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

#!/usr/bin/env perl

$output=shift;
$masm=1 if ($output =~ /\.asm/);
open STDOUT,">$output" || die "can't open $output: $!";

print<<___ if(defined($masm));
_TEXT	SEGMENT
PUBLIC	OPENSSL_rdtsc

PUBLIC	OPENSSL_atomic_add
ALIGN	16
OPENSSL_atomic_add	PROC
	mov	eax,DWORD PTR[rcx]
\$Lspin:	lea	r8,DWORD PTR[rdx+rax]
lock	cmpxchg	DWORD PTR[rcx],r8d
	jne	\$Lspin
	mov	eax,r8d
	cdqe    
	ret
OPENSSL_atomic_add	ENDP

PUBLIC	OPENSSL_wipe_cpu
ALIGN	16
OPENSSL_wipe_cpu	PROC
	pxor	xmm0,xmm0
	pxor	xmm1,xmm1
	pxor	xmm2,xmm2
	pxor	xmm3,xmm3
	pxor	xmm4,xmm4
	pxor	xmm5,xmm5
	xor	rcx,rcx
	xor	rdx,rdx
	xor	r8,r8
	xor	r9,r9
	xor	r10,r10
	xor	r11,r11
	lea	rax,QWORD PTR[rsp+8]
	ret
OPENSSL_wipe_cpu	ENDP
_TEXT	ENDS

CRT\$XIU	SEGMENT
EXTRN	OPENSSL_cpuid_setup:PROC
DQ	OPENSSL_cpuid_setup
CRT\$XIU	ENDS

___
print<<___ if(!defined($masm));
.text

.globl	OPENSSL_atomic_add
.type	OPENSSL_atomic_add,\@function
.align	16
OPENSSL_atomic_add:
	movl	(%rdi),%eax
.Lspin:	leaq	(%rsi,%rax),%r8
lock;	cmpxchgl	%r8d,(%rdi)
	jne	.Lspin
	movl	%r8d,%eax
	.byte	0x48,0x98
	ret
.size	OPENSSL_atomic_add,.-OPENSSL_atomic_add

.globl	OPENSSL_wipe_cpu
.type	OPENSSL_wipe_cpu,\@function
.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

.section	.init
	call	OPENSSL_cpuid_setup

___

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

print<<___;
.text

.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,\@function,2
.align  16
OPENSSL_cleanse:
	xor	%rax,%rax
	cmp	\$15,%rsi
	jae	.Lot
.Little:
	mov	%al,(%rdi)
	sub	\$1,%rsi
	lea	1(%rdi),%rdi
	jnz	.Little
	ret
.align	16
.Lot:
	test	\$7,%rdi
	jz	.Laligned
	mov	%al,(%rdi)
	lea	-1(%rsi),%rsi
	lea	1(%rdi),%rdi
	jmp	.Lot
.Laligned:
	mov	%rax,(%rdi)
	lea	-8(%rsi),%rsi
	test	\$-8,%rsi
	lea	8(%rdi),%rdi
	jnz	.Laligned
	cmp	\$0,%rsi
	jne	.Little
	ret
.size	OPENSSL_cleanse,.-OPENSSL_cleanse
___
close STDOUT;	# flush
Commit	Line	Data
14e21f86 AP	1	#!/usr/bin/env perl
	2
	3	$output=shift;
932cc129	4	$masm=1 if ($output =~ /\.asm/);
14e21f86 AP	5	open STDOUT,">$output" \|\| die "can't open $output: $!";
14e21f86 AP	6
932cc129	7	print<<___ if(defined($masm));
376729e1	8	_TEXT SEGMENT
14e21f86	9	PUBLIC OPENSSL_rdtsc
376729e1 AP	10
	11	PUBLIC OPENSSL_atomic_add
	12	ALIGN 16
	13	OPENSSL_atomic_add PROC
	14	mov eax,DWORD PTR[rcx]
	15	\$Lspin: lea r8,DWORD PTR[rdx+rax]
	16	lock cmpxchg DWORD PTR[rcx],r8d
	17	jne \$Lspin
	18	mov eax,r8d
	19	cdqe
	20	ret
	21	OPENSSL_atomic_add ENDP
	22
	23	PUBLIC OPENSSL_wipe_cpu
	24	ALIGN 16
	25	OPENSSL_wipe_cpu PROC
	26	pxor xmm0,xmm0
	27	pxor xmm1,xmm1
	28	pxor xmm2,xmm2
	29	pxor xmm3,xmm3
	30	pxor xmm4,xmm4
	31	pxor xmm5,xmm5
	32	xor rcx,rcx
	33	xor rdx,rdx
	34	xor r8,r8
	35	xor r9,r9
	36	xor r10,r10
	37	xor r11,r11
	38	lea rax,QWORD PTR[rsp+8]
	39	ret
	40	OPENSSL_wipe_cpu ENDP
376729e1 AP	41	_TEXT ENDS
	42
	43	CRT\$XIU SEGMENT
	44	EXTRN OPENSSL_cpuid_setup:PROC
	45	DQ OPENSSL_cpuid_setup
	46	CRT\$XIU ENDS
932cc129	47
14e21f86	48	___
932cc129	49	print<<___ if(!defined($masm));
14e21f86	50	.text
376729e1 AP	51
	52	.globl OPENSSL_atomic_add
	53	.type OPENSSL_atomic_add,\@function
	54	.align 16
	55	OPENSSL_atomic_add:
	56	movl (%rdi),%eax
e442c362 AP	57	.Lspin: leaq (%rsi,%rax),%r8
e442c362 AP	58	lock; cmpxchgl %r8d,(%rdi)
376729e1	59	jne .Lspin
e442c362 AP	60	movl %r8d,%eax
e442c362 AP	61	.byte 0x48,0x98
376729e1 AP	62	ret
	63	.size OPENSSL_atomic_add,.-OPENSSL_atomic_add
	64
	65	.globl OPENSSL_wipe_cpu
	66	.type OPENSSL_wipe_cpu,\@function
	67	.align 16
	68	OPENSSL_wipe_cpu:
	69	pxor %xmm0,%xmm0
	70	pxor %xmm1,%xmm1
	71	pxor %xmm2,%xmm2
	72	pxor %xmm3,%xmm3
	73	pxor %xmm4,%xmm4
	74	pxor %xmm5,%xmm5
	75	pxor %xmm6,%xmm6
	76	pxor %xmm7,%xmm7
	77	pxor %xmm8,%xmm8
	78	pxor %xmm9,%xmm9
	79	pxor %xmm10,%xmm10
	80	pxor %xmm11,%xmm11
	81	pxor %xmm12,%xmm12
	82	pxor %xmm13,%xmm13
	83	pxor %xmm14,%xmm14
	84	pxor %xmm15,%xmm15
e442c362 AP	85	xorq %rcx,%rcx
	86	xorq %rdx,%rdx
	87	xorq %rsi,%rsi
	88	xorq %rdi,%rdi
	89	xorq %r8,%r8
	90	xorq %r9,%r9
	91	xorq %r10,%r10
	92	xorq %r11,%r11
	93	leaq 8(%rsp),%rax
376729e1 AP	94	ret
	95	.size OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
	96
932cc129 AP	97	.section .init
	98	call OPENSSL_cpuid_setup
	99
	100	___
	101
abe7f8b4	102	$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
89778b7f	103	open STDOUT,"\| $^X ${dir}perlasm/x86_64-xlate.pl $output";
abe7f8b4	104
932cc129 AP	105	print<<___;
	106	.text
	107
	108	.globl OPENSSL_rdtsc
	109	.type OPENSSL_rdtsc,\@abi-omnipotent
	110	.align 16
	111	OPENSSL_rdtsc:
	112	rdtsc
	113	shl \$32,%rdx
	114	or %rdx,%rax
	115	ret
	116	.size OPENSSL_rdtsc,.-OPENSSL_rdtsc
	117
376729e1	118	.globl OPENSSL_ia32_cpuid
932cc129	119	.type OPENSSL_ia32_cpuid,\@abi-omnipotent
376729e1 AP	120	.align 16
376729e1 AP	121	OPENSSL_ia32_cpuid:
932cc129	122	mov %rbx,%r8
9babf392 AP	123
	124	xor %eax,%eax
	125	cpuid
	126	xor %eax,%eax
	127	cmp \$0x756e6547,%ebx # "Genu"
	128	setne %al
	129	mov %eax,%r9d
	130	cmp \$0x49656e69,%edx # "ineI"
	131	setne %al
	132	or %eax,%r9d
	133	cmp \$0x6c65746e,%ecx # "ntel"
	134	setne %al
	135	or %eax,%r9d
	136
932cc129	137	mov \$1,%eax
376729e1	138	cpuid
932cc129	139	cmp \$0,%r9d
9babf392	140	jne .Lnotintel
55eab3b7	141	or \$0x00100000,%edx # use reserved 20th bit to engage RC4_CHAR
9babf392 AP	142	and \$15,%ah
	143	cmp \$15,%ah # examine Family ID
	144	je .Lnotintel
55eab3b7	145	or \$0x40000000,%edx # use reserved bit to skip unrolled loop
9babf392	146	.Lnotintel:
3df2eff4 AP	147	bt \$28,%edx # test hyper-threading bit
3df2eff4 AP	148	jnc .Ldone
9babf392 AP	149	shr \$16,%ebx
	150	cmp \$1,%bl # see if cache is shared
	151	ja .Ldone
932cc129	152	and \$0xefffffff,%edx # ~(1<<28)
9babf392	153	.Ldone:
932cc129 AP	154	shl \$32,%rcx
	155	mov %edx,%eax
	156	mov %r8,%rbx
	157	or %rcx,%rax
376729e1 AP	158	ret
376729e1 AP	159	.size OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
b2dba9bf AP	160
	161	.globl OPENSSL_cleanse
	162	.type OPENSSL_cleanse,\@function,2
	163	.align 16
	164	OPENSSL_cleanse:
	165	xor %rax,%rax
	166	cmp \$15,%rsi
	167	jae .Lot
	168	.Little:
	169	mov %al,(%rdi)
	170	sub \$1,%rsi
	171	lea 1(%rdi),%rdi
	172	jnz .Little
	173	ret
	174	.align 16
	175	.Lot:
	176	test \$7,%rdi
	177	jz .Laligned
	178	mov %al,(%rdi)
	179	lea -1(%rsi),%rsi
	180	lea 1(%rdi),%rdi
	181	jmp .Lot
	182	.Laligned:
	183	mov %rax,(%rdi)
	184	lea -8(%rsi),%rsi
	185	test \$-8,%rsi
	186	lea 8(%rdi),%rdi
	187	jnz .Laligned
	188	cmp \$0,%rsi
	189	jne .Little
	190	ret
	191	.size OPENSSL_cleanse,.-OPENSSL_cleanse
14e21f86	192	___
5d863367	193	close STDOUT; # flush