From: Muchun Song <songmuchun@bytedance.com>
To: Michal Hocko <mhocko@suse.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>,
Andrew Morton <akpm@linux-foundation.org>,
Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>,
Andi Kleen <ak@linux.intel.com>,
Linux Memory Management List <linux-mm@kvack.org>,
LKML <linux-kernel@vger.kernel.org>
Subject: Re: [External] Re: [PATCH v2 3/6] mm: hugetlb: fix a race between freeing and dissolving the page
Date: Fri, 8 Jan 2021 17:01:03 +0800 [thread overview]
Message-ID: <CAMZfGtUy740SbsCW_h1NaP5O=ahSZniezkC+62pxZ5bW+vZpBg@mail.gmail.com> (raw)
In-Reply-To: <20210108084330.GW13207@dhcp22.suse.cz>
On Fri, Jan 8, 2021 at 4:43 PM Michal Hocko <mhocko@suse.com> wrote:
>
> On Thu 07-01-21 23:11:22, Muchun Song wrote:
> > On Thu, Jan 7, 2021 at 10:11 PM Michal Hocko <mhocko@suse.com> wrote:
> > >
> > > On Thu 07-01-21 20:59:33, Muchun Song wrote:
> > > > On Thu, Jan 7, 2021 at 8:38 PM Michal Hocko <mhocko@suse.com> wrote:
> > > [...]
> > > > > Right. Can we simply back off in the dissolving path when ref count is
> > > > > 0 && PageHuge() if list_empty(page->lru)? Is there any other scenario
> > > > > when the all above is true and the page is not being freed?
> > > >
> > > > The list_empty(&page->lru) may always return false.
> > > > The page before freeing is on the active list
> > > > (hstate->hugepage_activelist).Then it is on the free list
> > > > after freeing. So list_empty(&page->lru) is always false.
> > >
> > > The point I was trying to make is that the page has to be enqueued when
> > > it is dissolved and freed. If the page is not enqueued then something
> > > racing. But then I have realized that this is not a great check to
> > > detect the race because pages are going to be released to buddy
> > > allocator and that will reuse page->lru again. So scratch that and sorry
> > > for the detour.
> > >
> > > But that made me think some more and one way to reliably detect the race
> > > should be PageHuge() check in the freeing path. This is what dissolve
> > > path does already. PageHuge becomes false during update_and_free_page()
> > > while holding the hugetlb_lock. So can we use that?
> >
> > It may make the thing complex. Apart from freeing it to the
> > buddy allocator, free_huge_page also does something else for
> > us. If we detect the race in the freeing path, if it is not a HugeTLB
> > page, the freeing path just returns. We also should move those
> > things to the dissolve path. Right?
>
> Not sure what you mean. Dissolving is a subset of the freeing path. It
> effectivelly only implements the update_and_free_page branch (aka free
> to buddy). It skips some of the existing steps because it believes it
> sees a freed page. But as you have pointed out this is racy and I
> strongly suspect it is simply wrong in those assumptions. E.g. hugetlb
> cgroup accounting can get wrong right?
OK. I know what you mean. The update_and_free_page should
do the freeing which is similar to __free_huge_page().
>
> The more I think about it the more I think that dissolving path should
> simply have a common helper with __free_huge_page that would release
> the huge page to the allocator. The only thing that should be specific
> to the dissolving path is HWpoison handling. It shouldn't be playing
> with accounting and what not. Btw the HWpoison handling is suspicious as
> well, a lost race would mean this doesn't happen. But maybe there is
> some fixup handled later on...
>
> > But I find a tricky problem to solve. See free_huge_page().
> > If we are in non-task context, we should schedule a work
> > to free the page. We reuse the page->mapping. If the page
> > is already freed by the dissolve path. We should not touch
> > the page->mapping. So we need to check PageHuge().
> > The check and llist_add() should be protected by
> > hugetlb_lock. But we cannot do that. Right? If dissolve
> > happens after it is linked to the list. We also should
> > remove it from the list (hpage_freelist). It seems to make
> > the thing more complex.
>
> I am not sure I follow you here but yes PageHuge under hugetlb_lock
> should be the reliable way to check for the race. I am not sure why we
> really need to care about mapping or other state.
CPU0: CPU1:
free_huge_page(page)
if (PageHuge(page))
dissolve_free_huge_page(page)
spin_lock(&hugetlb_lock)
update_and_free_page(page)
spin_unlock(&hugetlb_lock)
llist_add(page->mapping)
// the mapping is corrupted
The PageHuge(page) and llist_add() should be protected by
hugetlb_lock. Right? If so, we cannot hold hugetlb_lock
in free_huge_page() path.
> --
> Michal Hocko
> SUSE Labs
next prev parent reply other threads:[~2021-01-08 9:02 UTC|newest]
Thread overview: 56+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-06 8:47 [PATCH v2 0/6] Fix some bugs about HugeTLB code Muchun Song
2021-01-06 8:47 ` [PATCH v2 1/6] mm: migrate: do not migrate HugeTLB page whose refcount is one Muchun Song
2021-01-06 16:13 ` Michal Hocko
2021-01-07 2:52 ` [External] " Muchun Song
2021-01-07 11:16 ` Michal Hocko
2021-01-07 11:24 ` Muchun Song
2021-01-07 11:48 ` Michal Hocko
2021-01-06 8:47 ` [PATCH v2 2/6] mm: hugetlbfs: fix cannot migrate the fallocated HugeTLB page Muchun Song
2021-01-06 16:35 ` Michal Hocko
2021-01-06 19:30 ` Mike Kravetz
2021-01-06 20:02 ` Michal Hocko
2021-01-06 21:07 ` Mike Kravetz
2021-01-07 8:36 ` Michal Hocko
2021-01-07 2:58 ` [External] " Muchun Song
2021-01-06 8:47 ` [PATCH v2 3/6] mm: hugetlb: fix a race between freeing and dissolving the page Muchun Song
2021-01-06 16:56 ` Michal Hocko
2021-01-06 20:58 ` Mike Kravetz
2021-01-07 3:08 ` [External] " Muchun Song
2021-01-07 8:40 ` Michal Hocko
2021-01-08 0:52 ` Mike Kravetz
2021-01-08 8:28 ` Michal Hocko
2021-01-07 5:39 ` [External] " Muchun Song
2021-01-07 8:41 ` Michal Hocko
2021-01-07 8:53 ` Muchun Song
2021-01-07 11:18 ` Michal Hocko
2021-01-07 11:38 ` Muchun Song
2021-01-07 12:38 ` Michal Hocko
2021-01-07 12:59 ` Muchun Song
2021-01-07 14:11 ` Michal Hocko
2021-01-07 15:11 ` Muchun Song
2021-01-08 1:06 ` Mike Kravetz
2021-01-08 2:38 ` Muchun Song
2021-01-08 8:43 ` Michal Hocko
2021-01-08 9:01 ` Muchun Song [this message]
2021-01-08 9:31 ` Michal Hocko
2021-01-08 10:08 ` Muchun Song
2021-01-08 11:44 ` Michal Hocko
2021-01-08 11:52 ` Muchun Song
2021-01-08 12:04 ` Michal Hocko
2021-01-08 12:24 ` Muchun Song
2021-01-06 8:47 ` [PATCH v2 4/6] mm: hugetlb: add return -EAGAIN for dissolve_free_huge_page Muchun Song
2021-01-06 17:07 ` Michal Hocko
2021-01-07 3:11 ` [External] " Muchun Song
2021-01-07 8:39 ` Michal Hocko
2021-01-07 9:01 ` Muchun Song
2021-01-07 11:22 ` Michal Hocko
2021-01-06 8:47 ` [PATCH v2 5/6] mm: hugetlb: fix a race between isolating and freeing page Muchun Song
2021-01-06 17:10 ` Michal Hocko
2021-01-06 8:47 ` [PATCH v2 6/6] mm: hugetlb: remove VM_BUG_ON_PAGE from page_huge_active Muchun Song
2021-01-06 17:16 ` Michal Hocko
2021-01-08 22:24 ` Mike Kravetz
2021-01-09 4:07 ` [External] " Muchun Song
2021-01-07 9:30 ` [PATCH v2 0/6] Fix some bugs about HugeTLB code David Hildenbrand
2021-01-07 9:40 ` [External] " Muchun Song
2021-01-07 10:10 ` David Hildenbrand
2021-01-07 10:16 ` Muchun Song
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAMZfGtUy740SbsCW_h1NaP5O=ahSZniezkC+62pxZ5bW+vZpBg@mail.gmail.com' \
--to=songmuchun@bytedance.com \
--cc=ak@linux.intel.com \
--cc=akpm@linux-foundation.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@suse.com \
--cc=mike.kravetz@oracle.com \
--cc=n-horiguchi@ah.jp.nec.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).