xref: /linux/arch/arm/lib/copy_page.S (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1/*
2 *  linux/arch/arm/lib/copypage.S
3 *
4 *  Copyright (C) 1995-1999 Russell King
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 *  ASM optimised string functions
11 */
12#include <linux/linkage.h>
13#include <asm/assembler.h>
14#include <asm/asm-offsets.h>
15#include <asm/cache.h>
16#include <asm/export.h>
17
18#define COPY_COUNT (PAGE_SZ / (2 * L1_CACHE_BYTES) PLD( -1 ))
19
20		.text
21		.align	5
22/*
23 * StrongARM optimised copy_page routine
24 * now 1.78bytes/cycle, was 1.60 bytes/cycle (50MHz bus -> 89MB/s)
25 * Note that we probably achieve closer to the 100MB/s target with
26 * the core clock switching.
27 */
28ENTRY(copy_page)
29		stmfd	sp!, {r4, lr}			@	2
30	PLD(	pld	[r1, #0]		)
31	PLD(	pld	[r1, #L1_CACHE_BYTES]		)
32		mov	r2, #COPY_COUNT			@	1
33		ldmia	r1!, {r3, r4, ip, lr}		@	4+1
341:	PLD(	pld	[r1, #2 * L1_CACHE_BYTES])
35	PLD(	pld	[r1, #3 * L1_CACHE_BYTES])
362:
37	.rept	(2 * L1_CACHE_BYTES / 16 - 1)
38		stmia	r0!, {r3, r4, ip, lr}		@	4
39		ldmia	r1!, {r3, r4, ip, lr}		@	4
40	.endr
41		subs	r2, r2, #1			@	1
42		stmia	r0!, {r3, r4, ip, lr}		@	4
43		ldmgtia	r1!, {r3, r4, ip, lr}		@	4
44		bgt	1b				@	1
45	PLD(	ldmeqia r1!, {r3, r4, ip, lr}	)
46	PLD(	beq	2b			)
47		ldmfd	sp!, {r4, pc}			@	3
48ENDPROC(copy_page)
49EXPORT_SYMBOL(copy_page)
50