Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [linuxtools-dev] CTF timestamps in big endian 
On 08/31/2012 04:17 PM, Matthew Khouzam wrote:


On 12-08-30 12:27 PM, Gerlando Falauto wrote:

Hi everyone,

I am trying to use the stock LTTng plugin that came with the Eclipse IDE
for C/C++ Developers, Juno (version number looks like
1.0.0.201206130106) in order to view Kernel CTF traces generated by
LTTng 2.0 on two different embedded targets (ARM and PowerPC).

While on ARM (little endian) everything seems to work fine, on PowerPC
(big endian) I get absolutely meaningless timestamps, ranging to years
2050, 2145, and so on...
I get huge gaps (as I see only the time of day part, it looks like
hours, but it might be as well be years!) between consecutive events.

Groups of adjacent events do sometimes get displayed with reasonably
close timestamps though (I guess for those events which got dumped with
a compact representation of the timestamp and/or header).

The output of babeltrace for the same trace looks OK.

Any other fields (pids, tids, task names, etc..) which would be affected
by a global endianness problem still look OK, so it's definitely
something related to timestamps.

Any ideas? Shall I send the affected trace?

I cannot say yes enough to having big endian traces for testing.
To be honest, we only had x86 traces to test on so far.


That's what I figured... :-)


Thanks for providing this info.



Please find attached the affected testcase.
It is a 10-second kernel trace on a PowerPC idle system running a 3.0.40 kernel. The output from babeltrace looks OK. 

Thanks!
Gerlando
P.S. If you could provide some pointers on how to patch/upgrade the tools (to me this whole Eclipse world looks as complicated as a galaxy!), I would really appreciate it! :-)

Attachment: trace-20120828-090002.tar.gz
Description: GNU Zip compressed data

Back to the top