Update, after more tests: - With all the "*DEBUG*" entries disabled in .config, "nohz=off" in the kernel command line and the i915 driver disabled in .config, I consistently get a worst-case latency below 40us (this is the order of magnitude I was expecting). Good - With everything as in the above configuration, but enabling the i915 driver in .config, I get a worst case latency of about 80us. Applying Sebastian's patch does not seem to affect the latency. I also disabled the various tracing options in .config, but the maximum latency is still around 80us. The latencies mentioned above have been measured with cyclictest -n -p 80 -i 50 -t 1 -m -a while running "du -h /" and a "make -j 2" of the kernel in background. So, the numbers I get now are much better, but the i915 driver still seems to create some real-time issues. Luca On Thu, 6 Dec 2018 at 12:16, Luca Abeni <lucabe72@xxxxxxxxx> wrote: > > Hi Sebastian, > > On Thu, 6 Dec 2018 at 11:41, Sebastian Andrzej Siewior > <bigeasy@xxxxxxxxxxxxx> wrote: > [...] > > > BTW, playing with some options I discovered that if I add "nohz=off" > > > to the kernel commandline I get a maximum latency of about 150us when > > > using the i915 driver, and less than 100us when not using it. I am > > > going to test your patch, and to disable debugging and tracing... I'll > > > let you know the results. > > > > Is it better or worse with nohz=off? And do you get the backtrace? > > The latency with "nohz=off" is about 150us smaller than without this > boot option. > I did not get any other backtrace. > > Some preliminary tests show that disabling debug options and .config > reduces the latency to less than 50us (but I am still running > experiments). I'll also check if your patch allows to get the same > latency values with and without the i915 driver. > > Thanks, > Luca
