The Linux kernel's drm/v3d driver exposes multiple independent hardware queues (BIN, RENDER, TFU, CSD) but only a single global reset. Timeout handlers for these queues previously ran concurrently on the system workqueue, which caused unsafe interactions during global resets. Specifically, if a reset occurs on one queue while another queue is handling its own timeout, the global reset stops and restarts the scheduler for all queues concurrently. This leads to inconsistent credit accounting in the CL queue's scheduler, causing more credits to be in flight than the limit permits, resulting in GPU hangs and UI freezes. The fix involves using a shared ordered workqueue to serialize timeout handlers, ensuring that resets do not interfere with each other.

The vulnerability can cause a complete GPU hang and user interface freeze due to inconsistent scheduler state and credit mismanagement in the drm/v3d driver. This impacts systems using the affected Linux kernel driver, such as Raspberry Pi 5 Model B devices running affected kernel versions. There is no indication of remote code execution or privilege escalation, but the denial of service caused by GPU hangs can disrupt normal system operation.

A fix is available that serializes the scheduler timeout handlers by using a shared ordered workqueue instead of the default system workqueue. This prevents concurrent resets from interfering with each other and eliminates the race condition causing GPU hangs. Users should update to a Linux kernel version that includes this fix. Patch status is not explicitly confirmed in the provided data; check the vendor advisory or Linux kernel changelogs for the exact fixed version and apply the update accordingly.