Re: [patch 21/32] NTB/msi: Convert to msi_on_each_desc()

[Jason Gunthorpe]

On Thu, Sep 22, 2022 at 10:42:46PM +0000, Tian, Kevin wrote:
> > From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > Sent: Thursday, September 22, 2022 8:14 PM
> > 
> > On Thu, Sep 22, 2022 at 05:11:00AM +0000, Tian, Kevin wrote:
> > 
> > > > Thinking of the interrupt routing as a host resource is the problem -
> > > > it is a device resource and just like PASID the ideal design would
> > > > have each RID have its own stable numberspace scoped within it. The
> > > > entire RID and all its stable numberspace would be copied over.
> > > >
> > >
> > > Unfortunately it is not the case at least on Intel VT-d. The interrupt
> > > remapping table is per VT-d instead of per RID.
> > 
> > Doesn't that just turn it into a SW problem how it manages it?
> > 
> 
> When IRTEs are shared cross RID's I'm not sure how to make the
> allocation stable between the src and the dest. They are allocated
> dynamically and hierarchically in request_irq().

So I suppose the issue is labeling, and we can't standardize where in
the shared space each RID will fall, and the HW can't virtualize this
for us.

Maybe the answer is that IMS just isn't supported on this HW if live
migration is also turned on.

Jason