2 * Copyright (C) 2008-2009 Michal Simek <monstr@monstr.eu>
3 * Copyright (C) 2008-2009 PetaLogix
4 * Copyright (C) 2007 John Williams
6 * Reasonably optimised generic C-code for memcpy on Microblaze
7 * This is generic C code to do efficient, alignment-aware memmove.
9 * It is based on demo code originally Copyright 2001 by Intel Corp, taken from
10 * http://www.embedded.com/showArticle.jhtml?articleID=19205567
12 * Attempts were made, unsuccessfully, to contact the original
13 * author of this code (Michael Morrow, Intel). Below is the original
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.
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>
33 #ifdef __HAVE_ARCH_MEMMOVE
34 #ifndef CONFIG_OPT_LIB_FUNCTION
35 void *memmove(void *v_dst
, const void *v_src
, __kernel_size_t c
)
37 const char *src
= v_src
;
43 /* Use memcpy when source is higher than dest */
45 return memcpy(v_dst
, v_src
, c
);
47 /* copy backwards, from end to beginning */
51 /* Simple, byte oriented memmove. */
57 #else /* CONFIG_OPT_LIB_FUNCTION */
58 void *memmove(void *v_dst
, const void *v_src
, __kernel_size_t c
)
60 const char *src
= v_src
;
62 const uint32_t *i_src
;
68 /* Use memcpy when source is higher than dest */
70 return memcpy(v_dst
, v_src
, c
);
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.
78 /* FIXME this part needs more test */
79 /* Do a descending copy - this is a bit trickier! */
84 unsigned value
, buf_hold
;
86 /* Align the destination to a word boundary. */
87 /* This is done in an endian independent manner. */
89 switch ((unsigned long)dst
& 3) {
102 /* Choose a copy scheme based on the source */
103 /* alignment relative to dstination. */
104 switch ((unsigned long)src
& 3) {
105 case 0x0: /* Both byte offsets are aligned */
107 i_src
= (const void *)src
;
109 for (; c
>= 4; c
-= 4)
112 src
= (const void *)i_src
;
114 case 0x1: /* Unaligned - Off by 1 */
115 /* Word align the source */
116 i_src
= (const void *) (((unsigned)src
+ 4) & ~3);
117 #ifndef __MICROBLAZEEL__
118 /* Load the holding buffer */
119 buf_hold
= *--i_src
>> 24;
121 for (; c
>= 4; c
-= 4) {
123 *--i_dst
= buf_hold
<< 8 | value
;
124 buf_hold
= value
>> 24;
127 /* Load the holding buffer */
128 buf_hold
= (*--i_src
& 0xFF) << 24;
130 for (; c
>= 4; c
-= 4) {
132 *--i_dst
= buf_hold
|
133 ((value
& 0xFFFFFF00) >> 8);
134 buf_hold
= (value
& 0xFF) << 24;
137 /* Realign the source */
138 src
= (const void *)i_src
;
141 case 0x2: /* Unaligned - Off by 2 */
142 /* Word align the source */
143 i_src
= (const void *) (((unsigned)src
+ 4) & ~3);
144 #ifndef __MICROBLAZEEL__
145 /* Load the holding buffer */
146 buf_hold
= *--i_src
>> 16;
148 for (; c
>= 4; c
-= 4) {
150 *--i_dst
= buf_hold
<< 16 | value
;
151 buf_hold
= value
>> 16;
154 /* Load the holding buffer */
155 buf_hold
= (*--i_src
& 0xFFFF) << 16;
157 for (; c
>= 4; c
-= 4) {
159 *--i_dst
= buf_hold
|
160 ((value
& 0xFFFF0000) >> 16);
161 buf_hold
= (value
& 0xFFFF) << 16;
164 /* Realign the source */
165 src
= (const void *)i_src
;
168 case 0x3: /* Unaligned - Off by 3 */
169 /* Word align the source */
170 i_src
= (const void *) (((unsigned)src
+ 4) & ~3);
171 #ifndef __MICROBLAZEEL__
172 /* Load the holding buffer */
173 buf_hold
= *--i_src
>> 8;
175 for (; c
>= 4; c
-= 4) {
177 *--i_dst
= buf_hold
<< 24 | value
;
178 buf_hold
= value
>> 8;
181 /* Load the holding buffer */
182 buf_hold
= (*--i_src
& 0xFFFFFF) << 8;
184 for (; c
>= 4; c
-= 4) {
186 *--i_dst
= buf_hold
|
187 ((value
& 0xFF000000) >> 24);
188 buf_hold
= (value
& 0xFFFFFF) << 8;
191 /* Realign the source */
192 src
= (const void *)i_src
;
199 /* simple fast copy, ... unless a cache boundary is crossed */
200 /* Finish off any remaining bytes */
213 #endif /* CONFIG_OPT_LIB_FUNCTION */
214 EXPORT_SYMBOL(memmove
);
215 #endif /* __HAVE_ARCH_MEMMOVE */