Re: [Qemu-devel] [PATCH v12 rebased] kvm: notify host when the guest is panicked

[Gleb Natapov]

On Mon, Mar 04, 2013 at 05:43:48PM -0300, Marcelo Tosatti wrote:
> On Mon, Mar 04, 2013 at 07:49:13PM +0200, Gleb Natapov wrote:
> > On Sun, Mar 03, 2013 at 07:29:53PM -0300, Marcelo Tosatti wrote:
> > > On Sun, Mar 03, 2013 at 03:00:22PM +0200, Gleb Natapov wrote:
> > > > On Fri, Mar 01, 2013 at 09:03:12PM -0300, Marcelo Tosatti wrote:
> > > > > On Thu, Feb 28, 2013 at 04:54:25PM +0800, Hu Tao wrote:
> > > > > > > > diff --git a/arch/x86/include/uapi/asm/kvm_para.h 
> > > > > > > > b/arch/x86/include/uapi/asm/kvm_para.h
> > > > > > > > index 06fdbd9..c15ef33 100644
> > > > > > > > --- a/arch/x86/include/uapi/asm/kvm_para.h
> > > > > > > > +++ b/arch/x86/include/uapi/asm/kvm_para.h
> > > > > > > > @@ -96,5 +96,7 @@ struct kvm_vcpu_pv_apf_data {
> > > > > > > >  #define KVM_PV_EOI_ENABLED KVM_PV_EOI_MASK
> > > > > > > >  #define KVM_PV_EOI_DISABLED 0x0
> > > > > > > >  
> > > > > > > > +#define KVM_PV_EVENT_PORT      (0x505UL)
> > > > > > > > +
> > > > > > > 
> > > > > > > No need for the ioport to be hard coded. What are the options to
> > > > > > > communicate an address to the guest? An MSR, via ACPI?
> > > > > > 
> > > > > > I'm not quite understanding here. By 'address', you mean an ioport?
> > > > > > how to communicate an address? (I have little knowledge about ACPI)
> > > > > 
> > > > > Yes, the ioport. The address of the ioport should not be fixed (for
> > > > > example future emulated board could use that fixed ioport address,
> > > > > 0x505UL).
> > > > > 
> > > > > One option is to pass the address via an MSR. Yes, that is probably 
> > > > > the
> > > > > best option because there is no dependency on ACPI.
> > > > > 
> > > > Why dependency on ACPI is problematic? ACPI is the standard way on x86
> > > > to enumerate platform devices. Passing it through MSR makes this panic
> > > > device CPU interface which it is not. And since relying on #GP to detect
> > > > valid MSRs is not good interface we will have to guard it by cpuid bit.
> > > > 
> > > > --
> > > >                         Gleb.
> > > 
> > > KVM guest <-> KVM host interface is not dependent on ACPI, so far. Say,
> > > its possible to use a Linux guest without ACPI and have KVM paravirt 
> > > fully functional.
> > This is not KVM guest <-> KVM host interface though. This is yet another
> > device. We could implement real impi device that have crash reporting
> > capability, but decided to go with something simpler. Without ACPI guest
> > will not be able to power down itself too, but this is not the reason
> > for us to introduce non-ACPI interface for power down.
> 
> Sure (its more of an aesthetic/organizational point, i guess).
> 
> Anyway, one problem with ACPI is whether its initialized early enough
> (which is the whole point of PIO the x86 specific interface).
ACPI is needed pretty early in the boot process.

--
                        Gleb.