I just committed this patch from Tristan and it is now
possible to re-enable and run with CONFIG_XEN_SMP (see
xen/include/asm-ia64/config.h). Thanks Tristan!
I see lots of "pending guest timer before its due" messages,
which I think are just indicating that without Kevin's
timer fix, dom0 would freeze. Kevin, is it OK to turn off this
message now? I am also seeing some Linux "Oops: timer tick
before it's due" messages.
At some point we may want to enable CONFIG_XEN_SMP by
default. (It has been enabled by default on Xen/x86 for
over a year.)
> -----Original Message-----
> From: Tristan Gingold [mailto:[hidden email]]
> Sent: Friday, January 13, 2006 6:34 AM
> To: Magenheimer, Dan (HP Labs Fort Collins)
> Subject: Re: FW: [Xen-ia64-devel] Good projects for someone
> Le Jeudi 12 Janvier 2006 23:05, Magenheimer, Dan (HP Labs
> Fort Collins) a
> écrit :
> > > Le Vendredi 06 Janvier 2006 22:13, vous avez écrit :
> > > > > 2) Turning CONFIG_SMP on doesn't currently work because
> > > > > the (new) routine flush_tlb_mask is unimplemented and panics.
> > > > > Once SMP is working again, perhaps we should turn it on
> > > > > by default?
> > > >
> > > > Have you looked at this? If it is easy to fix, it would
> > > > be nice to get SMP working again...
> > >
> > > Not yet.
> > >
> > > It is currently difficult to me to work on SMP because I
> > > don't have an ITP
> > > (ie, the machine with an ITP is busy).
> > > I'd plan to work on SMP as soon as I have the ITP. We are
> > > working on this ITP
> > > issue.
> > >
> > > Maybe the new routine can be written and checked easily ? If
> > > so I can work on
> > > it now.
> > I'm not sure. It looks like the routine is doing an IPI to flush
> > the TLB on one or more processors. Some of the places where
> > it is called may not apply on ia64. It may not even be necessary
> > at all (especially before we have SMP guests). Or it likely
> > will be rarely necessary, but when it IS necessary will result
> > in a crash if it's not done correctly.
> > How hard is it do an IPI to flush the TLB on another processor?
> > If it's easy, we should probably just add the code to do it.
> > If it's hard, we should look more carefully at the code to see
> > exactly when it is necessary.
> Humm, it is rather easy. I should have looked on this issue before.
> Here is the patch.
> Note: the signed-off-by line is included in the patch. I
> hope this is the
> right place.
> See you next week.