Re: [Xen-devel] [PATCH 3/3] x86_64: allow more vCPU-s per guest

[Keir Fraser <keir.fraser@xxxxxxxxxxxxx>]

On 18/06/2009 10:27, "Jan Beulich" <JBeulich@xxxxxxxxxx> wrote:

>> Applied. Is the vcpu[] array re-allocation in XEN_DOMCTL_max_vcpus an
>> example of over-optimistic-concurrency-control? It can't really be 100% safe
>> without extra locking on all users of that array (not good), or using rcu
>> (better), can it. The wmb() is a nice try. ;-)
> 
> Agreed. Originally I intended to add at least a comment, but after realizing
> that the path currently is only reached during domain creation (and I believe
> there are issues elsewhere if one would really allow increasing the # of
> vCPU-s in a domain on the fly), I decided to leave it as is (and the wmb()
> can be considered sort of a comment to that effect).

Then I think it would be better to make that domctl really singleshot (i.e.,
fail if d->max_vcpus is already non-zero), with a comment explaining why it
is implemented this way. Buggily implementing an unused case can't be good.
Can you generate a patch for this, please?

Is it safe do you think to go from d->max_vcpus=0 to d->max_vcpus!=0 on a
live domain? (i.e, the specific case we do use in your general observation
that increasing d->max_vcpus may be dangerous)?

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel