On AMD Phenom(tm) II X3 720 Processor with 2.6.30-gentoo-r8 (which should be equivalent to 2.6.30.9) seems to always report frequency of zeroth core as the "asserted by call to hardware" frequency cpufreq gets from /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq, while frequency reported in /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq differs for different cores in a way which is confirmed by framerate of mencoder (so it really can change frequency of cores independently). (Noticed with governer ondemand as "frequency stays at low limit" but experiments with switching to performance governer points to reporting incorrect core.)
I'll reassign this to cpufreq.
The problem you reported seems related to the changes below: [PATCH 0/2] powernow-k8: fix incorrect value reported in cpuinfo_cur_freq: http://www.spinics.net/lists/cpufreq/msg00766.html [PATCH 1/2] powernow-k8: read P-state from HW: http://www.spinics.net/lists/cpufreq/msg00767.html [PATCH 2/2] powernow-k8: get drv data for correct CPU: http://www.spinics.net/lists/cpufreq/msg00768.html These changes went into 2.6.31.
Do the changes that Naga Chumbalkar suggested resolve this issue?
Yes, the attached patches resolved the problem. Note that I can't try 2.6.31 because I have ATI card on that machine and must wait until the kernel mode-setting stabilise ...