Re: [PATCH v3] mm/hugetlb: fix set_max_huge_pages() when there are surplus pages

[Jinjiang Tu <tujinjiang@huawei.com>]

在 2025/4/8 21:10, Oscar Salvador 写道:
> On Mon, Apr 07, 2025 at 08:47:06PM +0800, Jinjiang Tu wrote:
>> In set_max_huge_pages(), min_count should mean the acquired persistent
>> huge pages, but it contains surplus huge pages. It will leads to failing
>> to freeing free huge pages for a Node.
>>
>> Steps to reproduce:
>> 1) create 5 hugetlb folios in Node0
>> 2) run a program to use all the hugetlb folios
>> 3) echo 0 > nr_hugepages for Node0 to free the hugetlb folios. Thus the 5
>> hugetlb folios in Node0 are accounted as surplus.
>> 4) create 5 hugetlb folios in Node1
>> 5) echo 0 > nr_hugepages for Node1 to free the hugetlb folios
>>
>> The result:
>>          Node0    Node1
>> Total     5         5
>> Free      0         5
>> Surp      5         5
> I would put this after the explanation, as otherwise is a bit hard to
> follow.
>
>> We couldn't subtract surplus_huge_pages from min_mount, since free hugetlb
>> folios may be surplus due to HVO. In __update_and_free_hugetlb_folio(),
>> hugetlb_vmemmap_restore_folio() may fail, add the folio back to pool and
>> treat it as surplus. If we directly subtract surplus_huge_pages from
>> min_mount, some free folios will be subtracted twice.
>>
>> To fix it, check if count is less than the num of free huge pages that
>> could be destroyed (i.e., available_huge_pages(h)), and remove hugetlb
>> folios if so.
> But this is not true, you are no longer comparing against
> available_huge_pages(h) as you did in v2.
>
> I would go with something along these lines as changelog.
>
> "In set_max_huge_pages(), min_count is computed taking into account also
>   surplushuge pages, which might lead in some cases to not be able to free
>   huge pages and end up accounting them as surplus intead.
>
>   One way to solve it is to substract surplus_huge_pages directly, but we
>   cannot do it blindly because there might be surplus pages thar are also
>   free pages, which might happen when we fail to restore the vmemmap for
>   optimized hvo pages.
>   So we could be subtracting the same page twice.
>
>   In order to work this around, let us first compute the number of
>   free persistent pages, and use that along with surplus pages
>   to compute min_count."
>
>   And then put the PoC.

Thanks very much, I will update the changelog and send a new version later.

>> Fixes: 9a30523066cd ("hugetlb: add per node hstate attributes")
>> Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
> Acked-by: Oscar Salvador <osalvador@suse.de>
>
>