Re: [PATCH] lockdep: Avoid triggering hardlockup from debug_show_all_locks()

[Tejun Heo]

Hello,

(cc'ing Steven, Sergey and Petr who are working on printk)

On Tue, Jan 23, 2018 at 02:03:57PM -0500, Rik van Riel wrote:
> On Mon, 2018-01-22 at 14:00 -0800, Tejun Heo wrote:
> > debug_show_all_locks() iterates all tasks and print held locks whole
> > holding tasklist_lock.  This can take a while on a slow console
> > device
> > and may end up triggering NMI hardlockup detector if someone else
> > ends
> > up waiting for tasklist_lock.
> > 
> > Touch the NMI watchdog while printing the held locks to avoid
> > spuriously triggering the hardlockup detector.
> > 
> > Signed-off-by: Tejun Heo <tj@xxxxxxxxxx>
> 
> On this patch:
> 
> Acked-by: Rik van Riel <riel@xxxxxxxxxxx>
> 
> 
> However, it seems like we run into things like
> this on a fairly regular (though not very frequent)
> basis. Would it make sense to go through the code
> and add sprinkle around a few more touch_nmi_watchdog()
> calls?
> 
> After all, there are maybe a few dozen places where
> we print out a lot of debugging information.

Yeah, it's ridiculous how often printk ends up escalating otherwise
recoverable situations into system crashes.  I don't know what the
right answer is.  For spurious NMI hardlockups, maybe auditing debug
paths and adding touch_nmi_watchdog() would be enough but that also is
a pretty leaky approach.

Thanks.

-- 
tejun