Re: [PATCH v6 2/4] KVM: x86: report negative values from wrmsr emulation to userspace

[Qian Cai <cai@redhat.com>]

On Mon, 2020-10-26 at 15:40 -0400, Qian Cai wrote:
> On Wed, 2020-09-23 at 00:10 +0300, Maxim Levitsky wrote:
> > This will allow the KVM to report such errors (e.g -ENOMEM)
> > to the userspace.
> > 
> > Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
> 
> Reverting this and its dependency:
> 
> 72f211ecaa80 KVM: x86: allow kvm_x86_ops.set_efer to return an error value
> 
> on the top of linux-next (they have also unfortunately merged into the
> mainline
> at the same time) fixed an issue that a simple Intel KVM guest is unable to
> boot
> below.

So I debug this a bit more. This also breaks nested virt (VMX). We have here:

[  345.504403] kvm [1491]: vcpu0 unhandled rdmsr: 0x4e data 0x0
[  345.758560] kvm [1491]: vcpu0 unhandled rdmsr: 0x1c9 data 0x0
[  345.758594] kvm [1491]: vcpu0 unhandled rdmsr: 0x1a6 data 0x0
[  345.758619] kvm [1491]: vcpu0 unhandled rdmsr: 0x1a7 data 0x0
[  345.758644] kvm [1491]: vcpu0 unhandled rdmsr: 0x3f6 data 0x0
[  345.951601] kvm [1493]: vcpu1 unhandled rdmsr: 0x4e data 0x0
[  351.857036] kvm [1493]: vcpu1 unhandled wrmsr: 0xc90 data 0xfffff

After this commit, -ENOENT is returned to vcpu_enter_guest() causes the
userspace to abort.

kvm_msr_ignored_check()
  kvm_set_msr()
    kvm_emulate_wrmsr()
      vmx_handle_exit()
        vcpu_enter_guest()

Something like below will unbreak the userspace, but does anyone has a better
idea?

--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1748,7 +1748,7 @@ int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
                return 0;
 
        /* Signal all other negative errors to userspace */
-       if (r < 0)
+       if (r < 0 && r != -ENOENT)
                return r;
 
        /* MSR write failed? Inject a #GP */