Re: [External] Re: [PATCH v2 3/6] mm: hugetlb: fix a race between freeing and dissolving the page

[Muchun Song <songmuchun@bytedance.com>]

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?

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.

> --
> Michal Hocko
> SUSE Labs