cpufreq regression due to dceff5c "cpufreq: fix serialization issues with freq change notifiers"

[Stephen Warren <swarren@wwwdotorg.org>]

One of the test cases I run (not often enough, hence I just noticed this
late) is the following, to make sure our cpufreq driver is present, and
adjusting the CPU clock as expected:

# echo 1000000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq ;
sleep 1 ; cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq ;
echo 216000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq  ;
sleep 1 ; cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq

which prints the following when it works:

1000000
216000

In next-20130909, that doesn't work, but instead prints:

216000
216000

I've also seen the following during git bisect:

608000
608000

... and perhaps other unexpected sets of values.

This is a regression. I found that if I revert dceff5c "cpufreq: fix
serialization issues with freq change notifiers", then the test case
works as expected.

Any idea what's up?