Re: [PATCH RFC] rtmutex: Permit rt_mutex_unlock() to be invoked with irqs disabled

[Thomas Gleixner <tglx@linutronix.de>]

On Sun, 24 Jul 2011, Thomas Gleixner wrote:

> On Sat, 23 Jul 2011, Paul E. McKenney wrote:
> 
> > On Thu, Jul 21, 2011 at 01:32:48PM +0200, Peter Zijlstra wrote:
> > > On Tue, 2011-07-19 at 13:14 -0700, Paul E. McKenney wrote:
> > > > Because rcu_read_unlock() can be invoked with interrupts disabled, it can
> > > > in turn invoke rt_mutex_unlock() with interrupts disabled.  This situation
> > > > results in lockdep splats (https://lkml.org/lkml/2011/7/7/362) because the
> > > > rt_mutex structure's ->lock_wait is acquired elsewhere without disabling
> > > > interrupts, which can result in deadlocks.
> > > > 
> > > > This commit therefore changes the rt_mutex structure's ->lock_wait
> > > > acquisitions to disable interrupts.
> > > > 
> > > > An alternative fix is to prohibit invoking rcu_read_unlock() with
> > > > interrupts disabled unless the entire preceding RCU read-side critical
> > > > section has run with interrupts disabled.  However, there is already
> > > > at least one case in mainline where this potential rule is violated,
> > > > and there might well be many more.  These would likely be found one at
> > > > a time using the lockdep-water-torture method, hence the alternative
> > > > fix in the form of this commit.
> > > 
> > > Thomas, I'm inclined to merge this, any objections?
> > 
> > FWIW, it has been passing tests here.
> 
> If it's only the unlock path, I'm fine with that change.
> 
> Acked-by-me

Hrmpft. That's requiring all places to take the lock irq safe. Not
really amused. For -RT that's a hotpath and we can really do without
the irq fiddling there. That needs a bit more thought.