LXR linux/arch/x86/include/asm/string

   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
  10extern char *strcpy(char *dest, const char *src);
  11
  12#define __HAVE_ARCH_STRNCPY
  13extern char *strncpy(char *dest, const char *src, size_t count);
  14
  15#define __HAVE_ARCH_STRCAT
  16extern char *strcat(char *dest, const char *src);
  17
  18#define __HAVE_ARCH_STRNCAT
  19extern char *strncat(char *dest, const char *src, size_t count);
  20
  21#define __HAVE_ARCH_STRCMP
  22extern int strcmp(const char *cs, const char *ct);
  23
  24#define __HAVE_ARCH_STRNCMP
  25extern int strncmp(const char *cs, const char *ct, size_t count);
  26
  27#define __HAVE_ARCH_STRCHR
  28extern char *strchr(const char *s, int c);
  29
  30#define __HAVE_ARCH_STRLEN
  31extern size_t strlen(const char *s);
  32
  33static __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 */
  52static __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
 146extern 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
 157static 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
 164static 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#ifndef CONFIG_KMEMCHECK
 183
 184#if (__GNUC__ >= 4)
 185#define memcpy(t, f, n) __builtin_memcpy(t, f, n)
 186#else
 187#define memcpy(t, f, n)                         \
 188        (__builtin_constant_p((n))              \
 189         ? __constant_memcpy((t), (f), (n))     \
 190         : __memcpy((t), (f), (n)))
 191#endif
 192#else
 193/*
 194 * kmemcheck becomes very happy if we use the REP instructions unconditionally,
 195 * because it means that we know both memory operands in advance.
 196 */
 197#define memcpy(t, f, n) __memcpy((t), (f), (n))
 198#endif
 199
 200#endif
 201#endif /* !CONFIG_FORTIFY_SOURCE */
 202
 203#define __HAVE_ARCH_MEMMOVE
 204void *memmove(void *dest, const void *src, size_t n);
 205
 206extern int memcmp(const void *, const void *, size_t);
 207#ifndef CONFIG_FORTIFY_SOURCE
 208#define memcmp __builtin_memcmp
 209#endif
 210
 211#define __HAVE_ARCH_MEMCHR
 212extern void *memchr(const void *cs, int c, size_t count);
 213
 214static inline void *__memset_generic(void *s, char c, size_t count)
 215{
 216        int d0, d1;
 217        asm volatile("rep\n\t"
 218                     "stosb"
 219                     : "=&c" (d0), "=&D" (d1)
 220                     : "a" (c), "1" (s), "0" (count)
 221                     : "memory");
 222        return s;
 223}
 224
 225/* we might want to write optimized versions of these later */
 226#define __constant_count_memset(s, c, count) __memset_generic((s), (c), (count))
 227
 228/*
 229 * memset(x, 0, y) is a reasonably common thing to do, so we want to fill
 230 * things 32 bits at a time even when we don't know the size of the
 231 * area at compile-time..
 232 */
 233static __always_inline
 234void *__constant_c_memset(void *s, unsigned long c, size_t count)
 235{
 236        int d0, d1;
 237        asm volatile("rep ; stosl\n\t"
 238                     "testb $2,%b3\n\t"
 239                     "je 1f\n\t"
 240                     "stosw\n"
 241                     "1:\ttestb $1,%b3\n\t"
 242                     "je 2f\n\t"
 243                     "stosb\n"
 244                     "2:"
 245                     : "=&c" (d0), "=&D" (d1)
 246                     : "a" (c), "q" (count), "0" (count/4), "1" ((long)s)
 247                     : "memory");
 248        return s;
 249}
 250
 251/* Added by Gertjan van Wingerde to make minix and sysv module work */
 252#define __HAVE_ARCH_STRNLEN
 253extern size_t strnlen(const char *s, size_t count);
 254/* end of additional stuff */
 255
 256#define __HAVE_ARCH_STRSTR
 257extern char *strstr(const char *cs, const char *ct);
 258
 259/*
 260 * This looks horribly ugly, but the compiler can optimize it totally,
 261 * as we by now know that both pattern and count is constant..
 262 */
 263static __always_inline
 264void *__constant_c_and_count_memset(void *s, unsigned long pattern,
 265                                    size_t count)
 266{
 267        switch (count) {
 268        case 0:
 269                return s;
 270        case 1:
 271                *(unsigned char *)s = pattern & 0xff;
 272                return s;
 273        case 2:
 274                *(unsigned short *)s = pattern & 0xffff;
 275                return s;
 276        case 3:
 277                *(unsigned short *)s = pattern & 0xffff;
 278                *((unsigned char *)s + 2) = pattern & 0xff;
 279                return s;
 280        case 4:
 281                *(unsigned long *)s = pattern;
 282                return s;
 283        }
 284
 285#define COMMON(x)                                                       \
 286        asm volatile("rep ; stosl"                                      \
 287                     x                                                  \
 288                     : "=&c" (d0), "=&D" (d1)                           \
 289                     : "a" (eax), "0" (count/4), "1" ((long)s)  \
 290                     : "memory")
 291
 292        {
 293                int d0, d1;
 294#if __GNUC__ == 4 && __GNUC_MINOR__ == 0
 295                /* Workaround for broken gcc 4.0 */
 296                register unsigned long eax asm("%eax") = pattern;
 297#else
 298                unsigned long eax = pattern;
 299#endif
 300
 301                switch (count % 4) {
 302                case 0:
 303                        COMMON("");
 304                        return s;
 305                case 1:
 306                        COMMON("\n\tstosb");
 307                        return s;
 308                case 2:
 309                        COMMON("\n\tstosw");
 310                        return s;
 311                default:
 312                        COMMON("\n\tstosw\n\tstosb");
 313                        return s;
 314                }
 315        }
 316
 317#undef COMMON
 318}
 319
 320#define __constant_c_x_memset(s, c, count)                      \
 321        (__builtin_constant_p(count)                            \
 322         ? __constant_c_and_count_memset((s), (c), (count))     \
 323         : __constant_c_memset((s), (c), (count)))
 324
 325#define __memset(s, c, count)                           \
 326        (__builtin_constant_p(count)                    \
 327         ? __constant_count_memset((s), (c), (count))   \
 328         : __memset_generic((s), (c), (count)))
 329
 330#define __HAVE_ARCH_MEMSET
 331extern void *memset(void *, int, size_t);
 332#ifndef CONFIG_FORTIFY_SOURCE
 333#if (__GNUC__ >= 4)
 334#define memset(s, c, count) __builtin_memset(s, c, count)
 335#else
 336#define memset(s, c, count)                                             \
 337        (__builtin_constant_p(c)                                        \
 338         ? __constant_c_x_memset((s), (0x01010101UL * (unsigned char)(c)), \
 339                                 (count))                               \
 340         : __memset((s), (c), (count)))
 341#endif
 342#endif /* !CONFIG_FORTIFY_SOURCE */
 343
 344#define __HAVE_ARCH_MEMSET16
 345static inline void *memset16(uint16_t *s, uint16_t v, size_t n)
 346{
 347        int d0, d1;
 348        asm volatile("rep\n\t"
 349                     "stosw"
 350                     : "=&c" (d0), "=&D" (d1)
 351                     : "a" (v), "1" (s), "0" (n)
 352                     : "memory");
 353        return s;
 354}
 355
 356#define __HAVE_ARCH_MEMSET32
 357static inline void *memset32(uint32_t *s, uint32_t v, size_t n)
 358{
 359        int d0, d1;
 360        asm volatile("rep\n\t"
 361                     "stosl"
 362                     : "=&c" (d0), "=&D" (d1)
 363                     : "a" (v), "1" (s), "0" (n)
 364                     : "memory");
 365        return s;
 366}
 367
 368/*
 369 * find the first occurrence of byte 'c', or 1 past the area if none
 370 */
 371#define __HAVE_ARCH_MEMSCAN
 372extern void *memscan(void *addr, int c, size_t size);
 373
 374#endif /* __KERNEL__ */
 375
 376#endif /* _ASM_X86_STRING_32_H */
 377