swab() is massively over complicated and can be simplified
by re-implementing using compiler builtins.
The compilers provide builtin function to swap bytes.
* gcc: https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html
* clang: https://clang.llvm.org/docs/LanguageExtensions.html
This patch introduces a new byteswapping infrastructure in
terms of compiler builtins and bswapXX(), so the swab() infrastructure
can be retired.
Signed-off-by: Lin Liu <lin....@citrix.com>
Reviewed-by: Andrew Cooper <andrew.coop...@citrix.com>
---
Cc: Stefano Stabellini <sstabell...@kernel.org>
Cc: Julien Grall <jul...@xen.org>
Cc: Bertrand Marquis <bertrand.marq...@arm.com>
Cc: Volodymyr Babchuk <volodymyr_babc...@epam.com>
Cc: Andrew Cooper <andrew.coop...@citrix.com>
Cc: George Dunlap <george.dun...@citrix.com>
Cc: Jan Beulich <jbeul...@suse.com>
Cc: Wei Liu <w...@xen.org>
Cc: "Roger Pau Monné" <roger....@citrix.com>
Changes in v4:
- Move `PASTE` definition to xen/compiler.h to support tools
- Revert emacs magics
Changes in v3:
- Check __has_builtin instead of GNUC version
Changes in v2:
- Add fallback for compilers without __builtin_bswap
- Implement with plain C instead of macros
---
xen/arch/arm/include/asm/byteorder.h | 6 ++-
xen/arch/x86/include/asm/byteorder.h | 34 ++---------------
xen/include/xen/byteorder.h | 56 ++++++++++++++++++++++++++++
xen/include/xen/byteswap.h | 44 ++++++++++++++++++++++
xen/include/xen/compiler.h | 24 ++++++++++++
xen/include/xen/lib.h | 4 --
6 files changed, 132 insertions(+), 36 deletions(-)
create mode 100644 xen/include/xen/byteorder.h
create mode 100644 xen/include/xen/byteswap.h
diff --git a/xen/arch/arm/include/asm/byteorder.h
b/xen/arch/arm/include/asm/byteorder.h
index 9c712c4788..b6a33b23c0 100644
--- a/xen/arch/arm/include/asm/byteorder.h
+++ b/xen/arch/arm/include/asm/byteorder.h
@@ -1,9 +1,11 @@
#ifndef __ASM_ARM_BYTEORDER_H__
#define __ASM_ARM_BYTEORDER_H__
-#define __BYTEORDER_HAS_U64__
+#ifndef __BYTE_ORDER__
+ #define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
+#endif
-#include <xen/byteorder/little_endian.h>
+#include <xen/byteorder.h>
#endif /* __ASM_ARM_BYTEORDER_H__ */
/*
diff --git a/xen/arch/x86/include/asm/byteorder.h
b/xen/arch/x86/include/asm/byteorder.h
index 1f77e502a5..82aadee7bd 100644
--- a/xen/arch/x86/include/asm/byteorder.h
+++ b/xen/arch/x86/include/asm/byteorder.h
@@ -1,36 +1,10 @@
#ifndef __ASM_X86_BYTEORDER_H__
#define __ASM_X86_BYTEORDER_H__
-#include <asm/types.h>
-#include <xen/compiler.h>
+#ifndef __BYTE_ORDER__
+ #define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
+#endif
-static inline __attribute_const__ __u32 ___arch__swab32(__u32 x)
-{
- asm("bswap %0" : "=r" (x) : "0" (x));
- return x;
-}
-
-static inline __attribute_const__ __u64 ___arch__swab64(__u64 val)
-{
- union {
- struct { __u32 a,b; } s;
- __u64 u;
- } v;
- v.u = val;
- asm("bswapl %0 ; bswapl %1 ; xchgl %0,%1"
- : "=r" (v.s.a), "=r" (v.s.b)
- : "0" (v.s.a), "1" (v.s.b));
- return v.u;
-}
-
-/* Do not define swab16. Gcc is smart enough to recognize "C" version and
- convert it into rotation or exhange. */
-
-#define __arch__swab64(x) ___arch__swab64(x)
-#define __arch__swab32(x) ___arch__swab32(x)
-
-#define __BYTEORDER_HAS_U64__
-
-#include <xen/byteorder/little_endian.h>
+#include <xen/byteorder.h>
#endif /* __ASM_X86_BYTEORDER_H__ */
diff --git a/xen/include/xen/byteorder.h b/xen/include/xen/byteorder.h
new file mode 100644
index 0000000000..2ec434e6a6
--- /dev/null
+++ b/xen/include/xen/byteorder.h
@@ -0,0 +1,56 @@
+#ifndef __XEN_BYTEORDER_H__
+#define __XEN_BYTEORDER_H__
+
+#include <xen/byteswap.h>
+
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+
+# ifndef __LITTLE_ENDIAN
+# define __LITTLE_ENDIAN 1234
+# endif
+
+# ifndef __LITTLE_ENDIAN_BITFIELD
+# define __LITTLE_ENDIAN_BITFIELD
+# endif
+
+# define cpu_to_le64(x) (x)
+# define le64_to_cpu(x) (x)
+# define cpu_to_le32(x) (x)
+# define le32_to_cpu(x) (x)
+# define cpu_to_le16(x) (x)
+# define le16_to_cpu(x) (x)
+# define cpu_to_be64(x) bswap64(x)
+# define be64_to_cpu(x) bswap64(x)
+# define cpu_to_be32(x) bswap32(x)
+# define be32_to_cpu(x) bswap32(x)
+# define cpu_to_be16(x) bswap16(x)
+# define be16_to_cpu(x) bswap16(x)
+
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+
+# ifndef __BIG_ENDIAN
+# define __BIG_ENDIAN 4321
+# endif
+
+# ifndef __BIG_ENDIAN_BITFIELD
+# define __BIG_ENDIAN_BITFIELD
+# endif
+
+# define cpu_to_le64(x) bswap64(x)
+# define le64_to_cpu(x) bswap64(x)
+# define cpu_to_le32(x) bswap32(x)
+# define le32_to_cpu(x) bswap32(x)
+# define cpu_to_le16(x) bswap16(x)
+# define le16_to_cpu(x) bswap16(x)
+# define cpu_to_be64(x) (x)
+# define be64_to_cpu(x) (x)
+# define cpu_to_be32(x) (x)
+# define be32_to_cpu(x) (x)
+# define cpu_to_be16(x) (x)
+# define be16_to_cpu(x) (x)
+
+#else
+# error "Unknown Endianness"
+#endif /* __BYTE_ORDER__ */
+
+#endif /* __XEN_BYTEORDER_H__ */
diff --git a/xen/include/xen/byteswap.h b/xen/include/xen/byteswap.h
new file mode 100644
index 0000000000..d2e371fbe7
--- /dev/null
+++ b/xen/include/xen/byteswap.h
@@ -0,0 +1,44 @@
+#ifndef __XEN_BYTESWAP_H__
+#define __XEN_BYTESWAP_H__
+
+#include <xen/types.h>
+#include <xen/compiler.h>
+
+#if !__has_builtin(__builtin_bswap16)
+static always_inline uint16_t __builtin_bswap16(uint16_t val)
+{
+ return ((val & 0x00FF) << 8) | ((val & 0xFF00) >> 8);
+}
+#endif
+
+#if !__has_builtin(__builtin_bswap32)
+static always_inline uint32_t __builtin_bswap32(uint32_t val)
+{
+ return ((val & 0x000000FF) << 24) |
+ ((val & 0x0000FF00) << 8) |
+ ((val & 0x00FF0000) >> 8) |
+ ((val & 0xFF000000) >> 24);
+}
+#endif
+
+#if !__has_builtin(__builtin_bswap64)
+static always_inline uint64_t __builtin_bswap64(uint64_t val)
+{
+ return ((val & 0x00000000000000FF) << 56) |
+ ((val & 0x000000000000FF00) << 40) |
+ ((val & 0x0000000000FF0000) << 24) |
+ ((val & 0x00000000FF000000) << 8) |
+ ((val & 0x000000FF00000000) >> 8) |
+ ((val & 0x0000FF0000000000) >> 24) |
+ ((val & 0x00FF000000000000) >> 40) |
+ ((val & 0xFF00000000000000) >> 56);
+}
+#endif
+
+#define bswap16(x) __builtin_bswap16(x)
+#define bswap32(x) __builtin_bswap32(x)
+#define bswap64(x) __builtin_bswap64(x)
+
+#define bswap_ul(x) PASTE(bswap,BITS_PER_LONG)(x)
+
+#endif /* __XEN_BYTESWAP_H__ */
diff --git a/xen/include/xen/compiler.h b/xen/include/xen/compiler.h
index 933aec09a9..ae029afa14 100644
--- a/xen/include/xen/compiler.h
+++ b/xen/include/xen/compiler.h
@@ -185,4 +185,28 @@
# define CLANG_DISABLE_WARN_GCC_COMPAT_END
#endif
+#ifndef __has_builtin
+/*
+ * Backwards compatibility for GCC < 10.
+ * All supported versions of Clang support __has_builtin
+ * */
+#define __has_builtin(x) GCC_has ## x
+
+#define GCC_has__builtin_bswap16 (CONFIG_GCC_VERSION >= 40800)
+#define GCC_has__builtin_bswap32 (CONFIG_GCC_VERSION >= 40400)
+#define GCC_has__builtin_bswap64 (CONFIG_GCC_VERSION >= 40400)
+#endif
+
+#ifndef __ORDER_LITTLE_ENDIAN__
+# define __ORDER_LITTLE_ENDIAN__ 1234
+#endif
+
+#ifndef __ORDER_BIG_ENDIAN__
+# define __ORDER_BIG_ENDIAN__ 4321
+#endif
+
+/* Indirect macros required for expanded argument pasting. */
+#define PASTE_(a, b) a ## b
+#define PASTE(a, b) PASTE_(a, b)
+
#endif /* __LINUX_COMPILER_H */
diff --git a/xen/include/xen/lib.h b/xen/include/xen/lib.h
index aab1fc7c4a..ebf2193569 100644
--- a/xen/include/xen/lib.h
+++ b/xen/include/xen/lib.h
@@ -15,10 +15,6 @@
#define count_args(args...) \
count_args_(., ## args, 8, 7, 6, 5, 4, 3, 2, 1, 0)
-/* Indirect macros required for expanded argument pasting. */
-#define PASTE_(a, b) a ## b
-#define PASTE(a, b) PASTE_(a, b)
-
#define __STR(...) #__VA_ARGS__
#define STR(...) __STR(__VA_ARGS__)
--
2.27.0
