x86/pgtable: explain constant sign extension problem

When the _PAGE_FOO constants are defined as (1ul << _PAGE_BIT_FOO), they
become unsigned longs.  In 32-bit PAE mode, these end up being
implicitly cast to 64-bit types when used to manipulate a pte, and
because they're unsigned the top 32-bits are 0, destroying the upper
bits of the pte.

When _PAGE_FOO constants are given a signed integer type, the cast to
64-bits will sign-extend so that the upper bits are all ones,
preserving the upper pte bits in manipulations.

Explain this in a prominent place.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

include/asm-x86/pgtable.h

@@ -19,6 +19,11 @@
 #define _PAGE_BIT_UNUSED3	11
 #define _PAGE_BIT_NX           63       /* No execute: only valid after cpuid check */
 
+/*
+ * Note: we use _AC(1, L) instead of _AC(1, UL) so that we get a
+ * sign-extended value on 32-bit with all 1's in the upper word,
+ * which preserves the upper pte values on 64-bit ptes:
+ */
 #define _PAGE_PRESENT	(_AC(1, L)<<_PAGE_BIT_PRESENT)
 #define _PAGE_RW	(_AC(1, L)<<_PAGE_BIT_RW)
 #define _PAGE_USER	(_AC(1, L)<<_PAGE_BIT_USER)