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 memmove.
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 #include <linux/string.h>
32 
33 #ifdef __HAVE_ARCH_MEMMOVE
34 #ifndef CONFIG_OPT_LIB_FUNCTION
memmove(void * v_dst,const void * v_src,__kernel_size_t c)35 void *memmove(void *v_dst, const void *v_src, __kernel_size_t c)
36 {
37 	const char *src = v_src;
38 	char *dst = v_dst;
39 
40 	if (!c)
41 		return v_dst;
42 
43 	/* Use memcpy when source is higher than dest */
44 	if (v_dst <= v_src)
45 		return memcpy(v_dst, v_src, c);
46 
47 	/* copy backwards, from end to beginning */
48 	src += c;
49 	dst += c;
50 
51 	/* Simple, byte oriented memmove. */
52 	while (c--)
53 		*--dst = *--src;
54 
55 	return v_dst;
56 }
57 #else /* CONFIG_OPT_LIB_FUNCTION */
memmove(void * v_dst,const void * v_src,__kernel_size_t c)58 void *memmove(void *v_dst, const void *v_src, __kernel_size_t c)
59 {
60 	const char *src = v_src;
61 	char *dst = v_dst;
62 	const uint32_t *i_src;
63 	uint32_t *i_dst;
64 
65 	if (!c)
66 		return v_dst;
67 
68 	/* Use memcpy when source is higher than dest */
69 	if (v_dst <= v_src)
70 		return memcpy(v_dst, v_src, c);
71 
72 	/* The following code tries to optimize the copy by using unsigned
73 	 * alignment. This will work fine if both source and destination are
74 	 * aligned on the same boundary. However, if they are aligned on
75 	 * different boundaries shifts will be necessary. This might result in
76 	 * bad performance on MicroBlaze systems without a barrel shifter.
77 	 */
78 	/* FIXME this part needs more test */
79 	/* Do a descending copy - this is a bit trickier! */
80 	dst += c;
81 	src += c;
82 
83 	if (c >= 4) {
84 		unsigned  value, buf_hold;
85 
86 		/* Align the destination to a word boundary. */
87 		/* This is done in an endian independent manner. */
88 
89 		switch ((unsigned long)dst & 3) {
90 		case 3:
91 			*--dst = *--src;
92 			--c;
93 			fallthrough;
94 		case 2:
95 			*--dst = *--src;
96 			--c;
97 			fallthrough;
98 		case 1:
99 			*--dst = *--src;
100 			--c;
101 		}
102 
103 		i_dst = (void *)dst;
104 		/* Choose a copy scheme based on the source */
105 		/* alignment relative to dstination. */
106 		switch ((unsigned long)src & 3) {
107 		case 0x0:	/* Both byte offsets are aligned */
108 
109 			i_src  = (const void *)src;
110 
111 			for (; c >= 4; c -= 4)
112 				*--i_dst = *--i_src;
113 
114 			src  = (const void *)i_src;
115 			break;
116 		case 0x1:	/* Unaligned - Off by 1 */
117 			/* Word align the source */
118 			i_src = (const void *) (((unsigned)src + 4) & ~3);
119 #ifndef __MICROBLAZEEL__
120 			/* Load the holding buffer */
121 			buf_hold = *--i_src >> 24;
122 
123 			for (; c >= 4; c -= 4) {
124 				value = *--i_src;
125 				*--i_dst = buf_hold << 8 | value;
126 				buf_hold = value >> 24;
127 			}
128 #else
129 			/* Load the holding buffer */
130 			buf_hold = (*--i_src & 0xFF) << 24;
131 
132 			for (; c >= 4; c -= 4) {
133 				value = *--i_src;
134 				*--i_dst = buf_hold |
135 						((value & 0xFFFFFF00) >> 8);
136 				buf_hold = (value  & 0xFF) << 24;
137 			}
138 #endif
139 			/* Realign the source */
140 			src = (const void *)i_src;
141 			src += 1;
142 			break;
143 		case 0x2:	/* Unaligned - Off by 2 */
144 			/* Word align the source */
145 			i_src = (const void *) (((unsigned)src + 4) & ~3);
146 #ifndef __MICROBLAZEEL__
147 			/* Load the holding buffer */
148 			buf_hold = *--i_src >> 16;
149 
150 			for (; c >= 4; c -= 4) {
151 				value = *--i_src;
152 				*--i_dst = buf_hold << 16 | value;
153 				buf_hold = value >> 16;
154 			}
155 #else
156 			/* Load the holding buffer */
157 			buf_hold = (*--i_src & 0xFFFF) << 16;
158 
159 			for (; c >= 4; c -= 4) {
160 				value = *--i_src;
161 				*--i_dst = buf_hold |
162 						((value & 0xFFFF0000) >> 16);
163 				buf_hold = (value & 0xFFFF) << 16;
164 			}
165 #endif
166 			/* Realign the source */
167 			src = (const void *)i_src;
168 			src += 2;
169 			break;
170 		case 0x3:	/* Unaligned - Off by 3 */
171 			/* Word align the source */
172 			i_src = (const void *) (((unsigned)src + 4) & ~3);
173 #ifndef __MICROBLAZEEL__
174 			/* Load the holding buffer */
175 			buf_hold = *--i_src >> 8;
176 
177 			for (; c >= 4; c -= 4) {
178 				value = *--i_src;
179 				*--i_dst = buf_hold << 24 | value;
180 				buf_hold = value >> 8;
181 			}
182 #else
183 			/* Load the holding buffer */
184 			buf_hold = (*--i_src & 0xFFFFFF) << 8;
185 
186 			for (; c >= 4; c -= 4) {
187 				value = *--i_src;
188 				*--i_dst = buf_hold |
189 						((value & 0xFF000000) >> 24);
190 				buf_hold = (value & 0xFFFFFF) << 8;
191 			}
192 #endif
193 			/* Realign the source */
194 			src = (const void *)i_src;
195 			src += 3;
196 			break;
197 		}
198 		dst = (void *)i_dst;
199 	}
200 
201 	/* simple fast copy, ... unless a cache boundary is crossed */
202 	/* Finish off any remaining bytes */
203 	switch (c) {
204 	case 4:
205 		*--dst = *--src;
206 		fallthrough;
207 	case 3:
208 		*--dst = *--src;
209 		fallthrough;
210 	case 2:
211 		*--dst = *--src;
212 		fallthrough;
213 	case 1:
214 		*--dst = *--src;
215 	}
216 	return v_dst;
217 }
218 #endif /* CONFIG_OPT_LIB_FUNCTION */
219 EXPORT_SYMBOL(memmove);
220 #endif /* __HAVE_ARCH_MEMMOVE */
221