Description
The latest POWER10 systems run unprivileged paravirtual linux guests
atop of a firmware hypervisor called PHYP. To enable workloads relying
on virtualization primitives it is necessary to allow those guests
to create "nested" guests themselves with minimal overhead.
We will discuss a new PHYP API that makes this possible and
the KVM PPC changes to make use of it and some of the implications for
things like IO.
Let's also discuss how this relates to a cross architecture trend in
virtualization, where KVM provides the scheduling, memory, etc and then
forwards calls to some other firmware that is responsible for actually
creating and running guests and what that means for the role of KVM.
Primary author
Jordan Niethe
(IBM)
