On Tue, 2012-08-14 at 19:26 +0200, Mike Galbraith wrote:

On Tue, 2012-08-14 at 17:18 +0200, Thomas Gleixner wrote: 

quoted

On Tue, 14 Aug 2012, Mike Galbraith wrote:

quoted

On Tue, 2012-08-14 at 16:39 +0200, Mike Galbraith wrote: 

quoted

On Tue, 2012-08-14 at 10:23 -0400, Steven Rostedt wrote: 

quoted

On Tue, 2012-08-14 at 15:06 +0200, Mike Galbraith wrote:

quoted

Greetings,

Using openSUSE's "partitioner" gizmo to set up a reiserfs partition with
quotas and whatnot enabled rewarded me with a deadlock.

Is this just a -rt bug? Can't this deadlock also in mainline?

It didn't with a NOPREEMPT kernel the one time I tried it, which seemed
mighty odd when I stared at the rt deadlock.  I'll give it another go, I
likely just munged q/d NOPREEMPT test.

Nope, no screw-up.  Maybe it can deadlock, but it refused to do so.

Lockdep should tell you if it could ...

I'll give that a shot.  I would expect the thing to gripe.. but then I
expected the damn thing to deadlock properly with NOPREEMPT too, so.. :)

But would it?

We have kworker blocked on the mutex where the owner did a flush_work(),
and waiting for kworker to finish because of a wait_for_completion(). I
don't see annotation here that will help lockdep catch such a thing.

-- Steve