Xen devel - Sep 2007 - [VTD-NEO][patch 6/6] Intel VT-d/Neocleus 1:1 mreged code for PCI passthrough

arch/x86/mm changes.

Signed-off-by: Allen Kay <allen.m.kay@intel.com>
Signed-off-by: Guy Zana <guy@neocleus.com>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

This is the last piece of initial vt-d code check-in:

mm.patch  : worked with Tim Deegan''s memory type changes in p2m/shadow.
dom0.patch: create vt-d 1:1 mapping for dom0; call iommu_domain_init()
in domain.c.

These changes were tested against 15909.  The changes made by cs# 15910
breaks even non-vtd vmx guest bring-up.  It causes xen crash complaining
about:

 "(XEN) Failed vm entry (exit reason 0x80000021) caused by invalid guest
state (0)"

Allen

Signed-off-by: Allen Kay <allen.m.kay@intel.com>
Signed-off-by: Guy Zana <guy@neocleus.com>
>-----Original Message-----
>From: Kay, Allen M 
>Sent: Tuesday, September 04, 2007 5:09 PM
>To: xen-devel@lists.xensource.com
>Cc: ''Keir Fraser''; Guy Zana
>Subject: [VTD-NEO][patch 6/6] Intel VT-d/Neocleus 1:1 mreged 
>code for PCI passthrough
>
>arch/x86/mm changes.
>
>Signed-off-by: Allen Kay <allen.m.kay@intel.com>
>Signed-off-by: Guy Zana <guy@neocleus.com>
>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 19/9/07 22:21, "Kay, Allen M" <allen.m.kay@intel.com> wrote:
> dom0.patch: create vt-d 1:1 mapping for dom0; call iommu_domain_init()
> in domain.c.
Do PV guests actually need per-domain iommu state? If they do, isn''t
release_devices() the partnering destructor function, and so should that not
be invoked from domain.c too?
> These changes were tested against 15909.  The changes made by cs# 15910
> breaks even non-vtd vmx guest bring-up.  It causes xen crash complaining
> about:
> 
>  "(XEN) Failed vm entry (exit reason 0x80000021) caused by invalid
guest
> state (0)"
How is the guest state broken? Does guest rflags have bit 1 set?

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 20/9/07 09:35, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk>
wrote:
>> These changes were tested against 15909.  The changes made by cs# 15910
>> breaks even non-vtd vmx guest bring-up.  It causes xen crash
complaining
>> about:
>> 
>>  "(XEN) Failed vm entry (exit reason 0x80000021) caused by invalid
guest
>> state (0)"
> 
> How is the guest state broken? Does guest rflags have bit 1 set?
Ah, it''s broken for 64-bit hypervisor. I''m just testing a fix.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 20/9/07 09:35, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk>
wrote:
>> dom0.patch: create vt-d 1:1 mapping for dom0; call iommu_domain_init()
>> in domain.c.
> 
> Do PV guests actually need per-domain iommu state? If they do,
isn''t
> release_devices() the partnering destructor function, and so should that
not
> be invoked from domain.c too?
I checked in everything except the extra invocation of iommu_domain_init().
I want to understand what iommu_domain_init() and release_devices() do
before deciding whether to move their invocations.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>On 20/9/07 09:35, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk>
wrote:
>
>>> dom0.patch: create vt-d 1:1 mapping for dom0; call 
>iommu_domain_init()
>>> in domain.c.
>> 
>> Do PV guests actually need per-domain iommu state? If they do,
isn''t
>> release_devices() the partnering destructor function, and so 
>should that not
>> be invoked from domain.c too?
The reason iommu_domain_init() initializes iommu domain strcture.
Device_release() frees various memory used by iommu domain structure +
it calls return_devices_to_dom0().  Return_devices_to_dom0() reassigns
devices back to dom0 in vt-d hw and moves the pdev''s to dom0''s
pdev_list.

Iommu_domain_init() needs to be called in domain.c because dom0 needs
it.  Device_release can potentially be called in domain.c also -
probably just need to add "if (d == dom0) return;" in
return_devices_to_dom0().

PV guest is not currently supported right now.  We had few pieces here
and there but I removed it from the submission until we do some tests.

Allen

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel