Re: BUG: credit=sched2 machine hang when using DRAKVUF

[Michał Leszczyński <michal.leszczynski@xxxxxxx>]

----- 23 paź, 2020 o 6:47, Jürgen Groß jgross@xxxxxxxx napisał(a):

> On 23.10.20 00:59, Michał Leszczyński wrote:
>> Hello,
>> 
>> when using DRAKVUF against a Windows 7 x64 DomU, the whole machine hangs 
>> after a
>> few minutes.
>> 
>> The chance for a hang seems to be correlated with number of PCPUs, in this 
>> case
>> we have 14 PCPUs and hang is very easily reproducible, while on other 
>> machines
>> with 2-4 PCPUs it's very rare (but still occurring sometimes). The issue is
>> observed with the default sched=credit2 and is no longer reproducible once
>> sched=credit is set.
> 
> Interesting. Can you please share some more information?
> 
> Which Xen version are you using?

RELEASE-4.14

> 
> Is there any additional information in the dom0 log which could be
> related to the hang (earlier WARN() splats, Oopses, Xen related
> messages, hardware failure messages, ...?

I will try to find something out next week and will come back to you.

> 
> Can you please try to get backtraces of all cpus at the time of the
> hang?
> 
> It would help to know which cpu was the target of the call of
> smp_call_function_single(), so a disassembly of that function would
> be needed to find that information from the dumped registers.
> 
> I'm asking because I've seen a similar problem recently and I was
> rather suspecting a fifo event channel issue than the Xen scheduler,
> but your data suggests it could be the scheduler after all (if it is
> the same issue, of course).
> 
> 
> Juergen