1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 *  linux/arch/arm/lib/memset.S
4 *
5 *  Copyright (C) 1995-2000 Russell King
6 *
7 *  ASM optimised string functions
8 */
9#include <linux/linkage.h>
10#include <asm/assembler.h>
11#include <asm/unwind.h>
12
13	.text
14	.align	5
15
16ENTRY(__memset)
17ENTRY(mmioset)
18WEAK(memset)
19UNWIND( .fnstart         )
20	ands	r3, r0, #3		@ 1 unaligned?
21	mov	ip, r0			@ preserve r0 as return value
22	bne	6f			@ 1
23/*
24 * we know that the pointer in ip is aligned to a word boundary.
25 */
261:	orr	r1, r1, r1, lsl #8
27	orr	r1, r1, r1, lsl #16
28	mov	r3, r1
297:	cmp	r2, #16
30	blt	4f
31
32#if ! CALGN(1)+0
33
34/*
35 * We need 2 extra registers for this loop - use r8 and the LR
36 */
37	stmfd	sp!, {r8, lr}
38UNWIND( .fnend              )
39UNWIND( .fnstart            )
40UNWIND( .save {r8, lr}      )
41	mov	r8, r1
42	mov	lr, r3
43
442:	subs	r2, r2, #64
45	stmiage	ip!, {r1, r3, r8, lr}	@ 64 bytes at a time.
46	stmiage	ip!, {r1, r3, r8, lr}
47	stmiage	ip!, {r1, r3, r8, lr}
48	stmiage	ip!, {r1, r3, r8, lr}
49	bgt	2b
50	ldmfdeq	sp!, {r8, pc}		@ Now <64 bytes to go.
51/*
52 * No need to correct the count; we're only testing bits from now on
53 */
54	tst	r2, #32
55	stmiane	ip!, {r1, r3, r8, lr}
56	stmiane	ip!, {r1, r3, r8, lr}
57	tst	r2, #16
58	stmiane	ip!, {r1, r3, r8, lr}
59	ldmfd	sp!, {r8, lr}
60UNWIND( .fnend              )
61
62#else
63
64/*
65 * This version aligns the destination pointer in order to write
66 * whole cache lines at once.
67 */
68
69	stmfd	sp!, {r4-r8, lr}
70UNWIND( .fnend                 )
71UNWIND( .fnstart               )
72UNWIND( .save {r4-r8, lr}      )
73	mov	r4, r1
74	mov	r5, r3
75	mov	r6, r1
76	mov	r7, r3
77	mov	r8, r1
78	mov	lr, r3
79
80	cmp	r2, #96
81	tstgt	ip, #31
82	ble	3f
83
84	and	r8, ip, #31
85	rsb	r8, r8, #32
86	sub	r2, r2, r8
87	movs	r8, r8, lsl #(32 - 4)
88	stmiacs	ip!, {r4, r5, r6, r7}
89	stmiami	ip!, {r4, r5}
90	tst	r8, #(1 << 30)
91	mov	r8, r1
92	strne	r1, [ip], #4
93
943:	subs	r2, r2, #64
95	stmiage	ip!, {r1, r3-r8, lr}
96	stmiage	ip!, {r1, r3-r8, lr}
97	bgt	3b
98	ldmfdeq	sp!, {r4-r8, pc}
99
100	tst	r2, #32
101	stmiane	ip!, {r1, r3-r8, lr}
102	tst	r2, #16
103	stmiane	ip!, {r4-r7}
104	ldmfd	sp!, {r4-r8, lr}
105UNWIND( .fnend                 )
106
107#endif
108
109UNWIND( .fnstart            )
1104:	tst	r2, #8
111	stmiane	ip!, {r1, r3}
112	tst	r2, #4
113	strne	r1, [ip], #4
114/*
115 * When we get here, we've got less than 4 bytes to set.  We
116 * may have an unaligned pointer as well.
117 */
1185:	tst	r2, #2
119	strbne	r1, [ip], #1
120	strbne	r1, [ip], #1
121	tst	r2, #1
122	strbne	r1, [ip], #1
123	ret	lr
124
1256:	subs	r2, r2, #4		@ 1 do we have enough
126	blt	5b			@ 1 bytes to align with?
127	cmp	r3, #2			@ 1
128	strblt	r1, [ip], #1		@ 1
129	strble	r1, [ip], #1		@ 1
130	strb	r1, [ip], #1		@ 1
131	add	r2, r2, r3		@ 1 (r2 = r2 - (4 - r3))
132	b	1b
133UNWIND( .fnend   )
134ENDPROC(memset)
135ENDPROC(mmioset)
136ENDPROC(__memset)
137
138ENTRY(__memset32)
139UNWIND( .fnstart         )
140	mov	r3, r1			@ copy r1 to r3 and fall into memset64
141UNWIND( .fnend   )
142ENDPROC(__memset32)
143ENTRY(__memset64)
144UNWIND( .fnstart         )
145	mov	ip, r0			@ preserve r0 as return value
146	b	7b			@ jump into the middle of memset
147UNWIND( .fnend   )
148ENDPROC(__memset64)
149