Re: [Xen-devel] [PATCH 0/4] Add posted interrupt supporting

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Wed, 10 Apr 2013, Konrad Rzeszutek Wilk wrote:
> On Wed, Apr 10, 2013 at 01:32:41PM +0000, Zhang, Yang Z wrote:
> > Konrad Rzeszutek Wilk wrote on 2013-04-10:
> > > On Wed, Apr 10, 2013 at 02:51:41AM +0000, Zhang, Yang Z wrote:
> > >> Konrad Rzeszutek Wilk wrote on 2013-04-10:
> > >>> On Tue, Apr 09, 2013 at 02:01:27PM +0800, Yang Zhang wrote:
> > >>>> From: Yang Zhang <yang.z.zhang@xxxxxxxxx>
> > >>>> 
> > >>>> The follwoing patches are adding the Posted Interrupt supporting to 
> > >>>> Xen:
> > >>>> Posted Interrupt allows vAPIC interrupts to inject into guest directly
> > >>>> without any vmexit.
> > >>> 
> > >>> Is there a mechanism for the guest to figure out whether this is 
> > >>> exported?
> > >> Expose it to guest? No.
> > >> 
> > >>> 
> > >>> Right now the Linux PVHVM guests will use the event channel mechanism
> > >>> (via the 0xf3 callback vector). This means that if we want to do IPIs
> > >>> (RESCHEDULE_VECTOR or CALL_FUNCTION_*) we end up doing a vmexit - but
> > >>> with this we should be able to do these inside the guest right?
> > >> I am not familiar with PV staffs. Will event channel mechanism touch 
> > >> vAPIC? If
> > > no, perhaps the answer is not.
> > > 
> > > No it won't. But any any inter-OS vectors (such as IPIs) would benefit
> > > from vAPIC right? As in posting those vectors (assuming the destination
> > > VCPU is running) would not cause an VMEXIT?
> > Yes.
> > 
> > >> 
> > >>> 
> > >>> If so, is there a mechanism inside the guest to detect this and
> > >>> use the HVM mechanisms for IPI?
> > >> Why guest should aware of this? Current implementation already covers the
> > > vIPI.
> > > 
> > > Not sure I understand you. What current implementation?
> > I thought you mean the vIPI. Obviously, I am wrong. :(
> > What does " HVM mechanisms for IPI " mean? 
> 
> '__default_send_IPI_dest_field' in the Linux code, which is
> writting in the ICR2 field in the APIC.

I am also interested in this question and I cannot find a clear answer:
when a guest vcpu wants to issue an IPI to another vcpu, will it trap
into Xen? Or will everything work automagically?

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