Back finally and I see lots of mails over cpufreq stuff.. :)
On 3 September 2013 18:50, Srivatsa S. Bhat srivatsa.bhat@linux.vnet.ibm.com wrote:
This doesn't solve the problem completely: it prevents the store_*() task from continuing *only* when it concurrently executes the __cpufreq_governor() function along with the CPU offline task. But if the two calls don't overlap, we will still have the possibility where the store_*() task tries to acquire the timer mutex after the CPU offline task has just finished destroying it.
How exactly? My brain is still on vacations :)
Anyway, this was one of the problem that I tried to solve with my patch. But there can be other race conditions where things can go wrong and so that patch may still be useful.
Call to __cpufreq_governor() must be serialized I believe.