Re: [Xen-devel] [PATCH 1/5] x86/dpci: allow hvm_irq_dpci to handle a variable number of GSIs

[Roger Pau Monne <roger.pau@xxxxxxxxxx>]

On Tue, Apr 18, 2017 at 06:13:54AM -0600, Jan Beulich wrote:
> >>> On 27.03.17 at 12:44, <roger.pau@xxxxxxxxxx> wrote:
> > --- a/xen/include/xen/hvm/irq.h
> > +++ b/xen/include/xen/hvm/irq.h
> > @@ -81,14 +81,16 @@ struct hvm_girq_dpci_mapping {
> >  
> >  /* Protected by domain's event_lock */
> >  struct hvm_irq_dpci {
> > -    /* Guest IRQ to guest device/intx mapping. */
> > -    struct list_head girq[NR_HVM_IRQS];
> >      /* Record of mapped ISA IRQs */
> >      DECLARE_BITMAP(isairq_map, NR_ISAIRQS);
> >      /* Record of mapped Links */
> >      uint8_t link_cnt[NR_LINK];
> > +    /* Guest IRQ to guest device/intx mapping. */
> > +    struct list_head girq[];
> >  };
> 
> Considering what you say in the overview mail I don't think the
> comment can be moved without adjusting it, as it doesn't seem
> to reflect Dom0 in any way. Which then puts under question
> whether struct hvm_girq_dpci_mapping is the right data format
> for Dom0 in the first place: With bus, device, and intx taken
> out, all that's left if machine_gsi, and iirc you identity map GSIs.

Yes, I also got the same feeling. I've done it this way so that I don't have to
touch a lot of code, it's mostly using the same paths as the ones a HVM guest
would use for pass-through.

OTHO, there's a lot of unneeded faff here. I could certainly do with simpler
structures for PVH Dom0 GSI passthrough, but then I would have to do more
changes to pt_irq_create_bind and probably the hvm/irq.c functions. I can look
into that.

> Even if the array needed to remain, the sparseness of the GSI
> space opens up the question whether using a simple array here
> is the right choice.
> 
> The patch needs re-basing anyway afaict.

Sure, most of those where sent a while back.

Thanks, Roger.

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