Re: [REGRESSION][v6.17-rc1]sched/fair: Bump sd->max_newidle_lb_cost when newidle balance fails

[Adam Li <adamli@os.amperecomputing.com>]

Hi Joseph,

On 10/18/2025 1:01 AM, Joseph Salisbury wrote:
> 
> 
> On 10/10/25 13:09, Peter Zijlstra wrote:
>> On Thu, Oct 09, 2025 at 09:04:40PM -0400, Joseph Salisbury wrote:
>>
[...]
>> diff --git a/kernel/sched/features.h b/kernel/sched/features.h
>> index 3c12d9f93331..5ded210701d0 100644
>> --- a/kernel/sched/features.h
>> +++ b/kernel/sched/features.h
>> @@ -121,3 +121,8 @@ SCHED_FEAT(WA_BIAS, true)
>>   SCHED_FEAT(UTIL_EST, true)
>>     SCHED_FEAT(LATENCY_WARN, false)
>> +
>> +SCHED_FEAT(NI_TARGET, false)
>> +SCHED_FEAT(NI_SPARE, false)
>> +SCHED_FEAT(NI_RUNNABLE, false)
>> +SCHED_FEAT(NI_RANDOM, false)
[...]
> 
> Thanks for providing the patch!
> 
> I performed the same testing with this patch applied to a v6.12 based kernel.  The performance regression still exists and is slightly worse.
> The drgn stats are attached in a file named 'apache_drgn_log_NewidleRework_From_Peter.log'.
> 
> 
Just wondering, did you enable NI_TARGET feature?
It is default false in Peter's patch.
In my test 'NI_TARGET' can fix the regression caused by commit 155213a2aed4.

Thanks,
-adam