Re: [PATCH 0/2] mm/compaction: some fix for the range passed to pageblock_pfn_to_page()

[David Hildenbrand <david@redhat.com>]

On 08.10.25 11:13, Wei Yang wrote:
> On Wed, Oct 08, 2025 at 09:52:51AM +0200, Michal Hocko wrote:
>> On Wed 08-10-25 02:32:18, Wei Yang wrote:
>>> On Tue, Oct 07, 2025 at 06:29:22PM -0700, Andrew Morton wrote:
>>>> On Thu,  2 Oct 2025 03:31:38 +0000 Wei Yang <richard.weiyang@gmail.com> wrote:
>>>>
>>>>> During the code reading of isolate_migratepages_range(), first spot the range
>>>>> passed to pageblock_pfn_to_page() is different from that to
>>>>> isolate_migratepages_block().
>>>>>
>>>>> This implies there is a chance that pageblock_pfn_to_page() thinks the range
>>>>> is in the same zone, but isolate_migratepages_block() will isolate range in
>>>>> two different zones. This is not what we expect.
>>>>>
>>>>> Then I found pageblock_pfn_to_page() has an optimization if zone->contiguous,
>>>>> this means even the range is across two different zones, it will think the
>>>>> range is within the same zone.
>>>>>
>>>>> So introduce two patches to fix it:
>>>>
>>>> What do you think might be the worst-case userspace-visible effects
>>>> of the bug?
>>>
>>> I don't see userspace-visible effect yet.
>>>
>>> Since I lack the knowledge of the consequence of isolating cross zone range,
>>> currently I can't tell the effect accurately.
>>>
>>> The worst case in my mind is we put some page on a different zone's freelist.
>>> But I don't totally understand the behavior now.
>>
>> Is there any actual problem to fix here then?
> 
> I don't see an actual problem in practice yet.

We should avoid talking about "fix" then. Let me take a look at the patches.

-- 
Cheers

David / dhildenb