On Tue, Aug 04, 2015 at 09:47:59AM +0200, Andrew Jones wrote: > On Tue, Aug 04, 2015 at 02:09:52PM +1000, David Gibson wrote: > > On Mon, Aug 03, 2015 at 07:08:17PM +0200, Paolo Bonzini wrote: > > > > > > > > > On 03/08/2015 16:41, Andrew Jones wrote: > > > > Add enough RTAS support to support power-off, and apply it to > > > > exit(). > > > > > > > > Signed-off-by: Andrew Jones <drjones@xxxxxxxxxx> > > > > > > Why not use virtio-mmio + testdev on ppc as well? Similar to how we're > > > not using PSCI on ARM or ACPI on x86. > > > > Strange as it seems, MMIO is actually a PITA for a simple pseries > > guest like this. Basically, you have to enable the MMU - which > > requires a whole bunch of setup - in order to perform cache-inhibited > > loads and stores, which is what you need for IO. > > > > There are hypercalls to sidestep this (H_LOGICAL_CI_LOAD and > > H_LOGICAL_CI_STORE), but having a hypercall and KVM exit for every IO > > access may be hideously slow. > > > > In early development we did have a hypercall mediated virtio model, > > but it was abandoned once we got PCI working. > > So I think by yours and Alex's responses, if we want testdev support > then we should target using pci to expose it. Um.. maybe. I'm not really familiar with these testdevs, so I can't answer directly. > I'm ok with that, but > prefer not to be distracted with it while getting ppc kickstarted. > So, question for Paolo, are you OK with the exitcode snooper "cheat"? If you wanted to add a special hypercall channel for use by the tests I'd be ok with that too. -- 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
Attachment:
pgpzoxG3kTxV1.pgp
Description: PGP signature
