Re: [Xen-devel] [PATCH v4 14/14] xen/x86: setup PVHv2 Dom0 ACPI tables

[Roger Pau Monne <roger.pau@xxxxxxxxxx>]

On Fri, Dec 23, 2016 at 12:13:47PM -0500, Boris Ostrovsky wrote:
> On 12/23/2016 11:02 AM, Konrad Rzeszutek Wilk wrote:
> > On Fri, Dec 23, 2016 at 10:35:10AM -0500, Boris Ostrovsky wrote:
> >> On 12/23/2016 10:31 AM, Konrad Rzeszutek Wilk wrote:
> >>>> But this still assumes that dom0 handles ACPI event for a pCPUs as well,
> >>>> right? And I am not sure this can work.
> >>>>
> >>>> Actually, how do we hotplug pCPUs now?
> >>> xen-hptool
> >> Yes, but this has nothing to do with an actual pCPU being hot-plugged
> >> into the system. It's similar to doing 'echo 1 > /sys/.../cpuX/online in
> >> Lnux.
> >>
> >> My question is --- how do we (hypervisor/dom0/toolstack) become aware of
> >> appearance of a new processor.
> > Ooooh. drivers/xen/xen-acpi-cpuhotplug.c
> 
> Looking at this code I think we should be able to at least differentiate
> between pCPU and vCPU and not add pCPU to dom0.

Right, but I don't really see any way to do this with the current ACPI
interface, and still be able to use ACPI CPU hotplug for both pCPUs and
vCPUs.

We can provide a crafted MADT that reflects the number of vCPUs available to
Dom0 at boot time, let Dom0 find the Processor objects in the ACPI namespace
and perform pCPU hotplug as it's been done until now. Then for Dom0 vCPU
hotplug we would need to use the hypercall interface as it's used by classic PV
guests, because AFAICT there's no way we can provide Dom0 with a PRST/PRSC
that's valid for both vCPUs and pCPUs.

Roger.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel