From mboxrd@z Thu Jan 1 00:00:00 1970 From: Simon Jeons Subject: Re: [PATCH] writeback: fix writeback cache thrashing Date: Sat, 05 Jan 2013 03:41:54 -0600 Message-ID: <1357378914.8716.3.camel@kernel.cn.ibm.com> References: <1356847190-7986-1-git-send-email-linkinjeon@gmail.com> <20121231113054.GC7564@quack.suse.cz> <20130102134334.GB30633@quack.suse.cz> <1357261151.5105.2.camel@kernel.cn.ibm.com> <1357346803.5273.10.camel@kernel.cn.ibm.com> <20130105032642.GA8188@localhost> <1357363603.5273.16.camel@kernel.cn.ibm.com> <20130105073846.GA11811@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Cc: Namjae Jeon , Jan Kara , Wanpeng Li , linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org, Namjae Jeon , Vivek Trivedi , Dave Chinner To: Fengguang Wu Return-path: In-Reply-To: <20130105073846.GA11811@localhost> Sender: owner-linux-mm@kvack.org List-Id: linux-fsdevel.vger.kernel.org On Sat, 2013-01-05 at 15:38 +0800, Fengguang Wu wrote: > On Fri, Jan 04, 2013 at 11:26:43PM -0600, Simon Jeons wrote: > > On Sat, 2013-01-05 at 11:26 +0800, Fengguang Wu wrote: > > > > > > Hi Namjae, > > > > > > > > > > > > Why use bdi_stat_error here? What's the meaning of its commen= t "maximal > > > > > > error of a stat counter"? > > > > > Hi Simon, > > > > >=20 > > > > > As you know bdi stats (BDI_RECLAIMABLE, BDI_WRITEBACK =E2=80=A6= ) are kept in > > > > > percpu counters. > > > > > When these percpu counters are incremented/decremented simultan= eously > > > > > on multiple CPUs by small amount (individual cpu counter less t= han > > > > > threshold BDI_STAT_BATCH), > > > > > it is possible that we get approximate value (not exact value) = of > > > > > these percpu counters. > > > > > In order, to handle these percpu counter error we have used > > > > > bdi_stat_error. bdi_stat_error is the maximum error which can h= appen > > > > > in percpu bdi stats accounting. > > > > >=20 > > > > > bdi_stat(bdi, BDI_RECLAIMABLE); > > > > > -> This will give approximate value of BDI_RECLAIMABLE by read= ing > > > > > previous value of percpu count. > > > > >=20 > > > > > bdi_stat_sum(bdi, BDI_RECLAIMABLE); > > > > > ->This will give exact value of BDI_RECLAIMABLE. It will take = lock > > > > > and add current percpu count of individual CPUs. > > > > > It is not recommended to use it frequently as it is expensiv= e. We > > > > > can better use =E2=80=9Cbdi_stat=E2=80=9D and work with approx = value of bdi stats. > > > > >=20 > > > >=20 > > > > Hi Namjae, thanks for your clarify. > > > >=20 > > > > But why compare error stat count to bdi_bground_thresh? What's th= e > > >=20 > > > It's not comparing bdi_stat_error to bdi_bground_thresh, but rather= , > > > in concept, comparing bdi_stat (with error bound adjustments) to > > > bdi_bground_thresh. > > >=20 > > > > relationship between them? I also see bdi_stat_error compare to > > > > bdi_thresh/bdi_dirty in function balance_dirty_pages.=20 > > >=20 > >=20 > > Hi Fengguang, > >=20 > > > Here, it's trying to use bdi_stat_sum(), the accurate (however more > > > costly) version of bdi_stat(), if the error would possibly be large= : > >=20 > > Why error is large use bdi_stat_sum and error is few use bdi_stat? >=20 Thanks for your response Fengguang! :) > It's the opposite. Please check this per-cpu counter routine to get an = idea: >=20 > /* > * Add up all the per-cpu counts, return the result. This is a more ac= curate > * but much slower version of percpu_counter_read_positive() > */ =20 > s64 __percpu_counter_sum(struct percpu_counter *fbc) >=20 > > >=20 > > > if (bdi_thresh < 2 * bdi_stat_error(bdi)) { > > > bdi_reclaimable =3D bdi_stat_sum(bdi, BDI_R= ECLAIMABLE); > > > //... > > > } else { > > > bdi_reclaimable =3D bdi_stat(bdi, BDI_RECLA= IMABLE); > > > //... > > > } > > >=20 The comment above these codes: * In order to avoid the stacked BDI deadlock we need * to ensure we accurately count the 'dirty' pages when * the threshold is low. Why your meaning threshold low is error large?=20 > > > Here the comment should have explained it well: > > >=20 > > > * In theory 1 page is enough to keep the comsumer-= producer > > > * pipe going: the flusher cleans 1 page =3D> the t= ask dirties 1 > > > * more page. However bdi_dirty has accounting erro= rs. So use > >=20 > > Why bdi_dirty has accounting errors? >=20 > Because it typically uses bdi_stat() to get the rough sum of the per-cp= u > counters. > =20 > Thanks, > Fengguang >=20 > > > * the larger and more IO friendly bdi_stat_error. > > > */ > > > if (bdi_dirty <=3D bdi_stat_error(bdi)) > > > break; > > >=20 > > >=20 > > > Thanks, > > > Fengguang > >=20 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org