Re: [PATCH v1] mm, hwpoison: add is_free_buddy_page() in HWPoisonHandlable()

From: HORIGUCHI NAOYA(堀口 直也)
Date: Wed Sep 08 2021 - 23:58:13 EST


On Wed, Sep 08, 2021 at 08:22:14PM -0700, Yang Shi wrote:
> On Wed, Sep 8, 2021 at 5:41 PM Naoya Horiguchi
> <naoya.horiguchi@xxxxxxxxx> wrote:
> >
> > From: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
> >
> > commit fcc00621d88b ("mm/hwpoison: retry with shake_page() for
> > unhandlable pages") changes the return value of __get_hwpoison_page() to
> > retry for transiently unhandlable cases. However, __get_hwpoison_page()
> > currently fails to properly judge buddy pages as handlable, so hard/soft
> > offline for buddy pages always fail as "unhandlable page". This is
> > totally regrettable.
> >
> > So let's add is_free_buddy_page() in HWPoisonHandlable(), so that
> > __get_hwpoison_page() returns different return values between buddy
> > pages and unhandlable pages as intended.
> >
> > Fixes: fcc00621d88b ("mm/hwpoison: retry with shake_page() for unhandlable pages")
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > Signed-off-by: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
> > ---
> > mm/memory-failure.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git v5.14-rc7-mmotm-2021-08-23-16-42/mm/memory-failure.c v5.14-rc7-mmotm-2021-08-23-16-42_patched/mm/memory-failure.c
> > index 60df8fcd0444..3416c55be810 100644
> > --- v5.14-rc7-mmotm-2021-08-23-16-42/mm/memory-failure.c
> > +++ v5.14-rc7-mmotm-2021-08-23-16-42_patched/mm/memory-failure.c
> > @@ -1126,7 +1126,7 @@ static int page_action(struct page_state *ps, struct page *p,
> > */
> > static inline bool HWPoisonHandlable(struct page *page)
> > {
> > - return PageLRU(page) || __PageMovable(page);
> > + return PageLRU(page) || __PageMovable(page) || is_free_buddy_page(page);
>
> It seems to work. Although this may change the return value of
> get_any_page() to 1 when MF_COUNT_INCREASED is set. This may cause
> soft offline to mishandle free buddy page, but MF_COUNT_INCREASED is
> only set when madvise is used, and madvise definitely can't soft
> offline free buddy page.

Yes, madvise(MADV_SOFT_OFFLINE) should not be called for free buddy pages,
because the interface covers only user-mapped pages.

> It did take me a while to figure out this
> trick. Maybe need some refactor?

I think so, the current code looks to me fragile for future change, but
I'm not sure how we can improve this.

One (maybe not related to MF_COUNT_INCREASED) room for refactoring is
"if (PageTransHuge(head))" block in __get_hwpoison_page(), which might
be old and useless code for now.

>
> Anyway this patch looks fine to me. Reviewed-by: Yang Shi <shy828301@xxxxxxxxx>

Thank you for the review.

- Naoya