Re: Information request about acpi

[Len Brown <lenb@kernel.org>]

> Sorry for disturbing, I am Tsafack Ghislain Landry, Ph.D. student at Ecole
> Normale Superieure of Lyon, France. I have been trying to scale the frequency
> on my Intel E5506 Xeon processor via the cpufreq_acpi driver unfortunately (I
> have tested that feature using the following linux kernel versions : 2.6.32,
> 2.6.38, 2.6.39, 2.6.39.3, and 3.0.rc7).  This a sample output
> 
> # cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
> 2128000
> #echo 1862000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed
> # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed
> 1862000
> # cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
> 2128000
> 
> I have tried using the cpufreq-set util as well, but the result is the same.
> 
> bellow is the content of the /proc/acpi/processor/CPU0/info file. I was
> wandering whether the "no" value of the parameters  throttling control, limit
> interface can fully explain that behavior.

they are unrelated to this issue.

> cat /proc/acpi/processor/CPU0/info
> processor id:            0
> acpi id:                 1
> bus mastering control:   no
> power management:        yes
> throttling control:      no
> limit interface:         no


It is likely that you are obsering the effects of "hardware coordination."
Although Linux treats all the cores as independent, and even treats
HT threads as independent, they actually have dependencies.

In particular, all threads inside a package share the same voltage 
regulator.

So in a package if one core asks to go fast and another asks to go slow,
the voltage will coordinated by hardware to support the most demanding
request.  Since it is always a good idea to go at he maximum speed
suported by the available voltage, the cpu that requests slow will
also go fast, the same speed as his peer.

boot with "maxcpus=1" and try again -- or control the speeds
of the other threads on the system accordingly.

cheers,
-Len Brown, Intel Open Source Technology Center