On Wed, Jun 01, 2011 at 12:34:43AM +0200, Segher Boessenkool wrote: > There's a loop of 128 iterations of 3 insns. > > I'm not saying it is actually bad, just that that 50 is slightly off ;-) That would be the TLB invalidation. On POWER7 we only need to do that if the virtual cpu last ran on a different physical cpu, or if this physical cpu last ran a different virtual cpu in the same partition. So hopefully we don't have to do it very often. The reason we have to do it in that case is to allow the optimization where we use tlbiel for TLB invalidations if the guest claims that the translation being invalidated was only ever used on this virtual cpu. That means that we have to guard against stale TLB entries left behind when a virtual cpu moves from one physical cpu to another. Paul. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
