GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel i40e: VF-supplied ring_len written to hardware HMC context without validation

CVSS 8.8CVE-2025-39973Kernel, userspace & hypervisorcurated

Impact

The ring_len a virtual function asks for was programmed straight into the hardware memory context with no upper bound and no alignment check; the fix caps Tx and Rx descriptor counts at the hardware maximum of 8k-32 and enforces the 8- and 32-descriptor alignment the hardware requires. A tenant VF can therefore put the NIC into a state the hardware does not define, which is reached through the host PF driver and recorded with a changed scope. On a GPU node where the data-plane NIC is shared by SR-IOV, that means one tenant can break host networking - and a NIC wedged at the hardware-context level typically needs a reboot, not a driver reload. Distinct from the other two i40e VF-validation CVEs in the same series: different field, different check.

Who can reach it

Local to the guest: any tenant that owns an i40e VF and can request ring sizes from the PF. No host credentials.

What to do

Move to a stable kernel with the linked fix and reboot the node. No configuration workaround exists; the interim mitigation is to withhold i40e VFs from untrusted tenants. Drain the node before rebooting.

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.