RE: [Xen-ia64-devel] Time for hybrid virtualization?
Quoting "Xu, Anthony" <anthony.xu@xxxxxxxxx>:
> tgingold@xxxxxxx wrote:
> > Hi,
> > here are my random thoughts:
> > * First, I am all for supporting dom0 VTi (as well as dom0 PV). This
> > should be an intermediate step. Intel did that in the past,
> > although I am not sure it was fully virtualized (what about GFW ?)
> Dom0 Vti didn't use GFW, mostly it used native FW, while Xen provide
> Fake information, such as memory map block, and Xen will inject some
> Sal/Pal call.
I am worried with Sal/Pal. How do you prevent dom0 from making host PAL calls?
> > * Building a dom0 vti means porting backend, balloon and all other
> > drivers to Vti.
> Dom0 Vti just mean that cpu virtulaization is using VT-i( privileged
> intruction emulation).
> All other should be same.
> We may need to modify backend, ballon and other drivers a little.
If we really want to reduce kernel maintenance, it would be great to be
able to run an unmodified kernel. But I am maybe going too far!
> > * What about performance on Montecito ? If Montecito Vti is ~55% of
> > Montvale, I'd prefer to keep PV.
> KB on Montecito Vti can get 85% of native,
> Roughly calculate, Montecito Vti is ~95% of Montvale.
> Tukwila Vti should be faster, it adds more VT hardware support.
Xen-ia64-devel mailing list