Re: workqueue - process context

[Roland Dreier <roland@topspin.com>]

    Vicente> I've been playing with workqueues, and I've found that
    Vicente> once I unload the module, if I don't call
    Vicente> destroy_workqueue(); then the workqueue I've created
    Vicente> stays in the process list, [my_wq], I don't know if
    Vicente> that's meant to be, or is it a bug, cause I believe there
    Vicente> can be two options in here:

    Vicente> 1) It's meant to be so you can unload your module and let
    Vicente> the works run some time after you're already gone, that
    Vicente> allows you to probe other modules or do whatever necesary
    Vicente> without the need to wait for the workqueue to be emtpy.

    Vicente> 2) It's a bug, cause the module allows to be unloaded,
    Vicente> destroying the structs but not removing the workqueue
    Vicente> from the process context.

Not destroying its workqueue is a bug in the module just like any
other resource leak.  It's analogous to a module allocating some
memory with kmalloc() and not calling kfree() when it's unloaded.  If
a module creates a workqueue, then it should call destroy_workqueue()
when it's unloaded.

By the way, the module (or any code calling destroy_workqueue()) must
make sure that it has race conditions that might result in work being
submitted to the queue while it is being destroyed.

 -R .