[libvirt] [PATCH v5 2/2] conf: List only online cpus for virsh emulatorpin

[Viktor Mihajlovski]

On 06.12.2016 13:10, Peter Krempa wrote:
> On Tue, Dec 06, 2016 at 13:01:48 +0100, Viktor Mihajlovski wrote:
>> On 06.12.2016 12:33, Peter Krempa wrote:
>>> On Fri, Nov 25, 2016 at 14:53:07 +0530, Nitesh Konkar wrote:
> 
> [...]
> 
>>> Additionally we historically treat the "all bits set" cpu map
>>> as that there is no pinning configured. This would change this 
>>> behavior. When the VM is started, the real cpu map should be 
>>> returned, but for the config time I think the current state is 
>>> correct and doing your change would be incorrect.
>> I have no strong opinion on whether the CPU mask should match the
>> host online mask for the defined case, but the "all bits set" map
>> will change with the old code as well, if new CPUs are added via
>> hotplug.
> 
> CPU hotplug on the host is very uncommon and thus not tested very
> well.
> 
I can assure you that is very common on Linux running on z Systems and
also widely used by customers. With the patchset in
https://www.redhat.com/archives/libvir-list/2016-November/msg01320.html
it can also be reliably uses with libvirt.
>> The crux is that we don't have any way to find out there's in
>> fact no pinning in place for a process nor a way to "unpin" it.
> 
> That is true indeed. It's not possible via the API. It's fully
> possible via XML though.
> 
> For unpinning I have an unfinished series adding a flag to remove 
> pinning since since any other method is not reliable enough.
> 
> Finally I think that the "all ones" bitmap should cover the full
> map the user sends to libvirt. This would differentiate it from the
> "pinned to current full set of cpus" case.
> 


-- 

Mit freundlichen Grüßen/Kind Regards
   Viktor Mihajlovski

IBM Deutschland Research & Development GmbH
Vorsitzender des Aufsichtsrats: Martina Köderitz
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294