[thirdparty/openssl.git] / crypto / x86cpuid.pl

#!/usr/bin/env perl

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $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	();
	&pop	("eax");
	&mov	("ecx","eax");
	&xor	("eax",1<<21);
	&push	("eax");
	&popf	();
	&pushf	();
	&pop	("eax");
	&xor	("ecx","eax");
	&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

	&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");		# 0 indicates Intel CPU
	&jz	(&label("intel"));

	&cmp	("ebx",0x68747541);	# "Auth"
	&setne	(&LB("eax"));
	&mov	("esi","eax");
	&cmp	("edx",0x69746E65);	# "enti"
	&setne	(&LB("eax"));
	&or	("esi","eax");
	&cmp	("ecx",0x444D4163);	# "cAMD"
	&setne	(&LB("eax"));
	&or	("esi","eax");		# 0 indicates AMD CPU
	&jnz	(&label("intel"));

	# AMD specific
	&mov	("eax",0x80000000);
	&cpuid	();
	&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);
	&cpuid	();
	&movz	("esi",&LB("ecx"));	# number of cores - 1
	&inc	("esi");		# number of cores

	&mov	("eax",1);
	&xor	("ecx","ecx");
	&cpuid	();
	&bt	("edx",28);
	&jnc	(&label("generic"));
	&shr	("ebx",16);
	&and	("ebx",0xff);
	&cmp	("ebx","esi");
	&ja	(&label("generic"));
	&and	("edx",0xefffffff);	# clear hyper-threading bit
	&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"));

	&mov	("eax",4);
	&mov	("ecx",0);		# query L1D
	&cpuid	();
	&mov	("edi","eax");
	&shr	("edi",14);
	&and	("edi",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);
	&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("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("generic"));
	&and	("edx",0xefffffff);
	&cmp	("edi",0);
	&je	(&label("generic"));

	&or	("edx",0x10000000);
	&shr	("ebx",16);
	&cmp	(&LB("ebx"),1);
	&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
	&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");

&function_begin_B("OPENSSL_rdtsc","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
	&xor	("eax","eax");
	&xor	("edx","edx");
	&picmeup("ecx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"ecx"),4);
	&jnc	(&label("notsc"));
	&rdtsc	();
&set_label("notsc");
	&ret	();
&function_end_B("OPENSSL_rdtsc");

# 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) {
		&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");
		&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 $max arguments]);
#endif
#
# (*)	it's designed to work even for __fastcall if number of
#	arguments is 1 or 2!
&function_begin_B("OPENSSL_indirect_call");
	{
	my ($max,$i)=(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"));
	&cmp	("ecx",0);
	&je	(&label("ret"));
&set_label("little");
	&mov	(&BP(0,"edx"),"al");
	&sub	("ecx",1);
	&lea	("edx",&DWP(1,"edx"));
	&jnz	(&label("little"));
&set_label("ret");
	&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");

{
my $lasttick = "esi";
my $lastdiff = "ebx";
my $out = "edi";
my $cnt = "ecx";
my $max = "ebp";

&function_begin("OPENSSL_instrument_bus");
    &mov	("eax",0);
    if ($sse2) {
	&picmeup("edx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"edx"),4);
	&jnc	(&label("nogo"));	# no TSC
	&bt	(&DWP(0,"edx"),19);
	&jnc	(&label("nogo"));	# no CLFLUSH

	&mov	($out,&wparam(0));	# load arguments
	&mov	($cnt,&wparam(1));

	# collect 1st tick
	&rdtsc	();
	&mov	($lasttick,"eax");	# lasttick = tick
	&mov	($lastdiff,0);		# lastdiff = 0
	&clflush(&DWP(0,$out));
	&data_byte(0xf0);		# lock
	&add	(&DWP(0,$out),$lastdiff);
	&jmp	(&label("loop"));

&set_label("loop",16);
	&rdtsc	();
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&clflush(&DWP(0,$out));
	&data_byte(0xf0);		# lock
	&add	(&DWP(0,$out),"eax");	# accumulate diff
	&lea	($out,&DWP(4,$out));	# ++$out
	&sub	($cnt,1);		# --$cnt
	&jnz	(&label("loop"));

	&mov	("eax",&wparam(1));
&set_label("nogo");
    }
&function_end("OPENSSL_instrument_bus");

&function_begin("OPENSSL_instrument_bus2");
    &mov	("eax",0);
    if ($sse2) {
	&picmeup("edx","OPENSSL_ia32cap_P");
	&bt	(&DWP(0,"edx"),4);
	&jnc	(&label("nogo"));	# no TSC
	&bt	(&DWP(0,"edx"),19);
	&jnc	(&label("nogo"));	# no CLFLUSH

	&mov	($out,&wparam(0));	# load arguments
	&mov	($cnt,&wparam(1));
	&mov	($max,&wparam(2));

	&rdtsc	();			# collect 1st tick
	&mov	($lasttick,"eax");	# lasttick = tick
	&mov	($lastdiff,0);		# lastdiff = 0

	&clflush(&DWP(0,$out));
	&data_byte(0xf0);		# lock
	&add	(&DWP(0,$out),$lastdiff);

	&rdtsc	();			# collect 1st diff
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&jmp	(&label("loop2"));

&set_label("loop2",16);
	&clflush(&DWP(0,$out));
	&data_byte(0xf0);		# lock
	&add	(&DWP(0,$out),"eax");	# accumulate diff

	&sub	($max,1);
	&jz	(&label("done2"));

	&rdtsc	();
	&mov	("edx","eax");		# put aside tick (yes, I neglect edx)
	&sub	("eax",$lasttick);	# diff
	&mov	($lasttick,"edx");	# lasttick = tick
	&cmp	("eax",$lastdiff);
	&mov	($lastdiff,"eax");	# lastdiff = diff
	&mov	("edx",0);
	&setne	("dl");
	&sub	($cnt,"edx");		# conditional --$cnt
	&lea	($out,&DWP(0,$out,"edx",4));	# conditional ++$out
	&jnz	(&label("loop2"));

&set_label("done2");
	&mov	("eax",&wparam(1));
	&sub	("eax",$cnt);
&set_label("nogo");
    }
&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();
Commit	Line	Data
14e21f86 AP	1	#!/usr/bin/env perl
14e21f86 AP	2
4d1f3f7a DSH	3	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
4d1f3f7a DSH	4	push(@INC, "${dir}perlasm", "perlasm");
14e21f86 AP	5	require "x86asm.pl";
	6
	7	&asm_init($ARGV[0],"x86cpuid");
	8
c5022659 AP	9	for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
c5022659 AP	10
14e21f86 AP	11	&function_begin("OPENSSL_ia32_cpuid");
	12	&xor ("edx","edx");
	13	&pushf ();
	14	&pop ("eax");
	15	&mov ("ecx","eax");
	16	&xor ("eax",1<<21);
	17	&push ("eax");
	18	&popf ();
	19	&pushf ();
	20	&pop ("eax");
	21	&xor ("ecx","eax");
162f677d	22	&xor ("eax","eax");
0f2ece87 AP	23	&bt ("ecx",21);
0f2ece87 AP	24	&jnc (&label("nocpuid"));
c5cd28bd AP	25	&mov ("esi",&wparam(0));
c5cd28bd AP	26	&mov (&DWP(8,"esi"),"eax"); # clear 3rd word
162f677d	27	&cpuid ();
761393bb AP	28	&mov ("edi","eax"); # max value for standard query level
761393bb AP	29
162f677d AP	30	&xor ("eax","eax");
	31	&cmp ("ebx",0x756e6547); # "Genu"
	32	&setne (&LB("eax"));
	33	&mov ("ebp","eax");
	34	&cmp ("edx",0x49656e69); # "ineI"
	35	&setne (&LB("eax"));
	36	&or ("ebp","eax");
	37	&cmp ("ecx",0x6c65746e); # "ntel"
	38	&setne (&LB("eax"));
58672313	39	&or ("ebp","eax"); # 0 indicates Intel CPU
58672313 AP	40	&jz (&label("intel"));
	41
	42	&cmp ("ebx",0x68747541); # "Auth"
	43	&setne (&LB("eax"));
	44	&mov ("esi","eax");
	45	&cmp ("edx",0x69746E65); # "enti"
	46	&setne (&LB("eax"));
	47	&or ("esi","eax");
	48	&cmp ("ecx",0x444D4163); # "cAMD"
	49	&setne (&LB("eax"));
	50	&or ("esi","eax"); # 0 indicates AMD CPU
	51	&jnz (&label("intel"));
	52
761393bb	53	# AMD specific
58672313 AP	54	&mov ("eax",0x80000000);
58672313 AP	55	&cpuid ();
b9064221 AP	56	&cmp ("eax",0x80000001);
	57	&jb (&label("intel"));
	58	&mov ("esi","eax");
	59	&mov ("eax",0x80000001);
	60	&cpuid ();
	61	&or ("ebp","ecx");
	62	&and ("ebp",1<<11\|1); # isolate XOP bit
	63	&cmp ("esi",0x80000008);
58672313 AP	64	&jb (&label("intel"));
	65
	66	&mov ("eax",0x80000008);
	67	&cpuid ();
	68	&movz ("esi",&LB("ecx")); # number of cores - 1
	69	&inc ("esi"); # number of cores
	70
761393bb	71	&mov ("eax",1);
5702e965	72	&xor ("ecx","ecx");
761393bb AP	73	&cpuid ();
761393bb AP	74	&bt ("edx",28);
b9064221	75	&jnc (&label("generic"));
761393bb AP	76	&shr ("ebx",16);
	77	&and ("ebx",0xff);
	78	&cmp ("ebx","esi");
b9064221	79	&ja (&label("generic"));
761393bb	80	&and ("edx",0xefffffff); # clear hyper-threading bit
b9064221	81	&jmp (&label("generic"));
761393bb	82
58672313	83	&set_label("intel");
1bc0b68d AP	84	&cmp ("edi",7);
	85	&jb (&label("cacheinfo"));
	86
	87	&mov ("esi",&wparam(0));
	88	&mov ("eax",7);
	89	&xor ("ecx","ecx");
	90	&cpuid ();
	91	&mov (&DWP(8,"esi"),"ebx");
	92
	93	&set_label("cacheinfo");
761393bb AP	94	&cmp ("edi",4);
	95	&mov ("edi",-1);
	96	&jb (&label("nocacheinfo"));
	97
	98	&mov ("eax",4);
	99	&mov ("ecx",0); # query L1D
	100	&cpuid ();
	101	&mov ("edi","eax");
	102	&shr ("edi",14);
	103	&and ("edi",0xfff); # number of cores -1 per L1D
	104
	105	&set_label("nocacheinfo");
14e21f86	106	&mov ("eax",1);
5702e965	107	&xor ("ecx","ecx");
14e21f86	108	&cpuid ();
18f5603c	109	&and ("edx",0xbfefffff); # force reserved bits #20, #30 to 0
162f677d	110	&cmp ("ebp",0);
4bb90087 AP	111	&jne (&label("notintel"));
4bb90087 AP	112	&or ("edx",1<<30); # set reserved bit#30 on Intel CPUs
3df2eff4 AP	113	&and (&HB("eax"),15); # familiy ID
3df2eff4 AP	114	&cmp (&HB("eax"),15); # P4?
4bb90087 AP	115	&jne (&label("notintel"));
	116	&or ("edx",1<<20); # set reserved bit#20 to engage RC4_CHAR
	117	&set_label("notintel");
3df2eff4	118	&bt ("edx",28); # test hyper-threading bit
b9064221	119	&jnc (&label("generic"));
761393bb AP	120	&and ("edx",0xefffffff);
761393bb AP	121	&cmp ("edi",0);
b9064221	122	&je (&label("generic"));
761393bb AP	123
761393bb AP	124	&or ("edx",0x10000000);
162f677d	125	&shr ("ebx",16);
761393bb	126	&cmp (&LB("ebx"),1);
b9064221	127	&ja (&label("generic"));
b1e8b4e6	128	&and ("edx",0xefffffff); # clear hyper-threading bit if not
b9064221 AP	129
	130	&set_label("generic");
	131	&and ("ebp",1<<11); # isolate AMD XOP flag
18f5603c	132	&and ("ecx",0xfffff7ff); # force 11th bit to 0
b9064221 AP	133	&mov ("esi","edx");
	134	&or ("ebp","ecx"); # merge AMD XOP flag
	135
b9064221	136	&bt ("ecx",27); # check OSXSAVE bit
e879dd43	137	&jnc (&label("clear_avx"));
b9064221 AP	138	&xor ("ecx","ecx");
	139	&data_byte(0x0f,0x01,0xd0); # xgetbv
	140	&and ("eax",6);
	141	&cmp ("eax",6);
	142	&je (&label("done"));
	143	&cmp ("eax",2);
	144	&je (&label("clear_avx"));
	145	&set_label("clear_xmm");
18f5603c AP	146	&and ("ebp",0xfdfffffd); # clear AESNI and PCLMULQDQ bits
18f5603c AP	147	&and ("esi",0xfeffffff); # clear FXSR
b9064221	148	&set_label("clear_avx");
18f5603c	149	&and ("ebp",0xefffe7ff); # clear AVX, FMA and AMD XOP bits
c5cd28bd AP	150	&mov ("edi",&wparam(0));
c5cd28bd AP	151	&and (&DWP(8,"edi"),0xffffffdf); # clear AVX2
3df2eff4	152	&set_label("done");
b9064221 AP	153	&mov ("eax","esi");
b9064221 AP	154	&mov ("edx","ebp");
0f2ece87	155	&set_label("nocpuid");
14e21f86 AP	156	&function_end("OPENSSL_ia32_cpuid");
14e21f86 AP	157
2b247cf8	158	&external_label("OPENSSL_ia32cap_P");
14e21f86	159
2b247cf8	160	&function_begin_B("OPENSSL_rdtsc","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
14e21f86 AP	161	&xor ("eax","eax");
14e21f86 AP	162	&xor ("edx","edx");
2b247cf8	163	&picmeup("ecx","OPENSSL_ia32cap_P");
14e21f86 AP	164	&bt (&DWP(0,"ecx"),4);
	165	&jnc (&label("notsc"));
	166	&rdtsc ();
	167	&set_label("notsc");
	168	&ret ();
	169	&function_end_B("OPENSSL_rdtsc");
	170
f642ebc1	171	# This works in Ring 0 only [read DJGPP+MS-DOS+privileged DPMI host],
c85c5c40 AP	172	# but it's safe to call it on any [supported] 32-bit platform...
	173	# Just check for [non-]zero return value...
	174	&function_begin_B("OPENSSL_instrument_halt","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
	175	&picmeup("ecx","OPENSSL_ia32cap_P");
	176	&bt (&DWP(0,"ecx"),4);
	177	&jnc (&label("nohalt")); # no TSC
	178
	179	&data_word(0x9058900e); # push %cs; pop %eax
	180	&and ("eax",3);
	181	&jnz (&label("nohalt")); # not enough privileges
	182
	183	&pushf ();
f9c5e5d9	184	&pop ("eax");
c85c5c40 AP	185	&bt ("eax",9);
	186	&jnc (&label("nohalt")); # interrupts are disabled
	187
	188	&rdtsc ();
	189	&push ("edx");
	190	&push ("eax");
	191	&halt ();
	192	&rdtsc ();
	193
	194	&sub ("eax",&DWP(0,"esp"));
	195	&sbb ("edx",&DWP(4,"esp"));
	196	&add ("esp",8);
	197	&ret ();
	198
	199	&set_label("nohalt");
	200	&xor ("eax","eax");
	201	&xor ("edx","edx");
	202	&ret ();
	203	&function_end_B("OPENSSL_instrument_halt");
	204
f642ebc1 RS	205	# Essentially there is only one use for this function. Under DJGPP:
	206	#
	207	# #include <go32.h>
	208	# ...
	209	# i=OPENSSL_far_spin(_dos_ds,0x46c);
	210	# ...
	211	# to obtain the number of spins till closest timer interrupt.
	212
	213	&function_begin_B("OPENSSL_far_spin");
	214	&pushf ();
	215	&pop ("eax");
	216	&bt ("eax",9);
	217	&jnc (&label("nospin")); # interrupts are disabled
	218
	219	&mov ("eax",&DWP(4,"esp"));
	220	&mov ("ecx",&DWP(8,"esp"));
	221	&data_word (0x90d88e1e); # push %ds, mov %eax,%ds
	222	&xor ("eax","eax");
	223	&mov ("edx",&DWP(0,"ecx"));
	224	&jmp (&label("spin"));
	225
	226	&align (16);
	227	&set_label("spin");
	228	&inc ("eax");
	229	&cmp ("edx",&DWP(0,"ecx"));
	230	&je (&label("spin"));
	231
	232	&data_word (0x1f909090); # pop %ds
	233	&ret ();
	234
	235	&set_label("nospin");
	236	&xor ("eax","eax");
	237	&xor ("edx","edx");
	238	&ret ();
	239	&function_end_B("OPENSSL_far_spin");
cee73df3 AP	240
	241	&function_begin_B("OPENSSL_wipe_cpu","EXTRN\t_OPENSSL_ia32cap_P:DWORD");
	242	&xor ("eax","eax");
	243	&xor ("edx","edx");
	244	&picmeup("ecx","OPENSSL_ia32cap_P");
	245	&mov ("ecx",&DWP(0,"ecx"));
	246	&bt (&DWP(0,"ecx"),1);
	247	&jnc (&label("no_x87"));
c5022659	248	if ($sse2) {
b9064221 AP	249	&and ("ecx",1<<26\|1<<24); # check SSE2 and FXSR bits
	250	&cmp ("ecx",1<<26\|1<<24);
	251	&jne (&label("no_sse2"));
c5022659 AP	252	&pxor ("xmm0","xmm0");
	253	&pxor ("xmm1","xmm1");
	254	&pxor ("xmm2","xmm2");
	255	&pxor ("xmm3","xmm3");
	256	&pxor ("xmm4","xmm4");
	257	&pxor ("xmm5","xmm5");
	258	&pxor ("xmm6","xmm6");
	259	&pxor ("xmm7","xmm7");
	260	&set_label("no_sse2");
	261	}
c49a0aa0 AP	262	# just a bunch of fldz to zap the fp/mm bank followed by finit...
c49a0aa0 AP	263	&data_word(0xeed9eed9,0xeed9eed9,0xeed9eed9,0xeed9eed9,0x90e3db9b);
cee73df3 AP	264	&set_label("no_x87");
	265	&lea ("eax",&DWP(4,"esp"));
	266	&ret ();
	267	&function_end_B("OPENSSL_wipe_cpu");
	268
	269	&function_begin_B("OPENSSL_atomic_add");
	270	&mov ("edx",&DWP(4,"esp")); # fetch the pointer, 1st arg
	271	&mov ("ecx",&DWP(8,"esp")); # fetch the increment, 2nd arg
	272	&push ("ebx");
	273	&nop ();
	274	&mov ("eax",&DWP(0,"edx"));
	275	&set_label("spin");
	276	&lea ("ebx",&DWP(0,"eax","ecx"));
	277	&nop ();
	278	&data_word(0x1ab10ff0); # lock; cmpxchg %ebx,(%edx) # %eax is envolved and is always reloaded
	279	&jne (&label("spin"));
	280	&mov ("eax","ebx"); # OpenSSL expects the new value
	281	&pop ("ebx");
	282	&ret ();
	283	&function_end_B("OPENSSL_atomic_add");
	284
061bebc0 AP	285	# This function can become handy under Win32 in situations when
	286	# we don't know which calling convention, __stdcall or __cdecl(*),
	287	# indirect callee is using. In C it can be deployed as
	288	#
	289	#ifdef OPENSSL_CPUID_OBJ
	290	# type OPENSSL_indirect_call(void *f,...);
	291	# ...
	292	# OPENSSL_indirect_call(func,[up to $max arguments]);
	293	#endif
	294	#
	295	# (*) it's designed to work even for __fastcall if number of
	296	# arguments is 1 or 2!
	297	&function_begin_B("OPENSSL_indirect_call");
	298	{
f9c5e5d9	299	my ($max,$i)=(7,); # $max has to be chosen as 4*n-1
061bebc0 AP	300	# in order to preserve eventual
	301	# stack alignment
	302	&push ("ebp");
	303	&mov ("ebp","esp");
	304	&sub ("esp",$max*4);
	305	&mov ("ecx",&DWP(12,"ebp"));
	306	&mov (&DWP(0,"esp"),"ecx");
	307	&mov ("edx",&DWP(16,"ebp"));
	308	&mov (&DWP(4,"esp"),"edx");
	309	for($i=2;$i<$max;$i++)
	310	{
	311	# Some copies will be redundant/bogus...
	312	&mov ("eax",&DWP(12+$i*4,"ebp"));
	313	&mov (&DWP(0+$i*4,"esp"),"eax");
	314	}
3ebf898e	315	&call_ptr (&DWP(8,"ebp"));# make the call...
061bebc0 AP	316	&mov ("esp","ebp"); # ... and just restore the stack pointer
	317	# without paying attention to what we called,
	318	# (__cdecl func) or (__stdcall one).
	319	&pop ("ebp");
	320	&ret ();
	321	}
	322	&function_end_B("OPENSSL_indirect_call");
	323
b2dba9bf AP	324	&function_begin_B("OPENSSL_cleanse");
	325	&mov ("edx",&wparam(0));
	326	&mov ("ecx",&wparam(1));
	327	&xor ("eax","eax");
	328	&cmp ("ecx",7);
	329	&jae (&label("lot"));
7676eebf AP	330	&cmp ("ecx",0);
7676eebf AP	331	&je (&label("ret"));
b2dba9bf AP	332	&set_label("little");
	333	&mov (&BP(0,"edx"),"al");
	334	&sub ("ecx",1);
	335	&lea ("edx",&DWP(1,"edx"));
	336	&jnz (&label("little"));
7676eebf	337	&set_label("ret");
b2dba9bf AP	338	&ret ();
	339
	340	&set_label("lot",16);
	341	&test ("edx",3);
	342	&jz (&label("aligned"));
	343	&mov (&BP(0,"edx"),"al");
	344	&lea ("ecx",&DWP(-1,"ecx"));
	345	&lea ("edx",&DWP(1,"edx"));
	346	&jmp (&label("lot"));
	347	&set_label("aligned");
	348	&mov (&DWP(0,"edx"),"eax");
	349	&lea ("ecx",&DWP(-4,"ecx"));
	350	&test ("ecx",-4);
	351	&lea ("edx",&DWP(4,"edx"));
	352	&jnz (&label("aligned"));
	353	&cmp ("ecx",0);
	354	&jne (&label("little"));
	355	&ret ();
	356	&function_end_B("OPENSSL_cleanse");
	357
5fabb88a AP	358	{
	359	my $lasttick = "esi";
	360	my $lastdiff = "ebx";
	361	my $out = "edi";
	362	my $cnt = "ecx";
	363	my $max = "ebp";
	364
	365	&function_begin("OPENSSL_instrument_bus");
	366	&mov ("eax",0);
	367	if ($sse2) {
	368	&picmeup("edx","OPENSSL_ia32cap_P");
	369	&bt (&DWP(0,"edx"),4);
	370	&jnc (&label("nogo")); # no TSC
	371	&bt (&DWP(0,"edx"),19);
	372	&jnc (&label("nogo")); # no CLFLUSH
	373
	374	&mov ($out,&wparam(0)); # load arguments
	375	&mov ($cnt,&wparam(1));
	376
	377	# collect 1st tick
	378	&rdtsc ();
	379	&mov ($lasttick,"eax"); # lasttick = tick
	380	&mov ($lastdiff,0); # lastdiff = 0
	381	&clflush(&DWP(0,$out));
b9064221	382	&data_byte(0xf0); # lock
5fabb88a AP	383	&add (&DWP(0,$out),$lastdiff);
	384	&jmp (&label("loop"));
	385
	386	&set_label("loop",16);
	387	&rdtsc ();
	388	&mov ("edx","eax"); # put aside tick (yes, I neglect edx)
	389	&sub ("eax",$lasttick); # diff
	390	&mov ($lasttick,"edx"); # lasttick = tick
	391	&mov ($lastdiff,"eax"); # lastdiff = diff
	392	&clflush(&DWP(0,$out));
b9064221	393	&data_byte(0xf0); # lock
5fabb88a AP	394	&add (&DWP(0,$out),"eax"); # accumulate diff
	395	&lea ($out,&DWP(4,$out)); # ++$out
	396	&sub ($cnt,1); # --$cnt
	397	&jnz (&label("loop"));
	398
	399	&mov ("eax",&wparam(1));
	400	&set_label("nogo");
	401	}
	402	&function_end("OPENSSL_instrument_bus");
	403
	404	&function_begin("OPENSSL_instrument_bus2");
	405	&mov ("eax",0);
	406	if ($sse2) {
	407	&picmeup("edx","OPENSSL_ia32cap_P");
	408	&bt (&DWP(0,"edx"),4);
	409	&jnc (&label("nogo")); # no TSC
	410	&bt (&DWP(0,"edx"),19);
	411	&jnc (&label("nogo")); # no CLFLUSH
	412
	413	&mov ($out,&wparam(0)); # load arguments
	414	&mov ($cnt,&wparam(1));
	415	&mov ($max,&wparam(2));
	416
	417	&rdtsc (); # collect 1st tick
	418	&mov ($lasttick,"eax"); # lasttick = tick
	419	&mov ($lastdiff,0); # lastdiff = 0
	420
	421	&clflush(&DWP(0,$out));
b9064221	422	&data_byte(0xf0); # lock
5fabb88a AP	423	&add (&DWP(0,$out),$lastdiff);
	424
	425	&rdtsc (); # collect 1st diff
	426	&mov ("edx","eax"); # put aside tick (yes, I neglect edx)
	427	&sub ("eax",$lasttick); # diff
	428	&mov ($lasttick,"edx"); # lasttick = tick
	429	&mov ($lastdiff,"eax"); # lastdiff = diff
	430	&jmp (&label("loop2"));
	431
	432	&set_label("loop2",16);
	433	&clflush(&DWP(0,$out));
b9064221	434	&data_byte(0xf0); # lock
5fabb88a AP	435	&add (&DWP(0,$out),"eax"); # accumulate diff
	436
	437	&sub ($max,1);
	438	&jz (&label("done2"));
	439
	440	&rdtsc ();
	441	&mov ("edx","eax"); # put aside tick (yes, I neglect edx)
	442	&sub ("eax",$lasttick); # diff
	443	&mov ($lasttick,"edx"); # lasttick = tick
	444	&cmp ("eax",$lastdiff);
	445	&mov ($lastdiff,"eax"); # lastdiff = diff
	446	&mov ("edx",0);
	447	&setne ("dl");
	448	&sub ($cnt,"edx"); # conditional --$cnt
	449	&lea ($out,&DWP(0,$out,"edx",4)); # conditional ++$out
	450	&jnz (&label("loop2"));
	451
	452	&set_label("done2");
	453	&mov ("eax",&wparam(1));
	454	&sub ("eax",$cnt);
	455	&set_label("nogo");
	456	}
	457	&function_end("OPENSSL_instrument_bus2");
	458	}
	459
301799b8 AP	460	&function_begin_B("OPENSSL_ia32_rdrand");
	461	&mov ("ecx",8);
	462	&set_label("loop");
	463	&rdrand ("eax");
	464	&jc (&label("break"));
	465	&loop (&label("loop"));
	466	&set_label("break");
	467	&cmp ("eax",0);
	468	&cmove ("eax","ecx");
	469	&ret ();
	470	&function_end_B("OPENSSL_ia32_rdrand");
	471
f4d45640 AP	472	&function_begin_B("OPENSSL_ia32_rdseed");
	473	&mov ("ecx",8);
	474	&set_label("loop");
	475	&rdseed ("eax");
	476	&jc (&label("break"));
	477	&loop (&label("loop"));
	478	&set_label("break");
	479	&cmp ("eax",0);
	480	&cmove ("eax","ecx");
	481	&ret ();
	482	&function_end_B("OPENSSL_ia32_rdseed");
	483
2b247cf8	484	&initseg("OPENSSL_cpuid_setup");
14e21f86	485
6206682a AP	486	&hidden("OPENSSL_cpuid_setup");
	487	&hidden("OPENSSL_ia32cap_P");
	488
14e21f86	489	&asm_finish();