linux-kernel - Re: [linux-pm] [PATCH] cpuidle: avoid singing capacitors

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:	Sat, 15 Mar 2008 01:41:43 +0100
From:	Pierre Ossman <drzeus-list@...eus.cx>
To:	"Pallipadi, Venkatesh" <venkatesh.pallipadi@...el.com>
Cc:	"Len Brown" <lenb@...nel.org>, linux-pm@...ts.linux-foundation.org,
	"Pavel Machek" <pavel@....cz>,
	"LKML" <linux-kernel@...r.kernel.org>,
	"Adam Belay" <abelay@...ell.com>,
	"Andi Kleen" <andi@...stfloor.org>,
	"Lee Revell" <rlrevell@...-job.com>
Subject: Re: [linux-pm] [PATCH] cpuidle: avoid singing capacitors

On Fri, 14 Mar 2008 14:15:46 -0700
"Pallipadi, Venkatesh" <venkatesh.pallipadi@...el.com> wrote:

> 
> Why go only one C-state down. Why not directly drop to max_hwcstate? We
> don't need to loop through that way.
> 

Good point. But we can't just jump there and be done with it. That state might not be available (then again, we could require the user to put in a valid max_hwcstate, but that isn't as user friendly).

> There are few other concerns which make me feel that the patch will be
> not as simple as this.
> 1) BIOS may already be exporting the lower C-states as separate states.
> In which case we just have to ignore this deep C-state and return. I
> mean, on your system if BIOS exports C1, C3 and C6 hardware C-states and
> you give max_hwcstate as C3, then we don't want to have C1, C3 and C3 in
> our list.
> 

Indeed. But the current system doesn't easily allows us to handle that. Any ideas?

> 2) On same lines the other information in ACPI like (low power of 100
> and higher latency for C6 on your system) corresponds to hardware C6
> state. If we change the hardware C-state here to C3, then continue to
> use latency info from ACPI for hw C6, that may lead to inefficient state
> selection in the governor. Also, ther are assumptions related DMA,
> lapic, TSC etc in upper level ACPI based on "ACPI C-state" and changing
> underlying hardware C-state to C1 for example will change some of those
> behavior.
> 

It is still way better going to C3 less than possible than not at all (i.e. the previous workaround of processor.max_cstate). As far as I can tell from the documentation, every C-state includes all of the negative side effects of the ones preceding it. So it should just be a matter of sub-optimal selection by the governor. And I don't see how we can fix that without a big table for each processor and possibly chipset.

Rgds
-- 
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org
--
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/