On 2025/10/20 10:37, Waiman Long wrote:
> On 9/28/25 3:12 AM, Chen Ridong wrote:
>> From: Chen Ridong <chenridong@huawei.com>
>>
>> For cgroup v2, when a cpuset is not a valid partition root, it inherits
>> the CS_SCHED_LOAD_BALANCE flag from its parent. The existing logic in
>> update_cpumasks_hier() manually handled this inheritance condition.
>>
>> This patch replaces the inline implementation with a call to the dedicated
>> update_partition_sd_lb() helper function, which already encapsulates the
>> same logic. The helper function comprehensively handles both the load
>> balance flag update and the necessary scheduling domain rebuild decision.
>>
>> Signed-off-by: Chen Ridong <chenridong@huawei.com>
>> ---
>> kernel/cgroup/cpuset.c | 16 +++-------------
>> 1 file changed, 3 insertions(+), 13 deletions(-)
>>
>> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
>> index 52468d2c178a..052f9e0c7a65 100644
>> --- a/kernel/cgroup/cpuset.c
>> +++ b/kernel/cgroup/cpuset.c
>> @@ -1257,6 +1257,8 @@ static void update_partition_sd_lb(struct cpuset *cs, int old_prs)
>> bool rebuild_domains = (new_prs > 0) || (old_prs > 0);
>> bool new_lb;
>> + if (!cpuset_v2())
>> + return;
>> /*
>> * If cs is not a valid partition root, the load balance state
>> * will follow its parent.
>> @@ -2276,19 +2278,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp,
>> !cpumask_equal(cp->cpus_allowed, cp->effective_cpus));
>> cpuset_update_tasks_cpumask(cp, cp->effective_cpus);
>> -
>> - /*
>> - * On default hierarchy, inherit the CS_SCHED_LOAD_BALANCE
>> - * from parent if current cpuset isn't a valid partition root
>> - * and their load balance states differ.
>> - */
>> - if (cpuset_v2() && !is_partition_valid(cp) &&
>> - (is_sched_load_balance(parent) != is_sched_load_balance(cp))) {
>> - if (is_sched_load_balance(parent))
>> - set_bit(CS_SCHED_LOAD_BALANCE, &cp->flags);
>> - else
>> - clear_bit(CS_SCHED_LOAD_BALANCE, &cp->flags);
>> - }
>> + update_partition_sd_lb(cp, old_prs);
>> /*
>> * On legacy hierarchy, if the effective cpumask of any non-
> Calling update_partition_sd_lb() directly from update_cpumasks_hier() may incorrectly force
> rebuidling sched domain when it is not really necessary.
>
> Cheers,
> Longman
Thank you Longman,
You are correct. Besides the change for the CS_SCHED_LOAD_BALANCE flag, which can lead to rebuilding
the sched domain, the condition of new_prs or old_prs being greater than 0 may also trigger a domain
rebuild, which is unnecessary. I will drop this patch in next version.
--
Best regards,
Ridong