Re: [PATCH 3/7] khugepaged: Drain LRU add pagevec to get rid of extra pins

[Yang Shi]

On Sat, Mar 28, 2020 at 5:18 AM Kirill A. Shutemov <kirill@xxxxxxxxxxxxx> wrote:
>
> On Fri, Mar 27, 2020 at 11:10:40AM -0700, Yang Shi wrote:
> > On Fri, Mar 27, 2020 at 10:06 AM Kirill A. Shutemov
> > <kirill@xxxxxxxxxxxxx> wrote:
> > >
> > > __collapse_huge_page_isolate() may fail due to extra pin in the LRU add
> > > pagevec. It's petty common for swapin case: we swap in pages just to
> > > fail due to the extra pin.
> > >
> > > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > ---
> > >  mm/khugepaged.c | 8 ++++++++
> > >  1 file changed, 8 insertions(+)
> > >
> > > diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> > > index 14d7afc90786..39e0994abeb8 100644
> > > --- a/mm/khugepaged.c
> > > +++ b/mm/khugepaged.c
> > > @@ -585,11 +585,19 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
> > >                  * The page must only be referenced by the scanned process
> > >                  * and page swap cache.
> > >                  */
> > > +               if (page_count(page) != 1 + PageSwapCache(page)) {
> > > +                       /*
> > > +                        * Drain pagevec and retry just in case we can get rid
> > > +                        * of the extra pin, like in swapin case.
> > > +                        */
> > > +                       lru_add_drain();
> >
> > This is definitely correct.
> >
> > I'm wondering if we need one more lru_add_drain() before PageLRU check
> > in khugepaged_scan_pmd() or not? The page might be in lru cache then
> > get skipped. This would improve the success rate.
>
> Could you elaborate on the scenario, I don't follow.

I mean the below change:

--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1195,6 +1195,9 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
                        goto out_unmap;
                }
                khugepaged_node_load[node]++;
+               if (!PageLRU(page))
+                       /* Flush the page out of lru cache */
+                       lru_add_drain();
                if (!PageLRU(page)) {
                        result = SCAN_PAGE_LRU;
                        goto out_unmap;

If the page is not on LRU we even can't reach collapse_huge_page(), right?

>
>
> --
>  Kirill A. Shutemov