Re: WARNING: at kernel/smp.c:385 smp_call_function_many

[Thomas Gleixner <tglx@linutronix.de>]

On Thu, 16 May 2013, Sasha Levin wrote:
> On 04/18/2013 08:29 PM, Sasha Levin wrote:
> > Hi Liu,
> > 
> > On 04/18/2013 08:10 PM, Liu, Chuansheng wrote:
> >>> It seems that with commit "smp: Give WARN()ing when calling
> >>>> smp_call_function_many()/single() in serving irq" I'm getting a bunch of these:
> >> This patch gives the warning even when the smp_call_function_many() is called in softirq context,
> >> because it will bring the possible deadlock cases, the example is below:
> >> CPUA                            CPUB
> >>                                 spin_lock(&spinlock)
> >>                                 Any irq coming, call the irq handler
> >>                                 irq_exit()
> >> spin_lock_irq(&spinlock)
> >> <== Blocking here due to
> >> CPUB hold it
> >>                                   __do_softirq()
> >>                                     run_timer_softirq()
> >>                                       timer_cb()
> >>                                         call smp_call_function_many()
> >>                                           send IPI interrupt to CPUA
> >>                                             wait_csd()
> >>
> >> Then both CPUA and CPUB will be deadlocked here.
> > 
> > Right.

Well, that's only true if LOCKDEP is enabled. The !LOCKDEP case
reenables interrupts if the lock is contended.

> > So commit f55a6faa3 ("hrtimer: Provide clock_was_set_delayed()") followed by one
> > of the commits that used clock_was_set_delayed() may trigger the deadlock you've
> > described. (cc John Stultz...)

I'll have a look. It's non trivial to fix :(

Thanks,

	tglx