On Mon, Jul 24, 2017 at 01:47:28PM +0200, Cédric Le Goater wrote: > On 07/24/2017 01:07 PM, Benjamin Herrenschmidt wrote: > > On Mon, 2017-07-24 at 19:50 +1000, David Gibson wrote: > >> On Mon, Jul 24, 2017 at 05:00:57PM +1000, Benjamin Herrenschmidt wrote: > >>> On Mon, 2017-07-24 at 14:36 +1000, David Gibson wrote: > >>>> On Wed, Jul 05, 2017 at 07:13:21PM +0200, Cédric Le Goater wrote: > >>>>> These flags define some characteristics of the source : > >>>>> > >>>>> - XIVE_SRC_H_INT_ESB the Event State Buffer are controlled with a > >>>>> specific hcall H_INT_ESB > >>>> > >>>> What's the other option? > >>> > >>> Direct MMIO access. Normally all interrupts use normal MMIOs, > >>> each interrupts has an associated MMIO page with special MMIOs > >>> to control the source state (PQ bits). This is something I added > >>> to the PAPR spec (and the OPAL <-> Linux interface) to allow firmware > >>> to work around broken HW (which happens on some P9 versions). > >> > >> Ok.. and that's something that can be decided at runtime? > > > > Well, at this point I think nothing will set that flag.... It's there > > for workaround around HW bugs on some chips. At least in full emu it > > shouldn't happen unless we try to emulate those bugs. Hopefully direct > > MMIO will just work. > > Nevertheless I have added support for the hcall in Linux and QEMU. > To use, I think we could create a specific source.
So, IIUC, it's host constraints that would make this one way or the
other. So what happens when a guest migrates from a host which has it
one way to one which has it the other way?
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
signature.asc
Description: PGP signature
