Skip to contentSkip to conversation

KVM: nVMX: Preserve registers modifications done before nested_svm_vmexit()

PATCH Superseded Vitaly Kuznetsov 2 participants 4 messages May 27, 2020

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: nVMX: Preserve registers modifications done before nested_svm_vmexit()

L2 guest hang is observed after 'exit_required' was dropped and nSVM switched to check_nested_events() completely. The hang is a busy loop when e.g. KVM is emulating an instruction (e.g. L2 is accessing MMIO space and we drop to userspace). After nested_svm_vmexit() and when L1 is doing VMRUN nested guest's RIP is not advanced so KVM goes into emulating the same instruction which cased nested_svm_vmexit() and the loop continues.

nested_svm_vmexit() is not new, however, with check_nested_events() we're now calling it later than before. In case by that time KVM has modified register state we may pick stale values from VMCS when trying to save nested guest state to nested VMCB.

VMX code handles this case correctly: sync_vmcs02_to_vmcs12() called from nested_vmx_vmexit() does 'vmcs12->guest_rip = kvm_rip_read(vcpu)' and this ensures KVM-made modifications are preserved. Do the same for nVMX.

Generally, nested_vmx_vmexit()/nested_svm_vmexit() need to pick up all nested guest state modifications done by KVM after vmexit. It would be great to find a way to express this in a way which would not require to manually track these changes, e.g. nested_{vmcb,vmcs}_get_field().

Co-debugged-with: Paolo Bonzini <[email protected]>

---
- To certain extent we're fixing currently-pending 'KVM: SVM: immediately inject INTR vmexit' commit but I'm not certain about that. We had so many problems with nested events before switching to check_nested_events() that what worked before could just be treated as a miracle. Miracles tend to appear and disappear all of a sudden.

---
 arch/x86/kvm/svm/nested.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/x86/kvm/svm/nested.c +3 -3
index 0f02521550b9..6b1049148c1b 100644
@@ -537,9 +537,9 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 	nested_vmcb->save.cr2    = vmcb->save.cr2;
 	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
-	nested_vmcb->save.rip    = vmcb->save.rip;
-	nested_vmcb->save.rsp    = vmcb->save.rsp;
-	nested_vmcb->save.rax    = vmcb->save.rax;
+	nested_vmcb->save.rip    = kvm_rip_read(&svm->vcpu);
+	nested_vmcb->save.rsp    = kvm_rsp_read(&svm->vcpu);
+	nested_vmcb->save.rax    = kvm_rax_read(&svm->vcpu);
 	nested_vmcb->save.dr7    = vmcb->save.dr7;
 	nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 	nested_vmcb->save.cpl    = vmcb->save.cpl;
Signature · 1 line
2.25.4
#1 lore raw

Shortlog says nVMX, code says nSVM :-)

Quoted 6 lines · Vitaly Kuznetsov

L2 guest hang is observed after 'exit_required' was dropped and nSVM switched to check_nested_events() completely. The hang is a busy loop when e.g. KVM is emulating an instruction (e.g. L2 is accessing MMIO space and we drop to userspace). After nested_svm_vmexit() and when L1 is doing VMRUN nested guest's RIP is not advanced so KVM goes into emulating the same instruction which cased nested_svm_vmexit() and the loop continues.

s/cased/caused?

Quoted 3 lines

nested_svm_vmexit() is not new, however, with check_nested_events() we're now calling it later than before. In case by that time KVM has modified register state we may pick stale values from VMCS when trying to save

s/VMCS/VMCB

Quoted 5 lines

nested guest state to nested VMCB.

VMX code handles this case correctly: sync_vmcs02_to_vmcs12() called from nested_vmx_vmexit() does 'vmcs12->guest_rip = kvm_rip_read(vcpu)' and this ensures KVM-made modifications are preserved. Do the same for nVMX.

s/nVMX/nSVM

Quoted 38 lines

Generally, nested_vmx_vmexit()/nested_svm_vmexit() need to pick up all nested guest state modifications done by KVM after vmexit. It would be great to find a way to express this in a way which would not require to manually track these changes, e.g. nested_{vmcb,vmcs}_get_field().

Co-debugged-with: Paolo Bonzini <[email protected]>

---
- To certain extent we're fixing currently-pending 'KVM: SVM: immediately inject INTR vmexit' commit but I'm not certain about that. We had so many problems with nested events before switching to check_nested_events() that what worked before could just be treated as a miracle. Miracles tend to appear and disappear all of a sudden.

---
 arch/x86/kvm/svm/nested.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/x86/kvm/svm/nested.c +3 -3
index 0f02521550b9..6b1049148c1b 100644
@@ -537,9 +537,9 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 	nested_vmcb->save.cr2    = vmcb->save.cr2;
 	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
-	nested_vmcb->save.rip    = vmcb->save.rip;
-	nested_vmcb->save.rsp    = vmcb->save.rsp;
-	nested_vmcb->save.rax    = vmcb->save.rax;
+	nested_vmcb->save.rip    = kvm_rip_read(&svm->vcpu);
+	nested_vmcb->save.rsp    = kvm_rsp_read(&svm->vcpu);
+	nested_vmcb->save.rax    = kvm_rax_read(&svm->vcpu);
 	nested_vmcb->save.dr7    = vmcb->save.dr7;
 	nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 	nested_vmcb->save.cpl    = vmcb->save.cpl;
Signature · 1 line
2.25.4
#2 lore raw
Quoted 2 lines · Sean Christopherson

Shortlog says nVMX, code says nSVM :-)

My brain is tainted, you know :-) Also, saying *VMX* always helps to get your review so I may use the trick again :-)

Quoted 10 lines · Vitaly Kuznetsov
Quoted 6 lines

L2 guest hang is observed after 'exit_required' was dropped and nSVM switched to check_nested_events() completely. The hang is a busy loop when e.g. KVM is emulating an instruction (e.g. L2 is accessing MMIO space and we drop to userspace). After nested_svm_vmexit() and when L1 is doing VMRUN nested guest's RIP is not advanced so KVM goes into emulating the same instruction which cased nested_svm_vmexit() and the loop continues.

s/cased/caused?

Yep.

Quoted 6 lines
Quoted 3 lines

nested_svm_vmexit() is not new, however, with check_nested_events() we're now calling it later than before. In case by that time KVM has modified register state we may pick stale values from VMCS when trying to save

s/VMCS/VMCB

Yep.

Quoted 8 lines
Quoted 5 lines

nested guest state to nested VMCB.

VMX code handles this case correctly: sync_vmcs02_to_vmcs12() called from nested_vmx_vmexit() does 'vmcs12->guest_rip = kvm_rip_read(vcpu)' and this ensures KVM-made modifications are preserved. Do the same for nVMX.

s/nVMX/nSVM

Yep.

I'll do v2 to avoid the confusion.

Quoted 39 lines
Quoted 38 lines

Generally, nested_vmx_vmexit()/nested_svm_vmexit() need to pick up all nested guest state modifications done by KVM after vmexit. It would be great to find a way to express this in a way which would not require to manually track these changes, e.g. nested_{vmcb,vmcs}_get_field().

Co-debugged-with: Paolo Bonzini <[email protected]>

---
- To certain extent we're fixing currently-pending 'KVM: SVM: immediately inject INTR vmexit' commit but I'm not certain about that. We had so many problems with nested events before switching to check_nested_events() that what worked before could just be treated as a miracle. Miracles tend to appear and disappear all of a sudden.

---
 arch/x86/kvm/svm/nested.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
arch/x86/kvm/svm/nested.c +3 -3
index 0f02521550b9..6b1049148c1b 100644
@@ -537,9 +537,9 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 	nested_vmcb->save.cr2    = vmcb->save.cr2;
 	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
-	nested_vmcb->save.rip    = vmcb->save.rip;
-	nested_vmcb->save.rsp    = vmcb->save.rsp;
-	nested_vmcb->save.rax    = vmcb->save.rax;
+	nested_vmcb->save.rip    = kvm_rip_read(&svm->vcpu);
+	nested_vmcb->save.rsp    = kvm_rsp_read(&svm->vcpu);
+	nested_vmcb->save.rax    = kvm_rax_read(&svm->vcpu);
 	nested_vmcb->save.dr7    = vmcb->save.dr7;
 	nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 	nested_vmcb->save.cpl    = vmcb->save.cpl;
Signature · 1 line
2.25.4
Signature · 1 line
Vitaly
#3 lore raw
Quoted 7 lines · Vitaly Kuznetsov
Quoted 2 lines · Sean Christopherson

Shortlog says nVMX, code says nSVM :-)

My brain is tainted, you know :-) Also, saying *VMX* always helps to get your review so I may use the trick again :-)

Beetlejuice...Beetlejuice...Beetlejuice!