Preemption
The kernel taking the CPU back from a process without asking — the opposite of waiting for it to yield voluntarily.
Early systems used cooperative multitasking, where each program was expected to relinquish control at intervals. A program stuck in an infinite loop froze the whole machine, as anyone who ran Windows 3.x or classic Mac OS remembers. Preemptive scheduling fixes this by having a timer interrupt regardless of what the process thinks.
The question applies to the kernel itself too. A kernel that cannot be interrupted mid-work gives worse latency, and Linux therefore has several modes — from server-oriented to fully preemptible with PREEMPT_RT for real-time use, trading throughput for predictable response times.