The Linux kernel's iommufd component previously allocated vevent queue memory inside a spinlock, which is problematic because the queue depth is user-controlled. This could lead to excessive atomic memory consumption, risking denial of service. The fix involves moving the memory allocation outside the spinlock and using GFP_NOWAIT to fail fast under memory pressure without using atomic reserves or triggering direct reclaim in IRQ context. If allocation fails, the kernel queues a lost_events_header and returns -ENOMEM to notify user space of memory pressure. This behavior differs from queue overflow, which is a normal pacing condition. A future update will limit the maximum vevent queue depth.

The vulnerability can cause denial of service by allowing user space to exhaust atomic memory reserves through large vevent queue allocations. It does not affect confidentiality or integrity but impacts availability. There are no known exploits in the wild. The CVSS score is 5.5 (medium severity), reflecting local attack vector with low complexity and no user interaction required.

A fix is available that moves the vevent queue memory allocation outside the spinlock and uses GFP_NOWAIT to prevent atomic memory exhaustion. Users should apply the official Linux kernel updates that address CVE-2026-64292. Since this is a kernel vulnerability, updating to the patched kernel version is the recommended remediation. Patch status is not explicitly confirmed in the provided data; users should consult the vendor or distribution advisory for the exact fixed versions and update instructions.