On Fri, 22 May 2020, Anshuman Khandual wrote:
> On 05/21/2020 07:19 PM, Hugh Dickins wrote:
> > Fix 5.7-rc6-mm1 page migration crash in unmap_and_move(): when the
> > page to be migrated has been freed from under us, that is considered
> > a MIGRATEPAGE_SUCCESS, but no newpage has been allocated (and I don't
> > think it would ever need to be counted as a successful THP migration).
> >
> > Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
> > ---
> > Fix to mm-vmstat-add-events-for-pmd-based-thp-migration-without-split.patch
> >
> > mm/migrate.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > --- 5.7-rc6-mm1/mm/migrate.c 2020-05-20 12:21:56.117693827 -0700
> > +++ linux/mm/migrate.c 2020-05-20 15:08:12.319476978 -0700
> > @@ -1248,7 +1248,7 @@ out:
> > * we want to retry.
> > */
> > if (rc == MIGRATEPAGE_SUCCESS) {
> > - if (PageTransHuge(newpage))
> > + if (newpage && PageTransHuge(newpage))
> > thp_migration_success(true);
> > put_page(page);
> > if (reason == MR_MEMORY_FAILURE) {
> >
>
> Thanks Hugh. I am preparing to respin the vmstat patch accommodating some
> earlier comments. Wondering if I should also fold these changes here after
> adding your signed-off-by ?
Thanks for asking, but please just fold this one-line fixup into your
respin without my signed-off-by: I'm not heavily invested in these stats,
just want to avoid the crash; and don't know if I would want to sign the
result. You could add something like [hughd: fixed oops on NULL newpage]
to confirm that the fix is on board, but the fix is what matters.
Hugh
