On Mon, Nov 16, 2015 at 8:09 AM, Miroslav Lichvar <[email protected]> wrote: > The problem with abs64() and the one in the thread were both actually > not related directly to the nohz code. The effect of disabling nohz in > 3.17+ kernels should be pretty small anyway and even if the nohz > option was broken it would probably not explain the difference in the > PPS jitter you see when the CPU is forced to wakeup. > > Is the CPU on BBB entering some deep sleep states like C2 and C3? > Maybe the idle=halt kernel option could reduce the observed jitter.
It was not entering cpuidle states, actually the beaglebone support for that just appeared this week. So I just tried to recompile a kernel and disabling dyntick. That's it. The jitter dropped to the best values I had before when having a task running every 1ms, so AFAIK nohz=off is really not working. And yes, beside dyntick, now with the cpuidle states, I also have to disable them. So thanks for the advise. Not sure if I should collect any more data regarding the nohz problem, on other platforms, before I refer this to the kernel people. Thanks, Nuno -- To unsubscribe email [email protected] with "unsubscribe" in the subject. For help email [email protected] with "help" in the subject. Trouble? Email [email protected].
