Re: hrtimer deadlock caused by nohz_full

[Frederic Weisbecker]

On Thu, Sep 25, 2014 at 04:35:49PM +0200, Thomas Gleixner wrote:
> On Thu, 25 Sep 2014, Dave Jones wrote:
> 
> > Got this on a box that had been fuzzing for 12 hours or so.
> > There's also some timer stuff going on htere, so cc'ing the usual suspects.
> 
> And it's a hrtimer lockup
>  
> >  [<ffffffffa10f95f8>] ? hrtimer_try_to_cancel+0x58/0x1f0
> >  [<ffffffffa10d103d>] ? lock_release+0x1d/0x300
> >  [<ffffffffa10d103d>] ? lock_release+0x1d/0x300
> >  [<ffffffffa10f95f8>] ? hrtimer_try_to_cancel+0x58/0x1f0
> >  [<ffffffffa10d103d>] ? lock_release+0x1d/0x300
> >  <<EOE>>  <IRQ>  [<ffffffffa1823ac4>] _raw_spin_unlock_irqrestore+0x24/0x70
> >  [<ffffffffa10f95f8>] hrtimer_try_to_cancel+0x58/0x1f0
> >  [<ffffffffa10f97aa>] hrtimer_cancel+0x1a/0x30
> >  [<ffffffffa110a0e7>] tick_nohz_restart+0x17/0x90
> >  [<ffffffffa110af38>] __tick_nohz_full_check+0xc8/0xe0
> >  [<ffffffffa110af5e>] nohz_full_kick_work_func+0xe/0x10
> >  [<ffffffffa117c9bf>] irq_work_run_list+0x4f/0x70
> >  [<ffffffffa117ca0a>] irq_work_run+0x2a/0x60
> >  [<ffffffffa10f82eb>] update_process_times+0x5b/0x70
> >  [<ffffffffa1109dc5>] tick_sched_handle.isra.21+0x25/0x60
> >  [<ffffffffa110a0b1>] tick_sched_timer+0x41/0x60
> >  [<ffffffffa10f8c71>] __run_hrtimer+0x81/0x480
> >  [<ffffffffa110a070>] ? tick_sched_do_timer+0x90/0x90
> >  [<ffffffffa10f9b27>] hrtimer_interrupt+0x107/0x260
> >  [<ffffffffa10331a4>] local_apic_timer_interrupt+0x34/0x60
> >  [<ffffffffa182734f>] smp_apic_timer_interrupt+0x3f/0x60
> >  [<ffffffffa182576f>] apic_timer_interrupt+0x6f/0x80
> 
> hrtimer_interrupt
>   tick_sched_timer
>     tick_sched_handle
>       update_process_times
>         irq_work_run
> 	  irq_work_run_list
> 	    nohz_full_kick_work_func
> 	      __tick_nohz_full_check
> 	        tick_nohz_restart
>                   hrtimer_cancel
> 
> And that hrtimer_cancel is:
> 
> static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
> {
> 	hrtimer_cancel(&ts->sched_timer);
> 
> Now, that's really bad because we are in the timer callback of
> ts->sched_timer. So hrtimer_cancel will loop forever waiting for the
> callback to complete.
> 
> Frederic !?!?

Right, this patchset fixes it: "[PATCH 0/8] nohz: Fix nohz kick irq work on tick v3"

I was about to make the pull request, the branch is acked by peterz.
Would you like to pull it? It's all merge window material.

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
	nohz/fixes-v3

Thanks.
--
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/