The Linux kernel s390 vfio_ap driver has three separate flaws that cause leaks of KVM GISC registrations, pinned guest pages containing the notification indicator byte, and orphaned IRQ resources when a queue is removed from a host's access‑port configuration while still attached to a mediated device. The defects stem from missing handling of the AP_RESPONSE_Q_NOT_AVAIL response, incomplete cleanup of KVM GISC registration, and improper clearing of the saved interrupt status field in vfio_ap_free_aqic_resources. These flaws can lead to memory exhaustion, unstable interrupt handling, and a potential double‑free during a later cleanup, all of which are symptomatic of a kernel resource leak.

This bug affects all Linux kernels running on the s390 architecture that incorporate the vfio_ap driver prior to the patches incorporated in commits 3c3433657c82f200dae36c29e2b80681c81b85be, 65e05ec252a9b79e75930d3c4dd42d8877db04c5, cf441a0db681dfad5e89b015d6be11365159d7a9, da69d3ac53490724eba0ade9deb4e539a6ae9034, and their successors. No specific version range is supplied, so any kernel prior to the application of these changes is considered vulnerable.

Exploitation requires locally privileged access that can trigger the removal of an AP queue from the host configuration while the queue remains bound to a mediated device, for example by manipulating the host's AP configuration or unloading a mediator. The likely attack vector is local privileged manipulation of the AP configuration. Based on the description, it is inferred that the attack is local in scope. The low CVSS score of 1.9 and an EPSS value below 1% indicate a low likelihood of exploitation and severity. The vulnerability is not currently listed in the CISA KEV catalog. A successful exploit would result in resource leaks that could potentially cause a local denial of service; the potential for denial of service is inferred from the described resource leaks, but the vulnerability does not provide a path to privilege escalation or remote code execution.