GPU VulnDB

Database/NVIDIA / GPU stack

Linux kernel accel/qaic: leaked DBC after owner exit hangs the next process that activates a network

CVSS 7.8CVE-2026-43007NVIDIA / GPU stackcurated

Impact

If a process using a Qualcomm Cloud AI 100 accelerator exits before the driver processes the device's DBC deactivation message, the host loses track of which DBCs are free and never releases that DBC's resources. When the device later gives the same DBC to another user's network, that user's process waits indefinitely and hangs. On a shared inference node this is a denial of service against other users of the accelerator, and it lasts until the device is removed. The record describes a resource leak and a hang, not memory corruption, so read the generic 7.8 score with that in mind.

Who can reach it

Local user with access to the qaic accelerator device node. A process only has to exit or be killed before the deactivation is handled. Affects only hosts that have Cloud AI 100 cards and load the qaic driver.

What to do

Install a kernel that includes the upstream stable fix and reboot each affected node. Until then, a hung DBC is only freed by removing the device (driver unbind or reload, or a reboot).

References

https://nvd.nist.gov/vuln/detail/CVE-2026-43007https://git.kernel.org/stable/c/08021f2d4a557d6491e3bcc288e96425f50aa3cfhttps://git.kernel.org/stable/c/2dd67966f39a2abf8ccb4865031c722e40e01b7fhttps://git.kernel.org/stable/c/ee0180e77e6c8482644569632065411de844c515

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.