The amdgpu driver in the Linux kernel acquired the cp_ecc_error_irq interrupt during late initialization but released it during hardware finalization, creating a mismatch that allows the interrupt handler to be released twice in a SR‑IOV virtual function context. The double release can decrement the handler reference count incorrectly, trigger a kernel panic, and result in a denial of service. This weakness corresponds to an improper release of a reference‑counted resource (CWE‑416).

All builds of the Linux kernel that include the amdgpu graphics driver are affected, excluding kernel configurations that use GFX generation 8.0 where the interrupt is managed locally. The problem exists in any kernel before the commit that introduces a ras_suspend callback and removes the premature cp_ecc_error_irq put, regardless of distribution or patch level.

No CVSS score is publicly available, the EPSS score is not reported, and the vulnerability is not listed in the CISA KEV catalog. Exploitation requires kernel‑level privileges or the ability to unload the amdgpu driver or trigger driver unload during a suspend scenario. Without such privileges the likelihood is low, but if privilege escalation is already achieved, the path to a kernel crash becomes straightforward and highly impactful.