[00/10] KVM: SVM: Attempt to cleanup SEV_FEATURES

Message

Sean Christopherson Feb. 19, 2025, 1:26 a.m. UTC

This is a hastily thrown together series, barely above RFC, to try and
address the worst of the issues that arise with guest controlled SEV
features (thanks AP creation)[1].

In addition to the initial flaws with DebugSwap, I came across a variety
of issues when trying to figure out how best to handle SEV features in
general.  E.g. AFAICT, KVM doesn't guard against userspace manually making
a vCPU RUNNABLE after it has been DESTROYED (or after a failed CREATE).

This is essentially compile-tested only, as I don't have easy access to a
system with SNP enabled.  I ran the SEV-ES selftests, but that's not much
in the way of test coverage.

AMD folks, I would greatly appreciate reviews, testing, and most importantly,
confirmation that all of this actually works the way I think it does.

[1] https://lore.kernel.org/all/[e-mail]

Sean Christopherson (10):
  KVM: SVM: Save host DR masks but NOT DRs on CPUs with DebugSwap
  KVM: SVM: Don't rely on DebugSwap to restore host DR0..DR3
  KVM: SVM: Terminate the VM if a SEV-ES+ guest is run with an invalid
    VMSA
  KVM: SVM: Don't change target vCPU state on AP Creation VMGEXIT error
  KVM: SVM: Require AP's "requested" SEV_FEATURES to match KVM's view
  KVM: SVM: Simplify request+kick logic in SNP AP Creation handling
  KVM: SVM: Use guard(mutex) to simplify SNP AP Creation error handling
  KVM: SVM: Mark VMCB dirty before processing incoming snp_vmsa_gpa
  KVM: SVM: Use guard(mutex) to simplify SNP vCPU state updates
  KVM: SVM: Invalidate "next" SNP VMSA GPA even on failure

 arch/x86/kvm/svm/sev.c | 218 +++++++++++++++++++----------------------
 arch/x86/kvm/svm/svm.c |   7 +-
 arch/x86/kvm/svm/svm.h |   2 +-
 3 files changed, 106 insertions(+), 121 deletions(-)


base-commit: fed48e2967f402f561d80075a20c5c9e16866e53

Comments

Tom Lendacky Feb. 20, 2025, 10:51 p.m. UTC | #1

On 2/18/25 19:26, Sean Christopherson wrote:
> This is a hastily thrown together series, barely above RFC, to try and
> address the worst of the issues that arise with guest controlled SEV
> features (thanks AP creation)[1].
> 
> In addition to the initial flaws with DebugSwap, I came across a variety
> of issues when trying to figure out how best to handle SEV features in
> general.  E.g. AFAICT, KVM doesn't guard against userspace manually making
> a vCPU RUNNABLE after it has been DESTROYED (or after a failed CREATE).
> 
> This is essentially compile-tested only, as I don't have easy access to a
> system with SNP enabled.  I ran the SEV-ES selftests, but that's not much
> in the way of test coverage.
> 
> AMD folks, I would greatly appreciate reviews, testing, and most importantly,
> confirmation that all of this actually works the way I think it does.

A quick test of a 64 vCPU SNP guest booted successfully, so that's a
good start. I'll take a closer look at these patches over the next few days.

Thanks,
Tom

> 
> [1] https://lore.kernel.org/all/[e-mail]
> 
> Sean Christopherson (10):
>   KVM: SVM: Save host DR masks but NOT DRs on CPUs with DebugSwap
>   KVM: SVM: Don't rely on DebugSwap to restore host DR0..DR3
>   KVM: SVM: Terminate the VM if a SEV-ES+ guest is run with an invalid
>     VMSA
>   KVM: SVM: Don't change target vCPU state on AP Creation VMGEXIT error
>   KVM: SVM: Require AP's "requested" SEV_FEATURES to match KVM's view
>   KVM: SVM: Simplify request+kick logic in SNP AP Creation handling
>   KVM: SVM: Use guard(mutex) to simplify SNP AP Creation error handling
>   KVM: SVM: Mark VMCB dirty before processing incoming snp_vmsa_gpa
>   KVM: SVM: Use guard(mutex) to simplify SNP vCPU state updates
>   KVM: SVM: Invalidate "next" SNP VMSA GPA even on failure
> 
>  arch/x86/kvm/svm/sev.c | 218 +++++++++++++++++++----------------------
>  arch/x86/kvm/svm/svm.c |   7 +-
>  arch/x86/kvm/svm/svm.h |   2 +-
>  3 files changed, 106 insertions(+), 121 deletions(-)
> 
> 
> base-commit: fed48e2967f402f561d80075a20c5c9e16866e53