A quick summary of the workflow

When using the kernel-global workqueue, a simple pattern (workflow) emerges:

1. Initialize the work task.
2. At the appropriate point in time, schedule it to execute (perhaps with a delay and/or on a particular CPU core).
3. Clean up. Typically, in the kernel module (or driver's) cleanup code path, cancel it. (Preferably, do this with synchronization so that any pending work tasks are completed first. Here, we will stick to employing the recommended cancel*work*() routines, avoiding the flush_*() ones).

Let's summarize this using a table:

Using the kernel-global workqueue	Regular work task	Delayed work task	Execute work task on given CPU
1. Initialization	INIT_WORK()	INIT_DELAYED_WORK()	< either immediate or delayed's fine >
2. Schedule work task to execute	schedule_work()	schedule_delayed_work()	schedule_delayed_work_on()
3. Cancel (or flush) it; foo_sync() to ensure it's complete	cancel_work_sync()	cancel_delayed_work_sync...