Re: [PATCH v2] vmpressure: make sure memcg stays alive until all users are signaled

[Li Zefan <lizefan-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>]

>> especially while shutting down objects as they enter a lingering stage
>> where they're de-registered but not destroyed and you should be
>> careful which parts of the object are still accessible.  I haven't
>> read it carefully but here I'm not sure whether it's safe to do event
>> related operations after removal.  From cgroup core side, event list
>> is shut down synchronously from cgroup_destroy_locked().  It doesn't
>> seem like that part is explicitly built to remain accessible
>> afterwards.
> 
> /me goes and checks the code
> 
> vmpressure_event sends signals to _registered_ events but those are
> unregistered from the work queue context by cgroup_event_remove (via
> vmpressure_unregister_event) queued from cgroup_destroy_locked.
> 
> I am not sure what are the guarantees for ordering on the workqueue but
> this all suggests that either vmpressure_event sees an empty vmpr->events
> or it can safely send signals as cgroup_event_remove is pending on the
> queue.
> 
> cgroup_event_remove drops a reference to cgrp->dentry after everything
> is unregistered and event->wait removed from the wait queue so
> cgroup_free_fn couldn't have been called yet and so memcg is still
> alive. This means that even css_get/put is not necessary.
> 
> So I guess we are safe with the code as is but this all is really
> _tricky_ and deserves a fat comment. So rather than adding flushing work
> item code we should document it properly.
> 
> Or am I missing something?
> 

But if I read the code correctly, even no one registers a vmpressure event,
vmpressure() is always running and queue the work item.