arm64: Use simpler arithmetics for the linear map macros

Because of the tagged addresses, the __is_lm_address() and
__lm_to_phys() macros grew to some harder to understand bitwise
operations using PAGE_OFFSET. Since these macros only accept untagged
addresses, use a simple subtract operation.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Link: https://lore.kernel.org/r/20210201190634.22942-3-catalin.marinas@arm.com

arch/arm64/include/asm/memory.h

@@ -251,9 +251,9 @@ static inline const void *__tag_set(const void *addr, u8 tag)
  * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
  * kernel's TTBR1 address range.
  */
-#define __is_lm_address(addr)	(((u64)(addr) ^ PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
+#define __is_lm_address(addr)	(((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
 
-#define __lm_to_phys(addr)	(((addr) & ~PAGE_OFFSET) + PHYS_OFFSET)
+#define __lm_to_phys(addr)	(((addr) - PAGE_OFFSET) + PHYS_OFFSET)
 #define __kimg_to_phys(addr)	((addr) - kimage_voffset)
 
 #define __virt_to_phys_nodebug(x) ({					\