From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755446AbaBUOZM (ORCPT ); Fri, 21 Feb 2014 09:25:12 -0500 Received: from forward5m.mail.yandex.net ([37.140.138.5]:52034 "EHLO forward5m.mail.yandex.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755151AbaBUOZK (ORCPT ); Fri, 21 Feb 2014 09:25:10 -0500 From: Kirill Tkhai To: Juri Lelli Cc: Peter Zijlstra , "linux-kernel@vger.kernel.org" , Steven Rostedt , Ingo Molnar In-Reply-To: <20140221134438.ab2d02c4165fa2a6cbd663be@gmail.com> References: <230991392848160@web13m.yandex.ru> <20140221103715.GP9987@twins.programming.kicks-ass.net> <128831392982418@web15j.yandex.ru> <311621392984565@web3h.yandex.ru> <20140221134438.ab2d02c4165fa2a6cbd663be@gmail.com> Subject: Re: [RFC] sched/deadline: Prevent rt_time growth to infinity MIME-Version: 1.0 Message-Id: <3231392992702@web26m.yandex.ru> X-Mailer: Yamail [ http://yandex.ru ] 5.0 Date: Fri, 21 Feb 2014 18:25:02 +0400 Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=koi8-r Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 21.02.2014, 16:44, "Juri Lelli" : > On Fri, 21 Feb 2014 16:09:25 +0400 > Kirill Tkhai wrote: > >> š21.02.2014, 15:39, "Kirill Tkhai" : >>> š21.02.2014, 14:37, "Peter Zijlstra" : >>>> ššOn Thu, Feb 20, 2014 at 02:16:00AM +0400, Kirill Tkhai wrote: >>>>> šššSince deadline tasks share rt bandwidth, we must care about >>>>> šššbandwidth timer set. Otherwise rt_time may grow up to infinity >>>>> šššin update_curr_dl(), if there are no other available RT tasks >>>>> šššon top level bandwidth. >>>>> >>>>> šššI'm going to decide the problem the way below. Almost untested >>>>> šššbecause of I skipped almost all of recent patches which haveto be applied from lkml. >>>>> >>>>> šššPlease say, if I skipped anything in idea. Maybe better put >>>>> šššstart_top_rt_bandwidth() into set_curr_task_dl()? >>>> ššHow about we only increment rt_time when there's an RT bandwidth timer >>>> ššactive? >>> šThis case RT and DL may eat all the time: >>> >>> š-------------- štime ------------------> >>> š|RT's working |DL's working| >>> š---------------------------- >>> š|rt_runtime šš| ššššššššššš| >>> š---------------------------- >>> š| šššššššrt_period šššššššš| >>> >>> šOr at least more, than it's allowed. >>> >>> šIt looks like, if we want to limit time of high priority classes >>> šexecution, we have to set the timer anyway. >> šOh, above is confusing. Sorry. >> >> šI mean one RT task and >> >> š----------------- štime --------------------------> >> š|DL's working ššš|RT's working| |DL's working ššš| >> š-------------------------------------------------- >> š| ššššššššššššššš|rt_runtime š| | ššššššššššššššš| >> š-------------------------------------------------- >> š| ššššššššššššššš| ššrt_period š| ššššššššššššššš| >> >> šIn this case FAIR receives less ratio, than (rt_period-rt_runtime)/rt_period. > > DL tasks won't be allowed to run in this situation, as their bw exceedes > rt_runtime/rt_period. Maybe, I don't uderstand. Where does DL control summary (DL+RT) runtime? RT does not do this too. But it looks like, RT has to do this. Kirill