Re: Target channel of state dumps

* Jan Kiszka ([email protected]) wrote:
> Hi,
> 
> my customer pointed out some oddity of the hybrid mode: Names of
> long-living process (ie. at least those that exist during the full trace
> period) are not known to lttv.
> 
> I dug deeper and learned that the initial state dumps, like those issued
> via list_process_state, go to the cpu channel by default and will thus
> be quickly overwritten. Reprogramming the target channel of
> list_process_state solves this for me:
> 
> echo "set_channel list_process_state default 0 processes" > /proc/ltt
> 
> And the above includes a second "oddity": The actual "set_channel"
> syntax does not meet the intended one as documented (only!) in the
> source code. Will check later if I can provide a patch for this.
> 
> But back to the actual issue: Isn't redirecting those state dumps to the
> channels under control/ the original intension? If so, should I fix up
> ltt-armall? What about redirecting exec events to "processes", in order
> to learn also about new threads over the trace runtime? I thought that
> this is what the hybrid mode is about, right?
> 

Yes, the intent is that events that are required to get a good knowledge
of the system processes should be written in the processes channel.

Namely :

exec
fork
exit
list_process_state

I may miss some. Basically, lttv/lttv/state.c is a good information
source to see what events are interesting.

I have not tested this since I have implemented the /proc/ltt interface
to activate the markers, this is relatively new.

I agree that the /proc/ltt interface needs some polishing.. I did it so
I could have something that "works", expecting to have to iterate a
little more on this when I would submit it to LKML. If you have fixes to
propose, I would take them gladly.

Thanks!

Mathieu

> Jan
> 
> -- 
> Siemens AG, Corporate Technology, CT SE 2
> Corporate Competence Center Embedded Linux
> _______________________________________________
> Ltt-dev mailing list
> [email protected]
> http://listserv.shafik.org/mailman/listinfo/ltt-dev
> 

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68