[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH 0/1] bhyve: Make LPC slot number configurable

Yes, that makes sense. I'll try to find some time next week to redo my code and send another patch. Since my time for working on libvirt is very limited can you confirm that the LPC configuration should look like this:

   <controller type="isa-bridge" index="0">
       <address type="pci" domain="0" bus="0" slot="NNN" function="0"/>

Also can you send me an example of how you imagine the <features> setting for unimplemented MSRs?


On 08/10/2018 06:24 PM, Daniel P. Berrangé wrote:
On Fri, Aug 10, 2018 at 06:11:57PM +0300, Ivan Mishonov wrote:
I'd like to hear Roman's opinion on this too since he wrote the initial
implementation. As for the command line arguments I was looking at
<qemu:commandline> since it's doing exactly the same thing and I thought it
would be nice to be consistent with it
It would still be reasonable to allow <bhvyve:commandline> for arbitrary
passthrough of new features which have no XML defined for them. I just
think it is reasonable to model these two example explicitly.

The namespaced passthrough is intended for short term hacks primarily.

On 08/10/2018 05:57 PM, Daniel P. Berrangé wrote:
On Fri, Aug 10, 2018 at 05:47:40PM +0300, Ivan Mishonov wrote:
Yes, this is totally doable. I just don't know if it's a good idea to add a
new device type specifically for bhyve LPC and nothing else. Even if we do
it like this I'll still have to send another patch including the bhyve XML
namespace as we need to be able to pass extra command line options to the
bhyve process related to unimplemented MSRs on AMD Zen systems. I thought
I'd do the 2 things in a similar manner as both of them are strictly bhyve
IMHO  the LPC thing is definitely in scope for correct modelling in
the XML.

For the MSRs option, it is probable we'd consider that in scope as
well. Currently KVM has a global "ignore unknown msrs" option in the
kernel module, but I think it is conceptually reasonable to expect
that to be settable on a per-VM basis.

Probably would do the MSRs thing as a <features> flag, as we stuff
lots of random feature toggles under there


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]