Re: [PATCH resend v5 6/6] KVM: nVMX: Enable nested posted interrupt processing

[Yong Wang <yong.y.wang@linux.intel.com>]

On Fri, Feb 27, 2015 at 04:56:06PM +0800, Wincy Van wrote:
> On Sun, Feb 15, 2015 at 2:27 PM, Yong Wang <yong.y.wang@linux.intel.com> wrote:
> >
> > Wincy, our QA found regressions with this patch that 64bit L2 linux guest
> > fails to boot up when running nested kvm on kvm.
> >
> > Environment:
> > ------------
> > Host OS (ia32/ia32e/IA64):ia32e
> > Guest OS (ia32/ia32e/IA64):ia32e
> > Guest OS Type (Linux/Windows):Linux
> > kvm.git Commit:6557bada461afeaa920a189fae2cff7c8fdce39f
> > qemu.kvm Commit:5c697ae74170d43928cb185f5ac1a9058adcae0b
> > Host Kernel Version:3.19.0-rc3
> > Hardware:Ivytown_EP, Haswell_EP
> >
> >
> > Bug detailed description:
> > --------------------------
> > create 64bit linux guest as L2 guest, the guest boot up fail
> >
> > note:
> > 1. create a 32bit linux guest as L2 guest, the guest boots up fine.
> > 2. create a 64bit windows guest as L2 guest, the guest boots up fine.
> > 3. this should be a kernel bug:
> > kvm       + qemu     = result
> > 6557bada  + 5c697ae7 = bad
> > 8fff5e37  + 5c697ae7 = good
> >
> > Reproduce steps:
> > ----------------
> > 1 create L1 guest:
> > qemu-system-x86_64 -enable-kvm -m 8G -smp 4 -net nic,macaddr=00:12:31:34:51:31 -net tap,script=/etc/kvm/qemu-ifup nested-kvm.qcow -cpu host
> >
> > 2. create L2 guest
> > qemu-system-x86_64 -enable-kvm -m 2G -smp 2 -net none rhel6u5.qcow
> >
> > Current result:
> > ----------------
> > create 64bit linux guest as L2 guest, the guest boots up fail
> >
> > Expected result:
> > ----------------
> > create 64bit linux guest as L2 guest, the guest boots up fine
> >
> > Please take a look.
> >
> 
> Yong, according to the logs, I found that L1 may have disabled x2apic,
> and the MSR_BITMAP field will be modified by following vmx_set_efer in
> prepare_vmcs02.
> So I think we can fix this issue by:
> 
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index f7b20b4..f6e3457 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2168,7 +2168,10 @@ static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
>  {
>         unsigned long *msr_bitmap;
> 
> -       if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
> +       if (is_guest_mode(vcpu))
> +               msr_bitmap = vmx_msr_bitmap_nested;
> +       else if (irqchip_in_kernel(vcpu->kvm) &&
> +               apic_x2apic_mode(vcpu->arch.apic)) {
>                 if (is_long_mode(vcpu))
>                         msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
>                 else
> 
> 

Our QA verified that your patch fixed the issue. Please prepare a formal patch
that Paolo can consider applying. Thanks a lot Wincy!