On Tue, May 27, 2014 at 05:00:26AM +0100, Al Viro wrote:
> On Tue, May 27, 2014 at 04:14:15AM +0100, Al Viro wrote:
> 
> > As the matter of fact, let's try this instead - retry the same sucker
> > immediately in case if trylocks fail.  Comments?
> 
> Better yet, let's take "move back to shrink list" into dentry_kill()
> itself.  Then we get consistent locking rules for dentry_kill() and
> instead of unlock_on_failure we simply pass it NULL or the shrink
> list to put the sucker back.  Mika, could you test this one and see
> if it fixes that livelock?  The difference in behaviour is that in
> case of trylock failure we hit that sucker again without letting
> it ride all the way around the list, same as we do for other dentry_kill()
> callers.

I tried this patch and unfortunately it still results the same sort of
livelock. I've attached the dmesg.

I also tried the serialization patch from Linus and it seemed to fix the
problem. After several rounds of USB memory stick plug/unplug I haven't
seen a single "soft lockup" warning in dmesg.

I'm able to reproduce the problem pretty easily, so if you have
something else to try I'm more than happy to give it a try.