Re: [Xen-devel] [RFC PATCH 09/10] ARM: vGIC: introduce vgic_get/put_pending_irq

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Thu, 4 May 2017, Julien Grall wrote:
> > @@ -28,11 +27,20 @@ static inline int
> > local_events_need_delivery_nomask(void)
> >       * case.
> >       */
> >      if ( gic_events_need_delivery() )
> > -        return 1;
> > +    {
> > +        ret = 1;
> > +    }
> > +    else
> > +    {
> > +        struct pending_irq *p;
> > 
> > -    if ( vcpu_info(current, evtchn_upcall_pending) &&
> > -        list_empty(&p->inflight) )
> > -        return 1;
> > +        p = vgic_get_pending_irq(current->domain, current,
> > +                                 current->domain->arch.evtchn_irq);
> > +        if ( vcpu_info(current, evtchn_upcall_pending) &&
> > +            list_empty(&p->inflight) )
> > +            ret = 1;
> > +        vgic_put_pending_irq(current->domain, p);
> 
> Looking at this code, I think there are a race condition. Because nothing
> protect list_empty(&p->inflight) (this could be modified by another physical
> vCPU at the same time).
> 
> Although, I don't know if this is really an issue. Stefano do you have an
> opinion?

I only gave a cursory look at the series, but I think you are right.
This access to inflight needs to be protected.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel