xen: make include/xen/unaligned.h usable on all architectures

Instead of defining get_unaligned() and put_unaligned() in a way that
is only supporting architectures allowing unaligned accesses, use the
same approach as the Linux kernel and let the compiler do the
decision how to generate the code for probably unaligned data accesses.

Update include/xen/unaligned.h from include/asm-generic/unaligned.h of
the Linux kernel.

The generated code has been checked to be the same on x86.

Modify the Linux variant to not use underscore prefixed identifiers,
avoid unneeded parentheses and drop the 24-bit accessors.

Add the definition of __packed to xg_dom_decompress_unsafe_zstd.c in
libxenguest as it is using a cruel hack to reuse the hypervisor's
decompressing code for stubdom.

Signed-off-by: Arnd Bergmann <[email protected]>
Origin: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 803f4e1eab7a
Signed-off-by: Juergen Gross <[email protected]>
Acked-by: Jan Beulich <[email protected]>

tools/libs/guest/xg_dom_decompress_unsafe_zstd.c

@@ -26,13 +26,13 @@ typedef uint64_t __be64;
 #define __force
 #define always_inline
 #define noinline
+#define __packed __attribute__((__packed__))
 
 #undef ERROR
 
 #define __BYTEORDER_HAS_U64__
 #define __TYPES_H__ /* xen/types.h guard */
 #include "../../xen/include/xen/byteorder/little_endian.h"
-#define __ASM_UNALIGNED_H__ /* asm/unaligned.h guard */
 #include "../../xen/include/xen/unaligned.h"
 #include "../../xen/include/xen/xxhash.h"
 #include "../../xen/lib/xxhash64.c"

xen/include/xen/unaligned.h

@@ -1,12 +1,4 @@
-/*
- * This header can be used by architectures where unaligned accesses work
- * without faulting, and at least reasonably efficiently.  Other architectures
- * will need to have a custom asm/unaligned.h.
- */
-#ifndef __ASM_UNALIGNED_H__
-#error "xen/unaligned.h should not be included directly - include asm/unaligned.h instead"
-#endif
-
+/* SPDX-License-Identifier: GPL-2.0 */
 #ifndef __XEN_UNALIGNED_H__
 #define __XEN_UNALIGNED_H__
 
@@ -15,67 +7,82 @@
 #include <asm/byteorder.h>
 #endif
 
-#define get_unaligned(p) (*(p))
-#define put_unaligned(val, p) (*(p) = (val))
+/*
+ * This is the most generic implementation of unaligned accesses
+ * and should work almost anywhere.
+ */
+
+#define get_unaligned_t(type, ptr) ({					\
+	const struct { type x; } __packed *ptr_ = (typeof(ptr_))(ptr);	\
+	ptr_->x;							\
+})
+
+#define put_unaligned_t(type, val, ptr) do {				\
+	struct { type x; } __packed *ptr_ = (typeof(ptr_))(ptr);	\
+	ptr_->x = val;							\
+} while (0)
+
+#define get_unaligned(ptr)	get_unaligned_t(typeof(*(ptr)), ptr)
+#define put_unaligned(val, ptr) put_unaligned_t(typeof(*(ptr)), val, ptr)
 
 static inline uint16_t get_unaligned_be16(const void *p)
 {
-	return be16_to_cpup(p);
+	return be16_to_cpu(get_unaligned_t(__be16, p));
 }
 
 static inline void put_unaligned_be16(uint16_t val, void *p)
 {
-	*(__force __be16*)p = cpu_to_be16(val);
+	put_unaligned_t(__be16, cpu_to_be16(val), p);
 }
 
 static inline uint32_t get_unaligned_be32(const void *p)
 {
-	return be32_to_cpup(p);
+	return be32_to_cpu(get_unaligned_t(__be32, p));
 }
 
 static inline void put_unaligned_be32(uint32_t val, void *p)
 {
-	*(__force __be32*)p = cpu_to_be32(val);
+	put_unaligned_t(__be32, cpu_to_be32(val), p);
 }
 
 static inline uint64_t get_unaligned_be64(const void *p)
 {
-	return be64_to_cpup(p);
+	return be64_to_cpu(get_unaligned_t(__be64, p));
 }
 
 static inline void put_unaligned_be64(uint64_t val, void *p)
 {
-	*(__force __be64*)p = cpu_to_be64(val);
+	put_unaligned_t(__be64, cpu_to_be64(val), p);
 }
 
 static inline uint16_t get_unaligned_le16(const void *p)
 {
-	return le16_to_cpup(p);
+	return le16_to_cpu(get_unaligned_t(__le16, p));
 }
 
 static inline void put_unaligned_le16(uint16_t val, void *p)
 {
-	*(__force __le16*)p = cpu_to_le16(val);
+	put_unaligned_t(__le16, cpu_to_le16(val), p);
 }
 
 static inline uint32_t get_unaligned_le32(const void *p)
 {
-	return le32_to_cpup(p);
+	return le32_to_cpu(get_unaligned_t(__le32, p));
 }
 
 static inline void put_unaligned_le32(uint32_t val, void *p)
 {
-	*(__force __le32*)p = cpu_to_le32(val);
+	put_unaligned_t(__le32, cpu_to_le32(val), p);
 }
 
 static inline uint64_t get_unaligned_le64(const void *p)
 {
-	return le64_to_cpup(p);
+	return le64_to_cpu(get_unaligned_t(__le64, p));
 }
 
 static inline void put_unaligned_le64(uint64_t val, void *p)
 {
-	*(__force __le64*)p = cpu_to_le64(val);
+	put_unaligned_t(__le64, cpu_to_le64(val), p);
 }
 
 #endif /* __XEN_UNALIGNED_H__ */