Hi,
On 17/08/18 11:27, Matt Fleming wrote:
> On Thu, 05 Jul, at 05:54:02PM, Valentin Schneider wrote:
>> On 05/07/18 14:27, Matt Fleming wrote:
>>> On Thu, 05 Jul, at 11:10:42AM, Valentin Schneider wrote:
>>>> Hi,
>>>>
>>>> On 04/07/18 15:24, Matt Fleming wrote:
>>>>> It's possible that the CPU doing nohz idle balance hasn't had its own
>>>>> load updated for many seconds. This can lead to huge deltas between
>>>>> rq->avg_stamp and rq->clock when rebalancing, and has been seen to
>>>>> cause the following crash:
>>>>>
>>>>> divide error: 0000 [#1] SMP
>>>>> Call Trace:
>>>>> [<ffffffff810bcba8>] update_sd_lb_stats+0xe8/0x560
>>
>> My confusion comes from not seeing where that crash happens. Would you mind
>> sharing the associated line number? I can feel the "how did I not see this"
>> from there but it can't be helped :(
>
> The divide by zero comes from scale_rt_capacity() where 'total' is a
> u64 but gets truncated when passed to div_u64() since the divisor
> parameter is u32.
>
Ah, nasty one. Interestingly enough that bit has been changed quite recently,
so I don't think you can get a div by 0 in there anymore - see
523e979d3164 ("sched/core: Use PELT for scale_rt_capacity()") and subsequent
cleanups.
> Sure, you could use div64_u64() instead, but the real issue is that
> the load hasn't been updated for a very long time and that we're
> trying to balance the domains with stale data.
>
Yeah I agree with that. However, the problem is with cpu_load - blocked load
on nohz CPUs will be periodically updated until entirely decayed. And if we
end up getting rid of cpu_load (depends on how [1] goes), then there's
nothing left to do. But we're not there yet...
[1]:
https://lore.kernel.org/lkml/20180809135753.21077-1-dietmar.eggem...@arm.com/
