x86: ASM fast memset()

arch/x86/kernel/include/arch-string.h

@@ -20,4 +20,23 @@ static inline void *__inline_memcpy(void *to, const void *from, size_t n)
 	return to;
 }
 
+#define ARCH_FAST_MEMSET
+
+static inline void *__inline_memset(void *s, unsigned long c, size_t count)
+{
+	int d0, d1;
+	asm volatile("rep ; stosl\n\t"
+		     "testb $2,%b3\n\t"
+		     "je 1f\n\t"
+		     "stosw\n"
+		     "1:\ttestb $1,%b3\n\t"
+		     "je 2f\n\t"
+		     "stosb\n"
+		     "2:"
+		     : "=&c" (d0), "=&D" (d1)
+		     : "a" (c), "q" (count), "0" (count/4), "1" ((long)s)
+		     : "memory");
+	return s;
+}
+
 #endif

lib/include/string.h

@@ -36,6 +36,10 @@ void *memset(void *s, int n, size_t l);
 #define fast_memcpy memcpy
 #endif
 
+#ifdef ARCH_FAST_MEMSET
+#define memset __inline_memset
+#endif
+
 extern int snprintf(char * buf, size_t size, const char *fmt, ...);
 extern int sprintf(char * buf, const char *fmt, ...);
 extern int sscanf(const char * buf, const char * fmt, ...);

lib/string.c

@@ -13,6 +13,7 @@
 #include <kmalloc.h>
 #include <string.h>
 #include <memory.h>
+#include <arch-string.h>
 
 size_t strlen(const char *p)
 {
@@ -153,6 +154,7 @@ void *memcpy_long(void *dest, const void *src, size_t n)
 	return dest;
 }
 
+#ifndef ARCH_FAST_MEMSET
 void *memset(void *s, int c, size_t n)
 {
 	char *s_aligned = (void *)(((unsigned long)s + 7) & ~7);
@@ -187,6 +189,7 @@ void *memset(void *s, int c, size_t n)
 
 	return s;
 }
+#endif
 
 int memcmp(const void *s1, const void *s2, size_t n)
 {