RE: [Xen-devel] cpuidle causing Dom0 soft lockups

["Tian, Kevin" <kevin.tian@xxxxxxxxx>]

>From: Jan Beulich [mailto:JBeulich@xxxxxxxxxx] 
>Sent: 2010年2月3日 21:21
>
>>>> "Tian, Kevin" <kevin.tian@xxxxxxxxx> 03.02.10 13:10 >>>
>> Possibly the real solution is to not have dom0 with large 
>virtual vCPUs. 
>
>But you realize that this is a Dom0-only issue only as long as DomU-s
>with more vCPU-s cannot be created? I.e. it'll become an issue 
>affecting
>any kind of guest as soon as that limitation gets out of the way. So I
>don't view this as a solution - it's at best a workaround 
>until a solution
>can be found.

Sure, but then it's all about how spinlock itself should be implemented 
in virtualization environment, or how xtime_lock should be used in this
specific case, to allow scale up with large vcpu numbers.

There's always limitation about how it may scale w/o triggering softlock
warning. We agree that cpuidle made it worse inadvertently, and then
come up patch to recover it back to the level of original limitation. And
we just realize that current ticket pv spinlock may still encounter such
limitation once system becomes hot when lock holder of xtime_lock
may stay in runqueue for relatively long time, even w/o cpuidle.


Thanks,
Kevin

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