GPU VulnDB

Database/NVIDIA / GPU stack

accel/amdxdna: device buffers could be exported over dma-buf when they were never meant to be

CVSS 7.8CVE-2026-98436NVIDIA / GPU stackcurated

Impact

The AMD XDNA accelerator driver allowed device buffers to be exported as dma-bufs although they were never designed to be shared that way; the fix disables exporting explicitly. The upstream record is one sentence and does not describe what an attacker gains, so treat the concrete impact as unestablished - the plausible class is a local user sharing a device buffer whose lifetime or backing the driver does not track across the export, in a driver for an on-die AI accelerator. NVD's 7.8 local-privilege-escalation vector is the generic kernel-CVE score, not a demonstrated escalation. Most GPU datacenter fleets do not have XDNA NPU hardware; this matters mainly to operators running AMD Ryzen AI parts.

Who can reach it

Local user holding the amdxdna accel device node who can issue the dma-buf export ioctl. No remote path, no authentication beyond access to the device.

What to do

Upgrade to a stable kernel carrying either of the two listed commits and reboot the affected nodes. If no XDNA NPU hardware is present the driver never binds and there is nothing to schedule. The record names no fixed distro version.

References

Related entries

All NVIDIA / GPU stack 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.