quoted

 stack backtrace:
 Pid: 2308, comm: NetworkManager Not tainted 2.6.25-0.161.rc7.fc9.i686 #1
  [print_circular_bug_tail+91/102] print_circular_bug_tail+0x5b/0x66
  [print_circular_bug_entry+57/67] ? print_circular_bug_entry+0x39/0x43
  [__lock_acquire+2488/3089] __lock_acquire+0x9b8/0xc11
  [_spin_unlock_irq+34/47] ? _spin_unlock_irq+0x22/0x2f
  [lock_acquire+106/144] lock_acquire+0x6a/0x90
  [flush_workqueue+0/133] ? flush_workqueue+0x0/0x85
  [flush_workqueue+68/133] flush_workqueue+0x44/0x85
  [flush_workqueue+0/133] ? flush_workqueue+0x0/0x85
  [flush_scheduled_work+13/15] flush_scheduled_work+0xd/0xf
  [<d096d80a>] tulip_down+0x20/0x1a3 [tulip]
  [trace_hardirqs_on+233/266] ? trace_hardirqs_on+0xe9/0x10a
  [dev_deactivate+177/222] ? dev_deactivate+0xb1/0xde

Yes, see for example:

http://www.mail-archive.com/netdev@vger.kernel.org/msg31718.html

You can't flush a workqueue in the device close handler
exactly because of this locking conflict.

Nobody has come up with a suitable way to fix this yet.

Maybe we should check which schedule_work users actually lock the rtnl
within the work function and move them to a uses-rtnl-in-work workqueue
so that everybody else can have rtnl around flush.

Depending on how many users there are that might not be feasible, but I
have so far only seen linkwatch_event() lock the rtnl within the work
function and everybody else seems to want to use the rtnl around the
flushing.

johannes