xen: properly account for _PAGE_NUMA during xen pte translations

Steven Noonan forwarded a users report where they had a problem starting
vsftpd on a Xen paravirtualized guest, with this in dmesg:

  BUG: Bad page map in process vsftpd  pte:8000000493b88165 pmd:e9cc01067
  page:ffffea00124ee200 count:0 mapcount:-1 mapping:     (null) index:0x0
  page flags: 0x2ffc0000000014(referenced|dirty)
  addr:00007f97eea74000 vm_flags:00100071 anon_vma:ffff880e98f80380 mapping:          (null) index:7f97eea74
  CPU: 4 PID: 587 Comm: vsftpd Not tainted 3.12.7-1-ec2 #1
  Call Trace:
    dump_stack+0x45/0x56
    print_bad_pte+0x22e/0x250
    unmap_single_vma+0x583/0x890
    unmap_vmas+0x65/0x90
    exit_mmap+0xc5/0x170
    mmput+0x65/0x100
    do_exit+0x393/0x9e0
    do_group_exit+0xcc/0x140
    SyS_exit_group+0x14/0x20
    system_call_fastpath+0x1a/0x1f
  Disabling lock debugging due to kernel taint
  BUG: Bad rss-counter state mm:ffff880e9ca60580 idx:0 val:-1
  BUG: Bad rss-counter state mm:ffff880e9ca60580 idx:1 val:1

The issue could not be reproduced under an HVM instance with the same
kernel, so it appears to be exclusive to paravirtual Xen guests.  He
bisected the problem to commit 1667918 ("mm: numa: clear numa
hinting information on mprotect") that was also included in 3.12-stable.

The problem was related to how xen translates ptes because it was not
accounting for the _PAGE_NUMA bit.  This patch splits pte_present to add
a pteval_present helper for use by xen so both bare metal and xen use
the same code when checking if a PTE is present.

[[email protected]: wrote changelog, proposed minor modifications]
[[email protected]: fix typo in comment]
Reported-by: Steven Noonan <[email protected]>
Tested-by: Steven Noonan <[email protected]>
Signed-off-by: Elena Ufimtseva <[email protected]>
Signed-off-by: Mel Gorman <[email protected]>
Reviewed-by: David Vrabel <[email protected]>
Acked-by: Konrad Rzeszutek Wilk <[email protected]>
Cc: <[email protected]>	[3.12+]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Author: Mel Gorman

arch/x86/include/asm/pgtable.h

@@ -445,10 +445,20 @@ static inline int pte_same(pte_t a, pte_t b)
 	return a.pte == b.pte;
 }
 
+static inline int pteval_present(pteval_t pteval)
+{
+	/*
+	 * Yes Linus, _PAGE_PROTNONE == _PAGE_NUMA. Expressing it this
+	 * way clearly states that the intent is that protnone and numa
+	 * hinting ptes are considered present for the purposes of
+	 * pagetable operations like zapping, protection changes, gup etc.
+	 */
+	return pteval & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_NUMA);
+}
+
 static inline int pte_present(pte_t a)
 {
-	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE |
-			       _PAGE_NUMA);
+	return pteval_present(pte_flags(a));
 }
 
 #define pte_accessible pte_accessible

arch/x86/xen/mmu.c

@@ -365,7 +365,7 @@ void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
 /* Assume pteval_t is equivalent to all the other *val_t types. */
 static pteval_t pte_mfn_to_pfn(pteval_t val)
 {
-	if (val & _PAGE_PRESENT) {
+	if (pteval_present(val)) {
 		unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
 		unsigned long pfn = mfn_to_pfn(mfn);
 
@@ -381,7 +381,7 @@ static pteval_t pte_mfn_to_pfn(pteval_t val)
 
 static pteval_t pte_pfn_to_mfn(pteval_t val)
 {
-	if (val & _PAGE_PRESENT) {
+	if (pteval_present(val)) {
 		unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
 		pteval_t flags = val & PTE_FLAGS_MASK;
 		unsigned long mfn;