[LITMUS^RT] DEMO plugin problems

[Björn Brandenburg]

> On 18. Jul 2017, at 10:10, Matteo Locatelli <matteo.locatelli1993 at gmail.com> wrote:
> 
> 
> >The plotting script still is pretty much beta, so it doesn’t handle unexpected data very gracefully. If you send me your traces I can take a >look. 
> 
> Ok, I send you the trace as an attachment. 

I had a look at the file. The file appears to miss almost all events after creation of the real-time task until a (final?) completion record. 

[         0] NAME            1984/0     on CPU  0 rtspin
[         0] PARAM           1984/0     on CPU  0 T=(cost: 10.00ms, period:100.00ms, phase:  0.00ms), part=0
[     80453] COMPLETION      1984/1     on CPU  0 


Have a look at where sched_trace_task_switch_to() and sched_trace_task_switch_away() are being called. These call sites are completely independent of whatever plugin happens to be active — a plugin cannot prevent these events from being recorded. So if you cannot see any scheduling events (look with st-dump), then something is else is happening to the trace. 

Could it be that your trace buffer is too small or that the tracing process is being starved?

You said your setup works without problem for PSN-EDF but not for the DEMO scheduler? It’s difficult to imagine a possible root cause. Anything else that seems to be broken?

Are you using rtspin -w and release_ts?

- Björn