linux-kernel - Re: [PATCH] cpuidle: menu: Fall back to polling if next timer event is near

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] [day] [month] [year] [list]

Date:	Mon, 21 Mar 2016 14:55:33 +0100
From:	"Rafael J. Wysocki" <rjw@...ysocki.net>
To:	Doug Smythies <dsmythies@...us.net>
Cc:	'Rik van Riel' <riel@...hat.com>,
	'Linux PM list' <linux-pm@...r.kernel.org>,
	'Linux Kernel Mailing List' <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] cpuidle: menu: Fall back to polling if next timer event is near

On Saturday, March 19, 2016 11:54:39 PM Doug Smythies wrote:
> On 2016.03.19 17:34 Rafael J. Wysocki wrote:
> 
> > Commit a9ceb78bc75c (cpuidle,menu: use interactivity_req to disable
> > polling) changed the behavior of the fallback state selection part
> > of menu_select() so it looks at interactivity_req instead of
> > data->next_timer_us when it makes its decision.  That effectively
> > caused polling to be used more often as fallback idle which led to
> > significant increases of energy consumption in some cases.
> >
> > Commit e132b9b3bc7f (cpuidle: menu: use high confidence factors
> > only when considering polling) changed that logic again to be more
> > predictable, but that didn't help with the increased energy
> > consumption problem.
> >
> > For this reason, go back to making decisions on which state to fall
> > back to based on data->next_timer_us which is the time we know for
> > sure something will happen rather than a prediction (which may be
> > inaccurate and turns out to be so often enough to be problematic).
> > However, take the target residency of the first proper idle state
> > (C1) into account, so that state is not used as the fallback one
> > if its target residency is greater than data->next_timer_us.
> >
> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> > ---
> > 
> > Doug, can you please see if this patch helps to address the problem
> > you're seeing?
> > 
> 
> Yes, it appears to work great.
> 
> >
> > Commit e132b9b3bc7f is in linux-next only ATM, but it is a rebased
> > version of https://patchwork.kernel.org/patch/8602101/.
> 
> So, I applied it to my existing kernel 4.5-rc7 work, i.e. the not
> rebased version.
> 
> My reference = rvr7:
> Kernel 4.5-rc7 +
> Rafael's 3 patch set version 10 "cpufreq: Replace timers with utilization update callbacks" +
> Rik's patch (rvr5) "cpuidle: menu: use high confidence factors only when considering polling" +
> Rafael's patch from herein.
> 
> My reference: k45rc7-rjw10-rvr7
> 
> Aggregate Idle States in minutes for 2000 second test (some old data re-stated for reference):
> 
> State	k45rc7-rjw10	k45rc7-rjw10-reverted	k45rc7-rjw10-rvr7
> 0.00	18.07			0.92				0.41
> 1.00	12.35			19.51				21.17
> 2.00	3.96			4.28				4.40
> 3.00	1.55			1.53				1.66
> 4.00	138.96		141.99			150.77
> 			
> total	174.90		168.24			178.41
> 
> Energy:
> >>>> Kernel 4.5-rc7-rjw10: 61983 Joules
> >>>> Kernel 4.5-rc7-rjw10-reverted: 48409 Joules (test 2 was 55040 Joules)
> k45rc7-rjw10-rvr7: 54748 Joules
> 
> The trace data has a record low number of long durations at 71.

OK, thanks!

I'm going to apply it for 4.6, then.

Thanks,
Rafael