GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux KVM arm64 vgic: use-after-free racing LPI disable against interrupt affinity change

CVSS 9.3CVE-2026-72288Kernel, userspace & hypervisorcurated

Impact

vgic_prune_ap_list() dropped the irq and ap_list locks and reacquired them without holding a reference, so a guest that disables another vCPU's LPIs while a third vCPU migrates a pending LPI can have the interrupt freed in the window - a use-after-free in host kernel memory. The fix takes a reference before dropping the locks and stops trying to migrate the pending bit. This matters on Arm-based GPU hosts (Grace-Hopper and similar) that run KVM guests: a guest controls the timing from inside, so a tenant VM can corrupt host memory, with host crash the reliable outcome and worse not ruled out. Taking the host down means every accelerator attached to it goes with it. x86 hosts are unaffected.

Who can reach it

Local to the host, driven from inside a guest - any tenant VM on an arm64 KVM host whose vGIC has LPIs (GICv3 ITS). No host authentication; the guest only needs to write its own redistributor and ITS.

What to do

Apply the stable kernel fix (three backport commits listed) and reboot each arm64 KVM host after draining its VMs. There is no runtime mitigation short of not running untrusted guests on affected arm64 hosts; the record names no fixed distro release.

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.