arch/x86/include/asm/string_32.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _ASM_X86_STRING_32_H
   3 #define _ASM_X86_STRING_32_H
   4
   5 #ifdef __KERNEL__
   6
   7 /* Let gcc decide whether to inline or use the out of line functions */
   8
   9 #define __HAVE_ARCH_STRCPY
  10 extern char *strcpy(char *dest, const char *src);
  11
  12 #define __HAVE_ARCH_STRNCPY
  13 extern char *strncpy(char *dest, const char *src, size_t count);
  14
  15 #define __HAVE_ARCH_STRCAT
  16 extern char *strcat(char *dest, const char *src);
  17
  18 #define __HAVE_ARCH_STRNCAT
  19 extern char *strncat(char *dest, const char *src, size_t count);
  20
  21 #define __HAVE_ARCH_STRCMP
  22 extern int strcmp(const char *cs, const char *ct);
  23
  24 #define __HAVE_ARCH_STRNCMP
  25 extern int strncmp(const char *cs, const char *ct, size_t count);
  26
  27 #define __HAVE_ARCH_STRCHR
  28 extern char *strchr(const char *s, int c);
  29
  30 #define __HAVE_ARCH_STRLEN
  31 extern size_t strlen(const char *s);
  32
  33 static __always_inline void *__memcpy(void *to, const void *from, size_t n)
  34 {
  35         int d0, d1, d2;
  36         asm volatile("rep ; movsl\n\t"
  37                      "movl %4,%%ecx\n\t"
  38                      "andl $3,%%ecx\n\t"
  39                      "jz 1f\n\t"
  40                      "rep ; movsb\n\t"
  41                      "1:"
  42                      : "=&c" (d0), "=&D" (d1), "=&S" (d2)
  43                      : "0" (n / 4), "g" (n), "1" ((long)to), "2" ((long)from)
  44                      : "memory");
  45         return to;
  46 }
  47
  48 /*
  49  * This looks ugly, but the compiler can optimize it totally,
  50  * as the count is constant.
  51  */
  52 static __always_inline void *__constant_memcpy(void *to, const void *from,
  53                                                size_t n)
  54 {
  55         long esi, edi;
  56         if (!n)
  57                 return to;
  58
  59         switch (n) {
  60         case 1:
  61                 *(char *)to = *(char *)from;
  62                 return to;
  63         case 2:
  64                 *(short *)to = *(short *)from;
  65                 return to;
  66         case 4:
  67                 *(int *)to = *(int *)from;
  68                 return to;
  69         case 3:
  70                 *(short *)to = *(short *)from;
  71                 *((char *)to + 2) = *((char *)from + 2);
  72                 return to;
  73         case 5:
  74                 *(int *)to = *(int *)from;
  75                 *((char *)to + 4) = *((char *)from + 4);
  76                 return to;
  77         case 6:
  78                 *(int *)to = *(int *)from;
  79                 *((short *)to + 2) = *((short *)from + 2);
  80                 return to;
  81         case 8:
  82                 *(int *)to = *(int *)from;
  83                 *((int *)to + 1) = *((int *)from + 1);
  84                 return to;
  85         }
  86
  87         esi = (long)from;
  88         edi = (long)to;
  89         if (n >= 5 * 4) {
  90                 /* large block: use rep prefix */
  91                 int ecx;
  92                 asm volatile("rep ; movsl"
  93                              : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
  94                              : "0" (n / 4), "1" (edi), "2" (esi)
  95                              : "memory"
  96                 );
  97         } else {
  98                 /* small block: don't clobber ecx + smaller code */
  99                 if (n >= 4 * 4)
 100                         asm volatile("movsl"
 101                                      : "=&D"(edi), "=&S"(esi)
 102                                      : "0"(edi), "1"(esi)
 103                                      : "memory");
 104                 if (n >= 3 * 4)
 105                         asm volatile("movsl"
 106                                      : "=&D"(edi), "=&S"(esi)
 107                                      : "0"(edi), "1"(esi)
 108                                      : "memory");
 109                 if (n >= 2 * 4)
 110                         asm volatile("movsl"
 111                                      : "=&D"(edi), "=&S"(esi)
 112                                      : "0"(edi), "1"(esi)
 113                                      : "memory");
 114                 if (n >= 1 * 4)
 115                         asm volatile("movsl"
 116                                      : "=&D"(edi), "=&S"(esi)
 117                                      : "0"(edi), "1"(esi)
 118                                      : "memory");
 119         }
 120         switch (n % 4) {
 121                 /* tail */
 122         case 0:
 123                 return to;
 124         case 1:
 125                 asm volatile("movsb"
 126                              : "=&D"(edi), "=&S"(esi)
 127                              : "0"(edi), "1"(esi)
 128                              : "memory");
 129                 return to;
 130         case 2:
 131                 asm volatile("movsw"
 132                              : "=&D"(edi), "=&S"(esi)
 133                              : "0"(edi), "1"(esi)
 134                              : "memory");
 135                 return to;
 136         default:
 137                 asm volatile("movsw\n\tmovsb"
 138                              : "=&D"(edi), "=&S"(esi)
 139                              : "0"(edi), "1"(esi)
 140                              : "memory");
 141                 return to;
 142         }
 143 }
 144
 145 #define __HAVE_ARCH_MEMCPY
 146 extern void *memcpy(void *, const void *, size_t);
 147
 148 #ifndef CONFIG_FORTIFY_SOURCE
 149 #ifdef CONFIG_X86_USE_3DNOW
 150
 151 #include <asm/mmx.h>
 152
 153 /*
 154  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 155  */
 156
 157 static inline void *__constant_memcpy3d(void *to, const void *from, size_t len)
 158 {
 159         if (len < 512)
 160                 return __constant_memcpy(to, from, len);
 161         return _mmx_memcpy(to, from, len);
 162 }
 163
 164 static inline void *__memcpy3d(void *to, const void *from, size_t len)
 165 {
 166         if (len < 512)
 167                 return __memcpy(to, from, len);
 168         return _mmx_memcpy(to, from, len);
 169 }
 170
 171 #define memcpy(t, f, n)                         \
 172         (__builtin_constant_p((n))              \
 173          ? __constant_memcpy3d((t), (f), (n))   \
 174          : __memcpy3d((t), (f), (n)))
 175
 176 #else
 177
 178 /*
 179  *      No 3D Now!
 180  */
 181
 182 #define memcpy(t, f, n) __builtin_memcpy(t, f, n)
 183
 184 #endif
 185 #endif /* !CONFIG_FORTIFY_SOURCE */
 186
 187 #define __HAVE_ARCH_MEMMOVE
 188 void *memmove(void *dest, const void *src, size_t n);
 189
 190 extern int memcmp(const void *, const void *, size_t);
 191 #ifndef CONFIG_FORTIFY_SOURCE
 192 #define memcmp __builtin_memcmp
 193 #endif
 194
 195 #define __HAVE_ARCH_MEMCHR
 196 extern void *memchr(const void *cs, int c, size_t count);
 197
 198 static inline void *__memset_generic(void *s, char c, size_t count)
 199 {
 200         int d0, d1;
 201         asm volatile("rep\n\t"
 202                      "stosb"
 203                      : "=&c" (d0), "=&D" (d1)
 204                      : "a" (c), "1" (s), "0" (count)
 205                      : "memory");
 206         return s;
 207 }
 208
 209 /* we might want to write optimized versions of these later */
 210 #define __constant_count_memset(s, c, count) __memset_generic((s), (c), (count))
 211
 212 /*
 213  * memset(x, 0, y) is a reasonably common thing to do, so we want to fill
 214  * things 32 bits at a time even when we don't know the size of the
 215  * area at compile-time..
 216  */
 217 static __always_inline
 218 void *__constant_c_memset(void *s, unsigned long c, size_t count)
 219 {
 220         int d0, d1;
 221         asm volatile("rep ; stosl\n\t"
 222                      "testb $2,%b3\n\t"
 223                      "je 1f\n\t"
 224                      "stosw\n"
 225                      "1:\ttestb $1,%b3\n\t"
 226                      "je 2f\n\t"
 227                      "stosb\n"
 228                      "2:"
 229                      : "=&c" (d0), "=&D" (d1)
 230                      : "a" (c), "q" (count), "0" (count/4), "1" ((long)s)
 231                      : "memory");
 232         return s;
 233 }
 234
 235 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 236 #define __HAVE_ARCH_STRNLEN
 237 extern size_t strnlen(const char *s, size_t count);
 238 /* end of additional stuff */
 239
 240 #define __HAVE_ARCH_STRSTR
 241 extern char *strstr(const char *cs, const char *ct);
 242
 243 /*
 244  * This looks horribly ugly, but the compiler can optimize it totally,
 245  * as we by now know that both pattern and count is constant..
 246  */
 247 static __always_inline
 248 void *__constant_c_and_count_memset(void *s, unsigned long pattern,
 249                                     size_t count)
 250 {
 251         switch (count) {
 252         case 0:
 253                 return s;
 254         case 1:
 255                 *(unsigned char *)s = pattern & 0xff;
 256                 return s;
 257         case 2:
 258                 *(unsigned short *)s = pattern & 0xffff;
 259                 return s;
 260         case 3:
 261                 *(unsigned short *)s = pattern & 0xffff;
 262                 *((unsigned char *)s + 2) = pattern & 0xff;
 263                 return s;
 264         case 4:
 265                 *(unsigned long *)s = pattern;
 266                 return s;
 267         }
 268
 269 #define COMMON(x)                                                       \
 270         asm volatile("rep ; stosl"                                      \
 271                      x                                                  \
 272                      : "=&c" (d0), "=&D" (d1)                           \
 273                      : "a" (eax), "0" (count/4), "1" ((long)s)  \
 274                      : "memory")
 275
 276         {
 277                 int d0, d1;
 278                 unsigned long eax = pattern;
 279
 280                 switch (count % 4) {
 281                 case 0:
 282                         COMMON("");
 283                         return s;
 284                 case 1:
 285                         COMMON("\n\tstosb");
 286                         return s;
 287                 case 2:
 288                         COMMON("\n\tstosw");
 289                         return s;
 290                 default:
 291                         COMMON("\n\tstosw\n\tstosb");
 292                         return s;
 293                 }
 294         }
 295
 296 #undef COMMON
 297 }
 298
 299 #define __constant_c_x_memset(s, c, count)                      \
 300         (__builtin_constant_p(count)                            \
 301          ? __constant_c_and_count_memset((s), (c), (count))     \
 302          : __constant_c_memset((s), (c), (count)))
 303
 304 #define __memset(s, c, count)                           \
 305         (__builtin_constant_p(count)                    \
 306          ? __constant_count_memset((s), (c), (count))   \
 307          : __memset_generic((s), (c), (count)))
 308
 309 #define __HAVE_ARCH_MEMSET
 310 extern void *memset(void *, int, size_t);
 311 #ifndef CONFIG_FORTIFY_SOURCE
 312 #define memset(s, c, count) __builtin_memset(s, c, count)
 313 #endif /* !CONFIG_FORTIFY_SOURCE */
 314
 315 #define __HAVE_ARCH_MEMSET16
 316 static inline void *memset16(uint16_t *s, uint16_t v, size_t n)
 317 {
 318         int d0, d1;
 319         asm volatile("rep\n\t"
 320                      "stosw"
 321                      : "=&c" (d0), "=&D" (d1)
 322                      : "a" (v), "1" (s), "0" (n)
 323                      : "memory");
 324         return s;
 325 }
 326
 327 #define __HAVE_ARCH_MEMSET32
 328 static inline void *memset32(uint32_t *s, uint32_t v, size_t n)
 329 {
 330         int d0, d1;
 331         asm volatile("rep\n\t"
 332                      "stosl"
 333                      : "=&c" (d0), "=&D" (d1)
 334                      : "a" (v), "1" (s), "0" (n)
 335                      : "memory");
 336         return s;
 337 }
 338
 339 /*
 340  * find the first occurrence of byte 'c', or 1 past the area if none
 341  */
 342 #define __HAVE_ARCH_MEMSCAN
 343 extern void *memscan(void *addr, int c, size_t size);
 344
 345 #endif /* __KERNEL__ */
 346
 347 #endif /* _ASM_X86_STRING_32_H */