On 12/01/15 08:57, Jan Beulich wrote:
> --- a/xen/include/xen/event.h
> +++ b/xen/include/xen/event.h
> @@ -152,10 +152,11 @@ static inline void evtchn_port_init(stru
>          d->evtchn_port_ops->init(d, evtchn);
>  }
>  
> -static inline void evtchn_port_set_pending(struct vcpu *v,
> +static inline void evtchn_port_set_pending(struct domain *d,
> +                                           unsigned int vcpu_id,
>                                             struct evtchn *evtchn)

I would rename this to the, now vacant, evtchn_set_pending().  It takes
an evtchn* not a port.  (Its sole caller was evtchn_set_pending(), so
the patch won't grow)

Furthermore, all callers except send_guest_vcpu_virq() currently use
evtchn->notify_vcpu_id to get a struct vcpu* to pass.  I think you can
drop the vcpu_id parameter and use evtchn->notify_vcpu_id directly,
which reduces the likelyhood of a bug where the evtchn is bound to one
vcpu but a caller gets the wrong id and raises the event channel on the
wrong vcpu.

~Andrew

>  {
> -    v->domain->evtchn_port_ops->set_pending(v, evtchn);
> +    d->evtchn_port_ops->set_pending(d->vcpu[vcpu_id], evtchn);
>  }
>  
>  static inline void evtchn_port_clear_pending(struct domain *d,
>
>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel