arch/x86/include/asm/string_32.h

   1 #ifndef _ASM_X86_STRING_32_H
   2 #define _ASM_X86_STRING_32_H
   3
   4 #ifdef __KERNEL__
   5
   6 /* Let gcc decide whether to inline or use the out of line functions */
   7
   8 #define __HAVE_ARCH_STRCPY
   9 extern char *strcpy(char *dest, const char *src);
  10
  11 #define __HAVE_ARCH_STRNCPY
  12 extern char *strncpy(char *dest, const char *src, size_t count);
  13
  14 #define __HAVE_ARCH_STRCAT
  15 extern char *strcat(char *dest, const char *src);
  16
  17 #define __HAVE_ARCH_STRNCAT
  18 extern char *strncat(char *dest, const char *src, size_t count);
  19
  20 #define __HAVE_ARCH_STRCMP
  21 extern int strcmp(const char *cs, const char *ct);
  22
  23 #define __HAVE_ARCH_STRNCMP
  24 extern int strncmp(const char *cs, const char *ct, size_t count);
  25
  26 #define __HAVE_ARCH_STRCHR
  27 extern char *strchr(const char *s, int c);
  28
  29 #define __HAVE_ARCH_STRLEN
  30 extern size_t strlen(const char *s);
  31
  32 static __always_inline void *__memcpy(void *to, const void *from, size_t n)
  33 {
  34         int d0, d1, d2;
  35         asm volatile("rep ; movsl\n\t"
  36                      "movl %4,%%ecx\n\t"
  37                      "andl $3,%%ecx\n\t"
  38                      "jz 1f\n\t"
  39                      "rep ; movsb\n\t"
  40                      "1:"
  41                      : "=&c" (d0), "=&D" (d1), "=&S" (d2)
  42                      : "0" (n / 4), "g" (n), "1" ((long)to), "2" ((long)from)
  43                      : "memory");
  44         return to;
  45 }
  46
  47 /*
  48  * This looks ugly, but the compiler can optimize it totally,
  49  * as the count is constant.
  50  */
  51 static __always_inline void *__constant_memcpy(void *to, const void *from,
  52                                                size_t n)
  53 {
  54         long esi, edi;
  55         if (!n)
  56                 return to;
  57
  58         switch (n) {
  59         case 1:
  60                 *(char *)to = *(char *)from;
  61                 return to;
  62         case 2:
  63                 *(short *)to = *(short *)from;
  64                 return to;
  65         case 4:
  66                 *(int *)to = *(int *)from;
  67                 return to;
  68         case 3:
  69                 *(short *)to = *(short *)from;
  70                 *((char *)to + 2) = *((char *)from + 2);
  71                 return to;
  72         case 5:
  73                 *(int *)to = *(int *)from;
  74                 *((char *)to + 4) = *((char *)from + 4);
  75                 return to;
  76         case 6:
  77                 *(int *)to = *(int *)from;
  78                 *((short *)to + 2) = *((short *)from + 2);
  79                 return to;
  80         case 8:
  81                 *(int *)to = *(int *)from;
  82                 *((int *)to + 1) = *((int *)from + 1);
  83                 return to;
  84         }
  85
  86         esi = (long)from;
  87         edi = (long)to;
  88         if (n >= 5 * 4) {
  89                 /* large block: use rep prefix */
  90                 int ecx;
  91                 asm volatile("rep ; movsl"
  92                              : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
  93                              : "0" (n / 4), "1" (edi), "2" (esi)
  94                              : "memory"
  95                 );
  96         } else {
  97                 /* small block: don't clobber ecx + smaller code */
  98                 if (n >= 4 * 4)
  99                         asm volatile("movsl"
 100                                      : "=&D"(edi), "=&S"(esi)
 101                                      : "0"(edi), "1"(esi)
 102                                      : "memory");
 103                 if (n >= 3 * 4)
 104                         asm volatile("movsl"
 105                                      : "=&D"(edi), "=&S"(esi)
 106                                      : "0"(edi), "1"(esi)
 107                                      : "memory");
 108                 if (n >= 2 * 4)
 109                         asm volatile("movsl"
 110                                      : "=&D"(edi), "=&S"(esi)
 111                                      : "0"(edi), "1"(esi)
 112                                      : "memory");
 113                 if (n >= 1 * 4)
 114                         asm volatile("movsl"
 115                                      : "=&D"(edi), "=&S"(esi)
 116                                      : "0"(edi), "1"(esi)
 117                                      : "memory");
 118         }
 119         switch (n % 4) {
 120                 /* tail */
 121         case 0:
 122                 return to;
 123         case 1:
 124                 asm volatile("movsb"
 125                              : "=&D"(edi), "=&S"(esi)
 126                              : "0"(edi), "1"(esi)
 127                              : "memory");
 128                 return to;
 129         case 2:
 130                 asm volatile("movsw"
 131                              : "=&D"(edi), "=&S"(esi)
 132                              : "0"(edi), "1"(esi)
 133                              : "memory");
 134                 return to;
 135         default:
 136                 asm volatile("movsw\n\tmovsb"
 137                              : "=&D"(edi), "=&S"(esi)
 138                              : "0"(edi), "1"(esi)
 139                              : "memory");
 140                 return to;
 141         }
 142 }
 143
 144 #define __HAVE_ARCH_MEMCPY
 145
 146 #ifdef CONFIG_X86_USE_3DNOW
 147
 148 #include <asm/mmx.h>
 149
 150 /*
 151  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 152  */
 153
 154 static inline void *__constant_memcpy3d(void *to, const void *from, size_t len)
 155 {
 156         if (len < 512)
 157                 return __constant_memcpy(to, from, len);
 158         return _mmx_memcpy(to, from, len);
 159 }
 160
 161 static inline void *__memcpy3d(void *to, const void *from, size_t len)
 162 {
 163         if (len < 512)
 164                 return __memcpy(to, from, len);
 165         return _mmx_memcpy(to, from, len);
 166 }
 167
 168 #define memcpy(t, f, n)                         \
 169         (__builtin_constant_p((n))              \
 170          ? __constant_memcpy3d((t), (f), (n))   \
 171          : __memcpy3d((t), (f), (n)))
 172
 173 #else
 174
 175 /*
 176  *      No 3D Now!
 177  */
 178
 179 #ifndef CONFIG_KMEMCHECK
 180
 181 #if (__GNUC__ >= 4)
 182 #define memcpy(t, f, n) __builtin_memcpy(t, f, n)
 183 #else
 184 #define memcpy(t, f, n)                         \
 185         (__builtin_constant_p((n))              \
 186          ? __constant_memcpy((t), (f), (n))     \
 187          : __memcpy((t), (f), (n)))
 188 #endif
 189 #else
 190 /*
 191  * kmemcheck becomes very happy if we use the REP instructions unconditionally,
 192  * because it means that we know both memory operands in advance.
 193  */
 194 #define memcpy(t, f, n) __memcpy((t), (f), (n))
 195 #endif
 196
 197 #endif
 198
 199 #define __HAVE_ARCH_MEMMOVE
 200 void *memmove(void *dest, const void *src, size_t n);
 201
 202 #define memcmp __builtin_memcmp
 203
 204 #define __HAVE_ARCH_MEMCHR
 205 extern void *memchr(const void *cs, int c, size_t count);
 206
 207 static inline void *__memset_generic(void *s, char c, size_t count)
 208 {
 209         int d0, d1;
 210         asm volatile("rep\n\t"
 211                      "stosb"
 212                      : "=&c" (d0), "=&D" (d1)
 213                      : "a" (c), "1" (s), "0" (count)
 214                      : "memory");
 215         return s;
 216 }
 217
 218 /* we might want to write optimized versions of these later */
 219 #define __constant_count_memset(s, c, count) __memset_generic((s), (c), (count))
 220
 221 /*
 222  * memset(x, 0, y) is a reasonably common thing to do, so we want to fill
 223  * things 32 bits at a time even when we don't know the size of the
 224  * area at compile-time..
 225  */
 226 static __always_inline
 227 void *__constant_c_memset(void *s, unsigned long c, size_t count)
 228 {
 229         int d0, d1;
 230         asm volatile("rep ; stosl\n\t"
 231                      "testb $2,%b3\n\t"
 232                      "je 1f\n\t"
 233                      "stosw\n"
 234                      "1:\ttestb $1,%b3\n\t"
 235                      "je 2f\n\t"
 236                      "stosb\n"
 237                      "2:"
 238                      : "=&c" (d0), "=&D" (d1)
 239                      : "a" (c), "q" (count), "0" (count/4), "1" ((long)s)
 240                      : "memory");
 241         return s;
 242 }
 243
 244 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 245 #define __HAVE_ARCH_STRNLEN
 246 extern size_t strnlen(const char *s, size_t count);
 247 /* end of additional stuff */
 248
 249 #define __HAVE_ARCH_STRSTR
 250 extern char *strstr(const char *cs, const char *ct);
 251
 252 /*
 253  * This looks horribly ugly, but the compiler can optimize it totally,
 254  * as we by now know that both pattern and count is constant..
 255  */
 256 static __always_inline
 257 void *__constant_c_and_count_memset(void *s, unsigned long pattern,
 258                                     size_t count)
 259 {
 260         switch (count) {
 261         case 0:
 262                 return s;
 263         case 1:
 264                 *(unsigned char *)s = pattern & 0xff;
 265                 return s;
 266         case 2:
 267                 *(unsigned short *)s = pattern & 0xffff;
 268                 return s;
 269         case 3:
 270                 *(unsigned short *)s = pattern & 0xffff;
 271                 *((unsigned char *)s + 2) = pattern & 0xff;
 272                 return s;
 273         case 4:
 274                 *(unsigned long *)s = pattern;
 275                 return s;
 276         }
 277
 278 #define COMMON(x)                                                       \
 279         asm volatile("rep ; stosl"                                      \
 280                      x                                                  \
 281                      : "=&c" (d0), "=&D" (d1)                           \
 282                      : "a" (eax), "0" (count/4), "1" ((long)s)  \
 283                      : "memory")
 284
 285         {
 286                 int d0, d1;
 287 #if __GNUC__ == 4 && __GNUC_MINOR__ == 0
 288                 /* Workaround for broken gcc 4.0 */
 289                 register unsigned long eax asm("%eax") = pattern;
 290 #else
 291                 unsigned long eax = pattern;
 292 #endif
 293
 294                 switch (count % 4) {
 295                 case 0:
 296                         COMMON("");
 297                         return s;
 298                 case 1:
 299                         COMMON("\n\tstosb");
 300                         return s;
 301                 case 2:
 302                         COMMON("\n\tstosw");
 303                         return s;
 304                 default:
 305                         COMMON("\n\tstosw\n\tstosb");
 306                         return s;
 307                 }
 308         }
 309
 310 #undef COMMON
 311 }
 312
 313 #define __constant_c_x_memset(s, c, count)                      \
 314         (__builtin_constant_p(count)                            \
 315          ? __constant_c_and_count_memset((s), (c), (count))     \
 316          : __constant_c_memset((s), (c), (count)))
 317
 318 #define __memset(s, c, count)                           \
 319         (__builtin_constant_p(count)                    \
 320          ? __constant_count_memset((s), (c), (count))   \
 321          : __memset_generic((s), (c), (count)))
 322
 323 #define __HAVE_ARCH_MEMSET
 324 #if (__GNUC__ >= 4)
 325 #define memset(s, c, count) __builtin_memset(s, c, count)
 326 #else
 327 #define memset(s, c, count)                                             \
 328         (__builtin_constant_p(c)                                        \
 329          ? __constant_c_x_memset((s), (0x01010101UL * (unsigned char)(c)), \
 330                                  (count))                               \
 331          : __memset((s), (c), (count)))
 332 #endif
 333
 334 /*
 335  * find the first occurrence of byte 'c', or 1 past the area if none
 336  */
 337 #define __HAVE_ARCH_MEMSCAN
 338 extern void *memscan(void *addr, int c, size_t size);
 339
 340 #endif /* __KERNEL__ */
 341
 342 #endif /* _ASM_X86_STRING_32_H */