Re: [for-xen-4.5 PATCH v2 2/2] dpci: Add ZOMBIE state to allow the softirq to finish with the dpci_pirq.

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Fri, Nov 21, 2014 at 01:51:24PM +0100, Sander Eikelenboom wrote:
> 
> Friday, November 21, 2014, 12:50:16 PM, you wrote:
> 
> > On November 21, 2014 2:51:33 AM EST, Jan Beulich <JBeulich@suse.com> wrote:
> >>>>> On 20.11.14 at 20:51, <konrad.wilk@oracle.com> wrote:
> >>> @@ -669,7 +670,7 @@ static void hvm_dirq_assist(struct domain *d,
> >>struct hvm_pirq_dpci *pirq_dpci)
> >>>      ASSERT(d->arch.hvm_domain.irq.dpci);
> >>>  
> >>>      spin_lock(&d->event_lock);
> >>> -    if ( pirq_dpci->state )
> >>> +    if ( test_and_clear_bool(pirq_dpci->masked) )
> >>>      {
> >>>          struct pirq *pirq = dpci_pirq(pirq_dpci);
> >>>          const struct dev_intx_gsi_link *digl;
> >>
> >>So this now guards solely against the timeout enforced EOI? Why do
> >>you no longer need to guard against cancellation (i.e. why isn't this
> >>looking at both ->state and ->masked)?
> >>
> 
> > The previous state check was superfluous as the dpci_softirq would check for the valid STATE_ before calling hvm_dirq_assist (and deal with cancellation).
> 
> > I actually had an cleanup patch that would have removed the 'if (pirq_dpci->state) ' and move the code for Xen 4.6.
> 
> > Anyhow waiting to see if Sander was able to test with this patch.
> 
> >>Jan
> 
> Hi Konrad / Jan,
> 
> I have tested it for 3 hours now, no host crash so far (even after applying some 
> extra stress to the guest).

Yeey! Thank you for being so flexible and willing to test these patches out!
> 
> --
> Sander
> 
>