[Top][All Lists]

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

Re: [Qemu-devel] [0/3] pseries: RFC: Advertise host CPU capabilties to g

From: David Gibson
Subject: Re: [Qemu-devel] [0/3] pseries: RFC: Advertise host CPU capabilties to guest
Date: Fri, 30 Sep 2011 19:00:13 +1000
User-agent: Mutt/1.5.21 (2010-09-15)

On Fri, Sep 30, 2011 at 10:20:14AM +0200, Alexander Graf wrote:
> Am 30.09.2011 um 09:50 schrieb David Gibson <address@hidden>:
> > This series contains some patches which, when using KVM, gather
> > information about the capabilities of the host CPU and advertise them
> > to the guest system when using the pseries machine.  Specifically it
> > does this for whether the CPU supports VMX, VSX and/or DFP
> > instructions, and for the CPUs supported page sizes.
> > 
> > The VSX and DFP portions of this were posted earlier, and I've fixed
> > the minor comments which people made.  This leaves one objection from
> > Alex Graf, that whether the features are advertised should also depend
> > on the target CPU selected in qemu.  A similar objection may apply to
> > the pagesizes patch.  I guess the idea is to "clamp" the advertised
> > capabilities to those permitted by the selected target CPU, but I'm
> > not entirely sure what the logic here should be.
> > 
> > Frankly, particularly in the case of KVM Book3S-HV, I'm not terribly
> > convinced that attempting to make the guest CPU appear different from
> > the host CPU is terribly meaningful.  These patches as they stand have
> > the advantage that future, roughly compatible CPUs should Just Work
> > with these capabilities advertised in the correct cases.  Alex, can
> > you advise what sort of logic you'd like here.
> Yes, very simple. I want you to create a CPU type 'host', similar to
> how x86 does it. That should be the default CPU type for KVM with
> the pseries machine.

Ah, ok.  I didn't realize x86 did that, but I'd been thinking
something like that would make more sense.  I'm away for the next
week, but I'll look at this when I get the chance.

> You can also add a check in-kernel that verifies if guest PVR ==
> host PVR for HV mode. That way you ensure that -cpu host is always
> used there. If you later add compat modes, you can check them there,
> but still have -cpu xxx available to tell all pieces of the kvm/qemu
> chain what to use.


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_!

reply via email to

[Prev in Thread] Current Thread [Next in Thread]