1/* SPDX-License-Identifier: GPL-2.0 */
2/* Written 2003 by Andi Kleen, based on a kernel by Evandro Menezes */
3
4#include <linux/linkage.h>
5#include <asm/cpufeatures.h>
6#include <asm/alternative-asm.h>
7#include <asm/export.h>
8
9/*
10 * Some CPUs run faster using the string copy instructions (sane microcode).
11 * It is also a lot simpler. Use this when possible. But, don't use streaming
12 * copy unless the CPU indicates X86_FEATURE_REP_GOOD. Could vary the
13 * prefetch distance based on SMP/UP.
14 */
15	ALIGN
16SYM_FUNC_START(copy_page)
17	ALTERNATIVE "jmp copy_page_regs", "", X86_FEATURE_REP_GOOD
18	movl	$4096/8, %ecx
19	rep	movsq
20	ret
21SYM_FUNC_END(copy_page)
22EXPORT_SYMBOL(copy_page)
23
24SYM_FUNC_START_LOCAL(copy_page_regs)
25	subq	$2*8,	%rsp
26	movq	%rbx,	(%rsp)
27	movq	%r12,	1*8(%rsp)
28
29	movl	$(4096/64)-5,	%ecx
30	.p2align 4
31.Loop64:
32	dec	%rcx
33	movq	0x8*0(%rsi), %rax
34	movq	0x8*1(%rsi), %rbx
35	movq	0x8*2(%rsi), %rdx
36	movq	0x8*3(%rsi), %r8
37	movq	0x8*4(%rsi), %r9
38	movq	0x8*5(%rsi), %r10
39	movq	0x8*6(%rsi), %r11
40	movq	0x8*7(%rsi), %r12
41
42	prefetcht0 5*64(%rsi)
43
44	movq	%rax, 0x8*0(%rdi)
45	movq	%rbx, 0x8*1(%rdi)
46	movq	%rdx, 0x8*2(%rdi)
47	movq	%r8,  0x8*3(%rdi)
48	movq	%r9,  0x8*4(%rdi)
49	movq	%r10, 0x8*5(%rdi)
50	movq	%r11, 0x8*6(%rdi)
51	movq	%r12, 0x8*7(%rdi)
52
53	leaq	64 (%rsi), %rsi
54	leaq	64 (%rdi), %rdi
55
56	jnz	.Loop64
57
58	movl	$5, %ecx
59	.p2align 4
60.Loop2:
61	decl	%ecx
62
63	movq	0x8*0(%rsi), %rax
64	movq	0x8*1(%rsi), %rbx
65	movq	0x8*2(%rsi), %rdx
66	movq	0x8*3(%rsi), %r8
67	movq	0x8*4(%rsi), %r9
68	movq	0x8*5(%rsi), %r10
69	movq	0x8*6(%rsi), %r11
70	movq	0x8*7(%rsi), %r12
71
72	movq	%rax, 0x8*0(%rdi)
73	movq	%rbx, 0x8*1(%rdi)
74	movq	%rdx, 0x8*2(%rdi)
75	movq	%r8,  0x8*3(%rdi)
76	movq	%r9,  0x8*4(%rdi)
77	movq	%r10, 0x8*5(%rdi)
78	movq	%r11, 0x8*6(%rdi)
79	movq	%r12, 0x8*7(%rdi)
80
81	leaq	64(%rdi), %rdi
82	leaq	64(%rsi), %rsi
83	jnz	.Loop2
84
85	movq	(%rsp), %rbx
86	movq	1*8(%rsp), %r12
87	addq	$2*8, %rsp
88	ret
89SYM_FUNC_END(copy_page_regs)
90