arch/microblaze/lib/memcpy.c

   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/types.h>
  28 #include <linux/stddef.h>
  29 #include <linux/compiler.h>
  30 #include <linux/module.h>
  31
  32 #include <linux/string.h>
  33 #include <asm/system.h>
  34
  35 #ifdef __HAVE_ARCH_MEMCPY
  36 #ifndef CONFIG_OPT_LIB_FUNCTION
  37 void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
  38 {
  39         const char *src = v_src;
  40         char *dst = v_dst;
  41
  42         /* Simple, byte oriented memcpy. */
  43         while (c--)
  44                 *dst++ = *src++;
  45
  46         return v_dst;
  47 }
  48 #else /* CONFIG_OPT_LIB_FUNCTION */
  49 void *memcpy(void *v_dst, const void *v_src, __kernel_size_t c)
  50 {
  51         const char *src = v_src;
  52         char *dst = v_dst;
  53
  54         /* The following code tries to optimize the copy by using unsigned
  55          * alignment. This will work fine if both source and destination are
  56          * aligned on the same boundary. However, if they are aligned on
  57          * different boundaries shifts will be necessary. This might result in
  58          * bad performance on MicroBlaze systems without a barrel shifter.
  59          */
  60         const uint32_t *i_src;
  61         uint32_t *i_dst;
  62
  63         if (likely(c >= 4)) {
  64                 unsigned  value, buf_hold;
  65
  66                 /* Align the dstination to a word boundry. */
  67                 /* This is done in an endian independant manner. */
  68                 switch ((unsigned long)dst & 3) {
  69                 case 1:
  70                         *dst++ = *src++;
  71                         --c;
  72                 case 2:
  73                         *dst++ = *src++;
  74                         --c;
  75                 case 3:
  76                         *dst++ = *src++;
  77                         --c;
  78                 }
  79
  80                 i_dst = (void *)dst;
  81
  82                 /* Choose a copy scheme based on the source */
  83                 /* alignment relative to dstination. */
  84                 switch ((unsigned long)src & 3) {
  85                 case 0x0:       /* Both byte offsets are aligned */
  86                         i_src  = (const void *)src;
  87
  88                         for (; c >= 4; c -= 4)
  89                                 *i_dst++ = *i_src++;
  90
  91                         src  = (const void *)i_src;
  92                         break;
  93                 case 0x1:       /* Unaligned - Off by 1 */
  94                         /* Word align the source */
  95                         i_src = (const void *) ((unsigned)src & ~3);
  96 #ifndef __MICROBLAZEEL__
  97                         /* Load the holding buffer */
  98                         buf_hold = *i_src++ << 8;
  99
 100                         for (; c >= 4; c -= 4) {
 101                                 value = *i_src++;
 102                                 *i_dst++ = buf_hold | value >> 24;
 103                                 buf_hold = value << 8;
 104                         }
 105 #else
 106                         /* Load the holding buffer */
 107                         buf_hold = (*i_src++ & 0xFFFFFF00) >>8;
 108
 109                         for (; c >= 4; c -= 4) {
 110                                 value = *i_src++;
 111                                 *i_dst++ = buf_hold | ((value & 0xFF) << 24);
 112                                 buf_hold = (value & 0xFFFFFF00) >>8;
 113                         }
 114 #endif
 115                         /* Realign the source */
 116                         src = (const void *)i_src;
 117                         src -= 3;
 118                         break;
 119                 case 0x2:       /* Unaligned - Off by 2 */
 120                         /* Word align the source */
 121                         i_src = (const void *) ((unsigned)src & ~3);
 122 #ifndef __MICROBLAZEEL__
 123                         /* Load the holding buffer */
 124                         buf_hold = *i_src++ << 16;
 125
 126                         for (; c >= 4; c -= 4) {
 127                                 value = *i_src++;
 128                                 *i_dst++ = buf_hold | value >> 16;
 129                                 buf_hold = value << 16;
 130                         }
 131 #else
 132                         /* Load the holding buffer */
 133                         buf_hold = (*i_src++ & 0xFFFF0000 )>>16;
 134
 135                         for (; c >= 4; c -= 4) {
 136                                 value = *i_src++;
 137                                 *i_dst++ = buf_hold | ((value & 0xFFFF)<<16);
 138                                 buf_hold = (value & 0xFFFF0000) >>16;
 139                         }
 140 #endif
 141                         /* Realign the source */
 142                         src = (const void *)i_src;
 143                         src -= 2;
 144                         break;
 145                 case 0x3:       /* Unaligned - Off by 3 */
 146                         /* Word align the source */
 147                         i_src = (const void *) ((unsigned)src & ~3);
 148 #ifndef __MICROBLAZEEL__
 149                         /* Load the holding buffer */
 150                         buf_hold = *i_src++ << 24;
 151
 152                         for (; c >= 4; c -= 4) {
 153                                 value = *i_src++;
 154                                 *i_dst++ = buf_hold | value >> 8;
 155                                 buf_hold = value << 24;
 156                         }
 157 #else
 158                         /* Load the holding buffer */
 159                         buf_hold = (*i_src++ & 0xFF000000) >> 24;
 160
 161                         for (; c >= 4; c -= 4) {
 162                                 value = *i_src++;
 163                                 *i_dst++ = buf_hold | ((value & 0xFFFFFF) << 8);
 164                                 buf_hold = (value & 0xFF000000) >> 24;
 165                         }
 166 #endif
 167                         /* Realign the source */
 168                         src = (const void *)i_src;
 169                         src -= 1;
 170                         break;
 171                 }
 172                 dst = (void *)i_dst;
 173         }
 174
 175         /* Finish off any remaining bytes */
 176         /* simple fast copy, ... unless a cache boundry is crossed */
 177         switch (c) {
 178         case 3:
 179                 *dst++ = *src++;
 180         case 2:
 181                 *dst++ = *src++;
 182         case 1:
 183                 *dst++ = *src++;
 184         }
 185
 186         return v_dst;
 187 }
 188 #endif /* CONFIG_OPT_LIB_FUNCTION */
 189 EXPORT_SYMBOL(memcpy);
 190 #endif /* __HAVE_ARCH_MEMCPY */