Re: [PATCH] sched: reset sysctl_sched_time_avg to default when

From: Ethan Zhao
Date: Sat Sep 02 2017 - 02:55:31 EST


Peter,

V2 sent, tested okay on stable 4.1, re-factor a little, could apply
to v4.13-rc5,
please review.

[root@ ~]# sysctl kernel.sched_time_avg_ms
kernel.sched_time_avg_ms = 1000
[root@ ~]# sysctl kernel.sched_time_avg_ms=0
sysctl: setting key "kernel.sched_time_avg_ms": Invalid argument
kernel.sched_time_avg_ms = 0
[root@ ~]# sysctl kernel.sched_time_avg_ms
kernel.sched_time_avg_ms = 1000
[root@ ~]# sysctl kernel.sched_time_avg_ms=900
kernel.sched_time_avg_ms = 900
[root@ ~]# sysctl kernel.sched_time_avg_ms
kernel.sched_time_avg_ms = 900

Thanks,
Ethan

On Sat, Sep 2, 2017 at 8:33 AM, Ethan Zhao <ethan.kernel@xxxxxxxxx> wrote:
> Yep, that is the first place I considered to set the limit, but that would
> break KABI ?
>
> Thanks,
> Ethan
>
> On Fri, Sep 1, 2017 at 8:32 PM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>> On Fri, Sep 01, 2017 at 07:31:54PM +0800, Ethan Zhao wrote:
>>> System will hang if user set sysctl_sched_time_avg to 0 by
>>>
>>> [root@XXX ~]# sysctl kernel.sched_time_avg_ms=0
>>>
>>> Stack traceback for pid 0
>>> 0xffff883f6406c600 0 0 1 3 R 0xffff883f6406cf50 *swapper/3
>>> ffff883f7ccc3ae8 0000000000000018 ffffffff810c4dd0 0000000000000000
>>> 0000000000017800 ffff883f7ccc3d78 0000000000000003 ffff883f7ccc3bf8
>>> ffffffff810c4fc9 ffff883f7ccc3c08 00000000810c5043 ffff883f7ccc3c08
>>> Call Trace:
>>> <IRQ> [<ffffffff810c4dd0>] ? update_group_capacity+0x110/0x200
>>> [<ffffffff810c4fc9>] ? update_sd_lb_stats+0x109/0x600
>>> [<ffffffff810c5507>] ? find_busiest_group+0x47/0x530
>>> [<ffffffff810c5b84>] ? load_balance+0x194/0x900
>>> [<ffffffff810ad5ca>] ? update_rq_clock.part.83+0x1a/0xe0
>>> [<ffffffff810c6d42>] ? rebalance_domains+0x152/0x290
>>> [<ffffffff810c6f5c>] ? run_rebalance_domains+0xdc/0x1d0
>>> [<ffffffff8108a75b>] ? __do_softirq+0xfb/0x320
>>> [<ffffffff8108ac85>] ? irq_exit+0x125/0x130
>>> [<ffffffff810b3a17>] ? scheduler_ipi+0x97/0x160
>>> [<ffffffff81052709>] ? smp_reschedule_interrupt+0x29/0x30
>>> [<ffffffff8173a1be>] ? reschedule_interrupt+0x6e/0x80
>>> <EOI> [<ffffffff815bc83c>] ? cpuidle_enter_state+0xcc/0x230
>>> [<ffffffff815bc80c>] ? cpuidle_enter_state+0x9c/0x230
>>> [<ffffffff815bc9d7>] ? cpuidle_enter+0x17/0x20
>>> [<ffffffff810cd6dc>] ? cpu_startup_entry+0x38c/0x420
>>> [<ffffffff81053373>] ? start_secondary+0x173/0x1e0
>>>
>>> Because divide-by-zero error happens in function
>>>
>>> update_group_capacity()
>>> update_cpu_capacity()
>>> scale_rt_capacity()
>>> {
>>> ...
>>> total = sched_avg_period() + delta;
>>> used = div_u64(avg, total);
>>> ...
>>> }
>>>
>>> Seems this issue could be reproduced on all I tried stable 4.1 - last
>>> kernel.
>>>
>>> To fix this issue, reset sysctl_sched_time_avg to default value
>>> MSEC_PER_SEC if user input invalid value in function
>>>
>>> sched_avg_period()
>>>
>>> Reported-by: James Puthukattukaran <james.puthukattukaran@xxxxxxxxxx>
>>> Signed-off-by: Ethan Zhao <ethan.zhao@xxxxxxxxxx>
>>> ---
>>> Tested on stable 4.1, compiled on stable 4.13-rc5
>>> kernel/sched/sched.h | 3 +++
>>> 1 file changed, 3 insertions(+)
>>>
>>> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
>>> index eeef1a3..b398560 100644
>>> --- a/kernel/sched/sched.h
>>> +++ b/kernel/sched/sched.h
>>> @@ -1620,6 +1620,9 @@ static inline void rq_last_tick_reset(struct rq *rq)
>>>
>>> static inline u64 sched_avg_period(void)
>>> {
>>> + if (unlikely(sysctl_sched_time_avg <= 0))
>>> + sysctl_sched_time_avg = MSEC_PER_SEC;
>>> +
>>
>> Sight, no.. you can set limits in the sysctl table.