On Fri, 26 Apr 2019 18:26:42 +0800 Yafang Shao <laoar.shao@xxxxxxxxx> wrote:
> Recently there're some hungtasks on our server due to
> wait_on_page_writeback, and we want to know the details of this
> PG_writeback, i.e. this page is writing back to which device.
> But it is not so convenient to get the details.
>
> I think it would be better to introduce a tracepoint for diagnosing
> the writeback details.
Fair enough, I guess.
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -537,15 +537,7 @@ static inline int wait_on_page_locked_killable(struct page *page)
>
> extern void put_and_wait_on_page_locked(struct page *page);
>
> -/*
> - * Wait for a page to complete writeback
> - */
> -static inline void wait_on_page_writeback(struct page *page)
> -{
> - if (PageWriteback(page))
> - wait_on_page_bit(page, PG_writeback);
> -}
> -
> +void wait_on_page_writeback(struct page *page);
> extern void end_page_writeback(struct page *page);
> void wait_for_stable_page(struct page *page);
>
> ...
>
> +/*
> + * Wait for a page to complete writeback
> + */
> +void wait_on_page_writeback(struct page *page)
> +{
> + if (PageWriteback(page)) {
> + trace_wait_on_page_writeback(page, page_mapping(page));
> + wait_on_page_bit(page, PG_writeback);
> + }
> +}
> +EXPORT_SYMBOL_GPL(wait_on_page_writeback);
But this is a stealth change to the wait_on_page_writeback() licensing.
I will get sad emails from developers of accidentally-broken
out-of-tree filesystems.
We can discuss changing the licensing, but this isn't the way to do it!
--- a/mm/page-writeback.c~mm-page-writeback-introduce-tracepoint-for-wait_on_page_writeback-fix
+++ a/mm/page-writeback.c
@@ -2818,7 +2818,7 @@ void wait_on_page_writeback(struct page
wait_on_page_bit(page, PG_writeback);
}
}
-EXPORT_SYMBOL_GPL(wait_on_page_writeback);
+EXPORT_SYMBOL(wait_on_page_writeback);
/**
* wait_for_stable_page() - wait for writeback to finish, if necessary.
_
