Re: [PATCH 4/4] futex: convert hash_bucket locks to raw_spinlock_t

[Darren Hart]

On 07/11/2010 06:33 AM, Mike Galbraith wrote:
> On Sat, 2010-07-10 at 21:41 +0200, Mike Galbraith wrote:
> > On Fri, 2010-07-09 at 15:33 -0700, Darren Hart wrote:
>
> > > If we can't move the unlock above before set_owner, then we may need a:
> > >
> > > retry:
> > > cur->lock()
> > > top_waiter = get_top_waiter()
> > > cur->unlock()
> > >
> > > double_lock(cur, topwaiter)
> > > if top_waiter != get_top_waiter()
> > > 	double_unlock(cur, topwaiter)
> > > 	goto retry
> > >
> > > Not ideal, but I think I prefer that to making all the hb locks raw.
>
> Another option: only scratch the itchy spot.
>
> futex: non-blocking synchronization point for futex_wait_requeue_pi() and futex_requeue().
>
> Problem analysis by Darren Hart;
> The requeue_pi mechanism introduced proxy locking of the rtmutex.  This creates
> a scenario where a task can wake-up, not knowing it has been enqueued on an
> rtmutex. In order to detect this, the task would have to be able to take either
> task->pi_blocked_on->lock->wait_lock and/or the hb->lock.  Unfortunately,
> without already holding one of these, the pi_blocked_on variable can change
> from NULL to valid or from valid to NULL. Therefor, the task cannot be allowed
> to take a sleeping lock after wakeup or it could end up trying to block on two
> locks, the second overwriting a valid pi_blocked_on value. This obviously
> breaks the pi mechanism.
>
> Rather than convert the bh-lock to a raw spinlock, do so only in the spot where
> blocking cannot be allowed, ie before we know that lock handoff has completed.

I like it. I especially like the change is only evident if you are using 
the code path that introduced the problem in the first place. If you're 
doing a lot of requeue_pi operations, then the waking waiters have an 
advantage over new pending waiters or other tasks with futex keyed on 
the same hash-bucket... but that seems acceptable to me.

I'd like to confirm that holding the pendowner->pi-lock across the 
wakeup in wakeup_next_waiter() isn't feasible first. If it can work, I 
think the impact would be lower. I'll have a look tomorrow.

Nice work Mike.

--
Darrem

> Signed-off-by: Mike Galbraith<efault@xxxxxx>
> Cc: Darren Hart<dvhltc@xxxxxxxxxx>
> Cc: Thomas Gleixner<tglx@xxxxxxxxxxxxx>
> Cc: Peter Zijlstra<peterz@xxxxxxxxxxxxx>
> Cc: Ingo Molnar<mingo@xxxxxxx>
> Cc: Eric Dumazet<eric.dumazet@xxxxxxxxx>
> Cc: John Kacur<jkacur@xxxxxxxxxx>
> Cc: Steven Rostedt<rostedt@xxxxxxxxxxx>
>
> diff --git a/kernel/futex.c b/kernel/futex.c
> index a6cec32..ef489f3 100644
> --- a/kernel/futex.c
> +++ b/kernel/futex.c
> @@ -2255,7 +2255,14 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
>   	/* Queue the futex_q, drop the hb lock, wait for wakeup. */
>   	futex_wait_queue_me(hb,&q, to);
>
> -	spin_lock(&hb->lock);
> +	/*
> +	 * Non-blocking synchronization point with futex_requeue().
> +	 *
> +	 * We dare not block here because this will alter PI state, possibly
> +	 * before our waker finishes modifying same in wakeup_next_waiter().
> +	 */
> +	while(!spin_trylock(&hb->lock))
> +		cpu_relax();
>   	ret = handle_early_requeue_pi_wakeup(hb,&q,&key2, to);
>   	spin_unlock(&hb->lock);
>   	if (ret)


--
Darren Hart
IBM Linux Technology Center
Real-Time Linux Team
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/