Re: [PATCH] watchdog: fix for lockup detector breakage on resume

[Sameer Nanda]

On Mon, Apr 30, 2012 at 2:10 PM, Sameer Nanda <snanda@xxxxxxxxxxxx> wrote:
> On Sun, Apr 29, 2012 at 11:12 PM, Srivatsa S. Bhat
> <srivatsa.bhat@xxxxxxxxxxxxxxxxxx> wrote:
>> On 04/27/2012 11:40 PM, Sameer Nanda wrote:
>>
>>> On the suspend/resume path the boot CPU does not go though an
>>> offline->online transition. ÂThis breaks the NMI detector
>>> post-resume since it depends on PMU state that is lost when
>>> the system gets suspended.
>>>
>>> Fix this by forcing a CPU offline->online transition for the
>>> lockup detector on the boot CPU during resume.
>>>
>>> Signed-off-by: Sameer Nanda <snanda@xxxxxxxxxxxx>
>>> ---
>>> To provide more context, we enable NMI watchdog on
>>> Chrome OS. ÂWe have seen several reports of systems freezing
>>> up completely which indicated that the NMI watchdog was not
>>> firing for some reason.
>>>
>>> Debugging further, we found a simple way of repro'ing system
>>> freezes -- issuing the command 'tasket 1 sh -c "echo nmilockup > /proc/breakme"'
>>> after the system has been suspended/resumed one or more times.
>>>
>>> With this patch in place, the system freeze result in panics,
>>> as expected. ÂThese panics provide a nice stack trace for us
>>> to debug the actual issue causing the freeze.
>>>
>>>
>>> Âinclude/linux/sched.h Â| Â Â4 ++++
>>> Âkernel/power/suspend.c | Â Â3 +++
>>> Âkernel/watchdog.c   Â|  16 ++++++++++++++++
>>> Â3 files changed, 23 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>>> index 81a173c..118cc38 100644
>>> --- a/include/linux/sched.h
>>> +++ b/include/linux/sched.h
>>> @@ -317,6 +317,7 @@ extern int proc_dowatchdog_thresh(struct ctl_table *table, int write,
>>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â size_t *lenp, loff_t *ppos);
>>> Âextern unsigned int Âsoftlockup_panic;
>>> Âvoid lockup_detector_init(void);
>>> +void lockup_detector_bootcpu_resume(void);
>>> Â#else
>>> Âstatic inline void touch_softlockup_watchdog(void)
>>> Â{
>>> @@ -330,6 +331,9 @@ static inline void touch_all_softlockup_watchdogs(void)
>>> Âstatic inline void lockup_detector_init(void)
>>> Â{
>>> Â}
>>> +static inline void lockup_detector_bootcpu_resume(void)
>>> +{
>>> +}
>>> Â#endif
>>>
>>> Â#ifdef CONFIG_DETECT_HUNG_TASK
>>> diff --git a/kernel/power/suspend.c b/kernel/power/suspend.c
>>> index 396d262..0d262a8 100644
>>> --- a/kernel/power/suspend.c
>>> +++ b/kernel/power/suspend.c
>>> @@ -177,6 +177,9 @@ static int suspend_enter(suspend_state_t state, bool *wakeup)
>>> Â Â Â arch_suspend_enable_irqs();
>>> Â Â Â BUG_ON(irqs_disabled());
>>>
>>> + Â Â /* Kick the lockup detector */
>>> + Â Â lockup_detector_bootcpu_resume();
>>> +
>>> Â Enable_cpus:
>>> Â Â Â enable_nonboot_cpus();
>>>
>>> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
>>> index df30ee0..dd2ac93 100644
>>> --- a/kernel/watchdog.c
>>> +++ b/kernel/watchdog.c
>>> @@ -585,6 +585,22 @@ static struct notifier_block __cpuinitdata cpu_nfb = {
>>> Â Â Â .notifier_call = cpu_callback
>>> Â};
>>>
>>> +void lockup_detector_bootcpu_resume(void)
>>> +{
>>> + Â Â void *cpu = (void *)(long)smp_processor_id();
>>> +
>>> + Â Â /*
>>> + Â Â Â* On the suspend/resume path the boot CPU does not go though the
>>> + Â Â Â* offline->online transition. This breaks the NMI detector post
>>> + Â Â Â* resume. Force an offline->online transition for the boot CPU on
>>> + Â Â Â* resume.
>>> + Â Â Â*/
>>> + Â Â cpu_callback(&cpu_nfb, CPU_DEAD, cpu);
>>> + Â Â cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
>>> +
>>
>>
>> I have a couple of comments about this:
>>
>> 1. Strictly speaking, we should be using the _FROZEN variants here (since the
>> tasks are still frozen).
>>
>> Like, cpu_callback(&cpu_nfb, CPU_DEAD_FROZEN, cpu);
>> and  cpu_callback(&cpu_nfb, CPU_ONLINE_FROZEN, cpu);
>>
>> Right now, since the same action is taken for either variant (ie., with or without
>> _FROZEN), it really doesn't matter. But still, good to be on the safer side no?
>
> Agreed that the _FROZEN counterparts are a better fit here since the
> tasks are still frozen. ÂLet me make this change.
>
>>
>> 2. Why are we skipping the CPU_UP_PREPARE_FROZEN callback?
>
> Mainly because the hrtimer_init has already been done at kernel init
> time. ÂBut, this seems to be a good idea since the non-boot CPUs do
> transition through the CPU_UP_PREPARE_FROZEN phase on the way up
> during resume so it makes sense to keep the boot CPU path symmetrical.
>
> Let me make this change also.

Just sent the updated patch incorporating these two changes as well as
the earlier feedback from akpm.

>
>>
>> 3. How about hibernation? We don't hit this problem there?
>
> I am not too familiar with hibernation path and don't have a setup to
> test it either so can't really answer this one.
>
>>
>>> + Â Â return;
>>> +}
>>> +
>>> Âvoid __init lockup_detector_init(void)
>>> Â{
>>> Â Â Â void *cpu = (void *)(long)smp_processor_id();
>>
>>
>>
>> Regards,
>> Srivatsa S. Bhat
>>
>
>
>
> --
> Sameer



-- 
Sameer
--
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/