xref: /openbmc/linux/arch/microblaze/lib/memcpy.c (revision 2cf1c348)
1 /*
2  * Copyright (C) 2008-2009 Michal Simek <monstr@monstr.eu>
3  * Copyright (C) 2008-2009 PetaLogix
4  * Copyright (C) 2007 John Williams
5  *
6  * Reasonably optimised generic C-code for memcpy on Microblaze
7  * This is generic C code to do efficient, alignment-aware memcpy.
8  *
9  * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
10  * http://www.embedded.com/showArticle.jhtml?articleID=19205567
11  *
12  * Attempts were made, unsuccessfully, to contact the original
13  * author of this code (Michael Morrow, Intel).  Below is the original
14  * copyright notice.
15  *
16  * This software has been developed by Intel Corporation.
17  * Intel specifically disclaims all warranties, express or
18  * implied, and all liability, including consequential and
19  * other indirect damages, for the use of this program, including
20  * liability for infringement of any proprietary rights,
21  * and including the warranties of merchantability and fitness
22  * for a particular purpose. Intel does not assume any
23  * responsibility for and errors which may appear in this program
24  * not any responsibility to update it.
25  */
26 
27 #include <linux/export.h>
28 #include <linux/types.h>
29 #include <linux/stddef.h>
30 #include <linux/compiler.h>
31 
32 #include <linux/string.h>
33 
34 #ifdef __HAVE_ARCH_MEMCPY
35 #ifndef CONFIG_OPT_LIB_FUNCTION
36 void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
37 {
38 	const char *src = v_src;
39 	char *dst = v_dst;
40 
41 	/* Simple, byte oriented memcpy. */
42 	while (c--)
43 		*dst++ = *src++;
44 
45 	return v_dst;
46 }
47 #else /* CONFIG_OPT_LIB_FUNCTION */
48 void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
49 {
50 	const char *src = v_src;
51 	char *dst = v_dst;
52 
53 	/* The following code tries to optimize the copy by using unsigned
54 	 * alignment. This will work fine if both source and destination are
55 	 * aligned on the same boundary. However, if they are aligned on
56 	 * different boundaries shifts will be necessary. This might result in
57 	 * bad performance on MicroBlaze systems without a barrel shifter.
58 	 */
59 	const uint32_t *i_src;
60 	uint32_t *i_dst;
61 
62 	if (likely(c >= 4)) {
63 		unsigned  value, buf_hold;
64 
65 		/* Align the destination to a word boundary. */
66 		/* This is done in an endian independent manner. */
67 		switch ((unsigned long)dst & 3) {
68 		case 1:
69 			*dst++ = *src++;
70 			--c;
71 			fallthrough;
72 		case 2:
73 			*dst++ = *src++;
74 			--c;
75 			fallthrough;
76 		case 3:
77 			*dst++ = *src++;
78 			--c;
79 		}
80 
81 		i_dst = (void *)dst;
82 
83 		/* Choose a copy scheme based on the source */
84 		/* alignment relative to destination. */
85 		switch ((unsigned long)src & 3) {
86 		case 0x0:	/* Both byte offsets are aligned */
87 			i_src  = (const void *)src;
88 
89 			for (; c >= 4; c -= 4)
90 				*i_dst++ = *i_src++;
91 
92 			src  = (const void *)i_src;
93 			break;
94 		case 0x1:	/* Unaligned - Off by 1 */
95 			/* Word align the source */
96 			i_src = (const void *) ((unsigned)src & ~3);
97 #ifndef __MICROBLAZEEL__
98 			/* Load the holding buffer */
99 			buf_hold = *i_src++ << 8;
100 
101 			for (; c >= 4; c -= 4) {
102 				value = *i_src++;
103 				*i_dst++ = buf_hold | value >> 24;
104 				buf_hold = value << 8;
105 			}
106 #else
107 			/* Load the holding buffer */
108 			buf_hold = (*i_src++ & 0xFFFFFF00) >> 8;
109 
110 			for (; c >= 4; c -= 4) {
111 				value = *i_src++;
112 				*i_dst++ = buf_hold | ((value & 0xFF) << 24);
113 				buf_hold = (value & 0xFFFFFF00) >> 8;
114 			}
115 #endif
116 			/* Realign the source */
117 			src = (const void *)i_src;
118 			src -= 3;
119 			break;
120 		case 0x2:	/* Unaligned - Off by 2 */
121 			/* Word align the source */
122 			i_src = (const void *) ((unsigned)src & ~3);
123 #ifndef __MICROBLAZEEL__
124 			/* Load the holding buffer */
125 			buf_hold = *i_src++ << 16;
126 
127 			for (; c >= 4; c -= 4) {
128 				value = *i_src++;
129 				*i_dst++ = buf_hold | value >> 16;
130 				buf_hold = value << 16;
131 			}
132 #else
133 			/* Load the holding buffer */
134 			buf_hold = (*i_src++ & 0xFFFF0000) >> 16;
135 
136 			for (; c >= 4; c -= 4) {
137 				value = *i_src++;
138 				*i_dst++ = buf_hold | ((value & 0xFFFF) << 16);
139 				buf_hold = (value & 0xFFFF0000) >> 16;
140 			}
141 #endif
142 			/* Realign the source */
143 			src = (const void *)i_src;
144 			src -= 2;
145 			break;
146 		case 0x3:	/* Unaligned - Off by 3 */
147 			/* Word align the source */
148 			i_src = (const void *) ((unsigned)src & ~3);
149 #ifndef __MICROBLAZEEL__
150 			/* Load the holding buffer */
151 			buf_hold = *i_src++ << 24;
152 
153 			for (; c >= 4; c -= 4) {
154 				value = *i_src++;
155 				*i_dst++ = buf_hold | value >> 8;
156 				buf_hold = value << 24;
157 			}
158 #else
159 			/* Load the holding buffer */
160 			buf_hold = (*i_src++ & 0xFF000000) >> 24;
161 
162 			for (; c >= 4; c -= 4) {
163 				value = *i_src++;
164 				*i_dst++ = buf_hold | ((value & 0xFFFFFF) << 8);
165 				buf_hold = (value & 0xFF000000) >> 24;
166 			}
167 #endif
168 			/* Realign the source */
169 			src = (const void *)i_src;
170 			src -= 1;
171 			break;
172 		}
173 		dst = (void *)i_dst;
174 	}
175 
176 	/* Finish off any remaining bytes */
177 	/* simple fast copy, ... unless a cache boundary is crossed */
178 	switch (c) {
179 	case 3:
180 		*dst++ = *src++;
181 		fallthrough;
182 	case 2:
183 		*dst++ = *src++;
184 		fallthrough;
185 	case 1:
186 		*dst++ = *src++;
187 	}
188 
189 	return v_dst;
190 }
191 #endif /* CONFIG_OPT_LIB_FUNCTION */
192 EXPORT_SYMBOL(memcpy);
193 #endif /* __HAVE_ARCH_MEMCPY */
194