On Tue, Jan 26, 2021 at 12:31 PM Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
>
> On 1/25/21 7:10 PM, Muchun Song wrote:
> > The VM_BUG_ON_PAGE avoids the generation of any code, even if that
> > expression has side-effects when !CONFIG_DEBUG_VM.
> >
> > Fixes: e5dfacebe4a4 ("mm/hugetlb.c: just use put_page_testzero() instead of page_count()")
> > Signed-off-by: Muchun Song <songmuchun@xxxxxxxxxxxxx>
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > ---
> > mm/hugetlb.c | 5 ++++-
> > 1 file changed, 4 insertions(+), 1 deletion(-)
>
> Thanks for finding and fixing this! My bad for not noticing when the bug
> was introduced.
>
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index a6bad1f686c5..082ed643020b 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -2047,13 +2047,16 @@ static int gather_surplus_pages(struct hstate *h, long delta)
> >
> > /* Free the needed pages to the hugetlb pool */
> > list_for_each_entry_safe(page, tmp, &surplus_list, lru) {
> > + int zeroed;
> > +
> > if ((--needed) < 0)
> > break;
> > /*
> > * This page is now managed by the hugetlb allocator and has
> > * no users -- drop the buddy allocator's reference.
> > */
> > - VM_BUG_ON_PAGE(!put_page_testzero(page), page);
> > + zeroed = put_page_testzero(page);
> > + VM_BUG_ON_PAGE(!zeroed, page);
> > enqueue_huge_page(h, page);
>
> I was wondering why this was not causing any problems. We are putting the
> hugetlb page on the free list with a count of 1. There is no check in the
> enqueue code. When we dequeue the page, set_page_refcounted() is used to
> set the count to 1 without looking at the current value. And, all the other
> VM_DEBUG macros are off so we mostly do not notice the bug.
Yeah. You are right. I also thought about this question.
It is a very hidden bug.
>
> Thanks again,
>
> Reviewed-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
>
> --
> Mike Kravetz
>
> > }
> > free:
> >
