Re: RT kernel testing with stress-ng and scheduling while in atomic bugs.

[Punit Agrawal <punit1.agrawal@toshiba.co.jp>]

Hi Colin,

Colin Ian King <colin.king@canonical.com> writes:

> Hi,
>
> I've recently been exercising the 5.10-rc6 kernel with the latest RT
> patches from linux-rt-devel (v5.10-rc6-rt13) using stress-ng V0.12.00
> [1][2] to see rugged the RT kernel is and I'm getting quite a few
> "scheduling while atomic:" bugs in various stress-tests.
>
> While I realize this is a rc6 kernel, I was surprised to be able to trip
> these issues so easily [3]
>
> Questions:
>
> 1. Are these issues expected?
> 2. Is there an official way to report my bug findings?
> 3. I am keen to debug and fix these issues, have RT folk got some advice
> on how to start debugging these kind of issues?
>
> On an 8 CPU SMP test box stress-ng is hitting these issues on the
> following stress tests:
>
> daemon enosys exec fork fstat madvise prctl pthread session sigpipe vfork
>
> ..it seems fork/clone + scheduling is the common feature in the stress
> tests that is triggering this issue.
>
> How to reproduce:
>
> git clone https://github.com/ColinIanKing/stress-ng.git
> cd stress-ng
> make clean
> make fast-test-all

I tried to reproduce the failures bug am not getting any errors
following the above steps.

    ~/src/stress-ng$ make fast-test-all
    ...
    stress-ng: 14:20:53.88 debug: [261118] metrics-check: all stressor
    metrics validated and sane
    zombie PASSED
    240 PASSED
    0 FAILED
    7 SKIPPED, bind-mount binderfs cpu-online mlockmany oom-pipe sysinval watchdog

Not sure if the problems you saw were fixed in the interim but I am
using 5.10.0-rc6-rt14 on a quad core Apollolake system.

Are there any configs that need enabling? I used defconfig + Intel LPSS
 for platform support + PREEMPT_RT.

Thanks,
Punit

[...]