Re: [linux-next-20190214] Free pages statistics is broken.

[Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>]

On 2019/02/16 3:13, Dan Williams wrote:
> On Fri, Feb 15, 2019 at 9:44 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>>
>> On 2/15/19 3:27 PM, Tetsuo Handa wrote:
>>> On 2019/02/15 22:01, Michal Hocko wrote:
>>>> On Fri 15-02-19 11:27:10, Tetsuo Handa wrote:
>>>>> I noticed that amount of free memory reported by DMA: / DMA32: / Normal: fields are
>>>>> increasing over time. Since 5.0-rc6 is working correctly, some change in linux-next
>>>>> is causing this problem.
>>>>
>>>> Just a shot into the dark. Could you try to disable the page allocator
>>>> randomization (page_alloc.shuffle kernel command line parameter)? Not
>>>> that I see any bug there but it is a recent change in the page allocator
>>>> I am aware of and it might have some anticipated side effects.
>>>>
>>>
>>> I tried CONFIG_SHUFFLE_PAGE_ALLOCATOR=n but problem still exists.
>>
>> I think it's the preparation patch [1], even with randomization off:
>>
>> @@ -1910,7 +1900,7 @@ static inline void expand(struct zone *zone, struct page *page,
>>                 if (set_page_guard(zone, &page[size], high, migratetype))
>>                         continue;
>>
>> -               list_add(&page[size].lru, &area->free_list[migratetype]);
>> +               add_to_free_area(&page[size], area, migratetype);
>>                 area->nr_free++;
>>                 set_page_order(&page[size], high);
>>         }
>>
>> This should have removed the 'area->nr_free++;' line, as add_to_free_area()
>> includes the increment.
> 
> Yes, good find! I'll send an incremental fixup patch in a moment
> unless someone beats me to it.
> 

Removing the 'area->nr_free++;' line solved the problem. Thank you.