On 02/28/2018 10:40 PM, Julian Viereck wrote: > Thanks for your pointer to hwlatdetect and possible SMI issues. > > It looks like the hwlatdetect script is not working with the latest > kernel anymore. However, it appears the hwlat is now part of the > kernel's ftrace [1]. Here is the kernel documentation about how to use > and understand the trace loge file [2]. > > Using a trace window of 1sec, trace width of 0.5 sec and tracing_thresh > of 20 microseconds, there are no unexpected latencies detected by the > hwlat tool. Therefore, I doubt the issue is due to SMI interrupts. > > In the meantime, I've also tested: > > 1) disable hyperthreading and running a single core
147 us is clearly pathological Xenomai-wise. This can't be a SMT/SMP overhead. > 2) run the xenomai latency test from the console before booting the x11 > desktop (to rule out any graphics issues) > > In both cases, the high latencies under no-workload remains. > What are the exact settings regarding the CPU_IDLE and CPU_FREQ (governor) frameworks? > Any other ideas what to look into / debug / run traces on? > At this point, we'd need a pipeline trace. Could you take a snapshot with this tool [1] by running the latency test with the -f option? [1] http://xenomai.org/2014/06/using-the-i-pipe-tracer/ -- Philippe. _______________________________________________ Xenomai mailing list Xenomai@xenomai.org https://xenomai.org/mailman/listinfo/xenomai