x86/mm: Allow userspace have mappings above 47-bit

All bits and pieces are now in place and we can allow userspace to have VMAs
above 47 bits.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-arch@vger.kernel.org
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/20170716225954.74185-8-kirill.shutemov@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/include/asm/processor.h

@@ -850,9 +850,9 @@ static inline void spin_lock_prefetch(const void *x)
  * particular problem by preventing anything from being mapped
  * at the maximum canonical address.
  */
-#define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
+#define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
 
-#define DEFAULT_MAP_WINDOW	TASK_SIZE_MAX
+#define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
 
 /* This decides where the kernel will search for a free chunk of vm
  * space during mmap's.