I also wonder if we’re using all of the performance capabilities of Xen/QEMU under R4.1 for Windows HVMs.
Does Xen 4.14 support any of the viridian/hyperv options to allow the Windows VM and/or QEMU to elect to use lower-impact non-emulated hardware access?
Looking at the qemu command line given to the stubdomain, it looks like we are neither using viridian nor hyperv enlightenments even though my read of xen.cfg is that viridian might be available in the current config, but I don’t see it constructed onto the command line.
Hmm, under R4.1 in /var/log/libvirt/libxl/win_vm_name.log, I see “viridian”: “false” for the windows HVMs. This may be non-optimal.
I also see references to PIT, APIC and HEPT timers, but not TSC in that file, and viridian would allow non-emulated tsc…I think.
Are either the viridian capabilities or hyperv enlightenment options usable from Xen 4.14 (directly or via QEMU) in such a way as to give the Windows HVMs a performance advantage?
If so, can this be done without losing the PV drivers? Concern is that at least the second set of capabilities, hyperv options, may conflict with the MSR masking that Xen PV drivers expect?