Re: [Bonding-devel] [PATCH] bonding: cancel_delayed_work() -> cancel_delayed_work_sync()

[Jay Vosburgh <fubar@us.ibm.com>]

Johannes Berg <johannes@sipsolutions.net> wrote:

>On Thu, 2009-12-17 at 13:36 +0000, Jarek Poplawski wrote:
>> On Thu, Dec 17, 2009 at 02:19:46PM +0100, Johannes Berg wrote:
>> > Jarek,
>> > 
>> > Sorry to mail you directly, but I only saw your reply on gmane and
>> > didn't want to break up the threading etc.
>> > 
>> > cancel_delayed_work_sync() should be ok in this case unless the work
>> > items themselves used the rtnl,
>> 
>> Hmm, I'm not sure I get your point, but e.g. bond_mii_monitor() work
>> function can get rtnl_lock().
>
>Ok in that case you can't cancel_sync() it under rtnl. I was thinking of
>the case where it's just not ok because of other work. Sorry for the
>confusion!

	There's already logic in the monitors (bond_mii_monitor, et al)
to check a sentinel (kill_timers) and do nothing (not acquire rtnl) and
return.

	What exactly is the nature of the race that doing cancel..sync
is fixing?  The bond_close function sets kill_timers prior to calling
the cancel functions, so the monitor function might run once, but it
should do nothing.

	-J

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com