Re: [External] [PATCH v3 7/8] hugetlb: make free_huge_page irq safe

[Muchun Song <songmuchun@bytedance.com>]

On Sat, Apr 3, 2021 at 4:56 AM Mike Kravetz <mike.kravetz@oracle.com> wrote:
>
> On 4/2/21 5:47 AM, Muchun Song wrote:
> > On Wed, Mar 31, 2021 at 11:42 AM Mike Kravetz <mike.kravetz@oracle.com> wrote:
> >>
> >> Commit c77c0a8ac4c5 ("mm/hugetlb: defer freeing of huge pages if in
> >> non-task context") was added to address the issue of free_huge_page
> >> being called from irq context.  That commit hands off free_huge_page
> >> processing to a workqueue if !in_task.  However, this doesn't cover
> >> all the cases as pointed out by 0day bot lockdep report [1].
> >>
> >> :  Possible interrupt unsafe locking scenario:
> >> :
> >> :        CPU0                    CPU1
> >> :        ----                    ----
> >> :   lock(hugetlb_lock);
> >> :                                local_irq_disable();
> >> :                                lock(slock-AF_INET);
> >> :                                lock(hugetlb_lock);
> >> :   <Interrupt>
> >> :     lock(slock-AF_INET);
> >>
> >> Shakeel has later explained that this is very likely TCP TX zerocopy
> >> from hugetlb pages scenario when the networking code drops a last
> >> reference to hugetlb page while having IRQ disabled. Hugetlb freeing
> >> path doesn't disable IRQ while holding hugetlb_lock so a lock dependency
> >> chain can lead to a deadlock.
> >>
> >> This commit addresses the issue by doing the following:
> >> - Make hugetlb_lock irq safe.  This is mostly a simple process of
> >>   changing spin_*lock calls to spin_*lock_irq* calls.
> >> - Make subpool lock irq safe in a similar manner.
> >> - Revert the !in_task check and workqueue handoff.
> >>
> >> [1] https://lore.kernel.org/linux-mm/000000000000f1c03b05bc43aadc@google.com/
> >>
> >> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
> >> Acked-by: Michal Hocko <mhocko@suse.com>
> >> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
> >
> > Hi Mike,
> >
> > Today I pulled the newest code (next-20210401). I found that
> > alloc_and_dissolve_huge_page is not updated. In this function,
> > hugetlb_lock is still non-irq safe. Maybe you or Oscar need
> > to fix.
> >
> > Thanks.
>
> Thank you Muchun,
>
> Oscar's changes were not in Andrew's tree when I started on this series
> and I failed to notice their inclusion.  In addition,
> isolate_or_dissolve_huge_page also needs updating as well as a change in
> set_max_huge_pages that was omitted while rebasing.
>
> Andrew, the following patch addresses those missing changes.  Ideally,
> the changes should be combined/included in this patch.  If you want me
> to sent another version of this patch or another series, let me know.
>
> From 450593eb3cea895f499ddc343c22424c552ea502 Mon Sep 17 00:00:00 2001
> From: Mike Kravetz <mike.kravetz@oracle.com>
> Date: Fri, 2 Apr 2021 13:18:13 -0700
> Subject: [PATCH] hugetlb: fix irq locking omissions
>
> The pach "hugetlb: make free_huge_page irq safe" changed spin_*lock
> calls to spin_*lock_irq* calls.  However, it missed several places
> in the file hugetlb.c.  Add the overlooked changes.
>
> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>

Thanks MIke. But there are still some places that need
improvement. See below.

> ---
>  mm/hugetlb.c | 16 ++++++++--------
>  1 file changed, 8 insertions(+), 8 deletions(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index c22111f3da20..a6bfc6bcbc81 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -2284,7 +2284,7 @@ static int alloc_and_dissolve_huge_page(struct hstate *h, struct page *old_page,
>          */
>         page_ref_dec(new_page);
>  retry:
> -       spin_lock(&hugetlb_lock);
> +       spin_lock_irq(&hugetlb_lock);
>         if (!PageHuge(old_page)) {
>                 /*
>                  * Freed from under us. Drop new_page too.
> @@ -2297,7 +2297,7 @@ static int alloc_and_dissolve_huge_page(struct hstate *h, struct page *old_page,
>                  * Fail with -EBUSY if not possible.
>                  */
>                 update_and_free_page(h, new_page);

Now update_and_free_page can be called without holding
hugetlb_lock. We can move it out of hugetlb_lock. In this
function, there are 3 places which call update_and_free_page().
We can move all of them out of hugetlb_lock. Right?

This optimization can squash to the commit:

    "hugetlb: call update_and_free_page without hugetlb_lock"

Thanks.

> -               spin_unlock(&hugetlb_lock);
> +               spin_unlock_irq(&hugetlb_lock);
>                 if (!isolate_huge_page(old_page, list))
>                         ret = -EBUSY;
>                 return ret;
> @@ -2307,7 +2307,7 @@ static int alloc_and_dissolve_huge_page(struct hstate *h, struct page *old_page,
>                  * freelist yet. Race window is small, so we can succed here if
>                  * we retry.
>                  */
> -               spin_unlock(&hugetlb_lock);
> +               spin_unlock_irq(&hugetlb_lock);
>                 cond_resched();
>                 goto retry;
>         } else {
> @@ -2323,7 +2323,7 @@ static int alloc_and_dissolve_huge_page(struct hstate *h, struct page *old_page,
>                 __enqueue_huge_page(&h->hugepage_freelists[nid], new_page);
>         }
>  unlock:
> -       spin_unlock(&hugetlb_lock);
> +       spin_unlock_irq(&hugetlb_lock);
>
>         return ret;
>  }
> @@ -2339,15 +2339,15 @@ int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list)
>          * to carefully check the state under the lock.
>          * Return success when racing as if we dissolved the page ourselves.
>          */
> -       spin_lock(&hugetlb_lock);
> +       spin_lock_irq(&hugetlb_lock);
>         if (PageHuge(page)) {
>                 head = compound_head(page);
>                 h = page_hstate(head);
>         } else {
> -               spin_unlock(&hugetlb_lock);
> +               spin_unlock_irq(&hugetlb_lock);
>                 return 0;
>         }
> -       spin_unlock(&hugetlb_lock);
> +       spin_unlock_irq(&hugetlb_lock);
>
>         /*
>          * Fence off gigantic pages as there is a cyclic dependency between
> @@ -2737,7 +2737,7 @@ static int set_max_huge_pages(struct hstate *h, unsigned long count, int nid,
>          * pages in hstate via the proc/sysfs interfaces.
>          */
>         mutex_lock(&h->resize_lock);
> -       spin_lock(&hugetlb_lock);
> +       spin_lock_irq(&hugetlb_lock);
>
>         /*
>          * Check for a node specific request.
> --
> 2.30.2
>