Re: [PATCH] mm/min_free_kbytes: modify min_free_kbytes calculation rules
From: Huang, Ying
Date: Tue Jun 20 2023 - 01:52:49 EST
liuq <liuq131@xxxxxxxxxxxxxxx> writes:
> 在 2023/6/19 11:25, Huang, Ying 写道:
>> Added Mel and Vlastimil.
>>
>> liuq <liuq131@xxxxxxxxxxxxxxx> writes:
>>
>>> 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.
>> This seems like a real problem per my understanding. Can you show the
>> contents of /proc/zoneinfo on a problem system?
>>
>> But, per my understanding, min_free_kbytes are used for __GFP_HIGH and
>> PF_MEMALLOC allocations. While ZONE_MOVABLE will not be used for them
>> usually. So I think we should treat ZONE_MOVABLE as ZONE_HIGHMEM in
>> __setup_per_zone_wmarks().
>>
>> Best Regards,
>> Huang, Ying
> On my testing machine with 16GB of memory (transparent hugepage is
> turned off by default), when movable zone is not configured,
> min_free_kbytes is 15806 (15806*15806/16=15614352kbytes, approximatel
> y 16G).
> The detailed info is as follows:
> [root@lq-workstation ~]# cat /proc/cmdline
> BOOT_IMAGE=/vmlinuz-6.2.0-rc7-00018-g0983f6bf2bfc-dirty
> root=/dev/mapper/ctyunos00-root ro resume=/dev/mapper/ctyunos00-swap
> rd.lvm.lv=ctyunos00/root rd.lvm.lv=ctyunos00/swap crashkernel=512M
> [root@lq-workstation ~]# cat /proc/zoneinfo |grep -A 5 min
> min 3
> low 6
> high 9
> spanned 4095
> present 3998
> managed 3840
> --
> min 328
> low 652
> high 976
> spanned 1044480
> present 478802
> managed 330969
> --
> min 3618
> low 7193
> high 10768
> spanned 3655680
> present 3655680
> managed 3575787
> --
> min 0
> low 0
> high 0
> spanned 0
> present 0
> managed 0
> [root@lq-workstation ~]# cat /proc/sys/vm/min_free_kbytes
> 15806
>
> If movablecore=12G is configured, at this time, min_free_kbytes is
> 7326 (7326 * 7326/16=3354392, approximately 16G-12G)
> The detailed info is as follows:
> [root@lq-workstation ~]# cat /proc/cmdline
> BOOT_IMAGE=/vmlinuz-6.2.0-rc7-00018-g0983f6bf2bfc-dirty
> root=/dev/mapper/ctyunos00-root ro resume=/dev/mapper/ctyunos00-swap
> rd.lvm.lv=ctyunos00/root rd.lvm.lv=ctyunos00/swap crashkernel=512M
> movablecore=12G
> [root@lq-workstation ~]# cat /proc/zoneinfo |grep -A 5 min
> min 1
> low 4
> high 7
> spanned 4095
> present 3998
> managed 3840
> --
> min 152
> low 476
> high 800
> spanned 1044480
> present 478802
> managed 330969
> --
> min 239
> low 748
> high 1257
> spanned 509952
> present 509952
> managed 509952
> --
> min 1437
> low 4502
> high 7567
> spanned 3145728
> present 3145728
> managed 3065833
> [root@lq-workstation ~]# cat /proc/sys/vm/min_free_kbytes
> 7326
Thank you very much for data! Per my understanding, this verifies a
real problem.
You patch can fix the too small "min" for ZONE_NORMAL/ZONE_DMA. But,
IMHO, it increases "min" for ZONE_MOVABLE unnecessarily. Because we
don't allocate from ZONE_MOVABLE for __GFP_HIGH or PF_MEMALLOC
allocations. So, IMHO, we should treat ZONE_MOVABLE as ZONE_HIGHMEM in
__setup_per_zone_wmarks().
Best Regards,
Huang, Ying
> After this patch is added, the configuration of the movable zone no
> longer affects the size of the min_free_kbytes, which is only affected
> by the size of the available memory.
>>> Signed-off-by: liuq <liuq131@xxxxxxxxxxxxxxx>
>>> ---
>>> include/linux/mm.h | 1 +
>>> mm/khugepaged.c | 2 +-
>>> mm/page_alloc.c | 15 ++++++++++++++-
>>> 3 files changed, 16 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>> index cf3d0d673f6b..1f91d035bcaf 100644
>>> --- a/include/linux/mm.h
>>> +++ b/include/linux/mm.h
>>> @@ -863,6 +863,7 @@ void split_page(struct page *page, unsigned int order);
>>> void folio_copy(struct folio *dst, struct folio *src);
>>> unsigned long nr_free_buffer_pages(void);
>>> +unsigned long nr_free_pagecache_pages(void);
>>> /*
>>> * Compound pages have a destructor function. Provide a
>>> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>>> index 16be62d493cd..6632264b951c 100644
>>> --- a/mm/khugepaged.c
>>> +++ b/mm/khugepaged.c
>>> @@ -2342,7 +2342,7 @@ static void set_recommended_min_free_kbytes(void)
>>> /* don't ever allow to reserve more than 5% of the lowmem */
>>> recommended_min = min(recommended_min,
>>> - (unsigned long) nr_free_buffer_pages() / 20);
>>> + (unsigned long) nr_free_pagecache_pages() / 20);
>>> recommended_min <<= (PAGE_SHIFT-10);
>>> if (recommended_min > min_free_kbytes) {
>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>> index e008a3df0485..489b564526dd 100644
>>> --- a/mm/page_alloc.c
>>> +++ b/mm/page_alloc.c
>>> @@ -5775,6 +5775,19 @@ unsigned long nr_free_buffer_pages(void)
>>> }
>>> EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
>>> +/**
>>> + * nr_free_pagecache_pages - count number of pages beyond high watermark
>>> + *
>>> + * nr_free_pagecache_pages() counts the number of pages which are beyond the
>>> + * high watermark within all zones.
>>> + *
>>> + * Return: number of pages beyond high watermark within all zones.
>>> + */
>>> +unsigned long nr_free_pagecache_pages(void)
>>> +{
>>> + return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
>>> +}
>>> +
>>> static inline void show_node(struct zone *zone)
>>> {
>>> if (IS_ENABLED(CONFIG_NUMA))
>>> @@ -8651,7 +8664,7 @@ void calculate_min_free_kbytes(void)
>>> unsigned long lowmem_kbytes;
>>> int new_min_free_kbytes;
>>> - lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
>>> + lowmem_kbytes = nr_free_pagecache_pages() * (PAGE_SIZE >> 10);
>>> new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
>>> if (new_min_free_kbytes > user_min_free_kbytes)