Xu, Anthony wrote:
> Isaku Yamahata wrote:
>> On Thu, Oct 23, 2008 at 10:31:20AM +0800, Xu, Anthony wrote:
>>> Iommu_un/map_page does nothing for PV domain.
>>> Why is there a race?
>> Oh sorry. I'm not very familiar with iommu code (yet).
>> My understanding is
>> iommu_domain_init() is called unconditionaly by
>> arch_domain_create() so that platform_ops is initialized.
>> iommu_(un)map_page() checks iommu_enabled and whether platform_ops
>> is initialized. Thus iommu_(un)map_page() manipulates iommu page
>> table even for PV domain.
>> Clearly I'm missing something important. Could you enlighten me?
> Acctually I missed something.
> First, PV domain can use VTD in x86 side.
> So before calling iommu_(un)map_page, Xen needs to check if it is
> if ( iommu_enabled && (is_hvm_domain(d) || need_iommu(d)) )
> I'll update the patch
There is an issue in VTD,
That's even if no device is assigned to hvm guest, VTD page table is still
Intel is working on that, to make sure setting up VTD page table only when
> Xen-ia64-devel mailing list
Xen-ia64-devel mailing list