Database/Kernel, userspace & hypervisor

Linux KVM TDX: Bus Lock VM exit reported as enabled for TDX guests but never programmed
Impact
On Intel TDX hosts, KVM advertises KVM_BUS_LOCK_DETECTION_EXIT based on hardware capability alone, so a VMM can enable it for a TDX guest and get no error, but the TD VMCS control is never set and no exit handler exists. The operator believes bus-lock detection (and any rate limiting built on it) is active for confidential guests when it is not. On a shared host, that removes the control meant to stop one guest's bus locks from degrading memory performance for co-tenant VMs on the same socket. The record describes a missing feature, not memory corruption.
Who can reach it
A tenant running code inside a TDX guest on a host whose VMM enables bus-lock detection. No host authentication needed beyond owning the guest.
What to do
Install a kernel carrying the stable fix (commits linked) on TDX hypervisor hosts, then live-migrate or drain guests and reboot each host. Until then, do not rely on bus-lock detection for TDX guests. The record does not list fixed version numbers.
References
Related entries
- Linux KVM TDX: Notify VM exit accepted for TDX guests but never enabled in the TD VMCSCVE-2026-99338 · Linux kernel KVM TDX (Notify VM exit / instruction timeout)Unscored
- Linux KVM x86: SynIC message delivery leaks seven bits of host kernel stack to the guestCVE-2026-99339 · Linux kernel KVM x86 Hyper-V SynIC emulationUnscored
- Linux KVM x86: kvm_zap_gfn_range races memslot deletion, use-after-free via KVM-VFIO deviceCVE-2026-99340 · Linux kernel KVM x86 MMU (kvm_zap_gfn_range, VFIO noncoherent DMA path)Unscored
- Linux KVM nSVM: L1 guest can enter L2 with EFER.LMA=1 and LME=0, breaking KVM MMU invariantsCVE-2026-99344 · Linux kernel KVM nSVM (nested AMD virtualization, VMRUN EFER handling)Unscored
- Linux KVM nSVM: KVM_SET_NESTED_STATE accepts LMA=1/LME=0, shadow MMU walks past its paging arraysCVE-2026-99345 · Linux kernel KVM nSVM (KVM_SET_NESTED_STATE)Unscored
- Linux KVM SVM: nested vmcb01/vmcb02 ASID tracking lets two vCPUs share one ASID on a pCPUCVE-2026-99346 · Linux kernel KVM SVM (nested ASID allocation)Unscored
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.