[PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[David Matlack]

kvm_skip_emulated_instruction() should not be called after emulating
a VM-entry failure during or after loading guest state
(nested_vmx_entry_failure()). Otherwise the L1 hypervisor is resumed
some number of bytes past vmcs->host_rip.

Fixes: eb2775621701e6ee3ea2a474437d04e93ccdcb2f
Signed-off-by: David Matlack <dmatlack@google.com>
---
 arch/x86/kvm/vmx.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index c41d7ff..0e7ad72 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -10474,12 +10474,12 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)) {
 		nested_vmx_entry_failure(vcpu, vmcs12,
 			EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
-		goto out;
+		return 1;
 	}
 	if (vmcs12->vmcs_link_pointer != -1ull) {
 		nested_vmx_entry_failure(vcpu, vmcs12,
 			EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
-		goto out;
+		return 1;
 	}
 
 	/*
@@ -10499,7 +10499,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
 			nested_vmx_entry_failure(vcpu, vmcs12,
 				EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
-			goto out;
+			return 1;
 		}
 	}
 
@@ -10517,7 +10517,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
 			nested_vmx_entry_failure(vcpu, vmcs12,
 				EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
-			goto out;
+			return 1;
 		}
 	}
 
-- 
2.8.0.rc3.226.g39d4020

Re: [PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[Kyle Huey]

On Mon, Dec 19, 2016 at 12:48 PM, David Matlack <dmatlack@google.com> wrote:
> kvm_skip_emulated_instruction() should not be called after emulating
> a VM-entry failure during or after loading guest state
> (nested_vmx_entry_failure()). Otherwise the L1 hypervisor is resumed
> some number of bytes past vmcs->host_rip.

Ah, I see.  Sorry for that regression.

These paths are supposed to trigger TF-induced singlestep exceptions
though.  Quoting from the Intel SDM (Vol 3, Chapter 26)

"EFLAGS.TF = 1 causes a VM-entry instruction to generate a single-step
debug exception only if failure of one of the checks in Section 26.1
and Section 26.2 causes control to pass to the following instruction.
A VM-entry does not generate a single-step debug exception in any of
the following cases: (1) the instruction generates a fault; (2)
failure of one of the checks in Section 26.3 or in loading MSRs causes
processor state to be loaded from the hoststate area of the VMCS; or
(3) the instruction passes all checks in Section 26.1, Section 26.2,
and Section 26.3 and there is no failure in loading MSRs"

- Kyle

> Fixes: eb2775621701e6ee3ea2a474437d04e93ccdcb2f
> Signed-off-by: David Matlack <dmatlack@google.com>
> ---
>  arch/x86/kvm/vmx.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index c41d7ff..0e7ad72 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -10474,12 +10474,12 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>             !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)) {
>                 nested_vmx_entry_failure(vcpu, vmcs12,
>                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
> -               goto out;
> +               return 1;
>         }
>         if (vmcs12->vmcs_link_pointer != -1ull) {
>                 nested_vmx_entry_failure(vcpu, vmcs12,
>                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
> -               goto out;
> +               return 1;
>         }
>
>         /*
> @@ -10499,7 +10499,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
>                         nested_vmx_entry_failure(vcpu, vmcs12,
>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
> -                       goto out;
> +                       return 1;
>                 }
>         }
>
> @@ -10517,7 +10517,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
>                         nested_vmx_entry_failure(vcpu, vmcs12,
>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
> -                       goto out;
> +                       return 1;
>                 }
>         }
>
> --
> 2.8.0.rc3.226.g39d4020
>

Re: [PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[Radim Krčmář]

2016-12-19 19:15-0800, Kyle Huey:
> On Mon, Dec 19, 2016 at 12:48 PM, David Matlack <dmatlack@google.com> wrote:
>> kvm_skip_emulated_instruction() should not be called after emulating
>> a VM-entry failure during or after loading guest state
>> (nested_vmx_entry_failure()). Otherwise the L1 hypervisor is resumed
>> some number of bytes past vmcs->host_rip.
> 
> Ah, I see.  Sorry for that regression.
> 
> These paths are supposed to trigger TF-induced singlestep exceptions
> though.  Quoting from the Intel SDM (Vol 3, Chapter 26)
> 
> "EFLAGS.TF = 1 causes a VM-entry instruction to generate a single-step
> debug exception only if failure of one of the checks in Section 26.1
> and Section 26.2 causes control to pass to the following instruction.
> A VM-entry does not generate a single-step debug exception in any of
> the following cases: (1) the instruction generates a fault; (2)
> failure of one of the checks in Section 26.3 or in loading MSRs causes
> processor state to be loaded from the hoststate area of the VMCS; or
> (3) the instruction passes all checks in Section 26.1, Section 26.2,
> and Section 26.3 and there is no failure in loading MSRs"

Changed cases are in section 26.3 => not generating #DB is correct,

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>

Sorry for missing that while applying, I wonder if there is a reason why
we didn't check them after enter_guest_mode() ...

>> Fixes: eb2775621701e6ee3ea2a474437d04e93ccdcb2f
>> Signed-off-by: David Matlack <dmatlack@google.com>
>> ---
>>  arch/x86/kvm/vmx.c | 8 ++++----
>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> index c41d7ff..0e7ad72 100644
>> --- a/arch/x86/kvm/vmx.c
>> +++ b/arch/x86/kvm/vmx.c
>> @@ -10474,12 +10474,12 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>>             !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)) {
>>                 nested_vmx_entry_failure(vcpu, vmcs12,
>>                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
>> -               goto out;
>> +               return 1;
>>         }
>>         if (vmcs12->vmcs_link_pointer != -1ull) {
>>                 nested_vmx_entry_failure(vcpu, vmcs12,
>>                         EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
>> -               goto out;
>> +               return 1;
>>         }
>>
>>         /*
>> @@ -10499,7 +10499,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>>                      ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
>>                         nested_vmx_entry_failure(vcpu, vmcs12,
>>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
>> -                       goto out;
>> +                       return 1;
>>                 }
>>         }
>>
>> @@ -10517,7 +10517,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>>                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
>>                         nested_vmx_entry_failure(vcpu, vmcs12,
>>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
>> -                       goto out;
>> +                       return 1;
>>                 }
>>         }
>>
>> --
>> 2.8.0.rc3.226.g39d4020
>>

Re: [PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[Radim Krčmář]

2016-12-20 16:24+0100, Radim Krčmář:
> 2016-12-19 19:15-0800, Kyle Huey:
> > On Mon, Dec 19, 2016 at 12:48 PM, David Matlack <dmatlack@google.com> wrote:
> >> kvm_skip_emulated_instruction() should not be called after emulating
> >> a VM-entry failure during or after loading guest state
> >> (nested_vmx_entry_failure()). Otherwise the L1 hypervisor is resumed
> >> some number of bytes past vmcs->host_rip.
> > 
> > Ah, I see.  Sorry for that regression.
> > 
> > These paths are supposed to trigger TF-induced singlestep exceptions
> > though.  Quoting from the Intel SDM (Vol 3, Chapter 26)
> > 
> > "EFLAGS.TF = 1 causes a VM-entry instruction to generate a single-step
> > debug exception only if failure of one of the checks in Section 26.1
> > and Section 26.2 causes control to pass to the following instruction.
> > A VM-entry does not generate a single-step debug exception in any of
> > the following cases: (1) the instruction generates a fault; (2)
> > failure of one of the checks in Section 26.3 or in loading MSRs causes
> > processor state to be loaded from the hoststate area of the VMCS; or
> > (3) the instruction passes all checks in Section 26.1, Section 26.2,
> > and Section 26.3 and there is no failure in loading MSRs"
> 
> Changed cases are in section 26.3 => not generating #DB is correct,
> 
> Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
> 
> Sorry for missing that while applying, I wonder if there is a reason why
> we didn't check them after enter_guest_mode() ...
> 
> >> @@ -10517,7 +10517,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
> >>                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
> >>                         nested_vmx_entry_failure(vcpu, vmcs12,
> >>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
> >> -                       goto out;
> >> +                       return 1;

My attention is in shambles these days ... this one looks like it is in
section 26.2.2 and should therefore be

  nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
  goto out;

Not a problem of this patch though, I'll go for a minor refactoring
after a beer.

Re: [PATCH] KVM: nVMX: fix instruction skipping during emulated vm-entry

[David Matlack]

On Tue, Dec 20, 2016 at 9:47 AM, Radim Krčmář <rkrcmar@redhat.com> wrote:
> 2016-12-20 16:24+0100, Radim Krčmář:
>> 2016-12-19 19:15-0800, Kyle Huey:
>> > On Mon, Dec 19, 2016 at 12:48 PM, David Matlack <dmatlack@google.com> wrote:
>> >> kvm_skip_emulated_instruction() should not be called after emulating
>> >> a VM-entry failure during or after loading guest state
>> >> (nested_vmx_entry_failure()). Otherwise the L1 hypervisor is resumed
>> >> some number of bytes past vmcs->host_rip.
>> >
>> > Ah, I see.  Sorry for that regression.
>> >
>> > These paths are supposed to trigger TF-induced singlestep exceptions
>> > though.  Quoting from the Intel SDM (Vol 3, Chapter 26)
>> >
>> > "EFLAGS.TF = 1 causes a VM-entry instruction to generate a single-step
>> > debug exception only if failure of one of the checks in Section 26.1
>> > and Section 26.2 causes control to pass to the following instruction.
>> > A VM-entry does not generate a single-step debug exception in any of
>> > the following cases: (1) the instruction generates a fault; (2)
>> > failure of one of the checks in Section 26.3 or in loading MSRs causes
>> > processor state to be loaded from the hoststate area of the VMCS; or
>> > (3) the instruction passes all checks in Section 26.1, Section 26.2,
>> > and Section 26.3 and there is no failure in loading MSRs"
>>
>> Changed cases are in section 26.3 => not generating #DB is correct,
>>
>> Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
>>
>> Sorry for missing that while applying, I wonder if there is a reason why
>> we didn't check them after enter_guest_mode() ...

I think this would be a nice cleanup.

>>
>> >> @@ -10517,7 +10517,7 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
>> >>                     ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
>> >>                         nested_vmx_entry_failure(vcpu, vmcs12,
>> >>                                 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
>> >> -                       goto out;
>> >> +                       return 1;
>
> My attention is in shambles these days ... this one looks like it is in
> section 26.2.2 and should therefore be
>
>   nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
>   goto out;

I agree. Good catch.

>
> Not a problem of this patch though, I'll go for a minor refactoring
> after a beer.