From: Hillf Danton <hdanton@sina.com>
To: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>,
Michal Hocko <mhocko@kernel.org>, Mel Gorman <mgorman@suse.de>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
linux-kernel <linux-kernel@vger.kernel.org>,
Johannes Weiner <hannes@cmpxchg.org>
Subject: Re: [Question] Should direct reclaim time be bounded?
Date: Thu, 11 Jul 2019 23:44:52 +0800 [thread overview]
Message-ID: <20190711154452.4940-1-hdanton@sina.com> (raw)
On Thu, 11 Jul 2019 02:42:56 +0800 Mike Kravetz wrote:
> On 7/7/19 10:19 PM, Hillf Danton wrote:
> > On Mon, 01 Jul 2019 20:15:51 -0700 Mike Kravetz wrote:
> >> On 7/1/19 1:59 AM, Mel Gorman wrote:
> >>>
> >>> I think it would be reasonable to have should_continue_reclaim allow an
> >>> exit if scanning at higher priority than DEF_PRIORITY - 2, nr_scanned is
> >>> less than SWAP_CLUSTER_MAX and no pages are being reclaimed.
> >>
> >> Thanks Mel,
> >>
> >> I added such a check to should_continue_reclaim. However, it does not
> >> address the issue I am seeing. In that do-while loop in shrink_node,
> >> the scan priority is not raised (priority--). We can enter the loop
> >> with priority == DEF_PRIORITY and continue to loop for minutes as seen
> >> in my previous debug output.
> >>
> > Does it help raise prioity in your case?
>
> Thanks Hillf, sorry for delay in responding I have been AFK.
>
> I am not sure if you wanted to try this somehow in addition to Mel's
> suggestion, or alone.
>
I wanted you might take a look at it if you continued to have difficulty
raising scanning priority.
> Unfortunately, such a change actually causes worse behavior.
>
> > --- a/mm/vmscan.c
> > +++ b/mm/vmscan.c
> > @@ -2543,11 +2543,18 @@ static inline bool should_continue_reclaim(struct pglist_data *pgdat,
> > unsigned long pages_for_compaction;
> > unsigned long inactive_lru_pages;
> > int z;
> > + bool costly_fg_reclaim = false;
> >
> > /* If not in reclaim/compaction mode, stop */
> > if (!in_reclaim_compaction(sc))
> > return false;
> >
> > + /* Let compact determine what to do for high order allocators */
> > + costly_fg_reclaim = sc->order > PAGE_ALLOC_COSTLY_ORDER &&
> > + !current_is_kswapd();
> > + if (costly_fg_reclaim)
> > + goto check_compact;
>
> This goto makes us skip the 'if (!nr_reclaimed && !nr_scanned)' test.
>
Correct.
> > +
> > /* Consider stopping depending on scan and reclaim activity */
> > if (sc->gfp_mask & __GFP_RETRY_MAYFAIL) {
> > /*
> > @@ -2571,6 +2578,7 @@ static inline bool should_continue_reclaim(struct pglist_data *pgdat,
> > return false;
> > }
> >
> > +check_compact:
> > /*
> > * If we have not reclaimed enough pages for compaction and the
> > * inactive lists are large enough, continue reclaiming
>
> It is quite easy to hit the condition where:
> nr_reclaimed == 0 && nr_scanned == 0 is true, but we skip the previous test
>
Erm it is becoming harder to imagine what prevented you from raising priority
when it was not difficult to hit the true condition above.
And I see the following in the mail thread,
===
Date: Fri, 28 Jun 2019 11:20:42 -0700
Message-ID: <dede2f84-90bf-347a-2a17-fb6b521bf573@oracle.com> (raw)
In-Reply-To: <04329fea-cd34-4107-d1d4-b2098ebab0ec@suse.cz>
I got back to looking into the direct reclaim/compaction stalls when
trying to allocate huge pages. As previously mentioned, the code is
looping for a long time in shrink_node(). The routine
should_continue_reclaim() returns true perhaps more often than it should.
As Vlastmil guessed, my debug code output below shows nr_scanned is remaining
non-zero for quite a while. This was on v5.2-rc6.
===
nr_scanned != 0 and the result of should_continue_reclaim() is not false, which
is unable to match the condition you easily hit.
> and the compaction check:
> sc->nr_reclaimed < pages_for_compaction &&
> inactive_lru_pages > pages_for_compaction
>
> is true, so we return true before the below check of costly_fg_reclaim
>
It is the price high order allocations pay: reclaiming enough pages for
compact to do its work. With plenty of inactive pages you got no pages
reclaimed and scanned. It is really hard to imagine. And costly_fg_reclaim
is not good for that imho.
> > @@ -2583,6 +2591,9 @@ static inline bool should_continue_reclaim(struct pglist_data *pgdat,
> > inactive_lru_pages > pages_for_compaction)
> > return true;
> >
> > + if (costly_fg_reclaim)
> > + return false;
> > +
> > /* If compaction would go ahead or the allocation would succeed, stop */
> > for (z = 0; z <= sc->reclaim_idx; z++) {
> > struct zone *zone = &pgdat->node_zones[z];
> > --
> >
>
> As Michal suggested, I'm going to do some testing to see what impact
> dropping the __GFP_RETRY_MAYFAIL flag for these huge page allocations
> will have on the number of pages allocated.
> --
> Mike Kravetz
>
next reply other threads:[~2019-07-11 15:45 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-07-11 15:44 Hillf Danton [this message]
-- strict thread matches above, loose matches on Subject: below --
2019-07-12 5:47 [Question] Should direct reclaim time be bounded? Hillf Danton
2019-07-13 1:11 ` Mike Kravetz
2019-04-23 4:07 Mike Kravetz
2019-04-23 7:19 ` Michal Hocko
2019-04-23 16:39 ` Mike Kravetz
2019-04-24 14:35 ` Vlastimil Babka
2019-06-28 18:20 ` Mike Kravetz
2019-07-01 8:59 ` Mel Gorman
2019-07-02 3:15 ` Mike Kravetz
2019-07-03 9:43 ` Mel Gorman
2019-07-03 23:54 ` Mike Kravetz
2019-07-04 11:09 ` Michal Hocko
2019-07-04 15:11 ` Mike Kravetz
2019-07-08 5:19 ` Hillf Danton
2019-07-10 18:42 ` Mike Kravetz
2019-07-10 19:44 ` Michal Hocko
2019-07-10 23:36 ` Mike Kravetz
2019-07-11 7:12 ` Michal Hocko
2019-07-12 9:49 ` Mel Gorman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20190711154452.4940-1-hdanton@sina.com \
--to=hdanton@sina.com \
--cc=hannes@cmpxchg.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mgorman@suse.de \
--cc=mhocko@kernel.org \
--cc=mike.kravetz@oracle.com \
--cc=vbabka@suse.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).