[Xen-devel] [PATCH for 4.6] x86/mm: make {set, clear}_identity_p2m_mapping() work for HVM guests as well

[Andrew Cooper]

From: Malcolm Crossley 

For Intel IGD passthrough, the guest driver programs the device to DMA
to/from its RMRR.

c/s 619ecf8 "make {set,clear}_identity_p2m_mapping() work for PV guests"
was incomplete for pre-Broadwell systems which did not support shared
EPT.

The correct check to use is iommu_use_hap_pt() not
paging_mode_translate() as IOMMU mappings also need to be created for
HVM guests with separate EPT and IOMMU tables.

Signed-off-by: Malcolm Crossley 
Signed-off-by: Andrew Cooper 
CC: Jan Beulich 
CC: George Dunlap 
CC: Wei Liu 
---
 xen/arch/x86/mm/p2m.c |4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
index c4329d2..c7f437c 100644
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -957,7 +957,7 @@ int set_identity_p2m_entry(struct domain *d, unsigned long 
gfn,
 struct p2m_domain *p2m = p2m_get_hostp2m(d);
 int ret;
 
-if ( !paging_mode_translate(p2m->domain) )
+if ( !iommu_use_hap_pt(d) )
 {
 if ( !need_iommu(d) )
 return 0;
@@ -1032,7 +1032,7 @@ int clear_identity_p2m_entry(struct domain *d, unsigned 
long gfn)
 struct p2m_domain *p2m = p2m_get_hostp2m(d);
 int ret;
 
-if ( !paging_mode_translate(d) )
+if ( !iommu_use_hap_pt(d) )
 {
 if ( !need_iommu(d) )
 return 0;
-- 
1.7.10.4


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

Re: [Xen-devel] [PATCH for 4.6] x86/mm: make {set, clear}_identity_p2m_mapping() work for HVM guests as well

[Jan Beulich]

>>> On 14.09.15 at 16:24,  wrote:
> For Intel IGD passthrough, the guest driver programs the device to DMA
> to/from its RMRR.
> 
> c/s 619ecf8 "make {set,clear}_identity_p2m_mapping() work for PV guests"
> was incomplete for pre-Broadwell systems which did not support shared
> EPT.
> 
> The correct check to use is iommu_use_hap_pt() not
> paging_mode_translate() as IOMMU mappings also need to be created for
> HVM guests with separate EPT and IOMMU tables.

No, at least not without a very good explanation:

> --- a/xen/arch/x86/mm/p2m.c
> +++ b/xen/arch/x86/mm/p2m.c
> @@ -957,7 +957,7 @@ int set_identity_p2m_entry(struct domain *d, unsigned 
> long gfn,
>  struct p2m_domain *p2m = p2m_get_hostp2m(d);
>  int ret;
>  
> -if ( !paging_mode_translate(p2m->domain) )
> +if ( !iommu_use_hap_pt(d) )

Not only have these checks no reason to be in any way IOMMU
dependent (no other function around here does so), but also
both ept_set_entry() and p2m_pt_set_entry() take care to call
iommu_{,un}map_page() as needed (as I had already pointed
out back when we discussed the patch this one tries to complete).
If anything, it would be those low level functions that may need
tweaking (and which have a reason to consider IOMMU flags) if
for some reason they don't actually do the expected calls.

Jan


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