[LITMUS^RT] kernelshark
Glenn Elliott
gelliott at cs.unc.edu
Mon Sep 16 23:39:39 CEST 2013
Jonathan helped me track down the issue. It appears that not all of Jonathan's ftrace were pulled into mainline Litmus. I think the container/server tracing was left out of the merge since containers/servers are not in mainline Litmus (https://github.com/LITMUS-RT/litmus-rt/commit/94edd278631ff219b4907322e620d9ffebceef4a). Unfortunately, an important change to ftrace.h was also in in 94edd2. Specifically, the bit of code that records a litmus_clock() timestamp to events. This timestamp is necessary for visualization.
I've pushed a fix to "prop/litmus-clock-ftrace" on github. The patch only modifies ftrace.h (container/server support is left out). Björn, could you pull this into staging?
Thanks to Jonathan for being so responsive to my questions.
-Glenn
On Sep 16, 2013, at 4:20 PM, Glenn Elliott <gelliott at cs.unc.edu> wrote:
> Has anyone tested kernelshark since the Litmus rebase? I am having trouble visualizing trace data.
>
> I use the command "trace-cmd record -e sched:* -e litmus:* -e timer:*" to capture data. I get about 60MB of trace data for 60 seconds of execution time (5 real-time tasks on a 4 CPU system). I can convert the binary trace data to text using "trace-cmd report". The resulting text shows that I have captured the desired events, including those from Litmus.
>
> However, when I try to visualize the data in kernelshark, a few events are drawn at the origin of the timeline (see attached), but nothing else is drawn. I've gone as far as rebasing rt-kernelshark on to the upstream kernelshark. The problem persists. (We can discuss incorporating these updates once this problem gets ironed out.) I've also tried capturing data using various versions of trace-cmd (the distro's, rt-kernelshark's, and upstreams).
>
> Comments? Suggestions?
>
> Thanks,
> Glenn
> <shark.png>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.litmus-rt.org/pipermail/litmus-dev/attachments/20130916/063276f3/attachment.html>
More information about the litmus-dev
mailing list