GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel Intel uncore PMU: die ID lookup bugs trip a warning and skip PMON units

CVSS 5.5CVE-2026-43344Kernel, userspace & hypervisorcurated

Impact

On Intel Sapphire Rapids and Emerald Rapids hosts - the CPUs under most current GPU nodes - uncore_device_to_die() returns -1 when every CPU behind a UBOX device is offline, and pcibus_to_node() returns NUMA_NO_NODE when NUMA is disabled on a NUMA-capable platform. The first case trips WARN_ON_ONCE and breaks out of a scan that pci_get_device does not order, so further UBOX devices are silently skipped; the second makes spr_update_device_location drop the PMON units entirely. The operator-visible result is missing uncore counters - memory bandwidth, UPI, home agent - in host telemetry, and a panic on a kernel booted with panic_on_warn. NVD rates it 5.5, local availability only. This is not a privilege boundary crossing.

Who can reach it

No remote path. The condition is reached by a local user with the privilege to offline CPUs, or simply by booting the platform with NUMA disabled - the CVSS vector requires local access with low privileges.

What to do

Take a stable kernel containing the linked commits and reboot the node to pick it up; drain GPU workloads first as with any kernel update. Not worth its own maintenance window unless you rely on SPR/EMR uncore counters for capacity or fabric telemetry, or you boot production kernels with panic_on_warn - in either case the missing PMON units or the panic are the symptom you are fixing.

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.