For each standard worker pool Linux allocates a workqueue-attrs object (via an initializer wrapped in a macro adding callbacks) with its CPU-mask vars, sets a `nice` attribute on those attrs & loads it into an array, & does so again for another array this time also setting an `ordered` flag.
To allocate a workqueue it (via wrappers callbacks, variadic args, & lock properties) it validates flags, calculates how much memory's needed in one of 2 ways, allocates it, allocates attributes, ...
3/3.2