x86, xen: Disable highmem PTE allocation even when CONFIG_HIGHPTE=y

There's a path in the pagefault code where the kernel deliberately
breaks its own locking rules by kmapping a high pte page without
holding the pagetable lock (in at least page_check_address). This
breaks Xen's ability to track the pinned/unpinned state of the
page. There does not appear to be a viable workaround for this
behaviour so simply disable HIGHPTE for all Xen guests.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
LKML-Reference: <1267204562-11844-1-git-send-email-ian.campbell@citrix.com>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Pasi Kärkkäinen <pasik@iki.fi>
Cc: <stable@kernel.org> # .32.x: 14315592: Allow highmem user page tables to be disabled at boot time
Cc: <stable@kernel.org> # .32.x
Cc: <xen-devel@lists.xensource.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>

arch/x86/xen/enlighten.c

@@ -50,6 +50,7 @@
 #include <asm/traps.h>
 #include <asm/setup.h>
 #include <asm/desc.h>
+#include <asm/pgalloc.h>
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
 #include <asm/reboot.h>
@@ -1094,6 +1095,12 @@ asmlinkage void __init xen_start_kernel(void)
 
 	__supported_pte_mask |= _PAGE_IOMAP;
 
+	/*
+	 * Prevent page tables from being allocated in highmem, even
+	 * if CONFIG_HIGHPTE is enabled.
+	 */
+	__userpte_alloc_gfp &= ~__GFP_HIGHMEM;
+
 	/* Work out if we support NX */
 	x86_configure_nx();