[dm-devel] 3.15-rc4: circular locking dependency triggered by dm-multipath

[Mike Snitzer]

On Mon, May 26 2014 at 10:54am -0400,
Bart Van Assche <bvanassche at acm.org> wrote:

> On 05/26/14 14:51, Hannes Reinecke wrote:
> > Or, better still, try the attached patch.
> > There's one more instance where lockdep might complain.
> 
> Does anyone know whether it's safe to revert commit 4cdd2ad780 ?

Yes, it should be reverted, otherwise multipath_ioctl is using
inconsistent lock order.  I should've waited for test confirmation
before sending the previous "fix" upstream.  It made the lock order
consistent (but consistently wrong).

> Anyway,
> I ran a test with v3.15-rc7 + your patch + 4cdd2ad780 reverted and that
> test ran fine. I didn't encounter any lockups nor any lockdep
> complaints. The test I ran was to run a fio I/O verification test on top
> of a dm-multipath device. That multipath device was created on top of
> several SCSI device nodes created by the ib_srp driver. About once every
> 30 seconds a cable removal / reinsertion was simulated via the
> ibportstate command.

Great, thanks.  I'll try to get this queued for Linus this week.