2023-03-03 15:13:14

by Peter Xu

Subject: [PATCH] mm/khugepaged: Cleanup memcg uncharge for failure path

Explicit memcg uncharging is not needed when the memcg accounting has the
same lifespan of the page/folio. That becomes the case for khugepaged
after Yang & Zach's recent rework so the hpage will be allocated for each
collapse rather than being cached.

Cleanup the explicit memcg uncharge in khugepaged failure path and leave
that for put_page().

Suggested-by: Zach O'Keefe <[email protected]>
Signed-off-by: Peter Xu <[email protected]>
---
mm/khugepaged.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 941d1c7ea910..dd5a7d9bc593 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1230,10 +1230,8 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
out_up_write:
mmap_write_unlock(mm);
out_nolock:
- if (hpage) {
- mem_cgroup_uncharge(page_folio(hpage));
+ if (hpage)
put_page(hpage);
- }
trace_mm_collapse_huge_page(mm, result == SCAN_SUCCEED, result);
return result;
}
@@ -2250,10 +2248,8 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
unlock_page(hpage);
out:
VM_BUG_ON(!list_empty(&pagelist));
- if (hpage) {
- mem_cgroup_uncharge(page_folio(hpage));
+ if (hpage)
put_page(hpage);
- }

trace_mm_khugepaged_collapse_file(mm, hpage, index, is_shmem, addr, file, nr, result);
return result;
--
2.39.1

2023-03-03 15:23:06

by Zach O'Keefe

Subject: Re: [PATCH] mm/khugepaged: Cleanup memcg uncharge for failure path

Thanks Peter!

On Mar 03 10:12, Peter Xu wrote:
> Explicit memcg uncharging is not needed when the memcg accounting has the
> same lifespan of the page/folio. That becomes the case for khugepaged
> after Yang & Zach's recent rework so the hpage will be allocated for each
> collapse rather than being cached.
>
> Cleanup the explicit memcg uncharge in khugepaged failure path and leave
> that for put_page().
>
> Suggested-by: Zach O'Keefe <[email protected]>
> Signed-off-by: Peter Xu <[email protected]>

Reviewed-by: Zach O'Keefe <[email protected]>

> ---
> mm/khugepaged.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 941d1c7ea910..dd5a7d9bc593 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1230,10 +1230,8 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
> out_up_write:
> mmap_write_unlock(mm);
> out_nolock:
> - if (hpage) {
> - mem_cgroup_uncharge(page_folio(hpage));
> + if (hpage)
> put_page(hpage);
> - }
> trace_mm_collapse_huge_page(mm, result == SCAN_SUCCEED, result);
> return result;
> }
> @@ -2250,10 +2248,8 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
> unlock_page(hpage);
> out:
> VM_BUG_ON(!list_empty(&pagelist));
> - if (hpage) {
> - mem_cgroup_uncharge(page_folio(hpage));
> + if (hpage)
> put_page(hpage);
> - }
>
> trace_mm_khugepaged_collapse_file(mm, hpage, index, is_shmem, addr, file, nr, result);
> return result;
> --
> 2.39.1
>

2023-03-03 19:00:42

by Yang Shi

Subject: Re: [PATCH] mm/khugepaged: Cleanup memcg uncharge for failure path

On Fri, Mar 3, 2023 at 7:12 AM Peter Xu <[email protected]> wrote:
>
> Explicit memcg uncharging is not needed when the memcg accounting has the
> same lifespan of the page/folio. That becomes the case for khugepaged
> after Yang & Zach's recent rework so the hpage will be allocated for each
> collapse rather than being cached.
>
> Cleanup the explicit memcg uncharge in khugepaged failure path and leave
> that for put_page().

Thanks for the cleanup. Reviewed-by: Yang Shi <[email protected]>

>
> Suggested-by: Zach O'Keefe <[email protected]>
> Signed-off-by: Peter Xu <[email protected]>
> ---
> mm/khugepaged.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 941d1c7ea910..dd5a7d9bc593 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1230,10 +1230,8 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
> out_up_write:
> mmap_write_unlock(mm);
> out_nolock:
> - if (hpage) {
> - mem_cgroup_uncharge(page_folio(hpage));
> + if (hpage)
> put_page(hpage);
> - }
> trace_mm_collapse_huge_page(mm, result == SCAN_SUCCEED, result);
> return result;
> }
> @@ -2250,10 +2248,8 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
> unlock_page(hpage);
> out:
> VM_BUG_ON(!list_empty(&pagelist));
> - if (hpage) {
> - mem_cgroup_uncharge(page_folio(hpage));
> + if (hpage)
> put_page(hpage);
> - }
>
> trace_mm_khugepaged_collapse_file(mm, hpage, index, is_shmem, addr, file, nr, result);
> return result;
> --
> 2.39.1
>