On Wed, 2017-12-06 at 20:20 +1100, David Gibson wrote: > On Tue, Dec 05, 2017 at 08:50:26AM -0600, Benjamin Herrenschmidt wrote: > > On Tue, 2017-12-05 at 18:00 +1100, David Gibson wrote: > > > > The CPU revision. But we won't introduce XIVE exploitation mode on > > > > anything else than DD2.0 which has full XIVE support. Even STORE_EOI > > > > that we should be adding. > > > > > > Hrm. Host CPU? That's a problem - if guest visible properties like > > > this vary with the host CPU, migration breaks. > > > > I don't think this is going to be a problem in practice. The > > availability of trigger comes from OPAL but in practice, all virtual > > interrupts are going to support it always, > > Ok. It still makes me nervous to derive guest visible features from > the host. I'd prefer to just hardwire the XIVE model to always/never > advertise it and simply fail if that isn't workable for the host kernel.
We could fail loudly if we see an migratable interrupt that doesn't have the flag. Cheers, Ben.
