Re: [PATCH] KVM: x86: Always set kvm_run->if_flag

[Tom Lendacky <thomas.lendacky@amd.com>]

On 12/7/21 9:14 AM, Marc Orr wrote:
> On Tue, Dec 7, 2021 at 6:43 AM Tom Lendacky <thomas.lendacky@amd.com> wrote:
>>
>> On 12/6/21 10:31 PM, Marc Orr wrote:
>>> The kvm_run struct's if_flag is apart of the userspace/kernel API. The
>>> SEV-ES patches failed to set this flag because it's no longer needed by
>>> QEMU (according to the comment in the source code). However, other
>>> hypervisors may make use of this flag. Therefore, set the flag for
>>> guests with encrypted regiesters (i.e., with guest_state_protected set).
>>>
>>> Fixes: f1c6366e3043 ("KVM: SVM: Add required changes to support intercepts under SEV-ES")
>>> Signed-off-by: Marc Orr <marcorr@google.com>
>>> ---
>>>    arch/x86/include/asm/kvm-x86-ops.h | 1 +
>>>    arch/x86/include/asm/kvm_host.h    | 1 +
>>>    arch/x86/kvm/svm/svm.c             | 8 ++++++++
>>>    arch/x86/kvm/vmx/vmx.c             | 6 ++++++
>>>    arch/x86/kvm/x86.c                 | 9 +--------
>>>    5 files changed, 17 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
>>> index cefe1d81e2e8..9e50da3ed01a 100644
>>> --- a/arch/x86/include/asm/kvm-x86-ops.h
>>> +++ b/arch/x86/include/asm/kvm-x86-ops.h
>>> @@ -47,6 +47,7 @@ KVM_X86_OP(set_dr7)
>>>    KVM_X86_OP(cache_reg)
>>>    KVM_X86_OP(get_rflags)
>>>    KVM_X86_OP(set_rflags)
>>> +KVM_X86_OP(get_if_flag)
>>>    KVM_X86_OP(tlb_flush_all)
>>>    KVM_X86_OP(tlb_flush_current)
>>>    KVM_X86_OP_NULL(tlb_remote_flush)
>>> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
>>> index 860ed500580c..a7f868ff23e7 100644
>>> --- a/arch/x86/include/asm/kvm_host.h
>>> +++ b/arch/x86/include/asm/kvm_host.h
>>> @@ -1349,6 +1349,7 @@ struct kvm_x86_ops {
>>>        void (*cache_reg)(struct kvm_vcpu *vcpu, enum kvm_reg reg);
>>>        unsigned long (*get_rflags)(struct kvm_vcpu *vcpu);
>>>        void (*set_rflags)(struct kvm_vcpu *vcpu, unsigned long rflags);
>>> +     bool (*get_if_flag)(struct kvm_vcpu *vcpu);
>>>
>>>        void (*tlb_flush_all)(struct kvm_vcpu *vcpu);
>>>        void (*tlb_flush_current)(struct kvm_vcpu *vcpu);
>>> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
>>> index d0f68d11ec70..91608f8c0cde 100644
>>> --- a/arch/x86/kvm/svm/svm.c
>>> +++ b/arch/x86/kvm/svm/svm.c
>>> @@ -1585,6 +1585,13 @@ static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
>>>        to_svm(vcpu)->vmcb->save.rflags = rflags;
>>>    }
>>>
>>> +static bool svm_get_if_flag(struct kvm_vcpu *vcpu)
>>> +{
>>> +     struct vmcb *vmcb = to_svm(vcpu)->vmcb;
>>> +
>>> +     return !!(vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK);
>>
>> I'm not sure if this is always valid to use for non SEV-ES guests. Maybe
>> the better thing would be:
>>
>>          return sev_es_guest(vcpu->kvm) ? vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK
>>                                         : kvm_get_rflags(vcpu) & X86_EFLAGS_IF;
>>
>> (Since this function returns a bool, I don't think you need the !!)
> 
> I had the same reservations when writing the patch. (Why fix what's
> not broken.) The reason I wrote the patch this way is based on what I
> read in APM vol2: Appendix B Layout of VMCB: "GUEST_INTERRUPT_MASK -
> Value of the RFLAGS.IF bit for the guest."

I just verified with the hardware team that this flag is indeed only set 
for a guest with protected state (SEV-ES / SEV-SNP). An update to the APM 
will be made.

Thanks,
Tom

> 
> Also, I had _thought_ that `svm_interrupt_allowed()` -- the
> AMD-specific function used to populate `ready_for_interrupt_injection`
> -- was relying on `GUEST_INTERRUPT_MASK`. But now I'm reading the code
> again, and I realized I was overly focused on the SEV-ES handling.
> That code is actually extracting the IF bit from the RFLAGS register
> in the same way you've proposed here.
> 
> Changing the patch as you've suggested SGTM. I can send out a v2. I'll
> wait a day or two to see if there are any other comments first. I
> guess the alternative would be to change `svm_interrupt_blocked()` to
> solely use the `SVM_GUEST_INTERRUPT_MASK`. If we were confident that
> it was sufficient, it would be a nice little cleanup. But regardless,
> I think we should keep the code introduced by this patch consistent
> with `svm_interrupt_blocked()`.
> 
> Also, noted on the `!!` not being needed when returning from a bool
> function. I'll keep this in mind in the future. Thanks!
>