2020-05-29 15:44:00

by Paolo Bonzini

Subject: [PATCH 20/30] KVM: SVM: preserve VGIF across VMCB switch

There is only one GIF flag for the whole processor, so make sure it is not clobbered
when switching to L2 (in which case we also have to include the V_GIF_ENABLE_MASK,
lest we confuse enable_gif/disable_gif/gif_set). When going back, L1 could in
theory have entered L2 without issuing a CLGI so make sure the svm_set_gif is
done last, after svm->vmcb->control.int_ctl has been copied back from hsave.

Signed-off-by: Paolo Bonzini <[email protected]>
---
arch/x86/kvm/svm/nested.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
index 7e4a506828c9..6c7f0bffdf01 100644
--- a/arch/x86/kvm/svm/nested.c
+++ b/arch/x86/kvm/svm/nested.c
@@ -293,6 +293,7 @@ static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *nested_v

static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
{
+ const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
if (svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE)
nested_svm_init_mmu_context(&svm->vcpu);

@@ -308,7 +309,10 @@ static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;

- svm->vmcb->control.int_ctl = svm->nested.ctl.int_ctl | V_INTR_MASKING_MASK;
+ svm->vmcb->control.int_ctl =
+ (svm->nested.ctl.int_ctl & ~mask) |
+ (svm->nested.hsave->control.int_ctl & mask);
+
svm->vmcb->control.virt_ext = svm->nested.ctl.virt_ext;
svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
svm->vmcb->control.int_state = svm->nested.ctl.int_state;
--
2.26.2

2020-05-31 23:16:24

by Krish Sadhukhan

Subject: Re: [PATCH 20/30] KVM: SVM: preserve VGIF across VMCB switch

On 5/29/20 8:39 AM, Paolo Bonzini wrote:
> There is only one GIF flag for the whole processor, so make sure it is not clobbered
> when switching to L2 (in which case we also have to include the V_GIF_ENABLE_MASK,
> lest we confuse enable_gif/disable_gif/gif_set). When going back, L1 could in
> theory have entered L2 without issuing a CLGI so make sure the svm_set_gif is
> done last, after svm->vmcb->control.int_ctl has been copied back from hsave.
>
> Signed-off-by: Paolo Bonzini <[email protected]>
> ---
> arch/x86/kvm/svm/nested.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
> index 7e4a506828c9..6c7f0bffdf01 100644
> --- a/arch/x86/kvm/svm/nested.c
> +++ b/arch/x86/kvm/svm/nested.c
> @@ -293,6 +293,7 @@ static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *nested_v
>
> static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
> {
> + const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
> if (svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE)
> nested_svm_init_mmu_context(&svm->vcpu);
>
> @@ -308,7 +309,10 @@ static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
> svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
> svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
>
> - svm->vmcb->control.int_ctl = svm->nested.ctl.int_ctl | V_INTR_MASKING_MASK;
> + svm->vmcb->control.int_ctl =
> + (svm->nested.ctl.int_ctl & ~mask) |
> + (svm->nested.hsave->control.int_ctl & mask);


If this is the very first VMRUN, do we have any int_ctl saved in hsave ?

> +
> svm->vmcb->control.virt_ext = svm->nested.ctl.virt_ext;
> svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
> svm->vmcb->control.int_state = svm->nested.ctl.int_state;

2020-06-01 07:33:33

by Paolo Bonzini

Subject: Re: [PATCH 20/30] KVM: SVM: preserve VGIF across VMCB switch

On 01/06/20 01:11, Krish Sadhukhan wrote:
>>
>> +    svm->vmcb->control.int_ctl             =
>> +        (svm->nested.ctl.int_ctl & ~mask) |
>> +        (svm->nested.hsave->control.int_ctl & mask);
>
>
> If this is the very first VMRUN, do we have any int_ctl saved in hsave ?

Yes, copy_vmcb_control_area(hsave, vmcb) is called before
enter_svm_guest_mode (which calls nested_prepare_vmcb_control).

Paolo