The job of the OS scheduler is to arbitrate access to the processor (CPU) resource, sharing it between competing entities (threads) that want to use it. But what if the system is busy, with many threads continually competing for and acquiring the processor? More correctly, what we're really getting at is: in order to ensure fair sharing of the CPU resource between tasks, you must ensure that the policeman in the picture, the scheduler itself, runs periodically on the processor. Sounds good, but how exactly can you ensure that?
Here's a (seemingly) logical way to go about it: invoke the scheduler when the timer interrupt fires; that is, it gets a chance to run CONFIG_HZ times a second (which is often set to the value 250)! Hang on, though, we learned a golden rule in Chapter 8, Kernel Memory Allocation for Module Authors – Part 1, in the Never sleep in interrupt or atomic contexts section: you cannot invoke the scheduler...
