Re: scheduler problems in -next (was: Re: [PATCH 6.4 000/227] 6.4.7-rc1 review)

[Guenter Roeck <linux@roeck-us.net>]

On 7/31/23 07:39, Peter Zijlstra wrote:
> On Mon, Jul 31, 2023 at 04:19:34PM +0200, Peter Zijlstra wrote:
>> On Sat, Jul 29, 2023 at 09:00:02PM -0700, Guenter Roeck wrote:
>>> On 7/27/23 16:18, Joel Fernandes wrote:
>>>
>>> [ ... ]
>>>
>>>>> I freely confess that I am having a hard time imagining what would
>>>>> be CPU dependent in that code.  Timing, maybe?  Whatever the reason,
>>>>> I am not seeing these failures in my testing.
>>>>>
>>>>> So which of the following Kconfig options is defined in your .config?
>>>>> CONFIG_TASKS_RCU, CONFIG_TASKS_RUDE_RCU, and CONFIG_TASKS_TRACE_RCU.
>>>>>
>>>>> If you have more than one of them, could you please apply this patch
>>>>> and show me the corresponding console output from the resulting hang?
>>>>
>>>> FWIW, I am not able to repro this issue either. If a .config can be shared of the problem system, I can try it out to see if it can be reproduced on my side.
>>>>
>>>
>>> I managed to bisect the problem. See bisect log below. Bisect repeated twice.
>>> so it should be reliable. I don't really understand it, but the following
>>> reverts fix the problem. This is on top of next-20230721 because next-20230728
>>> crashes immediately in my tests.
>>>
>>> 0caafe9b94ab (HEAD) Revert "sched/fair: Remove sched_feat(START_DEBIT)"
>>> 518bdbd39fdb Revert "sched/fair: Add lag based placement"
>>> a011162c3e32 Revert "sched/fair: Implement an EEVDF-like scheduling policy"
>>> df579720bf98 Revert "sched/fair: Commit to lag based placement"
>>> aac459a7e738 Revert "sched/smp: Use lag to simplify cross-runqueue placement"
>>> 8d686eb173e1 Revert "sched/fair: Commit to EEVDF"
>>> 486474c50f95 Revert "sched/debug: Rename sysctl_sched_min_granularity to sysctl_sched_base_slice"
>>> 79e94d67d08a Revert "sched/fair: Propagate enqueue flags into place_entity()"
>>> ae867bc97b71 (tag: next-20230721) Add linux-next specific files for 20230721
>>>
>>> For context: x86 images (32 and 64 bit) in -next tend to hang at
>>>
>>> [    2.309323] RCU Tasks: Setting shift to 0 and lim to 1 rcu_task_cb_adjust=1.
>>> [    2.311634] Running RCU-tasks wait API self tests
>>>
>>> The hang is not seen with every boot; it happens roughly about once every
>>> 10 boot attempts. It is not CPU dependent as I initially thought.
>>>
>>> Configuration file is at http://server.roeck-us.net/qemu/x86-next/config.
>>> Example qemu command line:
>>
>> Hurmph, let me see if I can reproduce on next-20230731 (not having the
>> older next thingies around).
> 
> I've taken your config above, and the rootfs.ext2 and run-sh from x86/.
> I've then modified run-sh to use:
> 
>    qemu-system-x86_64 -enable-kvm -cpu host
> 
> What I'm seeing is that some boots get stuck at:
> 
> [    0.608230] Running RCU-tasks wait API self tests
> 
> Is this the right 'problem' ?
> 


Yes, exactly.

Thanks,
Guenter