Hi,

i found the cause for the bug: it was commit "[CPUFREQ] remove rwsem 
lock from CPUFREQ_GOV_STOP call (second call site)".
It release the rwsem lock in the middle of __cpufreq_set_policy, that is 
not a good idea. There are no dead lock, which that patch solved,
because cpufreq_ondemand and cpufreq_conservative don't use the rwsem
lock in do_dbs_timer anymore.

Regards,
Andrej Gelenberg