GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel arm_ffa: out-of-bounds writes when building FF-A memory-sharing descriptors

CVSS 7.8CVE-2026-68401Kernel, userspace & hypervisorcurated

Impact

The FF-A driver builds its memory-sharing descriptor without accounting for the smaller endpoint-memory-access record used by FF-A versions below 1.2, so zeroing the reserved field writes past the allocated structure, and the constituent array is written without a size check. A local attacker able to drive the FF-A memory-share path corrupts kernel heap memory, which is the usual starting point for local privilege escalation on the host. This matters on Arm-based accelerator hosts (Grace, Ampere and similar) that use FF-A to talk to secure-world partitions; x86 GPU nodes do not load this driver at all, so most fleets are unaffected. The fix memsets the structure according to the negotiated FF-A version and bounds the constituent write.

Who can reach it

Local only, on an Arm host where the arm_ffa driver is loaded and a userspace or in-kernel consumer can reach the FF-A memory-sharing path. Low-privileged local access is enough per the vendor vector (AV:L/PR:L); no remote or network exposure.

What to do

Take the stable-kernel fix (backported across several stable branches - see the git.kernel.org commits) and reboot each affected Arm node. There is no runtime mitigation other than keeping the FF-A driver out of the kernel on platforms that do not need it, so plan a rolling drain-and-reboot across the Arm part of the fleet.

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.