Re: [PATCH] mm: thp: check total_mapcount instead of page_mapcount

[Yang Shi]

On Fri, Apr 30, 2021 at 2:30 PM Zi Yan <ziy@xxxxxxxxxx> wrote:
>
> On 30 Apr 2021, at 17:07, Yang Shi wrote:
>
> > When debugging the bug reported by Wang Yugui [1], try_to_unmap() may
> > return false positive for PTE-mapped THP since page_mapcount() is used
> > to check if the THP is unmapped, but it just checks compound mapount and
> > head page's mapcount.  If the THP is PTE-mapped and head page is not
> > mapped, it may return false positive.
> >
> > Use total_mapcount() instead of page_mapcount() and do so for the
> > VM_BUG_ON_PAGE in split_huge_page_to_list as well.
> >
> > [1] https://lore.kernel.org/linux-mm/20210412180659.B9E3.409509F4@xxxxxxxxxxxx/
> >
> > Signed-off-by: Yang Shi <shy828301@xxxxxxxxx>
> > ---
> >  mm/huge_memory.c | 2 +-
> >  mm/rmap.c        | 2 +-
> >  2 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > index 63ed6b25deaa..2122c3e853b9 100644
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -2718,7 +2718,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
> >       }
> >
> >       unmap_page(head);
> > -     VM_BUG_ON_PAGE(compound_mapcount(head), head);
> > +     VM_BUG_ON_PAGE(total_mapcount(head), head);
>
> I am not sure about this change. The code below also checks total_mapcount(head)
> and returns EBUSY if the count is non-zero. This change makes the code dead.

It is actually dead if CONFIG_DEBUG_VM is enabled and total_mapcount
is not 0 regardless of this change due to the below code, right?

if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) {
                        pr_alert("total_mapcount: %u, page_count(): %u\n",
                                        mapcount, count);
                        if (PageTail(page))
                                dump_page(head, NULL);
                        dump_page(page, "total_mapcount(head) > 0");
                        BUG();
                }

> On the other hand, the change will force all mappings to the page have to be
> successfully unmapped all the time. I am not sure if we want to do that.
> Maybe it is better to just check total_mapcount() and fail the split.
> The same situation happens with the code change below.

IIUC, the code did force all mappings to the page to be unmapped in
order to split it.

>
> >
> >       /* block interrupt reentry in xa_lock and spinlock */
> >       local_irq_disable();
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index 693a610e181d..2e547378ab5f 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1777,7 +1777,7 @@ bool try_to_unmap(struct page *page, enum ttu_flags flags)
> >       else
> >               rmap_walk(page, &rwc);
> >
> > -     return !page_mapcount(page) ? true : false;
> > +     return !total_mapcount(page) ? true : false;
> >  }
>
> In unmap_page(), VM_BUG_ON_PAGE(!unmap_success, page) will force all mappings
> to the page have to be all unmapped, which might not be the case we want.

AFAICT, I don't see such a case from all the callers of
try_to_unmap(). Imay miss something, but I do have a hard time
thinking of a usecase which can proceed safely with "not fully
unmapped" page.

> Maybe you will want to remove the VM_BUG_ON_PAGE here, check total_mapcount()
> above, and fail the split if not all mappings to the pages are unmapped.
>
>
>
> —
> Best Regards,
> Yan Zi