Re: [PATCH 10/10] mm/migrate.c: call detach_page_private to cleanup code

[Andrew Morton]

On Fri, 22 May 2020 09:18:25 +0200 Guoqing Jiang <guoqing.jiang@xxxxxxxxxxxxxxx> wrote:

> >> -	ClearPagePrivate(page);
> >> -	set_page_private(newpage, page_private(page));
> >> -	set_page_private(page, 0);
> >> -	put_page(page);
> >> +	set_page_private(newpage, detach_page_private(page));
> > attach_page_private(newpage, detach_page_private(page));
> 
> Mattew had suggested it as follows, but not sure if we can reorder of 
> the setup of
> the bh and setting PagePrivate, so I didn't want to break the original 
> syntax.
> 
> @@ -797,11 +797,7 @@ static int __buffer_migrate_page(struct address_space *mapping,
>          if (rc != MIGRATEPAGE_SUCCESS)
>                  goto unlock_buffers;
>   
> -       ClearPagePrivate(page);
> -       set_page_private(newpage, page_private(page));
> -       set_page_private(page, 0);
> -       put_page(page);
> -       get_page(newpage);
> +       attach_page_private(newpage, detach_page_private(page));
>   
>          bh = head;
>          do {
> @@ -810,8 +806,6 @@ static int __buffer_migrate_page(struct address_space *mapping,
>   
>          } while (bh != head);
>   
> -       SetPagePrivate(newpage);
> -
>          if (mode != MIGRATE_SYNC_NO_COPY)

This is OK - coherency between PG_private and the page's buffer
ring is maintained by holding lock_page().

I have (effectively) applied the above change.