[PATCH] hardlockup: detect hard lockups without NMIs using secondary cpus

[Colin Cross]

On Thu, Jan 10, 2013 at 9:57 PM, Liu, Chuansheng
<chuansheng.liu at intel.com> wrote:
>
>
>> -----Original Message-----
>> From: ccross at google.com [mailto:ccross at google.com] On Behalf Of Colin
>> Cross
>> Sent: Friday, January 11, 2013 1:34 PM
>> To: Liu, Chuansheng
>> Cc: linux-kernel at vger.kernel.org; Andrew Morton; Don Zickus; Ingo Molnar;
>> Thomas Gleixner; linux-arm-kernel at lists.infradead.org
>> Subject: Re: [PATCH] hardlockup: detect hard lockups without NMIs using
>> secondary cpus
>>
>> On Thu, Jan 10, 2013 at 5:39 PM, Liu, Chuansheng
>> <chuansheng.liu at intel.com> wrote:
>> >
>> >
>> >> -----Original Message-----
>> >> From: Colin Cross [mailto:ccross at android.com]
>> >> Sent: Thursday, January 10, 2013 9:58 AM
>> >> To: linux-kernel at vger.kernel.org
>> >> Cc: Andrew Morton; Don Zickus; Ingo Molnar; Thomas Gleixner; Liu,
>> >> Chuansheng; linux-arm-kernel at lists.infradead.org; Colin Cross
>> >> Subject: [PATCH] hardlockup: detect hard lockups without NMIs using
>> >> secondary cpus
>> >>
>> >> Emulate NMIs on systems where they are not available by using timer
>> >> interrupts on other cpus.  Each cpu will use its softlockup hrtimer
>> >> to check that the next cpu is processing hrtimer interrupts by
>> >> verifying that a counter is increasing.
>> >>
>> >> This patch is useful on systems where the hardlockup detector is not
>> >> available due to a lack of NMIs, for example most ARM SoCs.
>> >> Without this patch any cpu stuck with interrupts disabled can
>> >> cause a hardware watchdog reset with no debugging information,
>> >> but with this patch the kernel can detect the lockup and panic,
>> >> which can result in useful debugging info.
>> >>
>> >> Signed-off-by: Colin Cross <ccross at android.com>
>> >> +static void watchdog_check_hardlockup_other_cpu(void)
>> >> +{
>> >> +     int cpu;
>> >> +     cpumask_t cpus = watchdog_cpus;
>> >> +
>> >> +     /*
>> >> +      * Test for hardlockups every 3 samples.  The sample period is
>> >> +      *  watchdog_thresh * 2 / 5, so 3 samples gets us back to slightly
>> over
>> >> +      *  watchdog_thresh (over by 20%).
>> >> +      */
>> >> +     if (__this_cpu_read(hrtimer_interrupts) % 3 != 0)
>> >> +             return;
>> >> +
> Another feeling is about __this_cpu_read(hrtimer_interrupts) % 3 != 0,
> It will cause the actual timeout value for hard lockup detection is not very fix, or even
> very short.
> Sometimes using 3 samples can detect the lockup case, but sometimes 1 sample.
> Is it the case?

I'm not sure what you mean.  The mod 3 will cause every 3rd timer (12
seconds, assuming watchdog_thresh = 10) to check hrtimer_interrupts
vs. hrtimer_interrupts_saved, and then update it.  The sampling should
be fixed and very accurate.  It will cause a panic/warning between 12
and 24 seconds after a cpu stops processing timer interrupts,
depending on the alignment of the hrtimers between the two cpus.

> And in NMI case, the NMI interrupt is coming at least every watchdog_thresh.

NMI interrupt will happen every 10 seconds instead of 12, meaning the
panic/warning will occur between 10 and 20 seconds after a cpu stops
processing timer interrupts, depending on the alignment of the NMI
with the hrtimer, but otherwise my patch should be very similar.