On Thu, Apr 18, 2024 at 04:00:42PM +0800, Miaohe Lin wrote:
> On 2024/4/18 12:05, Oscar Salvador wrote:
> > On Thu, Apr 18, 2024 at 10:19:59AM +0800, Miaohe Lin wrote:
> >> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> >> index 26ab9dfc7d63..1da9a14a5513 100644
> >> --- a/mm/hugetlb.c
> >> +++ b/mm/hugetlb.c
> >> @@ -1788,7 +1788,8 @@ static void __update_and_free_hugetlb_folio(struct hstate *h,
> >> destroy_compound_gigantic_folio(folio, huge_page_order(h));
> >> free_gigantic_folio(folio, huge_page_order(h));
> >> } else {
> >> - INIT_LIST_HEAD(&folio->_deferred_list);
> >> + if (!folio_test_hugetlb(folio))
> >> + INIT_LIST_HEAD(&folio->_deferred_list);
> >
> > Ok, it took me a bit to figure this out.
> >
> > So we basically init __deferred_list when we know that
> > folio_put will not end up calling free_huge_folio
> > because a previous call to remove_hugetlb_folio has already cleared the
> > bit.
> >
> > Maybe Matthew thought that any folio ending here would not end up in
> > free_huge_folio (which is the one fiddling subpool).
> >
> > I mean, fix looks good because if hugetlb flag is cleared,
> > destroy_large_folio will go straight to free_the_page, but the
> > whole thing is a bit subtle.
>
> AFAICS, this is the most straightforward way to fix the issue. Do you have any suggestions
> on how to fix this in a more graceful way?
Not from the top of my head.
Anyway, I have been thinking for a while that this code needs some love,
so I will check how this can be untangled.
--
Oscar Salvador
SUSE Labs
