[ltp] Re: Linux-Thinkpad digest, Vol 1 #3310 - 3 msgs
Wed, 10 Jul 2013 12:08:11 -0500
On Wed, 2013-07-10 at 12:01 +0200,
> Message: 1
> Date: Tue, 9 Jul 2013 15:12:09 +0200
> From: Marius Gedminas <email@example.com>
> To: firstname.lastname@example.org
> Subject: [ltp] Idle X220 temperature =?iso-8859-1?Q?incre?=
> Reply-To: email@example.com
> (Prompted by the other thread, and a recurrence of the event.)
> Sometimes (but not always), when I resume my X220 (4291WJF), its
> temperature starts increasing from somewhere around +42=B0C up to around
> +74=B0C, where it stays for a while. Although today it fell down to +68=B0C
> and is now holding steady.
> All CPUs are idle. powertop says both cores spend over 90% in C7.
> Frequency-wise, oh look: powertop claims all four CPUs are running at
> 2.7=B10.1 GHz. Meanwhile /proc/cpuinfo says 800 MHz for all cores, most
> of the time, with brief spikes up to 2501 on random ones.
> This is Ubuntu 13.04 (GNOME Remix).
> $ uname -a
> Linux platonas 3.8.0-26-generic #38-Ubuntu SMP Mon Jun 17 21:43:33 UTC 2013=
> x86_64 x86_64 x86_64 GNU/Linux
> I've now recorded lspci -vvv (https://gist.github.com/mgedmin/5957200),
> will compare with normal one when laptop starts behaving normally again.
> Marius Gedminas
I have so much trouble with suspend-to that I don't try to use it at
There is at least one bug in the *buntu world with CPU scaling. Sadly, I
cannot find the reference as I write this. I'll keep looking and report
back when I track things down.
You don't say which flavor of "resume" you mean. Unfortunately, I find
the language used consistently. Do you mean continue after
suspend-to-RAM (sleep) or suspend-to-Disk (hibernate)? In the latter
case, after saving state, the hardware does power-off. Continuation
requires power-on-restart followed by state recovery.
After suspend-to-RAM, all of the bits stay where they are, moving parts
are stopped, and the CPU throttles back to near zero. Continuation thus
requires ramping the CPU back to full operation -- something very
similar to "performance scaling". This sounds like a good opportunity to
exercise that bug I mentioned,