idprio(1) broken

[Eugene Grosbein]

On 26.12.2017 18:37, Andriy Gapon wrote:

>>> Is idprio(1) broken in stable/11?
>>>
>>> As root, start one bzip2 instance with idprio and one additional bzip2 intance per CPU core:
>>>
>>> # idprio 5 bzip2 -9 </dev/zero >/dev/null &
>>> # n=$(sysctl -n kern.smp.cpus)
>>> # i=1; while [ $i -le $n ]; do bzip2 -9 </dev/zero >/dev/null & i=$(($i+1)); done
>>> # top
>>>
>>> For dual core system, I see that idprio'd bzip2 takes all cycles of first core
>>> and two "normal" bzip2's share cycles of second core each taking ~50% of CPU time.
>>>
>>> It is expected that idprio'd bzip2 get no CPU time at all and each of "normal" bzip2's
>>> get ~100% of single CPU core for such setup.
>>
>> This works as expected for stable/10.
> 
> Seems to work as expected on head as well.

I have a report that head r323607 has the problem and head r326729 has not.
I can't check it myself as I have do not run head yet.