A race condition in the Linux kernel's AMDGPU driver triggers the switch_power_profile function during GPU reset when the reset semaphore has not been released, causing an assertion failure. The kernel panics, halting the system and effectively denying service until a reboot occurs. The underlying weakness is a concurrency flaw that can be exploited if an attacker can induce a GPU reset while an idle work handler is pending.

All Linux kernel builds that include the AMDGPU driver and do not contain the commit that protects against the race condition. The data does not list specific version ranges, so any kernel released before the patch is vulnerable and should be updated to the latest stable version that incorporates the fix.

The CVSS score is 5.5, and the EPSS score of less than 1% indicates a very low exploitation probability. KEV has not listed this vulnerability. The attack vector is inferred to be local or requires privileged access to trigger GPU resets, making it a low-to-moderate exploitation risk for publicly exposed systems. While no public exploit exists, the kernel crash can be leveraged by an attacker who gains local privilege to cause a denial of service.