xref: /linux/arch/x86/lib/memmove_64.S (revision 37744feebc086908fd89760650f458ab19071750)
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-asm.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 */
27.weak memmove
28
29SYM_FUNC_START_ALIAS(memmove)
30SYM_FUNC_START(__memmove)
31
32	mov %rdi, %rax
33
34	/* Decide forward/backward copy mode */
35	cmp %rdi, %rsi
36	jge .Lmemmove_begin_forward
37	mov %rsi, %r8
38	add %rdx, %r8
39	cmp %rdi, %r8
40	jg 2f
41
42	/* FSRM implies ERMS => no length checks, do the copy directly */
43.Lmemmove_begin_forward:
44	ALTERNATIVE "cmp $0x20, %rdx; jb 1f", "", X86_FEATURE_FSRM
45	ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
46
47	/*
48	 * movsq instruction have many startup latency
49	 * so we handle small size by general register.
50	 */
51	cmp  $680, %rdx
52	jb	3f
53	/*
54	 * movsq instruction is only good for aligned case.
55	 */
56
57	cmpb %dil, %sil
58	je 4f
593:
60	sub $0x20, %rdx
61	/*
62	 * We gobble 32 bytes forward in each loop.
63	 */
645:
65	sub $0x20, %rdx
66	movq 0*8(%rsi), %r11
67	movq 1*8(%rsi), %r10
68	movq 2*8(%rsi), %r9
69	movq 3*8(%rsi), %r8
70	leaq 4*8(%rsi), %rsi
71
72	movq %r11, 0*8(%rdi)
73	movq %r10, 1*8(%rdi)
74	movq %r9, 2*8(%rdi)
75	movq %r8, 3*8(%rdi)
76	leaq 4*8(%rdi), %rdi
77	jae 5b
78	addq $0x20, %rdx
79	jmp 1f
80	/*
81	 * Handle data forward by movsq.
82	 */
83	.p2align 4
844:
85	movq %rdx, %rcx
86	movq -8(%rsi, %rdx), %r11
87	lea -8(%rdi, %rdx), %r10
88	shrq $3, %rcx
89	rep movsq
90	movq %r11, (%r10)
91	jmp 13f
92.Lmemmove_end_forward:
93
94	/*
95	 * Handle data backward by movsq.
96	 */
97	.p2align 4
987:
99	movq %rdx, %rcx
100	movq (%rsi), %r11
101	movq %rdi, %r10
102	leaq -8(%rsi, %rdx), %rsi
103	leaq -8(%rdi, %rdx), %rdi
104	shrq $3, %rcx
105	std
106	rep movsq
107	cld
108	movq %r11, (%r10)
109	jmp 13f
110
111	/*
112	 * Start to prepare for backward copy.
113	 */
114	.p2align 4
1152:
116	cmp $0x20, %rdx
117	jb 1f
118	cmp $680, %rdx
119	jb 6f
120	cmp %dil, %sil
121	je 7b
1226:
123	/*
124	 * Calculate copy position to tail.
125	 */
126	addq %rdx, %rsi
127	addq %rdx, %rdi
128	subq $0x20, %rdx
129	/*
130	 * We gobble 32 bytes backward in each loop.
131	 */
1328:
133	subq $0x20, %rdx
134	movq -1*8(%rsi), %r11
135	movq -2*8(%rsi), %r10
136	movq -3*8(%rsi), %r9
137	movq -4*8(%rsi), %r8
138	leaq -4*8(%rsi), %rsi
139
140	movq %r11, -1*8(%rdi)
141	movq %r10, -2*8(%rdi)
142	movq %r9, -3*8(%rdi)
143	movq %r8, -4*8(%rdi)
144	leaq -4*8(%rdi), %rdi
145	jae 8b
146	/*
147	 * Calculate copy position to head.
148	 */
149	addq $0x20, %rdx
150	subq %rdx, %rsi
151	subq %rdx, %rdi
1521:
153	cmpq $16, %rdx
154	jb 9f
155	/*
156	 * Move data from 16 bytes to 31 bytes.
157	 */
158	movq 0*8(%rsi), %r11
159	movq 1*8(%rsi), %r10
160	movq -2*8(%rsi, %rdx), %r9
161	movq -1*8(%rsi, %rdx), %r8
162	movq %r11, 0*8(%rdi)
163	movq %r10, 1*8(%rdi)
164	movq %r9, -2*8(%rdi, %rdx)
165	movq %r8, -1*8(%rdi, %rdx)
166	jmp 13f
167	.p2align 4
1689:
169	cmpq $8, %rdx
170	jb 10f
171	/*
172	 * Move data from 8 bytes to 15 bytes.
173	 */
174	movq 0*8(%rsi), %r11
175	movq -1*8(%rsi, %rdx), %r10
176	movq %r11, 0*8(%rdi)
177	movq %r10, -1*8(%rdi, %rdx)
178	jmp 13f
17910:
180	cmpq $4, %rdx
181	jb 11f
182	/*
183	 * Move data from 4 bytes to 7 bytes.
184	 */
185	movl (%rsi), %r11d
186	movl -4(%rsi, %rdx), %r10d
187	movl %r11d, (%rdi)
188	movl %r10d, -4(%rdi, %rdx)
189	jmp 13f
19011:
191	cmp $2, %rdx
192	jb 12f
193	/*
194	 * Move data from 2 bytes to 3 bytes.
195	 */
196	movw (%rsi), %r11w
197	movw -2(%rsi, %rdx), %r10w
198	movw %r11w, (%rdi)
199	movw %r10w, -2(%rdi, %rdx)
200	jmp 13f
20112:
202	cmp $1, %rdx
203	jb 13f
204	/*
205	 * Move data for 1 byte.
206	 */
207	movb (%rsi), %r11b
208	movb %r11b, (%rdi)
20913:
210	retq
211SYM_FUNC_END(__memmove)
212SYM_FUNC_END_ALIAS(memmove)
213EXPORT_SYMBOL(__memmove)
214EXPORT_SYMBOL(memmove)
215