Add 64-bit byte-swapping operations.

src/arch/i386/include/bits/byteswap.h

@@ -1,7 +1,8 @@
 #ifndef ETHERBOOT_BITS_BYTESWAP_H
 #define ETHERBOOT_BITS_BYTESWAP_H
 
-static inline uint16_t __i386_bswap_16(uint16_t x)
+static inline __attribute__ ((always_inline)) uint16_t
+__i386_bswap_16(uint16_t x)
 {
 	__asm__("xchgb %b0,%h0\n\t"
 		: "=q" (x)
@@ -9,7 +10,8 @@ static inline uint16_t __i386_bswap_16(uint16_t x)
 	return x;
 }
 
-static inline uint32_t __i386_bswap_32(uint32_t x)
+static inline __attribute__ ((always_inline)) uint32_t
+__i386_bswap_32(uint32_t x)
 {
 	__asm__("xchgb %b0,%h0\n\t"
 		"rorl $16,%0\n\t"
@@ -19,27 +21,56 @@ static inline uint32_t __i386_bswap_32(uint32_t x)
 	return x;
 }
 
+static inline __attribute__ ((always_inline)) uint64_t
+__i386_bswap_64(uint64_t x)
+{
+	union {
+		uint64_t qword;
+		uint32_t dword[2]; 
+	} u;
+
+	u.qword = x;
+	u.dword[0] = __i386_bswap_32(u.dword[0]);
+	u.dword[1] = __i386_bswap_32(u.dword[1]);
+	__asm__("xchgl %0,%1"
+		: "=r" ( u.dword[0] ), "=r" ( u.dword[1] )
+		: "0" ( u.dword[0] ), "1" ( u.dword[1] ) );
+	return u.qword;
+}
 
 #define __bswap_constant_16(x) \
 	((uint16_t)((((uint16_t)(x) & 0x00ff) << 8) | \
-		(((uint16_t)(x) & 0xff00) >> 8)))
+		    (((uint16_t)(x) & 0xff00) >> 8)))
 
 #define __bswap_constant_32(x) \
 	((uint32_t)((((uint32_t)(x) & 0x000000ffU) << 24) | \
-		(((uint32_t)(x) & 0x0000ff00U) <<  8) | \
-		(((uint32_t)(x) & 0x00ff0000U) >>  8) | \
-		(((uint32_t)(x) & 0xff000000U) >> 24)))
+		    (((uint32_t)(x) & 0x0000ff00U) <<  8) | \
+		    (((uint32_t)(x) & 0x00ff0000U) >>  8) | \
+		    (((uint32_t)(x) & 0xff000000U) >> 24)))
+
+#define __bswap_constant_64(x) \
+	((uint64_t)((((uint64_t)(x) & 0x00000000000000ffULL) << 56) | \
+		    (((uint64_t)(x) & 0x000000000000ff00ULL) << 40) | \
+		    (((uint64_t)(x) & 0x0000000000ff0000ULL) << 24) | \
+		    (((uint64_t)(x) & 0x00000000ff000000ULL) <<  8) | \
+		    (((uint64_t)(x) & 0x000000ff00000000ULL) >>  8) | \
+		    (((uint64_t)(x) & 0x0000ff0000000000ULL) >> 24) | \
+		    (((uint64_t)(x) & 0x00ff000000000000ULL) >> 40) | \
+		    (((uint64_t)(x) & 0xff00000000000000ULL) >> 56)))
 
 #define __bswap_16(x) \
 	((uint16_t)(__builtin_constant_p(x) ? \
 	__bswap_constant_16(x) : \
 	__i386_bswap_16(x)))
 
-
 #define __bswap_32(x) \
 	((uint32_t)(__builtin_constant_p(x) ? \
 	__bswap_constant_32(x) : \
 	__i386_bswap_32(x)))
 
+#define __bswap_64(x) \
+	((uint64_t)(__builtin_constant_p(x) ? \
+	__bswap_constant_64(x) : \
+	__i386_bswap_64(x)))
 
 #endif /* ETHERBOOT_BITS_BYTESWAP_H */

src/include/big_bswap.h

@@ -5,12 +5,16 @@
 #define htonl(x) 	(x)
 #define ntohs(x) 	(x)
 #define htons(x) 	(x)
+#define cpu_to_le64(x)	__bswap_64(x)
 #define cpu_to_le32(x)	__bswap_32(x)
 #define cpu_to_le16(x)	__bswap_16(x)
+#define cpu_to_be64(x)	(x)
 #define cpu_to_be32(x)	(x)
 #define cpu_to_be16(x)	(x)
+#define le64_to_cpu(x)	__bswap_64(x)
 #define le32_to_cpu(x)	__bswap_32(x)
 #define le16_to_cpu(x)	__bswap_16(x)
+#define be64_to_cpu(x)	(x)
 #define be32_to_cpu(x)	(x)
 #define be16_to_cpu(x)	(x)
 

src/include/byteswap.h

@@ -12,8 +12,10 @@
 #endif
 
 /* Make routines available to all */
+#define swap64(x)	__bswap_64(x)
 #define swap32(x)	__bswap_32(x)
 #define swap16(x)	__bswap_16(x)
+#define bswap_64(x)	__bswap_64(x)
 #define bswap_32(x)	__bswap_32(x)
 #define bswap_16(x)	__bswap_16(x)
 	

src/include/little_bswap.h

@@ -5,12 +5,16 @@
 #define htonl(x) 	__bswap_32(x)
 #define ntohs(x) 	__bswap_16(x)
 #define htons(x) 	__bswap_16(x)
+#define cpu_to_le64(x)	(x)
 #define cpu_to_le32(x)	(x)
 #define cpu_to_le16(x)	(x)
+#define cpu_to_be64(x)	__bswap_64(x)
 #define cpu_to_be32(x)	__bswap_32(x)
 #define cpu_to_be16(x)	__bswap_16(x)
+#define le64_to_cpu(x)	(x)
 #define le32_to_cpu(x)	(x)
 #define le16_to_cpu(x)	(x)
+#define be64_to_cpu(x)	__bswap_64(x)
 #define be32_to_cpu(x)	__bswap_32(x)
 #define be16_to_cpu(x)	__bswap_16(x)