Re: [PATCH v2] cgroup: Add preemption protection to css_rstat_updated()

[Waiman Long <llong@redhat.com>]

On 11/21/25 1:21 AM, Jiayuan Chen wrote:
> November 21, 2025 at 13:07, "Waiman Long" <llong@redhat.com mailto:llong@redhat.com?to=%22Waiman%20Long%22%20%3Cllong%40redhat.com%3E > wrote:
>
>
>> On 11/20/25 11:06 PM, Jiayuan Chen wrote:
>>
>>> BPF programs do not disable preemption, they only disable migration.
>>>   Therefore, when running the cgroup_hierarchical_stats selftest, a
>>>   warning [1] is generated.
>>>
>>>   The css_rstat_updated() function is lockless and reentrant. However,
>>>   as Tejun pointed out [2], preemption-related considerations need to
>>>   be considered. Since css_rstat_updated() can be called from BPF where
>>>   preemption is not disabled by its framework and it has already been
>>>   exposed as a kfunc to BPF programs, introducing a new kfunc like bpf_xx
>>>   will break existing uses. Thus, we directly make css_rstat_updated()
>>>   preempt-safe here.
>>>
>> My understand of Tejun's comment is to add bpf_preempt_disable() and bpf_preempt_enable() calls around the css_rstat_updated() call in the bpf program defined in tools/testing/selftests/bpf/prog_tests/cgroup_hierarchical_stats.c instead of adding that in the css_rstat_updated() function itself. But I may be wrong.
>>
>> Cheers, Longman
>>
> If that's really the case, then I'd rather add a new wrapper kfunc for BPF
> to replace css_rstat_updated(). Otherwise, whether it gets triggered would
> depend entirely on users behavior.
>
> Right now, this WARNING is showing up in all BPF selftests. Although it's not
> treated as an error that fails the tests,it's visible in the action runs:
> https://github.com/kernel-patches/bpf/actions

All the existing callers of css_rstat_updated() except the bpf selftest 
has preemption disabled. So it doesn't make sense to impose a cost 
(though small) on kernel code that are in production kernel in order to 
make a selftest pass with no change.

Cheers,
Longman

>