xref: /linux/arch/x86/lib/memcpy_32.c (revision 24bce201d79807b668bf9d9e0aca801c5c0d5f78)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/string.h>
3 #include <linux/export.h>
4 
5 #undef memcpy
6 #undef memset
7 #undef memmove
8 
9 __visible void *memcpy(void *to, const void *from, size_t n)
10 {
11 	return __memcpy(to, from, n);
12 }
13 EXPORT_SYMBOL(memcpy);
14 
15 __visible void *memset(void *s, int c, size_t count)
16 {
17 	return __memset(s, c, count);
18 }
19 EXPORT_SYMBOL(memset);
20 
21 __visible void *memmove(void *dest, const void *src, size_t n)
22 {
23 	int d0,d1,d2,d3,d4,d5;
24 	char *ret = dest;
25 
26 	__asm__ __volatile__(
27 		/* Handle more 16 bytes in loop */
28 		"cmp $0x10, %0\n\t"
29 		"jb	1f\n\t"
30 
31 		/* Decide forward/backward copy mode */
32 		"cmp %2, %1\n\t"
33 		"jb	2f\n\t"
34 
35 		/*
36 		 * movs instruction have many startup latency
37 		 * so we handle small size by general register.
38 		 */
39 		"cmp  $680, %0\n\t"
40 		"jb 3f\n\t"
41 		/*
42 		 * movs instruction is only good for aligned case.
43 		 */
44 		"mov %1, %3\n\t"
45 		"xor %2, %3\n\t"
46 		"and $0xff, %3\n\t"
47 		"jz 4f\n\t"
48 		"3:\n\t"
49 		"sub $0x10, %0\n\t"
50 
51 		/*
52 		 * We gobble 16 bytes forward in each loop.
53 		 */
54 		"3:\n\t"
55 		"sub $0x10, %0\n\t"
56 		"mov 0*4(%1), %3\n\t"
57 		"mov 1*4(%1), %4\n\t"
58 		"mov  %3, 0*4(%2)\n\t"
59 		"mov  %4, 1*4(%2)\n\t"
60 		"mov 2*4(%1), %3\n\t"
61 		"mov 3*4(%1), %4\n\t"
62 		"mov  %3, 2*4(%2)\n\t"
63 		"mov  %4, 3*4(%2)\n\t"
64 		"lea  0x10(%1), %1\n\t"
65 		"lea  0x10(%2), %2\n\t"
66 		"jae 3b\n\t"
67 		"add $0x10, %0\n\t"
68 		"jmp 1f\n\t"
69 
70 		/*
71 		 * Handle data forward by movs.
72 		 */
73 		".p2align 4\n\t"
74 		"4:\n\t"
75 		"mov -4(%1, %0), %3\n\t"
76 		"lea -4(%2, %0), %4\n\t"
77 		"shr $2, %0\n\t"
78 		"rep movsl\n\t"
79 		"mov %3, (%4)\n\t"
80 		"jmp 11f\n\t"
81 		/*
82 		 * Handle data backward by movs.
83 		 */
84 		".p2align 4\n\t"
85 		"6:\n\t"
86 		"mov (%1), %3\n\t"
87 		"mov %2, %4\n\t"
88 		"lea -4(%1, %0), %1\n\t"
89 		"lea -4(%2, %0), %2\n\t"
90 		"shr $2, %0\n\t"
91 		"std\n\t"
92 		"rep movsl\n\t"
93 		"mov %3,(%4)\n\t"
94 		"cld\n\t"
95 		"jmp 11f\n\t"
96 
97 		/*
98 		 * Start to prepare for backward copy.
99 		 */
100 		".p2align 4\n\t"
101 		"2:\n\t"
102 		"cmp  $680, %0\n\t"
103 		"jb 5f\n\t"
104 		"mov %1, %3\n\t"
105 		"xor %2, %3\n\t"
106 		"and $0xff, %3\n\t"
107 		"jz 6b\n\t"
108 
109 		/*
110 		 * Calculate copy position to tail.
111 		 */
112 		"5:\n\t"
113 		"add %0, %1\n\t"
114 		"add %0, %2\n\t"
115 		"sub $0x10, %0\n\t"
116 
117 		/*
118 		 * We gobble 16 bytes backward in each loop.
119 		 */
120 		"7:\n\t"
121 		"sub $0x10, %0\n\t"
122 
123 		"mov -1*4(%1), %3\n\t"
124 		"mov -2*4(%1), %4\n\t"
125 		"mov  %3, -1*4(%2)\n\t"
126 		"mov  %4, -2*4(%2)\n\t"
127 		"mov -3*4(%1), %3\n\t"
128 		"mov -4*4(%1), %4\n\t"
129 		"mov  %3, -3*4(%2)\n\t"
130 		"mov  %4, -4*4(%2)\n\t"
131 		"lea  -0x10(%1), %1\n\t"
132 		"lea  -0x10(%2), %2\n\t"
133 		"jae 7b\n\t"
134 		/*
135 		 * Calculate copy position to head.
136 		 */
137 		"add $0x10, %0\n\t"
138 		"sub %0, %1\n\t"
139 		"sub %0, %2\n\t"
140 
141 		/*
142 		 * Move data from 8 bytes to 15 bytes.
143 		 */
144 		".p2align 4\n\t"
145 		"1:\n\t"
146 		"cmp $8, %0\n\t"
147 		"jb 8f\n\t"
148 		"mov 0*4(%1), %3\n\t"
149 		"mov 1*4(%1), %4\n\t"
150 		"mov -2*4(%1, %0), %5\n\t"
151 		"mov -1*4(%1, %0), %1\n\t"
152 
153 		"mov  %3, 0*4(%2)\n\t"
154 		"mov  %4, 1*4(%2)\n\t"
155 		"mov  %5, -2*4(%2, %0)\n\t"
156 		"mov  %1, -1*4(%2, %0)\n\t"
157 		"jmp 11f\n\t"
158 
159 		/*
160 		 * Move data from 4 bytes to 7 bytes.
161 		 */
162 		".p2align 4\n\t"
163 		"8:\n\t"
164 		"cmp $4, %0\n\t"
165 		"jb 9f\n\t"
166 		"mov 0*4(%1), %3\n\t"
167 		"mov -1*4(%1, %0), %4\n\t"
168 		"mov  %3, 0*4(%2)\n\t"
169 		"mov  %4, -1*4(%2, %0)\n\t"
170 		"jmp 11f\n\t"
171 
172 		/*
173 		 * Move data from 2 bytes to 3 bytes.
174 		 */
175 		".p2align 4\n\t"
176 		"9:\n\t"
177 		"cmp $2, %0\n\t"
178 		"jb 10f\n\t"
179 		"movw 0*2(%1), %%dx\n\t"
180 		"movw -1*2(%1, %0), %%bx\n\t"
181 		"movw %%dx, 0*2(%2)\n\t"
182 		"movw %%bx, -1*2(%2, %0)\n\t"
183 		"jmp 11f\n\t"
184 
185 		/*
186 		 * Move data for 1 byte.
187 		 */
188 		".p2align 4\n\t"
189 		"10:\n\t"
190 		"cmp $1, %0\n\t"
191 		"jb 11f\n\t"
192 		"movb (%1), %%cl\n\t"
193 		"movb %%cl, (%2)\n\t"
194 		".p2align 4\n\t"
195 		"11:"
196 		: "=&c" (d0), "=&S" (d1), "=&D" (d2),
197 		  "=r" (d3),"=r" (d4), "=r"(d5)
198 		:"0" (n),
199 		 "1" (src),
200 		 "2" (dest)
201 		:"memory");
202 
203 	return ret;
204 
205 }
206 EXPORT_SYMBOL(memmove);
207