GPU VulnDB

Database/Kernel, userspace & hypervisor

KVM PPC Book3S HV: preemptible HPTE and rmap lock holders can leave host CPUs spinning forever

UnscoredCVE-2026-99334Kernel, userspace & hypervisorcurated

Impact

Two virtual-mode paths on POWER KVM HV hosts took HPTE bit-locks, rmap locks or the raw kvm->mmu_lock with preemption still enabled. These are hash-mode guest page faults after guest exit, and the HPT hcalls (H_ENTER, H_REMOVE, H_READ, H_PROTECT, H_BULK_REMOVE and others). If the vCPU thread is preempted while it holds one of these locks, any other thread on that CPU that wants the same lock spins forever. The result is a host CPU lockup that hits every guest on the node. Only POWER Book3S HV hosts running hash-MMU guests are affected. x86 and Arm GPU nodes are not.

Who can reach it

A guest on a powerpc KVM HV host that uses the hashed page table, through ordinary page faults or HPT hcalls. Guest code execution is enough. No host authentication is needed, but hitting the hang depends on host scheduling timing.

What to do

Apply the stable kernel fix, which adds preempt_disable()/preempt_enable() around the two kvmppc_hpte_hv_fault() call sites and the HPT hcall dispatch. This is a host kernel update, so evacuate or shut down guests and reboot each POWER host. The record has no CVSS score or vendor advisory.

References

Related entries

All Kernel, userspace & hypervisor entries

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.