GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel iommu/amd: every handled PPR fault leaks a PCI device reference

UnscoredCVE-2026-89453Kernel, userspace & hypervisorcurated

Impact

iommu_call_iopf_notifier() resolves the faulting requester with pci_get_domain_bus_and_slot(), which takes a reference, and neither the success path nor the abort path drops it. On an AMD host running accelerators that use PPR/IOPF page faulting, every single handled page-request leaks one struct pci_dev reference. The practical consequence is that the device can never be released: hot-unplug, driver unbind, VFIO teardown between tenants and SR-IOV VF teardown will hang or leave stale state, and the reference count grows without bound for the life of the boot. This is an availability and operational problem on a shared node rather than a memory-corruption primitive - the leak is upward, not a use-after-free. The same ownership bug was fixed years ago in the older ppr_notifier() path; the newer IOPF notifier was added separately and missed it.

Who can reach it

Local - any workload on the host that drives an accelerator using PPR/IOPF page faults triggers the leak simply by running. No authentication distinction and no deliberate attack is needed; ordinary tenant use of an AMD IOMMU PPR-capable device is enough.

What to do

Update to a stable kernel carrying the fix (four stable commits linked) and reboot the affected nodes. No module reload or runtime workaround clears already-leaked references. Affects only hosts with AMD IOMMU and PPR-capable devices in use, which narrows the reboot list. No fixed release number is given in the record.

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.