strange time drift if freq<=50%

I can confirm another case of what Dominik reports here; I don't see any other cases in the archive. I have a Fujitsu Lifebook 2010 with an AMD Athlon-M 1700+, running 2.5.72; if I set the clock to the minimum (via either

echo -n "0:63:100:powersave" > /proc/cpufreq

or

echo "powersave" > /system/devices/system/cpu/cpu0/cpufreq/scaling_governor

then I get "repeats" when I hit keys on my keyboard, my fan cranks up to high speed, and it seems like the system clock is running fast.

At 63% it seems to work properly.

joe

Dominik Brodowski wrote (on 4 Jun 2003):

On my p4 desktop I see a strange "time drift" if I set the frequency to 50% or below using the p4-clockmod driver.=20

Five runs of "time make" of a previously compiled kernel returned in average the following times:

freq: 100% 75% 62.5% 50%
time (avg.): 5.9s 7.8s 9.2s 22.5s=09

However, on my wall clock "time make" took only 11s on frequency=3D=3D50%. =
If I reduce the frequency even further, pressing any key on the keyboard results in many "repeats" appearing on the screen; and the system time running much too fast. If it's >=3D 62.5%, there's no problem at all.

The small patch at the bottom disables all frequency states <62.5% for the p4-clockmod driver until the real cause of this problem can be found. What I'd be interested in is the following, though: does the same bug appear on other cpufreq drivers and architectures? If so, this workaround should be abstracted and moved to a proper place (e.g. kernel/cpufreq.c or drivers/cpufreq/freq_table.c)

Dominik