Re: [PATCH v3] mm/min_free_kbytes: modify min_free_kbytes calculation rules

[liuq <liuq131@chinatelecom.cn>]

[-- Attachment #1: Type: text/plain, Size: 3749 bytes --]


在 2023/6/21 18:00, Kefeng Wang 写道:
> Hi Liuq,
>
> On 2023/6/21 17:20, liuq wrote:
>> The current calculation of min_free_kbytes only uses ZONE_DMA and
>> ZONE_NORMAL pages,but the ZONE_MOVABLE zone->_watermark[WMARK_MIN]
>> will also divide part of min_free_kbytes.This will cause the min
>> watermark of ZONE_NORMAL to be too small in the presence of 
>> ZONE_MOVEABLE.
>>
>> __GFP_HIGH and PF_MEMALLOC allocations usually don't need movable
>> zone pages, so just like ZONE_HIGHMEM, cap pages_min to a small
>> value in __setup_per_zone_wmarks.
>>
>> On my testing machine with 16GB of memory (transparent hugepage is
>> turned off by default, and movablecore=12G is configured)
>> The following is a comparative test data of watermark_min
>>
>>         no patch    add patch
>> ZONE_DMA    1        8
>> ZONE_DMA32    151        709
>> ZONE_NORMAL    233        1113
>> ZONE_MOVABLE    1434        128
>> min_free_kbytes    7288        7326
>>
>
> We see this issue and do the same change[1], and we add a per zone 
> watermark configuration too, but both of them is not accepted,
> let's add Mel and wupeng to see more comments.
>
> [1]https://lore.kernel.org/linux-mm/20220905032858.1462927-1-mawupeng1@huawei.com/ 
>
>
Hi wupeng,

Thank you for your reply. Actually, in my version 1, in the presence of ZONE_MOVABLE, 


the calculation of min_free_kbytes can be improved when the ZONE_MOVABLE pages were added to the lowmem_kbytes. 


However, whether the value of watermark_min should be maintained at high level is unclear.

https://lore.kernel.org/all/20230609032552.218010-1-liuq131@chinatelecom.cn/#r


Best Regards

liuq

>> Signed-off-by: liuq <liuq131@chinatelecom.cn>
>> ---
>>   mm/page_alloc.c | 12 ++++++------
>>   1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 47421bedc12b..590ed8725e09 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -6362,9 +6362,9 @@ static void __setup_per_zone_wmarks(void)
>>       struct zone *zone;
>>       unsigned long flags;
>>   -    /* Calculate total number of !ZONE_HIGHMEM pages */
>> +    /* Calculate total number of !ZONE_HIGHMEM and !ZONE_MOVABLE 
>> pages */
>>       for_each_zone(zone) {
>> -        if (!is_highmem(zone))
>> +        if (!is_highmem(zone) && zone_idx(zone) != ZONE_MOVABLE)
>>               lowmem_pages += zone_managed_pages(zone);
>>       }
>>   @@ -6374,15 +6374,15 @@ static void __setup_per_zone_wmarks(void)
>>           spin_lock_irqsave(&zone->lock, flags);
>>           tmp = (u64)pages_min * zone_managed_pages(zone);
>>           do_div(tmp, lowmem_pages);
>> -        if (is_highmem(zone)) {
>> +        if (is_highmem(zone) || zone_idx(zone) == ZONE_MOVABLE) {
>>               /*
>>                * __GFP_HIGH and PF_MEMALLOC allocations usually don't
>> -             * need highmem pages, so cap pages_min to a small
>> -             * value here.
>> +             * need highmem and movable zones pages, so cap pages_min
>> +             * to a small  value here.
>>                *
>>                * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
>>                * deltas control async page reclaim, and so should
>> -             * not be capped for highmem.
>> +             * not be capped for highmem and movable zones.
>>                */
>>               unsigned long min_pages;

[-- Attachment #2: Type: text/html, Size: 42197 bytes --]