xref: /linux/arch/x86/lib/memmove_64.S (revision f055268e3946555deb9bb80b2c8c9798c64dbc47)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Normally compiler builtins are used, but sometimes the compiler calls out
4 * of line code. Based on asm-i386/string.h.
5 *
6 * This assembly file is re-written from memmove_64.c file.
7 *	- Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
8 */
9#include <linux/linkage.h>
10#include <asm/cpufeatures.h>
11#include <asm/alternative.h>
12#include <asm/export.h>
13
14#undef memmove
15
16/*
17 * Implement memmove(). This can handle overlap between src and dst.
18 *
19 * Input:
20 * rdi: dest
21 * rsi: src
22 * rdx: count
23 *
24 * Output:
25 * rax: dest
26 */
27SYM_FUNC_START(__memmove)
28
29	mov %rdi, %rax
30
31	/* Decide forward/backward copy mode */
32	cmp %rdi, %rsi
33	jge .Lmemmove_begin_forward
34	mov %rsi, %r8
35	add %rdx, %r8
36	cmp %rdi, %r8
37	jg 2f
38
39	/* FSRM implies ERMS => no length checks, do the copy directly */
40.Lmemmove_begin_forward:
41	ALTERNATIVE "cmp $0x20, %rdx; jb 1f", "", X86_FEATURE_FSRM
42	ALTERNATIVE "", "jmp .Lmemmove_erms", X86_FEATURE_ERMS
43
44	/*
45	 * movsq instruction have many startup latency
46	 * so we handle small size by general register.
47	 */
48	cmp  $680, %rdx
49	jb	3f
50	/*
51	 * movsq instruction is only good for aligned case.
52	 */
53
54	cmpb %dil, %sil
55	je 4f
563:
57	sub $0x20, %rdx
58	/*
59	 * We gobble 32 bytes forward in each loop.
60	 */
615:
62	sub $0x20, %rdx
63	movq 0*8(%rsi), %r11
64	movq 1*8(%rsi), %r10
65	movq 2*8(%rsi), %r9
66	movq 3*8(%rsi), %r8
67	leaq 4*8(%rsi), %rsi
68
69	movq %r11, 0*8(%rdi)
70	movq %r10, 1*8(%rdi)
71	movq %r9, 2*8(%rdi)
72	movq %r8, 3*8(%rdi)
73	leaq 4*8(%rdi), %rdi
74	jae 5b
75	addq $0x20, %rdx
76	jmp 1f
77	/*
78	 * Handle data forward by movsq.
79	 */
80	.p2align 4
814:
82	movq %rdx, %rcx
83	movq -8(%rsi, %rdx), %r11
84	lea -8(%rdi, %rdx), %r10
85	shrq $3, %rcx
86	rep movsq
87	movq %r11, (%r10)
88	jmp 13f
89.Lmemmove_end_forward:
90
91	/*
92	 * Handle data backward by movsq.
93	 */
94	.p2align 4
957:
96	movq %rdx, %rcx
97	movq (%rsi), %r11
98	movq %rdi, %r10
99	leaq -8(%rsi, %rdx), %rsi
100	leaq -8(%rdi, %rdx), %rdi
101	shrq $3, %rcx
102	std
103	rep movsq
104	cld
105	movq %r11, (%r10)
106	jmp 13f
107
108	/*
109	 * Start to prepare for backward copy.
110	 */
111	.p2align 4
1122:
113	cmp $0x20, %rdx
114	jb 1f
115	cmp $680, %rdx
116	jb 6f
117	cmp %dil, %sil
118	je 7b
1196:
120	/*
121	 * Calculate copy position to tail.
122	 */
123	addq %rdx, %rsi
124	addq %rdx, %rdi
125	subq $0x20, %rdx
126	/*
127	 * We gobble 32 bytes backward in each loop.
128	 */
1298:
130	subq $0x20, %rdx
131	movq -1*8(%rsi), %r11
132	movq -2*8(%rsi), %r10
133	movq -3*8(%rsi), %r9
134	movq -4*8(%rsi), %r8
135	leaq -4*8(%rsi), %rsi
136
137	movq %r11, -1*8(%rdi)
138	movq %r10, -2*8(%rdi)
139	movq %r9, -3*8(%rdi)
140	movq %r8, -4*8(%rdi)
141	leaq -4*8(%rdi), %rdi
142	jae 8b
143	/*
144	 * Calculate copy position to head.
145	 */
146	addq $0x20, %rdx
147	subq %rdx, %rsi
148	subq %rdx, %rdi
1491:
150	cmpq $16, %rdx
151	jb 9f
152	/*
153	 * Move data from 16 bytes to 31 bytes.
154	 */
155	movq 0*8(%rsi), %r11
156	movq 1*8(%rsi), %r10
157	movq -2*8(%rsi, %rdx), %r9
158	movq -1*8(%rsi, %rdx), %r8
159	movq %r11, 0*8(%rdi)
160	movq %r10, 1*8(%rdi)
161	movq %r9, -2*8(%rdi, %rdx)
162	movq %r8, -1*8(%rdi, %rdx)
163	jmp 13f
164	.p2align 4
1659:
166	cmpq $8, %rdx
167	jb 10f
168	/*
169	 * Move data from 8 bytes to 15 bytes.
170	 */
171	movq 0*8(%rsi), %r11
172	movq -1*8(%rsi, %rdx), %r10
173	movq %r11, 0*8(%rdi)
174	movq %r10, -1*8(%rdi, %rdx)
175	jmp 13f
17610:
177	cmpq $4, %rdx
178	jb 11f
179	/*
180	 * Move data from 4 bytes to 7 bytes.
181	 */
182	movl (%rsi), %r11d
183	movl -4(%rsi, %rdx), %r10d
184	movl %r11d, (%rdi)
185	movl %r10d, -4(%rdi, %rdx)
186	jmp 13f
18711:
188	cmp $2, %rdx
189	jb 12f
190	/*
191	 * Move data from 2 bytes to 3 bytes.
192	 */
193	movw (%rsi), %r11w
194	movw -2(%rsi, %rdx), %r10w
195	movw %r11w, (%rdi)
196	movw %r10w, -2(%rdi, %rdx)
197	jmp 13f
19812:
199	cmp $1, %rdx
200	jb 13f
201	/*
202	 * Move data for 1 byte.
203	 */
204	movb (%rsi), %r11b
205	movb %r11b, (%rdi)
20613:
207	RET
208
209.Lmemmove_erms:
210	movq %rdx, %rcx
211	rep movsb
212	RET
213SYM_FUNC_END(__memmove)
214EXPORT_SYMBOL(__memmove)
215
216SYM_FUNC_ALIAS_WEAK(memmove, __memmove)
217EXPORT_SYMBOL(memmove)
218