[ofa-general] Re: [PATCH][RFC] IB/uverbs: Export ib_umem_get()/ib_umem_release() to modules

[Roland Dreier]

 > Error handling looks a bit bogus here - we'll never give the task
 > it's rlimit back. Wouldn't it be a bit cleaner to allocate
 > the work object together with umem?

Yes, that's a good point.  Mainline already has this issue, so I'll
make that a separate patch before this one.

 > > +		schedule_work(&work->work);
 > 
 > We never flush the work queue.
 > I wonder whether there's a potential issue with module unloading.

Actually ib_cache_cleanup_one() has a flush_scheduled_work() so
there's no actual problem here ;) ... but I agree it's far from
obvious that the work always gets flushed.  I'll just move the
flush_scheduled_work() from the uverbs cleanup function (where it's no
longer needed) to the core cleanup function.

 - R.