xref: /openbmc/linux/arch/x86/lib/memmove_64.S (revision d9fd5a71)
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 */
27SYM_FUNC_START_WEAK(memmove)
28SYM_FUNC_START(__memmove)
29
30	mov %rdi, %rax
31
32	/* Decide forward/backward copy mode */
33	cmp %rdi, %rsi
34	jge .Lmemmove_begin_forward
35	mov %rsi, %r8
36	add %rdx, %r8
37	cmp %rdi, %r8
38	jg 2f
39
40	/* FSRM implies ERMS => no length checks, do the copy directly */
41.Lmemmove_begin_forward:
42	ALTERNATIVE "cmp $0x20, %rdx; jb 1f", "", X86_FEATURE_FSRM
43	ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
44
45	/*
46	 * movsq instruction have many startup latency
47	 * so we handle small size by general register.
48	 */
49	cmp  $680, %rdx
50	jb	3f
51	/*
52	 * movsq instruction is only good for aligned case.
53	 */
54
55	cmpb %dil, %sil
56	je 4f
573:
58	sub $0x20, %rdx
59	/*
60	 * We gobble 32 bytes forward in each loop.
61	 */
625:
63	sub $0x20, %rdx
64	movq 0*8(%rsi), %r11
65	movq 1*8(%rsi), %r10
66	movq 2*8(%rsi), %r9
67	movq 3*8(%rsi), %r8
68	leaq 4*8(%rsi), %rsi
69
70	movq %r11, 0*8(%rdi)
71	movq %r10, 1*8(%rdi)
72	movq %r9, 2*8(%rdi)
73	movq %r8, 3*8(%rdi)
74	leaq 4*8(%rdi), %rdi
75	jae 5b
76	addq $0x20, %rdx
77	jmp 1f
78	/*
79	 * Handle data forward by movsq.
80	 */
81	.p2align 4
824:
83	movq %rdx, %rcx
84	movq -8(%rsi, %rdx), %r11
85	lea -8(%rdi, %rdx), %r10
86	shrq $3, %rcx
87	rep movsq
88	movq %r11, (%r10)
89	jmp 13f
90.Lmemmove_end_forward:
91
92	/*
93	 * Handle data backward by movsq.
94	 */
95	.p2align 4
967:
97	movq %rdx, %rcx
98	movq (%rsi), %r11
99	movq %rdi, %r10
100	leaq -8(%rsi, %rdx), %rsi
101	leaq -8(%rdi, %rdx), %rdi
102	shrq $3, %rcx
103	std
104	rep movsq
105	cld
106	movq %r11, (%r10)
107	jmp 13f
108
109	/*
110	 * Start to prepare for backward copy.
111	 */
112	.p2align 4
1132:
114	cmp $0x20, %rdx
115	jb 1f
116	cmp $680, %rdx
117	jb 6f
118	cmp %dil, %sil
119	je 7b
1206:
121	/*
122	 * Calculate copy position to tail.
123	 */
124	addq %rdx, %rsi
125	addq %rdx, %rdi
126	subq $0x20, %rdx
127	/*
128	 * We gobble 32 bytes backward in each loop.
129	 */
1308:
131	subq $0x20, %rdx
132	movq -1*8(%rsi), %r11
133	movq -2*8(%rsi), %r10
134	movq -3*8(%rsi), %r9
135	movq -4*8(%rsi), %r8
136	leaq -4*8(%rsi), %rsi
137
138	movq %r11, -1*8(%rdi)
139	movq %r10, -2*8(%rdi)
140	movq %r9, -3*8(%rdi)
141	movq %r8, -4*8(%rdi)
142	leaq -4*8(%rdi), %rdi
143	jae 8b
144	/*
145	 * Calculate copy position to head.
146	 */
147	addq $0x20, %rdx
148	subq %rdx, %rsi
149	subq %rdx, %rdi
1501:
151	cmpq $16, %rdx
152	jb 9f
153	/*
154	 * Move data from 16 bytes to 31 bytes.
155	 */
156	movq 0*8(%rsi), %r11
157	movq 1*8(%rsi), %r10
158	movq -2*8(%rsi, %rdx), %r9
159	movq -1*8(%rsi, %rdx), %r8
160	movq %r11, 0*8(%rdi)
161	movq %r10, 1*8(%rdi)
162	movq %r9, -2*8(%rdi, %rdx)
163	movq %r8, -1*8(%rdi, %rdx)
164	jmp 13f
165	.p2align 4
1669:
167	cmpq $8, %rdx
168	jb 10f
169	/*
170	 * Move data from 8 bytes to 15 bytes.
171	 */
172	movq 0*8(%rsi), %r11
173	movq -1*8(%rsi, %rdx), %r10
174	movq %r11, 0*8(%rdi)
175	movq %r10, -1*8(%rdi, %rdx)
176	jmp 13f
17710:
178	cmpq $4, %rdx
179	jb 11f
180	/*
181	 * Move data from 4 bytes to 7 bytes.
182	 */
183	movl (%rsi), %r11d
184	movl -4(%rsi, %rdx), %r10d
185	movl %r11d, (%rdi)
186	movl %r10d, -4(%rdi, %rdx)
187	jmp 13f
18811:
189	cmp $2, %rdx
190	jb 12f
191	/*
192	 * Move data from 2 bytes to 3 bytes.
193	 */
194	movw (%rsi), %r11w
195	movw -2(%rsi, %rdx), %r10w
196	movw %r11w, (%rdi)
197	movw %r10w, -2(%rdi, %rdx)
198	jmp 13f
19912:
200	cmp $1, %rdx
201	jb 13f
202	/*
203	 * Move data for 1 byte.
204	 */
205	movb (%rsi), %r11b
206	movb %r11b, (%rdi)
20713:
208	retq
209SYM_FUNC_END(__memmove)
210SYM_FUNC_END_ALIAS(memmove)
211EXPORT_SYMBOL(__memmove)
212EXPORT_SYMBOL(memmove)
213