Re: [PATCH 1/3] mm: don't call activate_page() on new ksm pages

From: Yang Shi
Date: Mon Aug 17 2020 - 16:48:52 EST


On Thu, Aug 13, 2020 at 12:34 AM Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
>
> On Wed, Aug 12, 2020 at 10:19:24PM -0700, Yang Shi wrote:
> > On Tue, Aug 11, 2020 at 9:04 PM Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
> > >
> > > lru_cache_add_active_or_unevictable() already adds new ksm pages to
> > > active lru. Calling activate_page() isn't really necessary in this
> > > case.
> > >
> > > Signed-off-by: Yu Zhao <yuzhao@xxxxxxxxxx>
> > > ---
> > > mm/swapfile.c | 10 +++++-----
> > > 1 file changed, 5 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/mm/swapfile.c b/mm/swapfile.c
> > > index 6c26916e95fd..cf115ea26a20 100644
> > > --- a/mm/swapfile.c
> > > +++ b/mm/swapfile.c
> > > @@ -1913,16 +1913,16 @@ static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
> > > pte_mkold(mk_pte(page, vma->vm_page_prot)));
> > > if (page == swapcache) {
> > > page_add_anon_rmap(page, vma, addr, false);
> > > + /*
> > > + * Move the page to the active list so it is not
> > > + * immediately swapped out again after swapon.
> > > + */
> > > + activate_page(page);
> >
> > Actually I think we could just remove this activate_page() call with
> > Joonsoo's anonymous page workingset series merged. The active bit will
> > be taken care by workingset_refault().
> >
> > > } else { /* ksm created a completely new copy */
> > > page_add_new_anon_rmap(page, vma, addr, false);
> > > lru_cache_add_active_or_unevictable(page, vma);
> >
> > And it looks the latest linus's tree already changed this to
> > lru_cache_add_inactive_or_unevictable() by commit b518154e59
> > ("mm/vmscan: protect the workingset on anonymous LRU")
>
> Oops, apparently my tree is out of date. I'll work on a new version
> that removes the superfluous activate_page(). Meanwhile, can you
> please take a look at the rest of this series and let me know if
> there is anything else that we might want to change? Thank you.

I took a look at those two patches. For the #2 I didn't spot anything
wrong, but I may miss something. For the #3, TBH I don't think the
justification is strong enough since you just moved the PG_waiters bit
cleared to allocation time, someone could argue it may hurt allocation
latency.