[ltp] Thinkpad R40 2722: various CPUFREQ troubles

flokno linux-thinkpad@linux-thinkpad.org
Thu, 11 Dec 2003 14:53:01 +0100


On Thu, 2003-12-11 at 14:31, Sebastian Kapfer wrote:
> Hello Thinkpad users,
> My Thinkpad R40 (2722-GDG) works quite well under Linux, except for a
> few obstacles. It would be nice if someone could shed a bit of light on
> the issue. I have found a mail with a single question on this list, but
> without a real solution. :-(

i have the same notebook.
did you upgrade your bios? maybe this helps...!

> * I Can't clock the machine at 1500 MHz when disconnected from external
> power.

i set speed to maximum in bios on all occasions, no powersave. always
performance. (that is also fixed max afair). if you need more exact
information i will have to reboot and look it up.

>  I know IBM wants to conserve battery power here. BUT I DON'T
> CARE. A useless (slow) system is worse than one running two hours less.

what is your current runtime on batteries?

mine is always around 3 hours compared to 5:30 hours under windows.
meaning the processor speed doesn't change much.

> Does this depend on BIOS settings? I think I tried everything possible,
> but never got past 600 MHz.

yes. i think it depends.

> * The system time runs slow. Order of magnitude is about 5 minutes slow
> per hour of operation. This doesn't seem to occur when it is turned off.
> I have to ntpdate several times a day. :-(  CPU frequency doesn't seem
> to have any influence. Time is lost at about the same speed no matter if
> 1500 or 150 MHz. Google says ACPI will fix this (and it does), but with
> ACPI the battery status doesn't work any more.

with 2.6.0-test11 battery is working.
but system shutdown is only possible when pressing the power button for
some seconds after linux shutdown.

any hints on this?

> * Strange message in the syslog (appears once, some minutes after
> booting):
> | Losing too many ticks!
> | TSC cannot be used as a timesource. (Are you running with SpeedStep?)
> | Falling back to a sane timesource.
> The answer to that question is probably "yes". The fallback timesource
> obviously isn't more sane either, because it's still wrong :-(
> After hours of googling, I found out that there's a kernel parameter
> which controls the default timesource. Available options are "pit",
> "tsc" and "cyclone".
> cyclone isn't accepted by the kernel at all; tsc is the default anyway,
> and when booted with clock=pit, CPUFREQ fails to detect a CPU. So I'm
> not sure what to do to get rid of this message.
> Summary of the system configuration:
> Pentium M CPU, Linux 2.6-test9 kernel with the CPUFREQ stuff, APM mode,
> acpi=off (you know, the battery).

test9 is some weeks old... why not use test11?

> BIOS settings: CPU frequency is set to "fixed max". (But still the CPU
> runs at 600 instead of 1500.)

you can set it two times somehow.

> Enabling or disabling SpeedStep in the BIOS doesn't make any difference
> whatsoever. CPUFREQ happily changes frequency in both modes, the system
> time is slow in both.