Re: Problem with commit 31e77c93e432 "sched/fair: Update blocked load when newly idle"

From: Vincent Guittot
Date: Thu Mar 29 2018 - 03:42:07 EST


On 28 March 2018 at 16:01, Vincent Guittot <vincent.guittot@xxxxxxxxxx> wrote:
> Hi,
>
> On 28 March 2018 at 12:37, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>> Hi,
>>
>> On 03/24/2018 01:47 PM, Heiner Kallweit wrote:
>>>
>>> Am 24.03.2018 um 07:46 schrieb Vincent Guittot:
>>>>
>>>> Hi Heiner,
>>>>
>>>> Le Friday 23 Mar 2018 Ã 22:28:09 (+0100), Heiner Kallweit a Ãcrit :
>>>>>
>>>>> Recently I started to get the following problems with linux-next:
>>>>>
>>>>> - When working via Putty/SSH on the system the console frequently
>>>>> freezes
>>>>> for few seconds. Sometimes only opening a second console makes the
>>>>> first one react again.
>>>>>
>>>>> - I get "INFO: rcu_sched detected stalls on CPUs/tasks:" warnings as
>>>>> described in [1].
>>>>>
>>
>> I can't catch this issue on my Juno r0 (arm64 big.Little).
>>
>> root@juno:~# uname -r
>> 4.16.0-rc4-00198-g31e77c93e432
>>
>> I'm using openssh-client and openssh-server though.
>
> I think that I have finally been able to reproduce it on my hikey
> (octo cortex-A53) after unplugging 6 cores and waiting for almost 2
> hours
> This seems to happen only on dual core system as I haven't faced that
> before on the hikey which I have used for my tests
>

I'm finally not so sure that i have the right set up to reproduce the
problem as I haven't been able to reproduce it since.

Heiner,

How fast the problem happens on your board ?
Are you doing anything specific on the console that trigger the problem ?

Regards,
Vincent

>>
>>>>> Bisecting the issue resulted in:
>>>>>
>>>>> 31e77c93e432dec79c7d90b888bbfc3652592741 is the first bad commit
>>>>> commit 31e77c93e432dec79c7d90b888bbfc3652592741
>>>>> Author: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
>>>>> Date: Wed Feb 14 16:26:46 2018 +0100
>>>>>
>>>>> sched/fair: Update blocked load when newly idle
>>>>>
>>>>> When NEWLY_IDLE load balance is not triggered, we might need to
>>>>> update the
>>>>> blocked load anyway. We can kick an ilb so an idle CPU will take
>>>>> care of
>>>>> updating blocked load or we can try to update them locally before
>>>>> entering
>>>>> idle. In the latter case, we reuse part of the nohz_idle_balance.
>>>>>
>>>>> After reversing this commit at least the issue with the freezing console
>>>>> is gone. The second one appeared only sporadically, I still have to see
>>>>> whether it pops up again.
>>
>>
>> [...]
>>