[LITMUS^RT] Running LITMUS-RT on ARM64

Andrii Anisov andrii_anisov at epam.com
Mon Aug 28 15:39:05 CEST 2017


Hello Björn,

Please see my comments below:


On 24.08.17 12:32, Björn Brandenburg wrote:
> > Within the work of evaluating XEN RTDS scheduler discussed here [1], I've built a generic ARMv8 
> yocto image (core-image-full-cmdline) with a kernel taken from [2], 
> the domu layer which builds a guest system with liblitmus and 
> feather-trace-tools you can find here [3]. I ran the image as a guest 
> domain on Salvator-X board setup compiled as described here [3].
>
> The ARM64 support in general is still experimental. That said, I’m not 
> aware of any major defects in the current master branch, so it should 
> actually work.
I got it.

> > In order check feather-trace functionality I run rtspin tasks and 
> trying to collect traces using st-trace-schedule. The problem is that 
> traces collected seems to be broken. First of all st-job-stats can not 
> get task parameters from trace files,
>
> You need to launch the tracer before creating the processes. If the 
> tracer is launched after the real-time processes have been launched, 
> then no task parameters will be included in the trace.
Yep, it works for me.

> > than st-draw builds graphs which looks like a mess.
>
> I don’t know what happened with your PDF, but if I run st-draw from 
> the current feather-trace-tools master branch (release 2017.1) on the 
> files that you sent, it looks ok to me (result attached).
Well, actually I'm a bit frustrated about the PDF I sent you, it is 
empty, and not one intended to be sent. Sorry for confusing you.
Actually I get the same PDF as you, from those binary traces.
Actually I ask for some clarifications about the graph I see in the 
document.
Could you please explain why do I see that some tasks execution time do 
not conform with one what was configured. E.g.:
     1. PID 1550 with WCET 5ms, in period 20-to-30ms is represented with 
~3,5ms execution time
     2. PID 1554 with WCET 3ms, in period 10-to-20ms is represented with 
~5ms execution time
and so on.


-- 

*Andrii Anisov*





More information about the litmus-dev mailing list