On 2016-10-19 10:55:11 [+0200], Christoph Mathys wrote: > I changed RCU_KTHREAD_PRIO to 99 from the default setting of 1. I now > observe cyclictest latencies of above 100us (16ms was the worst so > far). Before the change the systems max latency is below 30us. The > problem shows up when cyclictest has an equal or lower priority than > the RCU threads. RCU boosting is enabled (default). Ehm. It is not yet April. So you give another thread / application (the RCU thread in this case) a higher priority than cyclictest and now you complain about high latencies of cyclictest because the other thread runs? I would say it works as expected / configured. > I don't see that behaviour on 3.12 with RCU_BOOST_PRIO=99. I could > reproduce the behaviour on a Intel i7 and an Intel Celeron Dualcore. > On i7 it usually happens within a couple of minutes. It should lead to the same behaviour on v3.12 as well. It is possible that the later kernel has more RCU users which trigger the thread to run longer. Sebastian -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
