Re: [PATCH -next] cpuset: treate root invalid trialcs as exclusive

[Waiman Long <llong@redhat.com>]

On 11/15/25 4:31 AM, Chen Ridong wrote:
> From: Chen Ridong <chenridong@huawei.com>
>
> A test scenario revealed inconsistent results based on operation order:
> Scenario 1:
> 	#cd /sys/fs/cgroup/
> 	#mkdir A1
> 	#mkdir B1
> 	#echo 1-2 > B1/cpuset.cpus
> 	#echo 0-1 > A1/cpuset.cpus
> 	#echo root > A1/cpuset.cpus.partition
> 	#cat A1/cpuset.cpus.partition
> 	root invalid (Cpu list in cpuset.cpus not exclusive)
>
> Scenario 2:
> 	#cd /sys/fs/cgroup/
> 	#mkdir A1
> 	#mkdir B1
> 	#echo 1-2 > B1/cpuset.cpus
> 	#echo root > A1/cpuset.cpus.partition
> 	#echo 0-1 > A1/cpuset.cpus
> 	#cat A1/cpuset.cpus.partition
> 	root
>
> The second scenario produces an unexpected result: A1 should be marked
> as invalid but is incorrectly recognized as valid. This occurs because
> when validate_change is invoked, A1 (in root-invalid state) may
> automatically transition to a valid partition, with non-exclusive state
> checks against siblings, leading to incorrect validation.
>
> To fix this inconsistency, treat trialcs in root-invalid state as exclusive
> during validation and set the corresponding exclusive flags, ensuring
> consistent behavior regardless of operation order.
>
> Signed-off-by: Chen Ridong <chenridong@huawei.com>
> ---
>   kernel/cgroup/cpuset.c | 19 ++++++++++++++-----
>   1 file changed, 14 insertions(+), 5 deletions(-)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index daf813386260..a189f356b5f1 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -2526,6 +2526,18 @@ static void partition_cpus_change(struct cpuset *cs, struct cpuset *trialcs,
>   	}
>   }
>   
> +static int init_trialcs(struct cpuset *cs, struct cpuset *trialcs)
> +{
> +	trialcs->prs_err = PERR_NONE;
> +	/*
> +	 * If partition_root_state != 0, it may automatically change to a partition,
> +	 * Therefore, we should treat trialcs as exclusive during validation
> +	 */
> +	if (trialcs->partition_root_state)
> +		set_bit(CS_CPU_EXCLUSIVE, &trialcs->flags);
Nit: We usually use the non-atomic version __set_bit() if concurrent 
access isn't possible which is true in this case.

> +	return compute_trialcs_excpus(trialcs, cs);
> +}
> +
>   /**
>    * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
>    * @cs: the cpuset to consider
> @@ -2551,9 +2563,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
>   	if (alloc_tmpmasks(&tmp))
>   		return -ENOMEM;
>   
> -	compute_trialcs_excpus(trialcs, cs);
> -	trialcs->prs_err = PERR_NONE;
> -
> +	init_trialcs(cs, trialcs);
>   	retval = cpus_allowed_validate_change(cs, trialcs, &tmp);
>   	if (retval < 0)
>   		goto out_free;
> @@ -2612,7 +2622,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
>   	 * Reject the change if there is exclusive CPUs conflict with
>   	 * the siblings.
>   	 */
> -	if (compute_trialcs_excpus(trialcs, cs))
> +	if (init_trialcs(cs, trialcs))
>   		return -EINVAL;
>   
>   	/*
> @@ -2628,7 +2638,6 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
>   	if (alloc_tmpmasks(&tmp))
>   		return -ENOMEM;
>   
> -	trialcs->prs_err = PERR_NONE;
>   	partition_cpus_change(cs, trialcs, &tmp);
>   
>   	spin_lock_irq(&callback_lock);
Acked-by: Waiman Long <longman@redhat.com>