Re: bugsplat 3.6.7-rt18

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Dec 03, 2012 at 10:54:02AM +0100, Thomas Gleixner wrote:
> On Fri, 30 Nov 2012, Paul E. McKenney wrote:
> > On Fri, Nov 30, 2012 at 04:15:09PM -0800, Frank Rowand wrote:
> > > >  Possible unsafe locking scenario:
> > > > 
> > > >        CPU0
> > > >        ----
> > > >   lock(rcu_kthread_wq.lock.lock.wait_lock);
> > > >   <Interrupt>
> > > >     lock(rcu_kthread_wq.lock.lock.wait_lock);
> > 
> > ???
> > 
> > These are all spin_lock_irqsave() calls, so how would the interrupt
> > happen?  Or do these need to be changed to raw_spinlock_t?
> > 
> > Hmmm...  In include/linux/wait.h, doesn't struct __wait_queue_head
> > need to have raw_spinlock_t rather than spinlock_t?  Please try that
> > change and let me know what happens.
> 
> No. We can't do that. The reason is that some wait queue wakeups have
> callbacks which can't be called with interrupts disabled on RT.
> 
> IIRC, I restructured the RCU code in 3.0-rt so it won't be affected by
> this. I'll have a look.

I guess I am once again getting out of touch with -rt...  Apologies for
the noise.

							Thanx, Paul