[v2] introduce pte_offset_map_{ro|rw}_nolock()

[PATCH v2 05/14] mm: khugepaged: __collapse_huge_page_swapin() use pte_offset_map_ro_nolock()

Posted by Qi Zheng 1 year, 5 months ago

In __collapse_huge_page_swapin(), we just use the ptl for pte_same() check
in do_swap_page(). In other places, we directly use pte_offset_map_lock(),
so convert it to using pte_offset_map_ro_nolock().

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
---
 mm/khugepaged.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 4a83c40d90538..53bfa7f4b7f82 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1011,7 +1011,11 @@ static int __collapse_huge_page_swapin(struct mm_struct *mm,
 		};
 
 		if (!pte++) {
-			pte = pte_offset_map_nolock(mm, pmd, address, &ptl);
+			/*
+			 * Here the ptl is only used to check pte_same() in
+			 * do_swap_page(), so readonly version is enough.
+			 */
+			pte = pte_offset_map_ro_nolock(mm, pmd, address, &ptl);
 			if (!pte) {
 				mmap_read_unlock(mm);
 				result = SCAN_PMD_NULL;
-- 
2.20.1

Re: [PATCH v2 05/14] mm: khugepaged: __collapse_huge_page_swapin() use pte_offset_map_ro_nolock()

Posted by Muchun Song 1 year, 5 months ago


> On Aug 22, 2024, at 15:13, Qi Zheng <zhengqi.arch@bytedance.com> wrote:
> 
> In __collapse_huge_page_swapin(), we just use the ptl for pte_same() check
> in do_swap_page(). In other places, we directly use pte_offset_map_lock(),
> so convert it to using pte_offset_map_ro_nolock().
> 
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>

Reviewed-by: Muchun Song <muchun.song@linux.dev>

Re: [PATCH v2 05/14] mm: khugepaged: __collapse_huge_page_swapin() use pte_offset_map_ro_nolock()

Posted by David Hildenbrand 1 year, 5 months ago

On 22.08.24 09:13, Qi Zheng wrote:
> In __collapse_huge_page_swapin(), we just use the ptl for pte_same() check
> in do_swap_page(). In other places, we directly use pte_offset_map_lock(),
> so convert it to using pte_offset_map_ro_nolock().
> 
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
>   mm/khugepaged.c | 6 +++++-
>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 4a83c40d90538..53bfa7f4b7f82 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1011,7 +1011,11 @@ static int __collapse_huge_page_swapin(struct mm_struct *mm,
>   		};
>   
>   		if (!pte++) {
> -			pte = pte_offset_map_nolock(mm, pmd, address, &ptl);
> +			/*
> +			 * Here the ptl is only used to check pte_same() in
> +			 * do_swap_page(), so readonly version is enough.
> +			 */
> +			pte = pte_offset_map_ro_nolock(mm, pmd, address, &ptl);
>   			if (!pte) {
>   				mmap_read_unlock(mm);
>   				result = SCAN_PMD_NULL;

Suboptimal that the pteval comparison + unmap is buried in 
do_swap_page(). Moving that to the caller is also not significantly 
better ...

Acked-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb