
+&function_end_B("OPENSSL_indirect_call");
+
+&function_begin_B("OPENSSL_cleanse");
+	&mov	("edx",&wparam(0));
+	&mov	("ecx",&wparam(1));
+	&xor	("eax","eax");
+	&cmp	("ecx",7);
+	&jae	(&label("lot"));
+&set_label("little");
+	&mov	(&BP(0,"edx"),"al");
+	&sub	("ecx",1);
+	&lea	("edx",&DWP(1,"edx"));
+	&jnz	(&label("little"));
+	&ret	();
+
+&set_label("lot",16);
+	&test	("edx",3);
+	&jz	(&label("aligned"));
+	&mov	(&BP(0,"edx"),"al");
+	&lea	("ecx",&DWP(-1,"ecx"));
+	&lea	("edx",&DWP(1,"edx"));
+	&jmp	(&label("lot"));
+&set_label("aligned");
+	&mov	(&DWP(0,"edx"),"eax");
+	&lea	("ecx",&DWP(-4,"ecx"));
+	&test	("ecx",-4);
+	&lea	("edx",&DWP(4,"edx"));
+	&jnz	(&label("aligned"));
+	&cmp	("ecx",0);
+	&jne	(&label("little"));
+	&ret	();
+&function_end_B("OPENSSL_cleanse");
+
+&initseg("OPENSSL_cpuid_setup");
 
 &asm_finish();