Re: [PATCH 09/17] gup: Convert gup_pte_range() to use a folio

[John Hubbard <jhubbard@nvidia.com>]

On 1/2/22 13:57, Matthew Wilcox (Oracle) wrote:
> We still call try_grab_folio() once per PTE; a future patch could
> optimise to just adjust the reference count for each page within
> the folio.
> 
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>   mm/gup.c | 16 +++++++---------
>   1 file changed, 7 insertions(+), 9 deletions(-)
> 
> diff --git a/mm/gup.c b/mm/gup.c
> index 2307b2917055..d8535f9d5622 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -2260,7 +2260,8 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
>   	ptem = ptep = pte_offset_map(&pmd, addr);
>   	do {
>   		pte_t pte = ptep_get_lockless(ptep);
> -		struct page *head, *page;
> +		struct page *page;
> +		struct folio *folio;
>   
>   		/*
>   		 * Similar to the PMD case below, NUMA hinting must take slow
> @@ -2287,22 +2288,20 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
>   		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
>   		page = pte_page(pte);
>   
> -		head = try_grab_compound_head(page, 1, flags);
> -		if (!head)
> +		folio = try_grab_folio(page, 1, flags);
> +		if (!folio)
>   			goto pte_unmap;
>   
>   		if (unlikely(page_is_secretmem(page))) {
> -			put_compound_head(head, 1, flags);
> +			gup_put_folio(folio, 1, flags);
>   			goto pte_unmap;
>   		}
>   
>   		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
> -			put_compound_head(head, 1, flags);
> +			gup_put_folio(folio, 1, flags);
>   			goto pte_unmap;
>   		}
>   
> -		VM_BUG_ON_PAGE(compound_head(page) != head, page);
> -
>   		/*
>   		 * We need to make the page accessible if and only if we are
>   		 * going to access its content (the FOLL_PIN case).  Please
> @@ -2316,10 +2315,9 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
>   				goto pte_unmap;
>   			}
>   		}
> -		SetPageReferenced(page);
> +		folio_set_referenced(folio);

For the case of a tail page, I *think* the above hunk changes the
behavior. Previously, the tail page's flags were affected, but now, the
head page's (folio's) page flags are being changed...right?


thanks,
-- 
John Hubbard
NVIDIA

>   		pages[*nr] = page;
>   		(*nr)++;
> -
>   	} while (ptep++, addr += PAGE_SIZE, addr != end);
>   
>   	ret = 1;