Re: [PATCH v5 1/2] mm,hwpoison: fix race with hugetlb page allocation
From: Oscar Salvador
Date: Tue May 25 2021 - 05:09:34 EST
On Tue, May 25, 2021 at 08:07:07AM +0000, HORIGUCHI NAOYA(堀口 直也) wrote:
> OK, here's the current draft.
> Naoya Horiguchi
> From: Naoya Horiguchi <naoya.horiguchi@xxxxxxx>
> Date: Tue, 18 May 2021 23:49:18 +0900
> Subject: [PATCH] mm,hwpoison: fix race with hugetlb page allocation
> When hugetlb page fault (under overcommitting situation) and
> memory_failure() race, VM_BUG_ON_PAGE() is triggered by the following race:
> CPU0: CPU1:
> page = alloc_surplus_huge_page()
> zero = put_page_testzero(page)
> VM_BUG_ON_PAGE(!zero, page)
> enqueue_huge_page(h, page)
> __get_hwpoison_page() only checks the page refcount before taking an
> additional one for memory error handling, which is wrong because there's
> a time window where compound pages have non-zero refcount during
> initialization. So make __get_hwpoison_page() check page status a bit
> more for hugetlb pages.
I think that this changelog would benefit from some information about the new
!PageLRU && !__PageMovable check.
> static int __get_hwpoison_page(struct page *page)
> struct page *head = compound_head(page);
> + int ret = 0;
> + bool hugetlb = false;
> + ret = get_hwpoison_huge_page(head, &hugetlb);
> + if (hugetlb)
> + return ret;
> + if (!PageLRU(head) && !__PageMovable(head))
> + return 0;
This definitely needs a comment hinting the reader why we need to check for this.
AFAICS, this is to close the race where a page is about to be a hugetlb page soon,
so we do not go for get_page_unless_zero(), right?
>From soft_offline_page's POV I __guess__ that's fine because we only deal with
pages we know about.
But what about memory_failure()? I think memory_failure() is less picky about that,
so it is okay to not take a refcount on that case?