xen/pat: Disable PAT using pat_enabled value.

The git commit 8eaffa6
(xen/pat: Disable PAT support for now) explains in details why
we want to disable PAT for right now. However that
change was not enough and we should have also disabled
the pat_enabled value. Otherwise we end up with:

mmap-example:3481 map pfn expected mapping type write-back for
[mem 0x00010000-0x00010fff], got uncached-minus
 ------------[ cut here ]------------
WARNING: at /build/buildd/linux-3.8.0/arch/x86/mm/pat.c:774 untrack_pfn+0xb8/0xd0()
mem 0x00010000-0x00010fff], got uncached-minus
------------[ cut here ]------------
WARNING: at /build/buildd/linux-3.8.0/arch/x86/mm/pat.c:774
untrack_pfn+0xb8/0xd0()
...
Pid: 3481, comm: mmap-example Tainted: GF 3.8.0-6-generic #13-Ubuntu
Call Trace:
 [<ffffffff8105879f>] warn_slowpath_common+0x7f/0xc0
 [<ffffffff810587fa>] warn_slowpath_null+0x1a/0x20
 [<ffffffff8104bcc8>] untrack_pfn+0xb8/0xd0
 [<ffffffff81156c1c>] unmap_single_vma+0xac/0x100
 [<ffffffff81157459>] unmap_vmas+0x49/0x90
 [<ffffffff8115f808>] exit_mmap+0x98/0x170
 [<ffffffff810559a4>] mmput+0x64/0x100
 [<ffffffff810560f5>] dup_mm+0x445/0x660
 [<ffffffff81056d9f>] copy_process.part.22+0xa5f/0x1510
 [<ffffffff81057931>] do_fork+0x91/0x350
 [<ffffffff81057c76>] sys_clone+0x16/0x20
 [<ffffffff816ccbf9>] stub_clone+0x69/0x90
 [<ffffffff816cc89d>] ? system_call_fastpath+0x1a/0x1f
---[ end trace 4918cdd0a4c9fea4 ]---

(a similar message shows up if you end up launching 'mcelog')

The call chain is (as analyzed by Liu, Jinsong):
do_fork
  --> copy_process
    --> dup_mm
      --> dup_mmap
       	--> copy_page_range
          --> track_pfn_copy
            --> reserve_pfn_range
              --> line 624: flags != want_flags
It comes from different memory types of page table (_PAGE_CACHE_WB) and MTRR
(_PAGE_CACHE_UC_MINUS).

Stefan Bader dug in this deep and found out that:
"That makes it clearer as this will do

reserve_memtype(...)
--> pat_x_mtrr_type
  --> mtrr_type_lookup
    --> __mtrr_type_lookup

And that can return -1/0xff in case of MTRR not being enabled/initialized. Which
is not the case (given there are no messages for it in dmesg). This is not equal
to MTRR_TYPE_WRBACK and thus becomes _PAGE_CACHE_UC_MINUS.

It looks like the problem starts early in reserve_memtype:

       	if (!pat_enabled) {
                /* This is identical to page table setting without PAT */
                if (new_type) {
                        if (req_type == _PAGE_CACHE_WC)
                                *new_type = _PAGE_CACHE_UC_MINUS;
                        else
                               	*new_type = req_type & _PAGE_CACHE_MASK;
               	}
                return 0;
        }

This would be what we want, that is clearing the PWT and PCD flags from the
supported flags - if pat_enabled is disabled."

This patch does that - disabling PAT.

CC: stable@vger.kernel.org # 3.3 and further
Reported-by: Sander Eikelenboom <linux@eikelenboom.it>
Reported-and-Tested-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reported-and-Tested-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

arch/x86/xen/enlighten.c

@@ -67,6 +67,7 @@
 #include <asm/hypervisor.h>
 #include <asm/mwait.h>
 #include <asm/pci_x86.h>
+#include <asm/pat.h>
 
 #ifdef CONFIG_ACPI
 #include <linux/acpi.h>
@@ -1417,7 +1418,14 @@ asmlinkage void __init xen_start_kernel(void)
 	 */
 	acpi_numa = -1;
 #endif
-
+#ifdef CONFIG_X86_PAT
+	/*
+	 * For right now disable the PAT. We should remove this once
+	 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1
+	 * (xen/pat: Disable PAT support for now) is reverted.
+	 */
+	pat_enabled = 0;
+#endif
 	/* Don't do the full vcpu_info placement stuff until we have a
 	   possible map and a non-dummy shared_info. */
 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];