Re: [Xen-ia64-devel] Linux bug with Xen

[Tristan Gingold <Tristan.Gingold@xxxxxxxx>]

Le Jeudi 23 Mars 2006 13:33, Tian, Kevin a écrit :
> From: Tristan Gingold
>
> >Sent: 2006年3月23日 20:31
> >Hi,
> >
> >I think I hit a linux kernel bug.
> >The situation is:
> >
> >* an interrupt happen
> >* interrupt IVT entry is executed with ic=0
> >* within the entry, SAVE_MIN_WITH_COVER try to access to the current
> >area,
> >  pointed by kr6
> >* Unfortunatly, the 'current' area is not mapped, thus xen has to handle
> >the
> >page fault.
> >* the translation fails inside Xen and Xen injects a nested dtlb miss
> > fault. * The linux nested dtlb miss don't know how to handle this fault
> > (only vmemmap
> >are handled).
> >
> >As far as I know/read the sources, the linux current area is not
> >TR-mapped.
> >So Xen appears to be correct and the linux kernel may be buggy here.
>
> No, the current stack is always mapped by TR (at a granule of 64M/16M),
> or else nothing can be forwarded since heavy weight exception handler
> always needs to save interrupt context to current stack when psr.ic off.
> You can check ia64_switch_to in entry.S, and so that's a Xen bug since
> xen should always hit translation in vTR area.
Ok, thank you.
I missed this information since CURRENT_STACK is mapped and not CURRENT.
But they are on the same page.

So I have to find why TR and KR doesn't match.

Thanks,
Tristan.



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