Linux iommu/amd: signed offset parsing in IOMMU debugfs allowed a latent out-of-bounds access

CVSS 7.1CVE-2026-64186Kernel, userspace & hypervisorcurated

Impact

The AMD IOMMU debugfs write handlers parsed user offsets into int variables via kstrtou32_from_user(), so a large value became negative and could index outside the IOMMU MMIO and capability ranges. The commit is explicit that the existing code was already protected by an 8-byte input-length cap, implicit promotion against a u64 bound, and negative-value checks in the show handlers, so this closes a latent hole rather than a reachable one. It is still worth knowing on GPU hosts, because the AMD IOMMU is what enforces DMA isolation for passthrough GPUs and VFIO device assignment, and debugfs pokes at it are something operators do while debugging passthrough. Reaching it needs root on the host, which already owns the IOMMU.

Who can reach it

Local root on the host: debugfs must be mounted and writable, which normally means privileged access only. Not reachable by a tenant in a container or VM, and not reachable over the network.

What to do

Update to a stable kernel with the switch to kstrtos32_from_user() plus explicit negative checks; three stable commits are linked. That means a kernel update and a node reboot. As an interim mitigation, keep debugfs unmounted or restricted to root on production GPU hosts, which most hardened builds already do.

References

https://nvd.nist.gov/vuln/detail/CVE-2026-64186 https://git.kernel.org/stable/c/488d2c76bd9f78433a70690d1054bfae3d39a407 https://git.kernel.org/stable/c/8dfd3d8d74435344ee8dc9237596959c8b2a6cbe

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.