Re: [PATCH -next v2] cpuset: Treat cpusets in attaching as populated

[Chen Ridong <chenridong@huaweicloud.com>]

On 2025/11/13 22:14, Waiman Long wrote:
> On 11/13/25 8:28 AM, Chen Ridong wrote:
>> From: Chen Ridong <chenridong@huawei.com>
>>
>> Currently, the check for whether a partition is populated does not
>> account for tasks in the cpuset of attaching. This is a corner case
>> that can leave a task stuck in a partition with no effective CPUs.
>>
>> The race condition occurs as follows:
>>
>> cpu0                cpu1
>>                 //cpuset A  with cpu N
>> migrate task p to A
>> cpuset_can_attach
>> // with effective cpus
>> // check ok
>>
>> // cpuset_mutex is not held    // clear cpuset.cpus.exclusive
>>                 // making effective cpus empty
>>                 update_exclusive_cpumask
>>                 // tasks_nocpu_error check ok
>>                 // empty effective cpus, partition valid
>> cpuset_attach
>> ...
>> // task p stays in A, with non-effective cpus.
>>
>> To fix this issue, this patch introduces cs_is_populated, which considers
>> tasks in the attaching cpuset. This new helper is used in validate_change
>> and partition_is_populated.
>>
>> Fixes: e2d59900d936 ("cgroup/cpuset: Allow no-task partition to have empty cpuset.cpus.effective")
>> Signed-off-by: Chen Ridong <chenridong@huawei.com>
>> ---
>>   kernel/cgroup/cpuset.c | 31 +++++++++++++++++++++++--------
>>   1 file changed, 23 insertions(+), 8 deletions(-)
>>
>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>> index daf813386260..bd273b1e09b0 100644
>> --- a/kernel/cgroup/cpuset.c
>> +++ b/kernel/cgroup/cpuset.c
>> @@ -356,6 +356,15 @@ static inline bool is_in_v2_mode(void)
>>             (cpuset_cgrp_subsys.root->flags & CGRP_ROOT_CPUSET_V2_MODE);
>>   }
>>   +static inline bool cs_is_populated(struct cpuset *cs)
> Could you name it as "cpuset_is_populated()" as it is a cpuset specific version of
> cgroup_is_populated()?

Sure, will update.

>> +{
>> +    lockdep_assert_held(&cpuset_mutex);
>> +
>> +    /* Cpusets in the process of attaching should be considered as populated */
>> +    return cgroup_is_populated(cs->css.cgroup) ||
>> +        cs->attach_in_progress;
>> +}
>> +
>>   /**
>>    * partition_is_populated - check if partition has tasks
>>    * @cs: partition root to be checked
>> @@ -373,19 +382,25 @@ static inline bool is_in_v2_mode(void)
>>   static inline bool partition_is_populated(struct cpuset *cs,
>>                         struct cpuset *excluded_child)
>>   {
>> -    struct cgroup_subsys_state *css;
>> -    struct cpuset *child;
>> +    struct cpuset *cp;
>> +    struct cgroup_subsys_state *pos_css;
>>   -    if (cs->css.cgroup->nr_populated_csets)
>> +    /*
>> +     * We cannot call cs_is_populated(cs) directly, as
>> +     * nr_populated_domain_children may include populated
>> +     * csets from descendants that are partitions.
>> +     */
>> +    if (cs->css.cgroup->nr_populated_csets ||
>> +        cs->attach_in_progress)
>>           return true;
>>         rcu_read_lock();
>> -    cpuset_for_each_child(child, css, cs) {
>> -        if (child == excluded_child)
>> +    cpuset_for_each_descendant_pre(cp, pos_css, cs) {
>> +        if (cp == cs || cp == excluded_child)
>>               continue;
>> -        if (is_partition_valid(child))
>> +        if (is_partition_valid(cp))
> 
> You should add " pos_css = css_rightmost_descendant(pos_css);" to skip the whole subtree.
> 
> Cheers,
> Longman
> 

Oh... you're right, I should have caught this.

Thank you so much, Longman!

> 
>>               continue;
>> -        if (cgroup_is_populated(child->css.cgroup)) {
>> +        if (cs_is_populated(cp)) {
>>               rcu_read_unlock();
>>               return true;
>>           }
>> @@ -670,7 +685,7 @@ static int validate_change(struct cpuset *cur, struct cpuset *trial)
>>        * be changed to have empty cpus_allowed or mems_allowed.
>>        */
>>       ret = -ENOSPC;
>> -    if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) {
>> +    if (cs_is_populated(cur)) {
>>           if (!cpumask_empty(cur->cpus_allowed) &&
>>               cpumask_empty(trial->cpus_allowed))
>>               goto out;
> 

-- 
Best regards,
Ridong