Re: [PATCH 08/17] cgroup/cpuset: Release CPUs when a type change is rejected

[Guopeng Zhang <guopeng.zhang@linux.dev>]

在 2026/8/24 22:37, Waiman Long 写道:
> On 8/20/26 8:41 AM, Guopeng Zhang wrote:
>> From: Guopeng Zhang <zhangguopeng@kylinos.cn>
>>
>> When a housekeeping check rejects a root-to-isolated or isolated-to-root
>> change, update_prstate() marks the partition invalid but leaves its
>> effective_xcpus allocated. Those CPUs remain missing from the partition
>> which owns the invalidated subtree.
>>
>> Disable a rejected remote partition with remote_partition_disable().
>> For a local partition, return effective_xcpus to its parent before
>> recording the invalid state. Use the complete mask because descendants
>> are invalidated along with the parent and their CPUs return at the same
>> time.
>>
>> Fixes: 103b08709e8a ("cgroup/cpuset: Fail if isolated and nohz_full don't leave any housekeeping")
>> Signed-off-by: Guopeng Zhang <zhangguopeng@kylinos.cn>
>> ---
>>   kernel/cgroup/cpuset.c | 15 +++++++++++++--
>>   1 file changed, 13 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>> index 853942c1afeb..a616012927be 100644
>> --- a/kernel/cgroup/cpuset.c
>> +++ b/kernel/cgroup/cpuset.c
>> @@ -3020,10 +3020,21 @@ static int update_prstate(struct cpuset *cs, int new_prs)
>>           compute_partition_owned_cpumask(cs, tmpmask.new_cpus);
>>           if (((new_prs == PRS_ISOLATED) &&
>>                !isolated_cpus_can_update(tmpmask.new_cpus, NULL)) ||
>> -            prstate_housekeeping_conflict(new_prs, tmpmask.new_cpus))
>> +            prstate_housekeeping_conflict(new_prs, tmpmask.new_cpus)) {
>>               err = PERR_HKEEPING;
>> -        else
>> +            if (is_remote_partition(cs)) {
>> +                WRITE_ONCE(cs->prs_err, err);
> This WRITE_ONCE() is not necessary as it will be done later in the function.
>> +                remote_partition_disable(cs, &tmpmask);
>> +            } else {
>> +                spin_lock_irq(&callback_lock);
>> +                partition_xcpus_del(old_prs, parent, cs->effective_xcpus);
>> +                spin_unlock_irq(&callback_lock);
>> +                cpuset_update_tasks_cpumask(parent, tmpmask.new_cpus);
>> +                update_sibling_cpumasks(parent, cs, &tmpmask);
>> +            }
> 
> This code to disable the partition is duplicating what is being done in the last else clause of the outer if, but in a somewhat different way. That is not good programming practice. You can either jump to the partition disabling code below or separate out the partition disabling code controlled by a separate control flag and set the flag to do it.
> 

Agreed. I removed the redundant WRITE_ONCE() and used a
disable_partition flag to share the existing partition-disable path in
v2.

Thanks,
Guopeng

> Cheers,
> Longman
> 
>> +        } else {
>>               isolcpus_updated = true;
>> +        }
>>       } else {
>>           /*
>>            * Switching back to member is always allowed even if it
>