On Wed, Jul 07, 2021 at 08:48:28AM +0800, Chao Yu wrote:
> On 2021/7/6 17:12, Mel Gorman wrote:
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index d6e94cc8066c..be87c4be481f 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3515,8 +3515,13 @@ void split_page(struct page *page, unsigned int order)
> VM_BUG_ON_PAGE(PageCompound(page), page);
> VM_BUG_ON_PAGE(!page_count(page), page);
>
> - for (i = 1; i < (1 << order); i++)
> - set_page_refcounted(page + i);
> + for (i = 1; i < (1 << order); i++) {
> + struct page *tail_page = page + i;
> +
> + set_page_refcounted(tail_page);
> + if (WARN_ON_ONCE(tail_page->private))
> + pr_info("order:%x, tailpage.private:%x", order, tail_page->private);
> + }
> split_page_owner(page, 1 << order);
> split_page_memcg(page, 1 << order);
> }
> --
> 2.22.1
>
> With above diff, I got this:
>
> ------------[ cut here ]------------
> WARNING: CPU: 3 PID: 57 at mm/page_alloc.c:3363 split_page.cold+0x8/0x3b
> CPU: 3 PID: 57 Comm: kcompactd0 Tainted: G O 5.13.0-rc1+ #67
> <SNIP>
> order:7, tailpage.private:d0000
> order:7, tailpage.private:d0000
> order:7, tailpage.private:d0000
> order:7, tailpage.private:200000
> order:7, tailpage.private:d0000
> order:7, tailpage.private:d0000
> order:7, tailpage.private:d0000
>
> So how about adding set_page_private(page, 0) in split_page() to clear
> stall data left in tailpages' private field?
>
I think it would work but it would be preferable to find out why the
tail page has an order set in the first place. I've looked over
mm/page_alloc.c and mm/compaction.c a few times and did not spot where
set_private_page(page, 0) is missed when it should be covered by
clear_page_guard or del_page_from_free_list :(
--
Mel Gorman
SUSE Labs
