[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH v2 01/11] x86/domctl: Add XEN_DOMCTL_set_avail_vcpus



On 11/15/2016 09:33 AM, Jan Beulich wrote:
>>>> On 15.11.16 at 15:28, <boris.ostrovsky@xxxxxxxxxx> wrote:
>> On 11/15/2016 03:34 AM, Jan Beulich wrote:
>>>>>> On 09.11.16 at 15:39, <boris.ostrovsky@xxxxxxxxxx> wrote:
>>>> This domctl is called when a VCPU is hot-(un)plugged to a guest (via
>>>> 'xl vcpu-set'). While this currently is only intended to be needed by
>>>> PVH guests we will call this domctl for all (x86) guests for consistency.
>>> The discussion on the actual change seems to have pointed out all
>>> needs of change, but what I wasn't able to understand yet is why
>>> this is needed in the first place. From hypervisor pov, so far it's been
>>> up to the guest which CPUs get onlined/offlined, and the interface
>>> to request offlining (not an issue for onlining) was - afaict - a purely
>>> voluntary one. Why does this change with PVH? Any such ratonale
>>> should be put in the commit message.
>> If the question is why we need to have hypervisor interface for PVH
>> guests then it's because we need someone to send an SCI and set GPE
>> registers and there is noone but the hypervisor to do that for PVH (I
>> will add it to the commit message).
> Yes, that was the primary question. And it took me until quite late
> in the series until I've seen the purpose, so I appreciate you
> extending the description, even if just slightly.
>
>> As for whether we want to enforce available VCPU count --- I think we
>> decided that we can't do this for PV and so the question is whether it's
>> worth doing only for some types of guests. And as you pointed out the
>> second question (or may be the first) is whether enforcing it is the
>> right thing in the first place.
>>
>> (BTW, I am thinking to move the domctl from x86-specific to common code
>> since if we are no longer saying that it's PVH-only then ARM should have
>> it available too)
> Well, without us wanting to enforce anything, would we still need the
> domctl for anything?

While we can send an SCI directly from the toolstack we need to set the
GPE registers for the guest and that's what domctl's primary use is (was).

-boris


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

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.