Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220705062953.914256-1-luofei@unicloud.com>
In-Reply-To: <20220705062953.914256-1-luofei@unicloud.com>
From:   Muchun Song <songmuchun@bytedance.com>
Date:   Tue, 5 Jul 2022 16:57:35 +0800
Message-ID: <CAMZfGtUOA81Srwp0OLDMx5Nnpz=s_qmG_a3zauzNyd9AYZzReQ@mail.gmail.com>
Subject: Re: [PATCH] mm,hwpoison,hugetlb: defer dissolve hwpoison hugepage
 when allocating vmemmap failed
To:     luofei <luofei@unicloud.com>
Cc:     Mike Kravetz <mike.kravetz@oracle.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jul 5, 2022 at 2:32 PM luofei <luofei@unicloud.com> wrote:
>
> When dissolving hwpoison hugepage, if the allocation of vmemmap page
> failed, the faulty page should not be put back on the hugepage free
> list, which will cause the faulty pages to be reused. It's better to

Hi luofei,

How did it happen? If a hugepage is poisoned, then the head page's
flag will be set to PageHWPoison. See the code of
dequeue_huge_page_node_exact() which will filter out hwpoisoned
page. So the hwpoisoned pages cannot be reused, hopefully, I am
not missing something important.

> postpone the reexecution of dissolve operation.
>
> Meanwhile when the page fault handling program calls
> dissolve_free_huge_page() to dissolve the faulty page, the basic page
> fault processing operation(such as migration pages and unmap etc)
> has actually completed. There is no need to return -ENOMEM error code
> to the upper layer for temporarily vmemmap page allocation failure,
> which will cause the caller to make a wrong judgment. So just defer
> dissolve and return success.
>
> Signed-off-by: luofei <luofei@unicloud.com>
> ---
>  mm/hugetlb.c | 34 +++++++++++++++++++++-------------
>  1 file changed, 21 insertions(+), 13 deletions(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index ca081078e814..db25458eb0a5 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -90,6 +90,9 @@ struct mutex *hugetlb_fault_mutex_table ____cacheline_aligned_in_smp;
>
>  /* Forward declaration */
>  static int hugetlb_acct_memory(struct hstate *h, long delta);
> +static LLIST_HEAD(hpage_freelist);
> +static void free_hpage_workfn(struct work_struct *work);
> +static DECLARE_DELAYED_WORK(free_hpage_work, free_hpage_workfn);
>
>  static inline bool subpool_is_free(struct hugepage_subpool *spool)
>  {
> @@ -1535,15 +1538,21 @@ static void __update_and_free_page(struct hstate *h, struct page *page)
>         if (hstate_is_gigantic(h) && !gigantic_page_runtime_supported())
>                 return;
>
> -       if (hugetlb_vmemmap_restore(h, page))
> +       if (hugetlb_vmemmap_restore(h, page)) {
> +               if (unlikely(PageHWPoison(page))) {
> +                       llist_add((struct llist_node *)&page->mapping, &hpage_freelist);
> +                       schedule_delayed_work(&free_hpage_work, HZ);
> +                       goto out;
> +               }
>                 goto fail;
> +       }
>
>         /*
>          * Move PageHWPoison flag from head page to the raw error pages,
>          * which makes any healthy subpages reusable.
>          */
>         if (unlikely(PageHWPoison(page) && hugetlb_clear_page_hwpoison(page)))
> -               goto fail;
> +               goto out;
>
>         for (i = 0; i < pages_per_huge_page(h);
>              i++, subpage = mem_map_next(subpage, page, i)) {
> @@ -1574,6 +1583,8 @@ static void __update_and_free_page(struct hstate *h, struct page *page)
>          */
>         add_hugetlb_page(h, page, true);
>         spin_unlock_irq(&hugetlb_lock);
> +out:
> +       return;
>  }
>
>  /*
> @@ -1587,8 +1598,6 @@ static void __update_and_free_page(struct hstate *h, struct page *page)
>   * to be cleared in free_hpage_workfn() anyway, it is reused as the llist_node
>   * structure of a lockless linked list of huge pages to be freed.
>   */
> -static LLIST_HEAD(hpage_freelist);
> -
>  static void free_hpage_workfn(struct work_struct *work)
>  {
>         struct llist_node *node;
> @@ -1616,12 +1625,11 @@ static void free_hpage_workfn(struct work_struct *work)
>                 cond_resched();
>         }
>  }
> -static DECLARE_WORK(free_hpage_work, free_hpage_workfn);
>
>  static inline void flush_free_hpage_work(struct hstate *h)
>  {
>         if (hugetlb_vmemmap_optimizable(h))
> -               flush_work(&free_hpage_work);
> +               flush_delayed_work(&free_hpage_work);
>  }
>
>  static void update_and_free_page(struct hstate *h, struct page *page,
> @@ -1634,13 +1642,9 @@ static void update_and_free_page(struct hstate *h, struct page *page,
>
>         /*
>          * Defer freeing to avoid using GFP_ATOMIC to allocate vmemmap pages.
> -        *
> -        * Only call schedule_work() if hpage_freelist is previously
> -        * empty. Otherwise, schedule_work() had been called but the workfn
> -        * hasn't retrieved the list yet.
>          */
> -       if (llist_add((struct llist_node *)&page->mapping, &hpage_freelist))
> -               schedule_work(&free_hpage_work);
> +       llist_add((struct llist_node *)&page->mapping, &hpage_freelist);
> +       schedule_delayed_work(&free_hpage_work, 0);
>  }
>
>  static void update_and_free_pages_bulk(struct hstate *h, struct list_head *list)
> @@ -2118,11 +2122,15 @@ int dissolve_free_huge_page(struct page *page)
>                 rc = hugetlb_vmemmap_restore(h, head);
>                 if (!rc) {
>                         update_and_free_page(h, head, false);
> -               } else {
> +               } else if (!PageHWPoison(head)) {
>                         spin_lock_irq(&hugetlb_lock);
>                         add_hugetlb_page(h, head, false);
>                         h->max_huge_pages++;
>                         spin_unlock_irq(&hugetlb_lock);
> +               } else {
> +                       llist_add((struct llist_node *)&head->mapping, &hpage_freelist);
> +                       schedule_delayed_work(&free_hpage_work, HZ);
> +                       rc = 0;
>                 }
>
>                 return rc;
> --
> 2.27.0
>