Re: [PATCH v3 0/8] mm/page_owner: Add PID/TGID/COMM and cgroup filtering

[Zi Yan]

On 22 Sep 2026, at 8:33, zhen.ni wrote:

> 在 2026/9/22 00:25, Zi Yan 写道:
>> On 19 Sep 2026, at 22:11, zhen.ni wrote:
>>
>>> 在 2026/9/11 19:32, Lorenzo Stoakes (ARM) 写道:
>>>> On Fri, Sep 11, 2026 at 01:06:46PM +0200, David Hildenbrand (Arm) wrote:
>>>>> On 9/11/26 09:58, zhen.ni wrote:
>>>>>>
>>>>>>
>>>>>> 在 2026/9/9 23:54, Zi Yan 写道:
>>>>>>> On 9 Sep 2026, at 6:28, David Hildenbrand (Arm) wrote:
>>>>>>>
>>>>>>>>
>>>>>>>> Because it results in less kernel code  :)
>>>>>>>>
>>>>>>>> And less kernel code is good. Unless unavoidable.
>>>>>>>
>>>>>>> An alternative is to add BPF hooks like bpf_iter to do the filtering
>>>>>>> and by default, when no BPF program is attached, everything is printed.
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> I’ve also thought about a similar approach—perhaps a new bpf_iter type
>>>>>> that could iterate over all pages by PFN. The advantage would be that
>>>>>> you can fully customize what you want to print and what you want to
>>>>>> filter. However, doing so would require refactoring the entire
>>>>>> page_owner, and I’m not sure whether it’s worth pursuing in this
>>>>>> direction.I’d also like to hear everyone’s thoughts on this.
>>>>>
>>>>> Certainly better than having more and more manual filters added to the code base.
>>>>
>>>> Agreed.
>>>>
>>>> bpf seems the ideal fit here I think?
>>>>
>>>> Means that those who want this information can get it and those who don't are
>>>> unaffected.
>>>>
>>>
>>> I have prototyped page_owner bpf_iter and confirmed it is a viable and
>>> valuable direction:
>>>
>>> 1.Most parameter parsing is offloaded to the existing BPF verifier,
>>> keeping the page_owner kernel-side logic simple.
>>>
>>> 2.Filtering and printing are delegated to the BPF program, providing
>>> greater flexibility and enabling more complex composite filtering.
>>
>> Without attaching a BPF program, page_owner could still print to
>> the console, right? Otherwise, it will break existing users without a
>> BPF program. You could add a hook to decide where to print the results.
>>
>
> The existing console/debugfs path is untouched -- the BPF iterator is a
> new, opt-in channel that coexists with it. Without a program attached,
> page_owner behaves exactly as before.

Sounds great. Thank you for the confirmation.

Best Regards,
Yan, Zi