\uD83D\uDDD3 Date
\uD83D\uDC65 Participants
@Catalin Udma (NXP)
Srinivas Ramana
\uD83E\uDD45 Goals
Mathieu: Xen Sys Crate/Goldstein
AoB
\uD83D\uDDE3 Discussion topics
MP:
SS:
there is a difference between KVM views of VMM and Xen’s
IOREQ just does emulation
“tools” does the VM control
SS:
Virtio already works on HVM guests with QEMU x86
but becomes non-secure
also very slow
active work from Xilinx
TPM emulation
two machine types (HVM with IOREQ) and PV (without IOREQ)
Vikram @ Xilinx
MP: still not sure what the rust-vmm work is doing
SS:
just the basic work for hypercalls?
SS:
the toolstack is a whole bunch of issues
libs to make hypercalls
binary to start VMs
large code base, not liked and rewitten at least once
also a OCaml toolstack
Can both be replaced by a rust one
would be the 4th iteration
From a safety cert viewpoint
not even tried with the tool stack
likey rewrite from scartch
should not be a linux toolstack
MP
just concentrating on the foundation
AJB
outlined xen vhost-master
VG
done some work on merging
SS
we have the two approaches
EPAM work has been held up by the war
Was a spec question being held up on SCMI
Spec says DT defines device ids for SCMI
should be in generic device tree right?
difficult to make changes not used by LKML
push back from ARM maintainers - should be in spec?
VG
just pushing back because not used by kernel
optee create their own bindings as well
SS
we get a lot of specific DT nodes (OPTEE, Xen, Uboot are examples)
VG
maybe SCMI section is the solution
SS
✅ Action items
- Alex Bennée show the disaggregation patches to Stefano/Vikram
- Alex Bennée add cards for kernel eventfd/xen glue
- Vincent Guittot : thread to start with the ARM guys on SCMI DT bindings