Re: [PATCH 06/10] hugetlbfs: Convert remove_inode_hugepages() to use filemap_get_folios()

[Mike Kravetz]

On 6/10/22 14:17, Matthew Wilcox wrote:
> On Fri, Jun 10, 2022 at 05:52:05PM +0200, Sumanth Korikkar wrote:
>> To reproduce:
>> * clone libhugetlbfs:
>> * Execute, PATH=$PATH:"obj64/" LD_LIBRARY_PATH=../obj64/ alloc-instantiate-race shared
> 
> ... it's a lot harder to set up hugetlb than that ...
> 
> anyway, i figured it out without being able to run the reproducer.
> 
> Can you try this?

I can confirm that libhugetlbfs tests do not trigger the BUG with the
below change.
-- 
Mike Kravetz

> 
> diff --git a/mm/filemap.c b/mm/filemap.c
> index a30587f2e598..8ef861297ffb 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -2160,7 +2160,11 @@ unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
>  		if (xa_is_value(folio))
>  			continue;
>  		if (!folio_batch_add(fbatch, folio)) {
> -			*start = folio->index + folio_nr_pages(folio);
> +			unsigned long nr = folio_nr_pages(folio);
> +
> +			if (folio_test_hugetlb(folio))
> +				nr = 1;
> +			*start = folio->index + nr;
>  			goto out;
>  		}
>  	}
>