Re: [Xen-ia64-devel] PATCH: cleanup of tlbflush

[Tristan Gingold <Tristan.Gingold@xxxxxxxx>]

Le Mercredi 10 Mai 2006 11:14, Tian, Kevin a écrit :
> From: Tristan Gingold [mailto:Tristan.Gingold@xxxxxxxx]
>
> >Sent: 2006年5月10日 16:53
> >
> >> >> At least one simple enhancement
> >> >> we can do is to change syntax of domain_dirty_cpumask. We can
> >> >> change it to indicate processors that domain is ever running on.
> >
> >Then
> >
> >> >> update point only happens at creation/destroy/migration, or even
> >> >> pause/unpause. Though this simple strategy is not fine-grained, we
> >> >> can still achieve benefit especially when domain is bound.
> >> >
> >> >One use of domain_dirty_cpumask is to flush vtlb when a page is
> >> >ungranted.
> >> >If this mask is ever set, doing IOs trash the machine.
> >> >IMHO, this is the next major Xen/ia64 challenge: dealing correctly
> >
> >with
> >
> >> >granted page.
> >>
> >> Not clear about this one. Could you elaborate more? How dose IO
> >
> >cause
> >
> >> to trash the machine?
> >
> >[By trashing I mean slowing down].
> >The current common code calls flush_tlb_mask after ungranting a page.
> >On xen/ia64, flush_tlb_mask should flush the tlb and the vhpt, which
> >means
> >clearing 16MB per vcpu.  This is quiet high.
> >Unfortunatly, flushing is required for correctness.  The current Isaku
> >work-around is to reduce VHPT size (64Kb).  But even with a small
> >size,
> >flushing tlb requires an IPI when SMP-g, which is quiet slow.
> >
> >Tristan.
>
> I see your concern about flush efficiency. However we still need set
> necessary mask bits for correctness, right? 
Not yet, because pages are not transfered.

> It would be difficult to track
> exact processors which have footprint about different ungranted pages.
> To track that list may instead pull down performance at other places.
> Then to set domain_dirty_cpumask as ones that domain is currently
> running on, can be a simple/safe way in current stage though
> performance may be affected.
Unfortunatly performance are so badly affected that using SMP-g is useless!

Tristan.

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