GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel ice: double free in the E810 subfunction activation error path

CVSS 7.8CVE-2026-46162Kernel, userspace & hypervisorcurated

Impact

When auxiliary_device_add() fails during activation of an ice subfunction, the error path calls auxiliary_device_uninit() - whose release callback already frees sf_dev - and then falls through to kfree(sf_dev), freeing it a second time. The ice driver runs the Intel E810 NICs used for host and storage networking on many GPU nodes, and subfunctions are how those ports are carved up for containers and VMs. Reaching the bug needs the activation to fail partway, so in practice this is a crash or slab corruption during a privileged devlink operation rather than a tenant-reachable escalation. On a GPU node the cost is the outage: a corrupted slab means the node goes down and has to be drained and rebooted.

Who can reach it

Local and privileged. Requires the ability to create and activate ice subfunctions over devlink (root / CAP_NET_ADMIN on the host), plus a failing auxiliary_device_add(). No tenant or network path.

What to do

Take the stable kernel that keeps kfree(sf_dev) only on the auxiliary_device_init() failure path. Kernel update means drain each affected node and reboot it; schedule with the normal kernel cadence rather than out of band. Until then, avoid repeated subfunction create/activate churn on nodes still running the old kernel.

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.