RT threads can be prempted by higher prio RT, and IIRC some kernel threads run at the highest prio. Plus you can be prempted by SMI, an hypervisor, etc