[ltp] Temperature increase after S2Ram

[Paul Bolle]

On Mon, 2013-07-08 at 09:48 +0200, Paul Bolle wrote:
> On Mon, 2013-07-08 at 00:07 -0300, Henrique de Moraes Holschuh wrote:
> > Nah, let's assume 3.10 is borked.  Please check whether the latest 3.9.x
> > works well for you...
> 
> I hope to do so shortly.

Tested the last version of v3.9 I had at hand (v3.9.8-rc1). Booted it
once, suspended and resumed a few times, and noticed no problems. So
that light test suggests this is a v3.10 specific problem.

That also means I've hijacked Axel's thread, unless Axel was running a
V3.10 release candidate.

> > I see.  Can powertop show idle states on that box?  If it does, keep an eye
> > on them.
> 
> But powertop (at least) masks the problem: while powertop is running
> CPU1 (the second core that seems to be stuck at going 2 GHz) does drop
> back to 800 MHz.

The effect of powertop currently seems to be:
- when running always allows the core previously stuck at 2 GHz to drop
back to 800 MHz;
- when stopped: sometimes that core jump backs to 2 GHz, sometimes it
still can drop back to 800 MHz. Ie, I notice no clear pattern when I
quit powertop.

(Offlining and onlining always allows the core to drop back to 800 MHz.)

> Another curious thing, perhaps related: acpi_cpufreq's refcount wraps
> after offlining and onlining this core a few times:
>     $ cat /proc/modules | grep acpi_cpufreq
>     acpi_cpufreq 19612 18446744073709551609 - Live 0xffffffffa0465000 (F)
>     mperf 12607 1 acpi_cpufreq, Live 0xffffffffa0418000 (F)
> 
> 18446744073709551609 equals (2 ^ 64) - 7. I guess I've offlined that
> core seven times now.

This can also happen by just suspending (I saw it on v3.9.8-rc1).
Current theory: references increase per CPU, and (sometimes?) decrease
per core.


Paul Bolle