Imagine you're running your kernel module or driver on a Linux kernel that's been configured to be preemptible (that is, CONFIG_PREEMPT is on; we covered this topic in the companion guide Linux Kernel Programming, Chapter 10, The CPU Scheduler – Part 1). Consider that a process, P1, is running the driver's read method code in the process context, working on the global array. Now, while it's within the critical section (between time t2 and t3), what if the kernel preempts process P1 and context switches to another process, P2, which is just waiting to execute this very code path? It's dangerous, and again, a data race. This could well happen on even a UP system!
Another scenario that's somewhat similar (and again, could occur on either a single core (UP) or multicore system): process P1 is running through the critical section of the driver method (between...
