Re: [Xen-devel] [PATCH 0 of 3] RFC Paging support for AMD NPT V2

[Hongkaixing <hongkaixing@xxxxxxxxxx>]

> -----Original Message-----
> From: Andres Lagar-Cavilla [mailto:andres@xxxxxxxxxxxxxxxx]
> Sent: Friday, March 02, 2012 12:34 AM
> To: Hongkaixing
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx; tim@xxxxxxx; andres@xxxxxxxxxxxxxx; 
> olaf@xxxxxxxxx; adin@xxxxxxxxxxxxxx; wei.wang2@xxxxxxx;
> yanqiangjun@xxxxxxxxxx
> Subject: RE: [PATCH 0 of 3] RFC Paging support for AMD NPT V2
> 
> >
> >
> >> -----Original Message-----
> >> From: Andres Lagar-Cavilla [mailto:andres@xxxxxxxxxxxxxxxx]
> >> Sent: Thursday, March 01, 2012 11:16 AM
> >> To: xen-devel@xxxxxxxxxxxxxxxxxxx
> >> Cc: tim@xxxxxxx; andres@xxxxxxxxxxxxxx; olaf@xxxxxxxxx;
> >> adin@xxxxxxxxxxxxxx; wei.wang2@xxxxxxx; hongkaixing@xxxxxxxxxx
> >> Subject: [PATCH 0 of 3] RFC Paging support for AMD NPT V2
> >>
> >> There has been some progress, but still no joy. Definitely not intended
> >> for
> >> inclusion at this point.
> >>
> >> Tim, Wei, I added a Xen command line toggle to disable IOMMU and P2M
> >> table
> >> sharing.
> >>
> >> Tim, I verified that changes to p2m-pt.c don't break shadow mode (64bit
> >> hypervisor and Win 7 guest).
> >>
> >> Hongkaixing, I incorporated your suggestion in patch 2, so I should add
> >> your
> >> Signed-off-by eventually. Please review.
> >
> >   I have checked the code, it looks like OK. But I don't have AMD machine
> > to run it.
> >   We used almost the same code in a AMD server months before, and it
> > worked fine.
> >   The only difference is we have not changed the clipped invalid mfn to
> > INVALID_MFN, just like EPT.
> 
> That is most interesting. Can you elaborate? do you have functional paging
> on AMD since "months before"? If you do, can you please post it to
> xen-devel for inclusion?
> 
> Thanks a lot,
> Andres

All what we have done is shown below.
The code is based on xen 4.0.2, it looks informal, but it works fine when 
xenpaging target is set at arbitrary number.. 
Now we do not have AMD platform to validate, so I am not sure it can work under 
unstable code.

Index: E:/xen-4.0.2/xen/arch/x86/mm/p2m.c
===================================================================
--- E:/xen-4.0.2/xen/arch/x86/mm/p2m.c  (revision 30771)
+++ E:/xen-4.0.2/xen/arch/x86/mm/p2m.c  (revision 30772)
@@ -77,6 +77,13 @@
 #define SUPERPAGE_PAGES (1UL << 9)
 #define superpage_aligned(_x)  (((_x)&(SUPERPAGE_PAGES-1))==0)
 
+#ifdef __x86_64__
+/* there are only 40bits for pfn in PTE */
+#define MASK_MFN(x)  ( x & ((1UL << 40) -1 ))
+#else
+#define MASK_MFN(x)  ( x )
+#endif
+
 static unsigned long p2m_type_to_flags(p2m_type_t t) 
 {
     unsigned long flags;
@@ -1306,6 +1313,8 @@
         
         if ( mfn_valid(mfn) || (p2mt == p2m_mmio_direct) )
             entry_content = l1e_from_pfn(mfn_x(mfn), p2m_type_to_flags(p2mt));
+        else if ( p2m_is_paging( p2mt ) )
+            entry_content = l1e_from_pfn( MASK_MFN(mfn_x(mfn)), 
p2m_type_to_flags(p2mt));
         else
             entry_content = l1e_empty();
         
@@ -1370,6 +1379,7 @@
     paddr_t addr = ((paddr_t)gfn) << PAGE_SHIFT;
     l2_pgentry_t *l2e;
     l1_pgentry_t *l1e;
+    unsigned long flags;
 
     ASSERT(paging_mode_translate(d));
 
@@ -1455,7 +1465,8 @@
     l1e = map_domain_page(mfn_x(mfn));
     l1e += l1_table_offset(addr);
 pod_retry_l1:
-    if ( (l1e_get_flags(*l1e) & _PAGE_PRESENT) == 0 )
+    flags = l1e_get_flags(*l1e);
+    if ((( flags & _PAGE_PRESENT) == 0) && 
(!p2m_is_paging(p2m_flags_to_type(flags))))
     {
         /* PoD: Try to populate */
         if ( p2m_flags_to_type(l1e_get_flags(*l1e)) == p2m_populate_on_demand )
Index: E: /xen-4.0.2/xen/arch/x86/mm/mem_event.c
===================================================================
--- E:/xen-4.0.2/xen/arch/x86/mm/mem_event.c    (revision 30771)
+++ E:/xen-4.0.2/xen/arch/x86/mm/mem_event.c    (revision 30772)
@@ -231,10 +231,9 @@
             if ( d->mem_event.ring_page )
                 break;
 
-            /* Currently only EPT is supported */
+            /* Both AMD NPT and intel EPT are supported */
             rc = -ENODEV;
-            if ( !(is_hvm_domain(d) && d->arch.hvm_domain.hap_enabled &&
-                  (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)) )
+            if ( !(is_hvm_domain(d) && d->arch.hvm_domain.hap_enabled ))
                 break;
 
             /* Get MFN of ring page */
> 
> >
> >
> >>
> >> Olaf, I do not see errors when mapping pages prior to eviction. Let me
> >> know how
> >> it goes for you.
> >>
> >> If I unleash xenpaging on a domain for a non-ambitious target (64 MiBs
> >> paged
> >> out for one minute), we sail by just fine. But if I kick things up a
> >> notch,
> >> still dying on an VMEXIT_SHUTDOWN (a.k.a. triple fault). Not sure at all
> >> what
> >> to do next.
> >>
> >> Signed-off-by: Andres Lagar-Cavilla <andres@xxxxxxxxxxxxxxxx>
> >> Signed-off-by: Adin Scannell <adin@xxxxxxxxxxx>
> >>
> >>  xen/drivers/passthrough/iommu.c |   2 +
> >>  xen/arch/x86/mm/p2m-pt.c        |  56
> >> ++++++++++++++++++++++++++++------------
> >>  xen/arch/x86/mm/mem_event.c     |   7 +++-
> >>  xen/arch/x86/mm/mem_sharing.c   |   7 +++++
> >>  4 files changed, 53 insertions(+), 19 deletions(-)
> >
> >



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel