[thirdparty/openssl.git] / crypto / x86_64cpuid.pl

#!/usr/bin/env perl

$flavour = shift;
$output  = shift;
if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }

$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
open STDOUT,"| $^X ${dir}perlasm/x86_64-xlate.pl $flavour $output";

if ($win64)	{ $arg1="%rcx"; $arg2="%rdx"; }
else		{ $arg1="%rdi"; $arg2="%rsi"; }
print<<___;
.extern		OPENSSL_cpuid_setup
.section	.init
	call	OPENSSL_cpuid_setup

.text

.globl	OPENSSL_atomic_add
.type	OPENSSL_atomic_add,\@abi-omnipotent
.align	16
OPENSSL_atomic_add:
	movl	($arg1),%eax
.Lspin:	leaq	($arg2,%rax),%r8
	.byte	0xf0		# lock
	cmpxchgl	%r8d,($arg1)
	jne	.Lspin
	movl	%r8d,%eax
	.byte	0x48,0x98	# cltq/cdqe
	ret
.size	OPENSSL_atomic_add,.-OPENSSL_atomic_add

.globl	OPENSSL_rdtsc
.type	OPENSSL_rdtsc,\@abi-omnipotent
.align	16
OPENSSL_rdtsc:
	rdtsc
	shl	\$32,%rdx
	or	%rdx,%rax
	ret
.size	OPENSSL_rdtsc,.-OPENSSL_rdtsc

.globl	OPENSSL_ia32_cpuid
.type	OPENSSL_ia32_cpuid,\@abi-omnipotent
.align	16
OPENSSL_ia32_cpuid:
	mov	%rbx,%r8

	xor	%eax,%eax
	cpuid
	xor	%eax,%eax
	cmp	\$0x756e6547,%ebx	# "Genu"
	setne	%al
	mov	%eax,%r9d
	cmp	\$0x49656e69,%edx	# "ineI"
	setne	%al
	or	%eax,%r9d
	cmp	\$0x6c65746e,%ecx	# "ntel"
	setne	%al
	or	%eax,%r9d

	mov	\$1,%eax
	cpuid
	cmp	\$0,%r9d
	jne	.Lnotintel
	or	\$0x00100000,%edx	# use reserved 20th bit to engage RC4_CHAR
	and	\$15,%ah
	cmp	\$15,%ah		# examine Family ID
	je	.Lnotintel
	or	\$0x40000000,%edx	# use reserved bit to skip unrolled loop
.Lnotintel:
	bt	\$28,%edx		# test hyper-threading bit
	jnc	.Ldone
	shr	\$16,%ebx
	cmp	\$1,%bl			# see if cache is shared
	ja	.Ldone
	and	\$0xefffffff,%edx	# ~(1<<28)
.Ldone:
	shl	\$32,%rcx
	mov	%edx,%eax
	mov	%r8,%rbx
	or	%rcx,%rax
	ret
.size	OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid

.globl  OPENSSL_cleanse
.type   OPENSSL_cleanse,\@abi-omnipotent
.align  16
OPENSSL_cleanse:
	xor	%rax,%rax
	cmp	\$15,$arg2
	jae	.Lot
.Little:
	mov	%al,($arg1)
	sub	\$1,$arg2
	lea	1($arg1),$arg1
	jnz	.Little
	ret
.align	16
.Lot:
	test	\$7,$arg1
	jz	.Laligned
	mov	%al,($arg1)
	lea	-1($arg2),$arg2
	lea	1($arg1),$arg1
	jmp	.Lot
.Laligned:
	mov	%rax,($arg1)
	lea	-8($arg2),$arg2
	test	\$-8,$arg2
	lea	8($arg1),$arg1
	jnz	.Laligned
	cmp	\$0,$arg2
	jne	.Little
	ret
.size	OPENSSL_cleanse,.-OPENSSL_cleanse
___

print<<___ if (!$win64);
.globl	OPENSSL_wipe_cpu
.type	OPENSSL_wipe_cpu,\@abi-omnipotent
.align	16
OPENSSL_wipe_cpu:
	pxor	%xmm0,%xmm0
	pxor	%xmm1,%xmm1
	pxor	%xmm2,%xmm2
	pxor	%xmm3,%xmm3
	pxor	%xmm4,%xmm4
	pxor	%xmm5,%xmm5
	pxor	%xmm6,%xmm6
	pxor	%xmm7,%xmm7
	pxor	%xmm8,%xmm8
	pxor	%xmm9,%xmm9
	pxor	%xmm10,%xmm10
	pxor	%xmm11,%xmm11
	pxor	%xmm12,%xmm12
	pxor	%xmm13,%xmm13
	pxor	%xmm14,%xmm14
	pxor	%xmm15,%xmm15
	xorq	%rcx,%rcx
	xorq	%rdx,%rdx
	xorq	%rsi,%rsi
	xorq	%rdi,%rdi
	xorq	%r8,%r8
	xorq	%r9,%r9
	xorq	%r10,%r10
	xorq	%r11,%r11
	leaq	8(%rsp),%rax
	ret
.size	OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
___
print<<___ if ($win64);
.globl	OPENSSL_wipe_cpu
.type	OPENSSL_wipe_cpu,\@abi-omnipotent
.align	16
OPENSSL_wipe_cpu:
	pxor	%xmm0,%xmm0
	pxor	%xmm1,%xmm1
	pxor	%xmm2,%xmm2
	pxor	%xmm3,%xmm3
	pxor	%xmm4,%xmm4
	pxor	%xmm5,%xmm5
	xorq	%rcx,%rcx
	xorq	%rdx,%rdx
	xorq	%r8,%r8
	xorq	%r9,%r9
	xorq	%r10,%r10
	xorq	%r11,%r11
	leaq	8(%rsp),%rax
	ret
.size	OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
___

close STDOUT;	# flush
Commit	Line	Data
14e21f86 AP	1	#!/usr/bin/env perl
14e21f86 AP	2
aa8f38e4 AP	3	$flavour = shift;
	4	$output = shift;
	5	if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
14e21f86	6
aa8f38e4	7	$win64=0; $win64=1 if ($flavour =~ /[nm]asm\|mingw64/ \|\| $output =~ /\.asm$/);
376729e1	8
aa8f38e4 AP	9	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
aa8f38e4 AP	10	open STDOUT,"\| $^X ${dir}perlasm/x86_64-xlate.pl $flavour $output";
376729e1	11
aa8f38e4 AP	12	if ($win64) { $arg1="%rcx"; $arg2="%rdx"; }
	13	else { $arg1="%rdi"; $arg2="%rsi"; }
	14	print<<___;
	15	.extern OPENSSL_cpuid_setup
	16	.section .init
	17	call OPENSSL_cpuid_setup
932cc129	18
14e21f86	19	.text
376729e1 AP	20
376729e1 AP	21	.globl OPENSSL_atomic_add
aa8f38e4	22	.type OPENSSL_atomic_add,\@abi-omnipotent
376729e1 AP	23	.align 16
376729e1 AP	24	OPENSSL_atomic_add:
aa8f38e4 AP	25	movl ($arg1),%eax
	26	.Lspin: leaq ($arg2,%rax),%r8
	27	.byte 0xf0 # lock
	28	cmpxchgl %r8d,($arg1)
376729e1	29	jne .Lspin
e442c362	30	movl %r8d,%eax
aa8f38e4	31	.byte 0x48,0x98 # cltq/cdqe
376729e1 AP	32	ret
	33	.size OPENSSL_atomic_add,.-OPENSSL_atomic_add
	34
932cc129 AP	35	.globl OPENSSL_rdtsc
	36	.type OPENSSL_rdtsc,\@abi-omnipotent
	37	.align 16
	38	OPENSSL_rdtsc:
	39	rdtsc
	40	shl \$32,%rdx
	41	or %rdx,%rax
	42	ret
	43	.size OPENSSL_rdtsc,.-OPENSSL_rdtsc
	44
376729e1	45	.globl OPENSSL_ia32_cpuid
932cc129	46	.type OPENSSL_ia32_cpuid,\@abi-omnipotent
376729e1 AP	47	.align 16
376729e1 AP	48	OPENSSL_ia32_cpuid:
932cc129	49	mov %rbx,%r8
9babf392 AP	50
	51	xor %eax,%eax
	52	cpuid
	53	xor %eax,%eax
	54	cmp \$0x756e6547,%ebx # "Genu"
	55	setne %al
	56	mov %eax,%r9d
	57	cmp \$0x49656e69,%edx # "ineI"
	58	setne %al
	59	or %eax,%r9d
	60	cmp \$0x6c65746e,%ecx # "ntel"
	61	setne %al
	62	or %eax,%r9d
	63
932cc129	64	mov \$1,%eax
376729e1	65	cpuid
932cc129	66	cmp \$0,%r9d
9babf392	67	jne .Lnotintel
55eab3b7	68	or \$0x00100000,%edx # use reserved 20th bit to engage RC4_CHAR
9babf392 AP	69	and \$15,%ah
	70	cmp \$15,%ah # examine Family ID
	71	je .Lnotintel
55eab3b7	72	or \$0x40000000,%edx # use reserved bit to skip unrolled loop
9babf392	73	.Lnotintel:
3df2eff4 AP	74	bt \$28,%edx # test hyper-threading bit
3df2eff4 AP	75	jnc .Ldone
9babf392 AP	76	shr \$16,%ebx
	77	cmp \$1,%bl # see if cache is shared
	78	ja .Ldone
932cc129	79	and \$0xefffffff,%edx # ~(1<<28)
9babf392	80	.Ldone:
932cc129 AP	81	shl \$32,%rcx
	82	mov %edx,%eax
	83	mov %r8,%rbx
	84	or %rcx,%rax
376729e1 AP	85	ret
376729e1 AP	86	.size OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
b2dba9bf AP	87
b2dba9bf AP	88	.globl OPENSSL_cleanse
aa8f38e4	89	.type OPENSSL_cleanse,\@abi-omnipotent
b2dba9bf AP	90	.align 16
	91	OPENSSL_cleanse:
	92	xor %rax,%rax
aa8f38e4	93	cmp \$15,$arg2
b2dba9bf AP	94	jae .Lot
b2dba9bf AP	95	.Little:
aa8f38e4 AP	96	mov %al,($arg1)
	97	sub \$1,$arg2
	98	lea 1($arg1),$arg1
b2dba9bf AP	99	jnz .Little
	100	ret
	101	.align 16
	102	.Lot:
aa8f38e4	103	test \$7,$arg1
b2dba9bf	104	jz .Laligned
aa8f38e4 AP	105	mov %al,($arg1)
	106	lea -1($arg2),$arg2
	107	lea 1($arg1),$arg1
b2dba9bf AP	108	jmp .Lot
b2dba9bf AP	109	.Laligned:
aa8f38e4 AP	110	mov %rax,($arg1)
	111	lea -8($arg2),$arg2
	112	test \$-8,$arg2
	113	lea 8($arg1),$arg1
b2dba9bf	114	jnz .Laligned
aa8f38e4	115	cmp \$0,$arg2
b2dba9bf AP	116	jne .Little
	117	ret
	118	.size OPENSSL_cleanse,.-OPENSSL_cleanse
14e21f86	119	___
aa8f38e4 AP	120
	121	print<<___ if (!$win64);
	122	.globl OPENSSL_wipe_cpu
	123	.type OPENSSL_wipe_cpu,\@abi-omnipotent
	124	.align 16
	125	OPENSSL_wipe_cpu:
	126	pxor %xmm0,%xmm0
	127	pxor %xmm1,%xmm1
	128	pxor %xmm2,%xmm2
	129	pxor %xmm3,%xmm3
	130	pxor %xmm4,%xmm4
	131	pxor %xmm5,%xmm5
	132	pxor %xmm6,%xmm6
	133	pxor %xmm7,%xmm7
	134	pxor %xmm8,%xmm8
	135	pxor %xmm9,%xmm9
	136	pxor %xmm10,%xmm10
	137	pxor %xmm11,%xmm11
	138	pxor %xmm12,%xmm12
	139	pxor %xmm13,%xmm13
	140	pxor %xmm14,%xmm14
	141	pxor %xmm15,%xmm15
	142	xorq %rcx,%rcx
	143	xorq %rdx,%rdx
	144	xorq %rsi,%rsi
	145	xorq %rdi,%rdi
	146	xorq %r8,%r8
	147	xorq %r9,%r9
	148	xorq %r10,%r10
	149	xorq %r11,%r11
	150	leaq 8(%rsp),%rax
	151	ret
	152	.size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
	153	___
	154	print<<___ if ($win64);
	155	.globl OPENSSL_wipe_cpu
	156	.type OPENSSL_wipe_cpu,\@abi-omnipotent
	157	.align 16
	158	OPENSSL_wipe_cpu:
	159	pxor %xmm0,%xmm0
	160	pxor %xmm1,%xmm1
	161	pxor %xmm2,%xmm2
	162	pxor %xmm3,%xmm3
	163	pxor %xmm4,%xmm4
	164	pxor %xmm5,%xmm5
	165	xorq %rcx,%rcx
	166	xorq %rdx,%rdx
	167	xorq %r8,%r8
	168	xorq %r9,%r9
	169	xorq %r10,%r10
	170	xorq %r11,%r11
	171	leaq 8(%rsp),%rax
	172	ret
	173	.size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
	174	___
	175
5d863367	176	close STDOUT; # flush