Re: [PATCH 0/12] Factor VT-d KVM functions into a generic API (with multiple device assignment support)

[Muli Ben-Yehuda <muli@il.ibm.com>]

On Wed, Dec 03, 2008 at 10:03:04AM +0100, Joerg Roedel wrote:

> > > > Have you tried porting any of the current iommu controllers to
> > > > this new framework to see if it works properly for them?
> > > 
> > > It works currently for VT-d. I also port it to AMD IOMMU
> > > currently. With some extensions (offset for start address, flags
> > > and size limitation) it is also suitable for IOMMUs like GART or
> > > similar ones.
> > 
> > What about the Calgary chipset?
> 
> Calgary is quite similar to GART (there is something like the
> aperture and a linear single-level pagetable).

Actually, Calgary has multiple per-bus address spaces (each of which
is a single-level linear pagetable limited to 4GB of addressable
memory), so I think it should work with your current approach pretty
much as is, once we take into account these two (per-bus and 32-bit
addressability) limitations.

Cheers,
Muli
-- 
The First Workshop on I/O Virtualization (WIOV '08)
Dec 2008, San Diego, CA, http://www.usenix.org/wiov08/
                       <->
SYSTOR 2009---The Israeli Experimental Systems Conference
http://www.haifa.il.ibm.com/conferences/systor2009/