From: Minchan Kim Subject: Re: [PATCH 4/4] mm: vmscan: If kswapd has been running too long, allow it to sleep Date: Tue, 17 May 2011 09:48:59 +0900 Message-ID: References: <1305295404-12129-1-git-send-email-mgorman@suse.de> <1305295404-12129-5-git-send-email-mgorman@suse.de> <4DCFAA80.7040109@jp.fujitsu.com> <1305519711.4806.7.camel@mulgrave.site> <20110516084558.GE5279@suse.de> <20110516102753.GF5279@suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: James Bottomley , KOSAKI Motohiro , akpm@linux-foundation.org, colin.king@canonical.com, raghu.prabhu13@gmail.com, jack@suse.cz, chris.mason@oracle.com, cl@linux.com, penberg@kernel.org, riel@redhat.com, hannes@cmpxchg.org, linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org, linux-ext4@vger.kernel.org To: Mel Gorman Return-path: Received: from mail-qw0-f46.google.com ([209.85.216.46]:45092 "EHLO mail-qw0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751633Ab1EQAtB convert rfc822-to-8bit (ORCPT ); Mon, 16 May 2011 20:49:01 -0400 In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: On Tue, May 17, 2011 at 8:50 AM, Minchan Kim wr= ote: > On Mon, May 16, 2011 at 7:27 PM, Mel Gorman wrote: >> On Mon, May 16, 2011 at 05:58:59PM +0900, Minchan Kim wrote: >>> On Mon, May 16, 2011 at 5:45 PM, Mel Gorman wrote= : >>> > On Mon, May 16, 2011 at 02:04:00PM +0900, Minchan Kim wrote: >>> >> On Mon, May 16, 2011 at 1:21 PM, James Bottomley >>> >> wrote: >>> >> > On Sun, 2011-05-15 at 19:27 +0900, KOSAKI Motohiro wrote: >>> >> >> (2011/05/13 23:03), Mel Gorman wrote: >>> >> >> > Under constant allocation pressure, kswapd can be in the si= tuation where >>> >> >> > sleeping_prematurely() will always return true even if kswa= pd has been >>> >> >> > running a long time. Check if kswapd needs to be scheduled. >>> >> >> > >>> >> >> > Signed-off-by: Mel Gorman >>> >> >> > --- >>> >> >> > =C2=A0 mm/vmscan.c | =C2=A0 =C2=A04 ++++ >>> >> >> > =C2=A0 1 files changed, 4 insertions(+), 0 deletions(-) >>> >> >> > >>> >> >> > diff --git a/mm/vmscan.c b/mm/vmscan.c >>> >> >> > index af24d1e..4d24828 100644 >>> >> >> > --- a/mm/vmscan.c >>> >> >> > +++ b/mm/vmscan.c >>> >> >> > @@ -2251,6 +2251,10 @@ static bool sleeping_prematurely(pg_= data_t *pgdat, int order, long remaining, >>> >> >> > =C2=A0 =C2=A0 unsigned long balanced =3D 0; >>> >> >> > =C2=A0 =C2=A0 bool all_zones_ok =3D true; >>> >> >> > >>> >> >> > + =C2=A0 /* If kswapd has been running too long, just sleep= */ >>> >> >> > + =C2=A0 if (need_resched()) >>> >> >> > + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 return false; >>> >> >> > + >>> >> >> >>> >> >> Hmm... I don't like this patch so much. because this code doe= s >>> >> >> >>> >> >> - don't sleep if kswapd got context switch at shrink_inactive= _list >>> >> > >>> >> > This isn't entirely true: =C2=A0need_resched() will be false, = so we'll follow >>> >> > the normal path for determining whether to sleep or not, in ef= fect >>> >> > leaving the current behaviour unchanged. >>> >> > >>> >> >> - sleep if kswapd didn't >>> >> > >>> >> > This also isn't entirely true: whether need_resched() is true = at this >>> >> > point depends on a whole lot more that whether we did a contex= t switch >>> >> > in shrink_inactive. It mostly depends on how long we've been r= unning >>> >> > without giving up the CPU. =C2=A0Generally that will mean we'v= e been round >>> >> > the shrinker loop hundreds to thousands of times without sleep= ing. >>> >> > >>> >> >> It seems to be semi random behavior. >>> >> > >>> >> > Well, we have to do something. =C2=A0Chris Mason first suspect= ed the hang was >>> >> > a kswapd rescheduling problem a while ago. =C2=A0We tried putt= ing >>> >> > cond_rescheds() in several places in the vmscan code, but to n= o avail. >>> >> >>> >> Is it a result of =C2=A0test with patch of Hannes(ie, !pgdat_bal= anced)? >>> >> >>> >> If it isn't, it would be nop regardless of putting cond_reshed a= t vmscan.c. >>> >> Because, although we complete zone balancing, kswapd doesn't sle= ep as >>> >> pgdat_balance returns wrong result. And at last VM calls >>> >> balance_pgdat. In this case, balance_pgdat returns without any w= ork as >>> >> kswap couldn't find zones which have not enough free pages and g= oto >>> >> out. kswapd could repeat this work infinitely. So you don't have= a >>> >> chance to call cond_resched. >>> >> >>> >> But if your test was with Hanne's patch, I am very curious how c= ome >>> >> kswapd consumes CPU a lot. >>> >> >>> >> > The need_resched() in sleeping_prematurely() seems to be about= the best >>> >> > option. =C2=A0The other option might be just to put a cond_res= ched() in >>> >> > kswapd_try_to_sleep(), but that will really have about the sam= e effect. >>> >> >>> >> I don't oppose it but before that, I think we have to know why k= swapd >>> >> consumes CPU a lot although we applied Hannes' patch. >>> >> >>> > >>> > Because it's still possible for processes to allocate pages at th= e same >>> > rate kswapd is freeing them leading to a situation where kswapd d= oes not >>> > consider the zone balanced for prolonged periods of time. >>> >>> We have cond_resched in shrink_page_list, shrink_slab and balance_p= gdat. >>> So I think kswapd can be scheduled out although it's scheduled in >>> after a short time as task scheduled also need page reclaim. Althou= gh >>> all task in system need reclaim, kswapd cpu 99% consumption is a >>> natural result, I think. >>> Do I miss something? >>> >> >> Lets see; >> >> shrink_page_list() only applies if inactive pages were isolated >> =C2=A0 =C2=A0 =C2=A0 =C2=A0which in turn may not happen if all_unrec= laimable is set in >> =C2=A0 =C2=A0 =C2=A0 =C2=A0shrink_zones(). If for whatver reason, al= l_unreclaimable is >> =C2=A0 =C2=A0 =C2=A0 =C2=A0set on all zones, we can miss calling con= d_resched(). >> >> shrink_slab only applies if we are reclaiming slab pages. If the fir= st >> =C2=A0 =C2=A0 =C2=A0 =C2=A0shrinker returns -1, we do not call cond_= resched(). If that >> =C2=A0 =C2=A0 =C2=A0 =C2=A0first shrinker is dcache and __GFP_FS is = not set, direct >> =C2=A0 =C2=A0 =C2=A0 =C2=A0reclaimers will not shrink at all. Howeve= r, if there are >> =C2=A0 =C2=A0 =C2=A0 =C2=A0enough of them running or if one of the o= ther shrinkers >> =C2=A0 =C2=A0 =C2=A0 =C2=A0is running for a very long time, kswapd c= ould be starved >> =C2=A0 =C2=A0 =C2=A0 =C2=A0acquiring the shrinker_rwsem and never re= aching the >> =C2=A0 =C2=A0 =C2=A0 =C2=A0cond_resched(). > > Don't we have to move cond_resched? > > diff --git a/mm/vmscan.c b/mm/vmscan.c > index 292582c..633e761 100644 > --- a/mm/vmscan.c > +++ b/mm/vmscan.c > @@ -231,8 +231,10 @@ unsigned long shrink_slab(struct shrink_control = *shrink, > =C2=A0 =C2=A0 =C2=A0 =C2=A0if (scanned =3D=3D 0) > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0scanned =3D SW= AP_CLUSTER_MAX; > > - =C2=A0 =C2=A0 =C2=A0 if (!down_read_trylock(&shrinker_rwsem)) > - =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 return 1; =C2=A0 =C2= =A0 =C2=A0 /* Assume we'll be able to shrink next time */ > + =C2=A0 =C2=A0 =C2=A0 if (!down_read_trylock(&shrinker_rwsem)) { > + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 ret =3D 1; > + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 goto out; /* Assum= e we'll be able to shrink next time */ > + =C2=A0 =C2=A0 =C2=A0 } > > =C2=A0 =C2=A0 =C2=A0 =C2=A0list_for_each_entry(shrinker, &shrinker_li= st, list) { > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0unsigned long = long delta; > @@ -280,12 +282,14 @@ unsigned long shrink_slab(struct shrink_control= *shrink, > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0= =C2=A0 =C2=A0count_vm_events(SLABS_SCANNED, this_scan); > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0= =C2=A0 =C2=A0total_scan -=3D this_scan; > > - =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0= =C2=A0 cond_resched(); > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0} > > =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0shrinker->nr += =3D total_scan; > + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 cond_resched(); > =C2=A0 =C2=A0 =C2=A0 =C2=A0} > =C2=A0 =C2=A0 =C2=A0 =C2=A0up_read(&shrinker_rwsem); > +out: > + =C2=A0 =C2=A0 =C2=A0 cond_resched(); > =C2=A0 =C2=A0 =C2=A0 =C2=A0return ret; > =C2=A0} > > >> >> balance_pgdat() only calls cond_resched if the zones are not >> =C2=A0 =C2=A0 =C2=A0 =C2=A0balanced. For a high-order allocation tha= t is balanced, it >> =C2=A0 =C2=A0 =C2=A0 =C2=A0checks order-0 again. During that window,= order-0 might have >> =C2=A0 =C2=A0 =C2=A0 =C2=A0become unbalanced so it loops again for o= rder-0 and returns >> =C2=A0 =C2=A0 =C2=A0 =C2=A0that was reclaiming for order-0 to kswapd= (). It can then find >> =C2=A0 =C2=A0 =C2=A0 =C2=A0that a caller has rewoken kswapd for a hi= gh-order and re-enters >> =C2=A0 =C2=A0 =C2=A0 =C2=A0balance_pgdat() without ever have called = cond_resched(). > > If kswapd reclaims order-o followed by high order, it would have a > chance to call cond_resched in shrink_page_list. But if all zones are > all_unreclaimable is set, balance_pgdat could return any work. Typo : without any work. --=20 Kind regards, Minchan Kim -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" i= n the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html