Re: [PATCH] cgroup/cpuset: Fix missing effective_xcpus accounting of child partitions

[Waiman Long]

On 10/28/26 9:52 PM, Guopeng Zhang wrote:
> 在 2026/10/28 23:30, Waiman Long 写道:
> > On 10/27/26 8:38 AM, Guopeng Zhang wrote:
> > > From: Guopeng Zhang <zhangguopeng@xxxxxxxxxx>
> > >
> > > When a partition root expands its exclusive CPUs, a valid child
> > > partition will be able to claim more CPUs from its parent. As
> > > update_cpumasks_hier() always updates a cpuset before its descendants,
> > > the parent has already been committed when the child partition is
> > > reached. Its effective_cpus is computed with the child's old
> > > effective_xcpus subtracted and no further adjustment is made
> > > after the child commits its new effective_xcpus.
> > >
> > > The additional CPUs are thus granted to both the parent and the child
> > > partitions. generate_sched_domains() then emits the warning below
> > > about the overlapping sched domains:
> > >
> > >     WARNING: kernel/cgroup/cpuset.c:919 at rebuild_sched_domains_locked+0x1ec/0x2e0, CPU#21: bash/1225
> > >
> > > Fix that by taking the CPUs newly added to a valid child partition
> > > away from its parent at the commit point of update_cpumasks_hier()
> > > right before the new effective_xcpus is committed. The tasks in the
> > > parent cpuset as well as its other children will be updated if its
> > > effective_cpus is changed.
> > >
> > > A new test case is added to test_cpuset_prs.sh to test this patch.
> > >
> > > Fixes: e2ffe502ba45 ("cgroup/cpuset: Add cpuset.cpus.exclusive for v2")
> > > Signed-off-by: Guopeng Zhang <zhangguopeng@xxxxxxxxxx>
> > Thanks for reporting this bug. Yes, the current code doesn't handle expansion of parent effective_xcpus correctly. I would prefer to handle it inside update_parent_effective_cpumask() and try to minimize partition related changes in update_cpumasks_hier(). I have posted an alternative patch for your review. Let me know how you think about it.
> Thanks for the alternative patch.
>
> Well, I am still learning how you approach this whole series of
> problems.
>
> One more thing I found yesterday: neither the current code nor my
> patch runs the housekeeping check when a valid child partition gains
> CPUs from its parent, e.g. when its parent expands the exclusive
> CPUs. With nohz_full, an isolated child partition can take over the
> last housekeeping CPU. I verified this on a nohz_full=1-7 boot.

This is a problem reported by sashiko as well. I am going to send 
additional patch to address this as well.

Cheers,
Longman

> I handled it roughly like the diff below. Please take a look and see
> if it is worth handling together in this series.
>
> @@ -2142,12 +2142,28 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
>   				deleting = !cpumask_empty(tmp->delmask);
>   			else
>   				part_error = PERR_NOTEXCL;
> +		} else if (cpumask_andnot(tmp->addmask, tmp->delmask,
> +					  cs->effective_xcpus) &&
> +			   prstate_housekeeping_conflict(cs->partition_root_state,
> +							parent_prs,
> +							tmp->delmask, NULL)) {
> +			/*
> +			 * A valid partition may be able to claim more CPUs
> +			 * from its parent, e.g. when its parent expands the
> +			 * exclusive CPUs. The newly computed effective xcpus
> +			 * may not have been housekeeping checked before.
> +			 */
> +			part_error = PERR_HKEEPING;
> +			adding = cpumask_and(tmp->addmask,
> +				       cs->effective_xcpus,
> +				       parent->effective_xcpus);
>   		}
>
> I have something urgent at hand today, so I will get to your patch a
> bit later and run it through the test cases I have, including the
> isolated_cpus accounting case I mentioned in my previous reply.