From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1760024AbdADA4a (ORCPT ); Tue, 3 Jan 2017 19:56:30 -0500 Received: from mx0b-001b2d01.pphosted.com ([148.163.158.5]:53617 "EHLO mx0a-001b2d01.pphosted.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S967040AbdADA4N (ORCPT ); Tue, 3 Jan 2017 19:56:13 -0500 Date: Tue, 3 Jan 2017 16:55:59 -0800 From: "Paul E. McKenney" To: Michal Hocko Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, peterz@infradead.org Subject: Re: Fw: [lkp-developer] [sched,rcu] cf7a2dca60: [No primary change] +186% will-it-scale.time.involuntary_context_switches Reply-To: paulmck@linux.vnet.ibm.com References: <20161213151408.GC3924@linux.vnet.ibm.com> <20161214095425.GE25573@dhcp22.suse.cz> <20161214110609.GK3924@linux.vnet.ibm.com> <20161214161540.GP25573@dhcp22.suse.cz> <20161214164827.GL3924@linux.vnet.ibm.com> <20161214173923.GA16763@dhcp22.suse.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20161214173923.GA16763@dhcp22.suse.cz> User-Agent: Mutt/1.5.21 (2010-09-15) X-TM-AS-GCONF: 00 X-Content-Scanned: Fidelis XPS MAILER x-cbid: 17010400-0004-0000-0000-00001136F550 X-IBM-SpamModules-Scores: X-IBM-SpamModules-Versions: BY=3.00006369; HX=3.00000240; KW=3.00000007; PH=3.00000004; SC=3.00000199; SDB=6.00803159; UDB=6.00390695; IPR=6.00581009; BA=6.00005028; NDR=6.00000001; ZLA=6.00000005; ZF=6.00000009; ZB=6.00000000; ZP=6.00000000; ZH=6.00000000; ZU=6.00000002; MB=3.00013815; XFM=3.00000011; UTC=2017-01-04 00:56:01 X-IBM-AV-DETECTION: SAVI=unused REMOTE=unused XFE=unused x-cbparentid: 17010400-0005-0000-0000-00007BE1FF6A Message-Id: <20170104005559.GD3742@linux.vnet.ibm.com> X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10432:,, definitions=2017-01-03_22:,, signatures=0 X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 spamscore=0 suspectscore=0 malwarescore=0 phishscore=0 adultscore=0 bulkscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1612050000 definitions=main-1701040014 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Dec 14, 2016 at 06:39:24PM +0100, Michal Hocko wrote: > On Wed 14-12-16 08:48:27, Paul E. McKenney wrote: > > On Wed, Dec 14, 2016 at 05:15:41PM +0100, Michal Hocko wrote: > > > On Wed 14-12-16 03:06:09, Paul E. McKenney wrote: > > > > On Wed, Dec 14, 2016 at 10:54:25AM +0100, Michal Hocko wrote: > > > > > On Tue 13-12-16 07:14:08, Paul E. McKenney wrote: > > > > > > Just FYI for the moment... > > > > > > > > > > > > So even with the slowed-down checking, making cond_resched() do what > > > > > > cond_resched_rcu_qs() does results in a smallish but quite measurable > > > > > > degradation according to 0day. > > > > > > > > > > So if I understand those results properly, the reason seems to be the > > > > > increased involuntary context switches, right? Or am I misreading the > > > > > data? > > > > > I am looking at your "sched,rcu: Make cond_resched() provide RCU > > > > > quiescent state" in linux-next and I am wondering whether rcu_all_qs has > > > > > to be called unconditionally and not only when should_resched failed few > > > > > times? I guess you have discussed that with Peter already but do not > > > > > remember the outcome. > > > > > > > > My first thought is to wait for the grace period to age further before > > > > checking, the idea being to avoid increasing cond_resched() overhead > > > > any further. But if that doesn't work, then yes, I may have to look at > > > > adding more checks to cond_resched(). > > > > > > This might be really naive but would something like the following work? > > > The overhead should be pretty much negligible, I guess. Ideally the pcp > > > variable could be set somewhere from check_cpu_stall() but I couldn't > > > wrap my head around that code to see how exactly. > > > > My concern (perhaps misplaced) with this approach is that there are > > quite a few tight loops containing cond_resched(). So I would still > > need to throttle the resulting grace-period acceleration to keep the > > context switches down to a dull roar. > > Yes, I see your point. Something based on the stall timeout would be > much better of course. I just failed to come up with something that > would make sense. This was more my lack of familiarity with the code so > I hope you will be more successful ;) Well, here is my current shot at this. And so do I. ;-) So now it ignores cond_resched_rcu_qs() until at least jiffies_till_sched_qs jiffies have elapsed since the start of the grace period. The jiffies_till_sched_qs variable defaults to HZ/20, which should slow the checks down by about a factor of seven. Plus I don't see a problem with changing the default to (say) HZ/10 if needed. Thoughts? Thanx, Paul ------------------------------------------------------------------------ commit 7acd02c9e62fb21e7466e7a99fc21bf6ed6cc3cf Author: Paul E. McKenney Date: Tue Jan 3 16:49:46 2017 -0800 squash! rcu: Check cond_resched_rcu_qs() state less often to reduce GP overhead Now polling only after jiffies_till_sched_qs jiffies have elapsed. diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 083cb8a6299c..0369e0e0fe00 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -1274,7 +1274,9 @@ static int dyntick_save_progress_counter(struct rcu_data *rdp, static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, bool *isidle, unsigned long *maxj) { + unsigned long jtsq; int *rcrmp; + unsigned long rjtsc; struct rcu_node *rnp; /* @@ -1291,6 +1293,17 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, return 1; } + /* Compute and saturate jiffies_till_sched_qs. */ + jtsq = jiffies_till_sched_qs; + rjtsc = rcu_jiffies_till_stall_check(); + if (jtsq > rjtsc / 2) { + WRITE_ONCE(jiffies_till_sched_qs, rjtsc); + jtsq = rjtsc / 2; + } else if (jtsq < 1) { + WRITE_ONCE(jiffies_till_sched_qs, 1); + jtsq = 1; + } + /* * Has this CPU encountered a cond_resched_rcu_qs() since the * beginning of the grace period? For this to be the case, @@ -1298,7 +1311,8 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, * might not be the case for nohz_full CPUs looping in the kernel. */ rnp = rdp->mynode; - if (READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_qs_ctr, rdp->cpu) && + if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && + READ_ONCE(rdp->rcu_qs_ctr_snap) != per_cpu(rcu_qs_ctr, rdp->cpu) && READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) { trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc")); return 1; @@ -1333,9 +1347,8 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, * warning delay. */ rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu); - if (ULONG_CMP_GE(jiffies, - rdp->rsp->gp_start + jiffies_till_sched_qs) || - ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) { + if (time_after(jiffies, rdp->rsp->gp_start + jtsq) || + time_after(jiffies, rdp->rsp->jiffies_resched)) { if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) { WRITE_ONCE(rdp->cond_resched_completed, READ_ONCE(rdp->mynode->completed));