Interrupt

An interrupt is how hardware devices and software tell the CPU that something has happened: a key was pressed, a network packet arrived, a disk finished reading, or a timer ran out. Without interrupts, the CPU would have to keep checking every device to see whether it needed anything, which wastes time and power.

When an interrupt arrives, the CPU finishes its current instruction, saves the state of what it was doing, switches to kernel mode, and looks up the matching handler in an interrupt table set up by the kernel. The handler, called an interrupt service routine, deals with the event, and then the CPU restores the saved state and resumes where it left off. Handlers must be very quick, so kernels usually split the work: a short first part acknowledges the device, and the rest is deferred to run a little later. Interrupts come from devices (hardware interrupts, or IRQs), from a timer that lets the scheduler preempt running tasks, and from the CPU itself when an instruction causes an exception such as a page fault or division by zero.

An interrupt works like a doorbell. You don't walk to the front door every minute to check for visitors; you keep cooking until the bell rings, answer it, and then return to the recipe. Interrupts are what make keyboards and mice feel instant, let the operating system share the CPU fairly, and let idle processors sleep until real work arrives.

Interrupts are often contrasted with polling, where software repeatedly checks a device instead of waiting to be notified; very fast network cards sometimes switch to polling under heavy load because handling millions of interrupts costs more than checking. An interrupt is also different from a Unix signal: interrupts are delivered to the CPU and handled by the kernel, while signals are notifications the kernel delivers to a user process. Exceptions in programming languages, handled with try and catch, are a separate, language-level mechanism.