Re: [PATCH] Handle vma regions with no backing page (v2)

[Anthony Liguori <aliguori@us.ibm.com>]

Hollis Blanchard wrote:
> On Tuesday 29 April 2008 17:17:49 Avi Kivity wrote:
>   
>> I like this very much, as it only affects accessors and not the mmu core 
>> itself.
>>
>> Hollis/Xiantao/Carsten, can you confirm that this approach works for 
>> you?  Carsten, I believe you don't have mmio, but at least this 
>> shouldn't interfere.
>>     
>
> OK, so the idea is to mmap /sys/bus/pci/.../region within the guest RAM area, 
> and just include it within the normal guest RAM memslot?
>   

In the case of x86, since the PCI IO region is pretty far away from 
normal RAM, we'll probably use a different memory slot, but yes, that's 
the general idea.

IIUC PPC correctly, all IO pages have corresponding struct pages.  This 
means that get_user_pages() would succeed and you can reference count 
them?  In this case, we would never take the VM_PFNMAP path.

Is that correct?

> How will the IOMMU be programmed? Wouldn't you still need to register a 
> special type of memslot for that?
>   

That's independent of this patchset.  For non-aware guests, we'll have 
to pin all of physical memory up front and then create an IOMMU table 
from the pinned physical memory.  For aware guests with a PV DMA window 
API, we'll be able to build that mapping on the fly (enforcing mlock 
allocation limits).

Regards,

Anthony Liguori


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone