H. Peter Anvin wrote:
> Jeremy Fitzhardinge wrote:
> 
>> (To be specific: I reshape the guest e820 table so that it doesn't have
>> memory in any forbidden areas of the host e820 table.  That may require
>> moving the pseudo-physical address of pages into a new overflow e820
>> entry, which would also require allocating pages for the p2m radix tree.)
>>
>> 
>
> Isn't that the domain builder's job?
> 

No.  It just provides a flat memory map from 0-max_pfn.  I want to avoid 
the overlap between pseudo-phys and machine memory so that the resource 
manager doesn't get confused by overlapping ranges, and so that it 
places the pci window in the same place as native.

    J