diff mbox

NOHZ: WARNING: at arch/x86/kernel/smp.c:123 native_smp_send_reschedule, round 2

Message ID 5199CB59.1020309@linux.vnet.ibm.com (mailing list archive)
State Not Applicable, archived
Headers show

Commit Message

Michael Wang May 20, 2013, 7:06 a.m. UTC 
On 05/20/2013 02:58 PM, Michael Wang wrote:
> On 05/20/2013 02:47 PM, Borislav Petkov wrote:
>> On Mon, May 20, 2013 at 02:23:37PM +0800, Michael Wang wrote:
>>> On 05/20/2013 12:50 PM, Borislav Petkov wrote:
>>>> On Mon, May 20, 2013 at 11:16:33AM +0800, Michael Wang wrote:
>>>>> I suppose the reason is that the cpu we passed to
>>>>> mod_delayed_work_on() has a chance to become offline before we
>>>>> disabled irq, what about check it before send resched ipi? like:
>>>>
>>>> I think this is only addressing the symptoms - what we should be doing
>>>> instead is asking ourselves why are we even scheduling work on a cpu if
>>>> the machine goes offline?
>>>>
>>>> I don't know though who should be responsible for killing all that
>>>> work - the workqueue itself or the guy who created it, i.e. cpufreq
>>>> governor...
>>>
>>> So there are two questions here:
>>> 1. Is gov_queue_work() want to queue the work on offline cpu?
>>> 2. Is mod_delayed_work_on() allow offline cpu?
>>>
>>> I guess both should be false?
>>
>> Well, if we don't allow queueing work on a cpu which goes offline, i.e.
>> #2, the problem should be solved.
> 
> I've take a look at the usage of queue_delayed_work_on() and
> mod_delayed_work_on(), mostly passed this_cpu, or those in online mask,
> I think offline cpu is not by designed.
> 
> Besides, the cpu gov_queue_work() is using 'policy->cpus' which seems to
> be updated during UP DOWN notify, I think they are supposed to be online.
> 
> But we need expert in cpufreq to confirm all these...

And I guess this may help to reduce the chance to trigger WARN:

 }

Well, disable irq will be better, anyway...still need folks who own that
driver to make the decision, so let's CC them :)

Regards,
Michael Wang


> 
> Regards,
> Michael Wang
> 
>>
>> Tejun?
>>
>> Here are the splats: http://marc.info/?l=linux-kernel&m=136879901425951
>>
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Viresh Kumar May 20, 2013, 7:12 a.m. UTC | #1 
Hi Michael,

I haven't followed this mail chain earlier and saw this mail only as I am
added in cc now. I probably have answers to few questions here:

On 20 May 2013 12:36, Michael Wang <wangyun@linux.vnet.ibm.com> wrote:
> On 05/20/2013 02:58 PM, Michael Wang wrote:
>> On 05/20/2013 02:47 PM, Borislav Petkov wrote:
>>> On Mon, May 20, 2013 at 02:23:37PM +0800, Michael Wang wrote:
>>>> On 05/20/2013 12:50 PM, Borislav Petkov wrote:
>>>> So there are two questions here:
>>>> 1. Is gov_queue_work() want to queue the work on offline cpu?

No. We are only working with online cpus now in cpufreq core and governors.

>> Besides, the cpu gov_queue_work() is using 'policy->cpus' which seems to
>> be updated during UP DOWN notify, I think they are supposed to be online.
>>
>> But we need expert in cpufreq to confirm all these...

I confirm this. policy->cpus only contains online cpus.. and
policy->related_cpus
always contain online+offline cpus.

> And I guess this may help to reduce the chance to trigger WARN:
>
> diff --git a/drivers/cpufreq/cpufreq_governor.c
> b/drivers/cpufreq/cpufreq_governor.c
> index 443442d..0f96013 100644
> --- a/drivers/cpufreq/cpufreq_governor.c
> +++ b/drivers/cpufreq/cpufreq_governor.c
> @@ -180,7 +180,7 @@ void gov_queue_work(struct dbs_data *dbs_data,
> struct cpufreq_policy *policy,
>         if (!all_cpus) {
>                 __gov_queue_work(smp_processor_id(), dbs_data, delay);
>         } else {
> -               for_each_cpu(i, policy->cpus)
> +               for_each_cpu_and(i, policy->cpus, cpu_online_mask)
>                         __gov_queue_work(i, dbs_data, delay);
>         }
>  }

Not required at all... policy->cpus is guaranteed to have only online cpus.
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Michael Wang May 20, 2013, 7:25 a.m. UTC | #2 
Hi, Viresh
On 05/20/2013 03:12 PM, Viresh Kumar wrote:
> Hi Michael,
> 
> I haven't followed this mail chain earlier and saw this mail only as I am
> added in cc now. I probably have answers to few questions here:

Thanks for your quick respond :)

> 
> On 20 May 2013 12:36, Michael Wang <wangyun@linux.vnet.ibm.com> wrote:
>> On 05/20/2013 02:58 PM, Michael Wang wrote:
>>> On 05/20/2013 02:47 PM, Borislav Petkov wrote:
>>>> On Mon, May 20, 2013 at 02:23:37PM +0800, Michael Wang wrote:
>>>>> On 05/20/2013 12:50 PM, Borislav Petkov wrote:
>>>>> So there are two questions here:
>>>>> 1. Is gov_queue_work() want to queue the work on offline cpu?
> 
> No. We are only working with online cpus now in cpufreq core and governors.
> 
>>> Besides, the cpu gov_queue_work() is using 'policy->cpus' which seems to
>>> be updated during UP DOWN notify, I think they are supposed to be online.
>>>
>>> But we need expert in cpufreq to confirm all these...
> 
> I confirm this. policy->cpus only contains online cpus.. and
> policy->related_cpus
> always contain online+offline cpus.

Nice to be confirmed :)

> 
>> And I guess this may help to reduce the chance to trigger WARN:
>>
>> diff --git a/drivers/cpufreq/cpufreq_governor.c
>> b/drivers/cpufreq/cpufreq_governor.c
>> index 443442d..0f96013 100644
>> --- a/drivers/cpufreq/cpufreq_governor.c
>> +++ b/drivers/cpufreq/cpufreq_governor.c
>> @@ -180,7 +180,7 @@ void gov_queue_work(struct dbs_data *dbs_data,
>> struct cpufreq_policy *policy,
>>         if (!all_cpus) {
>>                 __gov_queue_work(smp_processor_id(), dbs_data, delay);
>>         } else {
>> -               for_each_cpu(i, policy->cpus)
>> +               for_each_cpu_and(i, policy->cpus, cpu_online_mask)
>>                         __gov_queue_work(i, dbs_data, delay);
>>         }
>>  }
> 
> Not required at all... policy->cpus is guaranteed to have only online cpus.

Yeah, that's right, I guess the issue is, although the policy->cpus is
correct at a given time, after get cpu from it, it's possible to be
changed, unless we disabled preempt or irq, or hotplug before we use it...

Like such issue cases:
				get x from policy->cpus
	DOWN notifier
	change policy->cpus
	do offline x
				send ipi to x

Will that happen?

Regards,
Michael Wang


> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/cpufreq/cpufreq_governor.c
b/drivers/cpufreq/cpufreq_governor.c
index 443442d..0f96013 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -180,7 +180,7 @@  void gov_queue_work(struct dbs_data *dbs_data,
struct cpufreq_policy *policy,
        if (!all_cpus) {
                __gov_queue_work(smp_processor_id(), dbs_data, delay);
        } else {
-               for_each_cpu(i, policy->cpus)
+               for_each_cpu_and(i, policy->cpus, cpu_online_mask)
                        __gov_queue_work(i, dbs_data, delay);
        }