Re: [PATCH v2 3/7] KVM: Add paravirt kvm_flush_tlb_others

[Nikunj A Dadhania <nikunj@linux.vnet.ibm.com>]

On Wed, 4 Jul 2012 23:09:10 -0300, Marcelo Tosatti <mtosatti@redhat.com> wrote:
> On Tue, Jul 03, 2012 at 01:49:49PM +0530, Nikunj A Dadhania wrote:
> > On Tue, 3 Jul 2012 04:55:35 -0300, Marcelo Tosatti <mtosatti@redhat.com> wrote:
> > > > 
> > > >    if (!zero_mask)
> > > > 	   goto again;
> > > 
> > > Can you please measure increased vmentry/vmexit overhead? x86/vmexit.c 
> > > of git://git.kernel.org/pub/scm/virt/kvm/kvm-unit-tests.git should 
> > > help.
> > >
> > Sure will get back with the result.
> > 
> > > > +	/* 
> > > > +	 * Guest might have seen us offline and would have set
> > > > +	 * flush_on_enter. 
> > > > +	 */
> > > > +	kvm_read_guest_cached(vcpu->kvm, ghc, vs, 2*sizeof(__u32));
> > > > +	if (vs->flush_on_enter) 
> > > > +		kvm_x86_ops->tlb_flush(vcpu);
> > > 
> > > 
> > > So flush_tlb_page which was an invlpg now flushes the entire TLB. Did
> > > you take that into account?
> > > 
> > When the vcpu is sleeping/pre-empted out, multiple request for flush_tlb
> > could have happened. And now when we are here, it is cleaning up all the
> > TLB.
> 
> Yes, cases where there are sufficient exits transforming one TLB entry
> invalidation into full TLB invalidation should go unnoticed.
> 
> > One other approach would be to queue the addresses, that brings us with
> > the question: how many request to queue? This would require us adding
> > more syncronization between guest and host for updating the area where
> > these addresses is shared.
> 
> Sounds unnecessarily complicated.
> 
Yes, I did give this a try earlier, but did not see much improvement
with the amount of complexity that it was bringing in.

Regards
Nikunj