GPU VulnDB

Database/Kernel, userspace & hypervisor

KVM x86 Hyper-V stimer: overflowed deadline livelocks the vCPU thread and stalls RCU on the host

CVSS 7.1CVE-2026-89927Kernel, userspace & hypervisorcurated

Impact

A guest (or the VMM) can program a Hyper-V synthetic timer with a COUNT near U64_MAX so the deadline calculation overflows into the past. The hrtimer then fires immediately, KVM re-arms it, the pending request aborts every guest entry, and vcpu_run() spins without yielding. The loop starves RCU grace-period kthreads - syzkaller reproduced multi-second rcu_preempt stalls - and on a SCHED_FIFO vCPU thread it can wedge the CPU outright. On a GPU host running Windows or Hyper-V-enlightened guests that is one tenant pinning a core and degrading or hanging the whole node, which then has to be rebooted; nothing is disclosed or modified, the damage is availability.

Who can reach it

Local to the host via a guest: any tenant VM on an x86 KVM host where Hyper-V synthetic timer emulation is enabled can write the MSRs itself. No host credentials required.

What to do

Apply the stable-kernel fix that clamps the stimer deadline (commits below) and reboot the affected hypervisor hosts after draining their VMs. As an interim measure, do not expose the Hyper-V SynIC/stimer enlightenments to untrusted guests, and avoid running vCPU threads at SCHED_FIFO, which is what turns the livelock into a full CPU stall.

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.