From mboxrd@z Thu Jan 1 00:00:00 1970 From: Steven Rostedt Subject: Re: [PATCH] sched/rt: RT_RUNTIME_GREED sched feature Date: Tue, 8 Nov 2016 12:17:10 -0500 Message-ID: <20161108121710.3e7eb664@gandalf.local.home> References: <20161108115958.GO3142@twins.programming.kicks-ass.net> <20161108090740.4226ffc9@gandalf.local.home> <20161108165133.GI3117@twins.programming.kicks-ass.net> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: Daniel Bristot de Oliveira , Ingo Molnar , Christoph Lameter , linux-rt-users , LKML To: Peter Zijlstra Return-path: In-Reply-To: <20161108165133.GI3117@twins.programming.kicks-ass.net> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-rt-users.vger.kernel.org On Tue, 8 Nov 2016 17:51:33 +0100 Peter Zijlstra wrote: > You really should already know this. I know what we want to do, but there's some momentous problems that need to be solved first. Until then, we may be forced to continue with hacks. > > As stands the current rt cgroup code (and all this throttling code) is a > giant mess (as in, its not actually correct from a RT pov). We should > not make it worse by adding random hacks to it. > > The right way to to about doing this is by replacing it with something > better; like the proposed DL server for FIFO tasks -- which is entirely > non-trivial as well, see existing discussion on that. Right. The biggest issue that I see is how to assign affinities to FIFO tasks and use a DL server to keep them from starving other tasks? > > I'm not entirely sure what this patch was supposed to fix, but it could > be running CFS tasks with higher priority than RT for a window, instead I'm a bit confused with the above sentence. Do you mean that this patch causes CFS tasks to run for a period with a higher priority than RT? Well, currently we have the both CFS tasks and the "idle" task run higher than RT, but this patch changes that to be just CFS tasks. > of throttling RT tasks. This seems fairly ill specified, but something > like that could easily done with an explicit or slack time DL server for > CFS tasks. If we can have a DL scheduler that can handle arbitrary affinities, then all could be solved with that. -- Steve