linux-kernel - Re: [patch 3/7] sched: set skip_clock_update in yield_task

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1322203171.8286.22.camel@marge.simson.net>
Date:	Fri, 25 Nov 2011 07:39:31 +0100
From:	Mike Galbraith <mgalbraith@...e.de>
To:	Peter Zijlstra <peterz@...radead.org>
Cc:	Suresh Siddha <suresh.b.siddha@...el.com>,
	linux-kernel <linux-kernel@...r.kernel.org>,
	Ingo Molnar <mingo@...e.hu>, Paul Turner <pjt@...gle.com>
Subject: Re: [patch 3/7] sched: set skip_clock_update in yield_task_fair()

On Thu, 2011-11-24 at 11:12 +0100, Peter Zijlstra wrote:
> On Thu, 2011-11-24 at 04:50 +0100, Mike Galbraith wrote:
> > On Wed, 2011-11-23 at 15:48 +0100, Peter Zijlstra wrote:
> > > On Tue, 2011-11-22 at 15:21 +0100, Mike Galbraith wrote:
> > > > This is another case where we are on our way to schedule(),
> > > > so can save a useless clock update and resulting microscopic
> > > > vruntime update.
> > > > 
> > > 
> > > Everytime I see one of these skip_clock_update thingies I get the idea
> > > we should do something smarter, because its most fragile and getting it
> > > wrong results in a subtle trainwreck..
> > > 
> > > Would something like the below help in validating this stuff?
> > 
> > No, because switch itself is irrelevant to fair-beans counting.
> 
> Hmm, right, but we can keep another counter that is more relevant, like
> all schedule(),ttwu() and various other calls, right?

Yeah, there's gotta be a way to be smarter about this.  When waking a
gaggle, we'll do useless microscopic updates and preempt checks for each
and every wakeup (not to mention waker can be preempted).  It's one
event per touched rq really, but we treat is as many.  The skip thing
just binds a couple obvious spots in a hacky way.

> Thing is, is that going to be less fragile than the current crap :-(

I agree it's fragile.

	-Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/