Re: [PATCH v9 0/7] Enable cpuset controller in default hierarchy

[Waiman Long]

On 05/30/2018 09:05 AM, Juri Lelli wrote:
> On 30/05/18 08:56, Waiman Long wrote:
>> On 05/30/2018 06:13 AM, Juri Lelli wrote:
>>> Hi,
>>>
>>> On 29/05/18 09:41, Waiman Long wrote:
>>>> v9:
>>>>  - Rename cpuset.sched.domain to cpuset.sched.domain_root to better
>>>>    identify its purpose as the root of a new scheduling domain or
>>>>    partition.
>>>>  - Clarify in the document about the purpose of domain_root and
>>>>    load_balance. Using domain_root is th only way to create new
>>>>    partition.
>>>>  - Fix a lockdep warning in update_isolated_cpumask() function.
>>>>  - Add a new patch to eliminate call to generate_sched_domains() for
>>>>    v2 when a change in cpu list does not touch a domain_root.
>>> I was playing with this and ended up with the situation below:
>>>
>>>  g1/cgroup.controllers:cpuset
>>>  g1/cgroup.events:populated 0
>>>  g1/cgroup.max.depth:max
>>>  g1/cgroup.max.descendants:max
>>>  g1/cgroup.stat:nr_descendants 1
>>>  g1/cgroup.stat:nr_dying_descendants 0
>>>  g1/cgroup.subtree_control:cpuset
>>>  g1/cgroup.type:domain
>>>  g1/cpuset.cpus:0-5                   <---
>>>  g1/cpuset.cpus.effective:0-5
>>>  g1/cpuset.mems.effective:0-1
>>>  g1/cpuset.sched.domain_root:1        <---
>>>  g1/cpuset.sched.load_balance:1
>>>  g1/cpu.stat:usage_usec 0
>>>  g1/cpu.stat:user_usec 0
>>>  g1/cpu.stat:system_usec 0
>>>  g1/g11/cgroup.events:populated 0
>>>  g1/g11/cgroup.max.descendants:max
>>>  g1/g11/cpu.stat:usage_usec 0
>>>  g1/g11/cpu.stat:user_usec 0
>>>  g1/g11/cpu.stat:system_usec 0
>>>  g1/g11/cgroup.type:domain
>>>  g1/g11/cgroup.stat:nr_descendants 0
>>>  g1/g11/cgroup.stat:nr_dying_descendants 0
>>>  g1/g11/cpuset.cpus.effective:0-5
>>>  g1/g11/cgroup.controllers:cpuset
>>>  g1/g11/cpuset.sched.load_balance:1
>>>  g1/g11/cpuset.mems.effective:0-1
>>>  g1/g11/cpuset.cpus:6-11              <---
>>>  g1/g11/cgroup.max.depth:max
>>>  g1/g11/cpuset.sched.domain_root:0
>>>
>>> Should this be allowed? I was expecting subgroup g11 should be
>>> restricted to a subset of g1's cpus.
>>>
>>> Best,
>>>
>>> - Juri
>> That shouldn't be allowed.The code is probably missing some checks that
>> should have been done. What was the sequence of commands leading to the
>> above configuration?
> This is a E5-2609 v3 (12 cores) Fedora Server box (with systemd, so
> first command is needed to be able to use cpuset controller with v2,
> IIUC):
>
> # umount /sys/fs/cgroup/cpuset 
> # cd /sys/fs/cgroup/unified/
> # echo "+cpuset" >cgroup.subtree_control 
> # mkdir g1
> # echo 0-5 >g1/cpuset.cpus
> # echo 6-11 >init.scope/cpuset.cpus
> # echo 6-11 >machine.slice/cpuset.cpus
> # echo 6-11 >system.slice/cpuset.cpus
> # echo 6-11 >user.slice/cpuset.cpus
> # echo 1 >g1/cpuset.sched.domain_root 
> # mkdir g1/g11
> # echo "+cpuset" > g1/cgroup.subtree_control 
> # echo 6-11 >g1/g11/cpuset.cpus
> # grep -R . g1/*
>
> That should be it. Am I doing something wrong?
>
> Thanks,
>
> - Juri


Yes, it is a bug in the existing code. I have sent out a patch to fix that.

-Longman