RE: [PATCH] allow vcpu to move between pcpus (was Re: [Xen-ia64-devel]credit sched on ia64)

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

>From: Isaku Yamahata
>Sent: 2006年7月3日 12:22
>
>Maybe a consideration on VHPT per pcpu is needed.
>- allocation of region id per domain.
>  Should rid region be allocated per vcpu?

If guest allocates RID globally like Linux, per vcpu rid region is difficult 
to work. If the higher 6 bit is used as vcpu_id instead of domain_id, 
the number of allowed domains reduce. If you use more highest bits, 
the minimal 18bits can't be satisfied and then you need non-linear 
mapping between mrid->vrid. Also the purge like Anthony mentioned 
will be a big issue.

>
>- other?

Just found a bad performance effect under some condition. For 
example, say one domain with 2 vcpus both running on same LP (no 
other domains). You can easily find that vhpt/tlb flush is issued at each 
context switch between these 2 vcpus even without migration.
      Last_vcpu_id     current vcpu id       result
           Invalid            0               N/A
              0              1               whole tlb/vhpt flush
              1              0               whole tlb/vhpt flush
              ...

Thanks,
Kevin

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel