Re: [Xen-devel] [xen-4.1-testing test] 9805: regressions - FAIL

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Thu, 17 Nov 2011, Keir Fraser wrote:
> On 17/11/2011 10:28, "Stefan Bader" <stefan.bader@xxxxxxxxxxxxx> wrote:
> 
> >>> Hm, yes we should. I am pretty sure I hit that code path often enough,
> >>> Wonder
> >>> why I never saw any dead lock there...
> >> 
> >> Perhaps your dom0 kernel doesn't register a pirq_eoi_map.
> >> 
> > Would be the only explanation. And quite possible. Heck, I would need to 
> > know
> > what that is used for anyway. :/ The kernel is 3.0 based the interrupt I was
> > looking at just was a normal apic emulated through events one...
> 
> Our automated tests still use 2.6.32. It wouldn't surprise me if upstream
> Linux 3 doesn't have the pirq_eoi_map stuff; it's an optimisation rather
> than core absolutely-required functionality.
> 
> It's the dom0 PV kernel that matters in this case by the way, not the kernel
> that you are running in HVM mode as a domU.

In any case considering that the PHYSDEVOP_eoi case wasn't protected by
a lock before this patch, we could just add the spinlock around the new
code only, to protect accesses to the pirq_emuirq array.

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