linux-kernel - Re: [PATCH 3/3] mm: vmscan: enforce inactive:active ratio at the reclaim root

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 12 Nov 2019 13:00:19 -0500
From:   Johannes Weiner <hannes@...xchg.org>
To:     Suren Baghdasaryan <surenb@...gle.com>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Andrey Ryabinin <aryabinin@...tuozzo.com>,
        Shakeel Butt <shakeelb@...gle.com>,
        Rik van Riel <riel@...riel.com>,
        Michal Hocko <mhocko@...e.com>, linux-mm <linux-mm@...ck.org>,
        cgroups mailinglist <cgroups@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>, kernel-team@...com
Subject: Re: [PATCH 3/3] mm: vmscan: enforce inactive:active ratio at the
 reclaim root

On Sun, Nov 10, 2019 at 06:15:50PM -0800, Suren Baghdasaryan wrote:
> On Thu, Nov 7, 2019 at 12:53 PM Johannes Weiner <hannes@...xchg.org> wrote:
> > @@ -2758,7 +2775,17 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
> >                         total_high_wmark += high_wmark_pages(zone);
> >                 }
> >
> > -               sc->file_is_tiny = file + free <= total_high_wmark;
> > +               /*
> > +                * Consider anon: if that's low too, this isn't a
> > +                * runaway file reclaim problem, but rather just
> > +                * extreme pressure. Reclaim as per usual then.
> > +                */
> > +               anon = node_page_state(pgdat, NR_INACTIVE_ANON);
> > +
> > +               sc->file_is_tiny =
> > +                       file + free <= total_high_wmark &&
> > +                       !(sc->may_deactivate & DEACTIVATE_ANON) &&
> > +                       anon >> sc->priority;
> 
> The name of file_is_tiny flag seems to not correspond with its actual
> semantics anymore. Maybe rename it into "skip_file"?

I'm not a fan of file_is_tiny, but I also don't like skip_file. IMO
it's better to have it describe a situation instead of an action, in
case we later want to take additional action for that situation.

Any other ideas? ;)

> I'm confused about why !(sc->may_deactivate & DEACTIVATE_ANON) should
> be a prerequisite for skipping file LRU reclaim. IIUC this means we
> will skip reclaiming from file LRU only when anonymous page
> deactivation is not allowed. Could you please add a comment explaining
> this?

The comment above this check tries to explain it: the definition of
file being "tiny" is dependent on the availability of anon. It's a
relative comparison.

If file only has a few pages, and anon is easily reclaimable (does not
require deactivation to reclaim pages), then file is "tiny" and we
should go after the more plentiful anon pages.

If anon is under duress, too, this preference doesn't make sense and
we should just reclaim both lists equally, as per usual.

Note that I'm not introducing this constraint, I'm just changing how
it's implemented. From the patch:

> >         /*
> >          * If the system is almost out of file pages, force-scan anon.
> > -        * But only if there are enough inactive anonymous pages on
> > -        * the LRU. Otherwise, the small LRU gets thrashed.
> >          */
> > -       if (sc->file_is_tiny &&
> > -           !inactive_list_is_low(lruvec, false, sc, false) &&
> > -           lruvec_lru_size(lruvec, LRU_INACTIVE_ANON,
> > -                           sc->reclaim_idx) >> sc->priority) {
> > +       if (sc->file_is_tiny) {
> >                 scan_balance = SCAN_ANON;
> >                 goto out;
> >         }

So it's always been checking whether reclaim would deactivate anon,
and whether inactive_anon has sufficient pages for this priority.