From: Chris Lalancette <clalance@redhat.com>
Date: Fri, 9 Jan 2009 12:07:41 +0100
Subject: [xen] allow > 4GB EPT guests on i386
Message-id: 49672FFD.1040307@redhat.com
O-Subject: [RHEL5.4 PATCH]: Allow > 4GB EPT guests on i386
Bugzilla: 478522
RH-Acked-by: Don Dutile <ddutile@redhat.com>
RH-Acked-by: Bill Burns <bburns@redhat.com>

All,
     When we initially put the HAP (i.e. NPT) patches into RHEL-5, we had a
piece of code that refused to start guests with > 4GB when NPT was enabled.
This is due to a hardware limitation in AMD's NPT implementation.  However,
Intel's EPT implementation does not have this limitation.  The problem is that
the check to start guests with > 4GB is in generic code.
     The following simple patch fixes that situation by only refusing to start >
4GB guests on i386 if the vendor == AMD.  This is a trivial re-diff of upstream
xen-unstable c/s 18943.
     Briefly tested by me on both i386 and x86_64.  This should resolve BZ
478522.  Please review and ACK.

Thanks,
Chris Lalancette

diff --git a/arch/x86/mm/p2m.c b/arch/x86/mm/p2m.c
index ad9ea0a..6f9e1d1 100644
--- a/arch/x86/mm/p2m.c
+++ b/arch/x86/mm/p2m.c
@@ -806,11 +806,12 @@ guest_physmap_add_entry(struct domain *d, unsigned long gfn,
 
 #if CONFIG_PAGING_LEVELS == 3
     /*
-     * 32bit PAE nested paging does not support over 4GB guest due to 
+     * 32bit AMD nested paging does not support over 4GB guest due to 
      * hardware translation limit. This limitation is checked by comparing
      * gfn with 0xfffffUL.
      */
-    if ( paging_mode_hap(d) && (gfn > 0xfffffUL) )
+    if ( paging_mode_hap(d) && (gfn > 0xfffffUL) &&
+         (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) )
     {
         if ( !test_and_set_bool(d->arch.hvm_domain.amd_npt_4gb_warning) )
             dprintk(XENLOG_WARNING, "Dom%d failed to populate memory beyond"