On 2018-12-11 08:28:12 [+0000], Andreas Hoefler wrote:

Hi all

Him

I finally found the root cause of my problem! What made me very suspicious was the fact, that the spikes occur pretty much exactly after 950ms each time. 
It turns out that throttling was activated having sched_rt_runtime_us set to 950000. Switching it to -1 and everything runs just fine!

Be aware that once a RT task forgets to go to sleep, it may lockup /
starve the whole system.

Just out of curiosity: in case the throttling kicks in, are all the rt task just in a suspended state?

All RT tasks. This includes threaded interrupts which usually run at
FIFO 50.

Again thanx to Sebastian for helping me out here!

BR
Andy

Sebastian