Re: [PATCH 04/17] cgroup/cpuset: Limit type-change accounting to owned CPUs

[Guopeng Zhang]

在 2026/8/24 22:24, Waiman Long 写道:
> On 8/20/26 8:41 AM, Guopeng Zhang wrote:
>> From: Guopeng Zhang <zhangguopeng@xxxxxxxxxx>
>>
>> effective_xcpus includes CPUs granted to valid child partitions. Passing
>> the whole mask to isolated_cpus_update() during a root-to-isolated or
>> isolated-to-root change applies the parent's new state to child-owned
>> CPUs as well.
>>
>> Build a mask of CPUs owned by the partition by subtracting the
>> effective_xcpus of valid children. Use this mask when updating
>> isolated_cpus for a partition type change.
>>
>> Fixes: 11e5f407b64a ("cgroup/cpuset: Keep track of CPUs in isolated partitions")
>> Signed-off-by: Guopeng Zhang <zhangguopeng@xxxxxxxxxx>
>> ---
>>   kernel/cgroup/cpuset.c | 30 ++++++++++++++++++++++++++++--
>>   1 file changed, 28 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>> index 2538faac9aba..468272baadb2 100644
>> --- a/kernel/cgroup/cpuset.c
>> +++ b/kernel/cgroup/cpuset.c
>> @@ -2155,6 +2155,30 @@ static void compute_partition_effective_cpumask(struct cpuset *cs,
>>       rcu_read_unlock();
>>   }
>>   +/*
>> + * Compute CPUs owned directly by a partition.
>> + *
>> + * effective_xcpus includes CPUs granted to valid child partitions. Exclude
>> + * those CPUs when changing only this partition's type.
>> + */
>> +static void compute_partition_owned_cpumask(struct cpuset *cs,
>> +                        struct cpumask *owned_cpus)
>> +{
>> +    struct cgroup_subsys_state *css;
>> +    struct cpuset *child;
>> +
>> +    lockdep_assert_held(&cpuset_mutex);
>> +    cpumask_copy(owned_cpus, cs->effective_xcpus);
>> +
>> +    rcu_read_lock();
>> +    cpuset_for_each_child(child, css, cs) {
>> +        if (is_partition_valid(child))
>> +            cpumask_andnot(owned_cpus, owned_cpus,
>> +                       child->effective_xcpus);
>> +    }
>> +    rcu_read_unlock();
>> +}
>> +
>>   /*
>>    * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree
>>    * @cs:  the cpuset to consider
>> @@ -2990,8 +3014,10 @@ static int update_prstate(struct cpuset *cs, int new_prs)
>>       } else if (old_prs && new_prs) {
>>           /*
>>            * A change in load balance state only, no change in cpumasks.
>> -         * Need to update isolated_cpus.
>> +         * Need to update isolated_cpus for CPUs owned by this partition,
>> +         * excluding CPUs distributed to valid child partitions.
>>            */
>> +        compute_partition_owned_cpumask(cs, tmpmask.new_cpus);
>>           if (((new_prs == PRS_ISOLATED) &&
>>                !isolated_cpus_can_update(cs->effective_xcpus, NULL)) ||
>>               prstate_housekeeping_conflict(new_prs, cs->effective_xcpus))
>> @@ -3030,7 +3056,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)
>>       if (!is_partition_valid(cs))
>>           reset_partition_data(cs);
>>       else if (isolcpus_updated)
>> -        isolated_cpus_update(old_prs, new_prs, cs->effective_xcpus);
>> +        isolated_cpus_update(old_prs, new_prs, tmpmask.new_cpus);
>>       spin_unlock_irq(&callback_lock);
>>         /* Force update if switching back to member & update effective_xcpus */
> 
> Your use of tmpmask.new_cpus in isolated_cpus_update() can be problematic. isolcpus_updated can be set when an isolated partition is enabled or disabled. In both cases, tmpmask.new_cpus can be used temporarily. So the content of this temporary cpumask may not be what we want to pass into isolated_cpus_update(). I will suggest you only use tmpmask.new_cpus if it is determined to be coming from partition state transition instead of enabling/disabling of partition.
> 

Thanks for the review, Longman.

I followed your suggestion and checked update_prstate() again.

isolcpus_updated starts as false and is set only in the
old_prs && new_prs branch, after tmpmask.new_cpus has been filled with
the CPUs owned directly by the partition. That branch handles only
root-to-isolated and isolated-to-root transitions. The
member-to-partition and partition-to-member paths do not set the flag.

Those enable and disable paths update isolated_cpus separately. For a
local partition, the path is:

        update_parent_effective_cpumask()
          -> partition_xcpus_add()/partition_xcpus_del()
             -> isolated_cpus_update()

For a remote partition, the path is:

        remote_partition_enable()/remote_partition_disable()
          -> partition_xcpus_add()/partition_xcpus_del()
             -> isolated_cpus_update()

There is no write to tmpmask.new_cpus between
compute_partition_owned_cpumask() and the isolated_cpus_update() call
guarded by isolcpus_updated. As far as I can tell, tmpmask.new_cpus
therefore still contains the directly owned CPUs whenever the flag is
set.

Please let me know if I have overlooked another path.

Thanks,
Guopeng

> Cheers,
> Longman