Re: bug? acpi p-state + ondemand keeps dropping max freq

["Björn Steinbrink" <B.Steinbrink@gmx.de>]

On 2008.06.16 07:43:37 -0700, Arjan van de Ven wrote:
> On Mon, 16 Jun 2008 12:42:00 +0200
> Pavel Machek <pavel@suse.cz> wrote:
> 
> > On Sat 2008-06-07 14:54:35, Arjan van de Ven wrote:
> > > On Sat, 7 Jun 2008 23:39:18 +0200
> > > Pavel Machek <pavel@suse.cz> wrote:
> > > 
> > > > > # echo 1866000 > scaling_max_freq ; cat scaling_max_freq
> > > > > 800000
> > > > > # echo 1866000 > scaling_max_freq ; cat scaling_max_freq
> > > > > 800000
> > > > > 
> > > > > 
> > > > > This renders my Dothan to utterly poor speeds. (standard T43)
> > > > > 
> > > > > performance cpufreq governor makes no difference - I still can't
> > > > > change the frequency upper/lower values.
> > > > 
> > > > Hmm, I have similar problem in Novell bugzilla, on very different
> > > > hw:
> > > > 
> > > > https://bugzilla.novell.com/show_bug.cgi?id=396311
> > 
> > > are either of you running gnome-power-manager or kpowersaved ?
> > > sometimes these programs (and more likely, the patches added by a
> > > distro maintainer who doesn't fully realize how power works) tend to
> > > muck with kernel settings around CPU frequency that they have
> > > absolutely no business touching...
> > 
> > In novel bugzilla case ignore_ppc=1 helped, so it seems to be BIOS
> > problem, not userland's...
> 
> well as long as the user doesn't use this for production use... the
> BIOS often reduces frequencies available to deal with thermal
> situations, so it's not a good idea to ignore that.

Yep, seems to be a thermal thing. I managed to find some time to play
around with it a bit, and running a cpu hog for a short period of time,
while watching temperature and scaling_max_freq showed this:

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             49 C
cpufreq# cat scaling_max_freq
1866000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             51 C
cpufreq# cat scaling_max_freq
1866000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             52 C
cpufreq# cat scaling_max_freq
1866000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             53 C
cpufreq# cat scaling_max_freq
1866000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             54 C
cpufreq# cat scaling_max_freq
800000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             53 C
cpufreq# cat scaling_max_freq
800000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             51 C
cpufreq# cat scaling_max_freq
800000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             51 C
cpufreq# cat scaling_max_freq
800000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             50 C
cpufreq# cat scaling_max_freq
800000

cpufreq# cat /proc/acpi/thermal_zone/THM0/temperature
temperature:             49 C
cpufreq# cat scaling_max_freq
1866000


So upon reaching 54°C some throttling kicks in and only when going back
to less then 50°C, that limit is lifted again. Too bad that with Linux,
this T43 already runs at about 47°C when idle, so as soon as there's any
load on the cpu, it will scale up for a few seconds and then get
throttled :-(

There's no userspace powersave foo involved here, just the plain
ondemand scaling governor, same happens with the performance governor.

Björn