Re: [PATCH 1/8] mm/compaction: avoid missing last page block in section after skip offline sections

[Kemeng Shi]

on 8/1/2023 11:53 AM, Baolin Wang wrote:
> 
> 
> On 8/1/2023 10:36 AM, Kemeng Shi wrote:
>>
>>
>> on 8/1/2023 10:18 AM, Kemeng Shi wrote:
>>>
>>>
>>> on 7/31/2023 8:01 PM, Baolin Wang wrote:
>>>>
>>>>
>>>> On 7/29/2023 1:10 AM, Kemeng Shi wrote:
>>>>> skip_offline_sections_reverse will return the last pfn in found online
>>>>> section. Then we set block_start_pfn to start of page block which
>>>>> contains the last pfn in section. Then we continue, move one page
>>>>> block forward and ignore the last page block in the online section.
>>>>> Make block_start_pfn point to first page block after online section to fix
>>>>> this:
>>>>> 1. make skip_offline_sections_reverse return end pfn of online section,
>>>>> i.e. pfn of page block after online section.
>>>>> 2. assign block_start_pfn with next_pfn.
>>>>>
>>>>> Fixes: f63224525309 ("mm: compaction: skip the memory hole rapidly when isolating free pages")
>>>>> Signed-off-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
>>>>> ---
>>>>>    mm/compaction.c | 5 ++---
>>>>>    1 file changed, 2 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>>>> index 9b7a0a69e19f..ce7841363b12 100644
>>>>> --- a/mm/compaction.c
>>>>> +++ b/mm/compaction.c
>>>>> @@ -259,7 +259,7 @@ static unsigned long skip_offline_sections_reverse(unsigned long start_pfn)
>>>>>          while (start_nr-- > 0) {
>>>>>            if (online_section_nr(start_nr))
>>>>> -            return section_nr_to_pfn(start_nr) + PAGES_PER_SECTION - 1;
>>>>> +            return section_nr_to_pfn(start_nr + 1);
>>>>
>>>> This is incorrect, you returned the start pfn of this section.
>>>>
>>>>>        }
>>>>>          return 0;
>>>>> @@ -1670,8 +1670,7 @@ static void isolate_freepages(struct compact_control *cc)
>>>>>                  next_pfn = skip_offline_sections_reverse(block_start_pfn);
>>>>>                if (next_pfn)
>>>>> -                block_start_pfn = max(pageblock_start_pfn(next_pfn),
>>>>> -                              low_pfn);
>>>>> +                block_start_pfn = max(next_pfn, low_pfn);
>>>>
>>>> 'block_start_pfn' should be pageblock aligned. If the 'next_pfn' is not pageblock-aligned (though this is not the common case), we should skip it.
>>>>
>>>> But if the 'next_pfn' is pageblock-aligned, yes, the commit f63224525309 still ignores the last pageblock, which is not right. So I think it should be:
>>>> block_start_pfn = pageblock_aligned(next_pfn) ? : pageblock_start_pfn(next_pfn);
>>>> block_start_pfn = max(block_start_pfn, low_pfn);
>>>>
>>> Hi Baolin, thanks for reply! As skip_offline_sections_reverse is based
>>> on skip_offline_sections. I make the assumption that section is pageblock
>>> aligned based on that we use section start from skip_offline_sections as
>>> block_start_fpn without align check.
>>> If section size is not pageblock aligned in real world, the pageblock aligned
>>> check should be added to skip_offline_sections and skip_offline_sections_reverse.
>>> If no one is against this, I will fix this in next version. THanks!
>>>
>> More information of aligment of section. For powerpc arch, we have SECTION_SIZE_BITS
>> with 24 while PAGE_SHIFT could be configured to 18.
>> Pageblock order is (18 + MAX_ORDER) which coule be 28 and is > SECTION_SZIE_BITS 24,
> 
> The maximum pageblock order is MAX_ORDER. But after thinking more, I think return the start pfn or end pfn of a section is okay, and it should be aligned to a pageblock order IIUC.
> 
Right, I mixed up the unit.
> So I think your change is good:
> + block_start_pfn = max(next_pfn, low_pfn);
> 
> But in skip_offline_sections_reverse(), we should still return the last pfn of the online section.
> 
Sure, then we should assign block_start_pfn with following change. Is this good to you?
-                block_start_pfn = max(pageblock_start_pfn(next_pfn),
+		 block_start_pfn = max(pageblock_end_pfn(next_pfn),
                              low_pfn);

-- 
Best wishes
Kemeng Shi