[lttng-dev] [PATCH] Expose kernel tracer to user-space (version 6)

[Mathieu Desnoyers]

* Francis Giraldeau (francis.giraldeau at gmail.com) wrote:
> Le 2012-07-03 13:39, Mathieu Desnoyers a écrit :
> > * Francis Giraldeau (francis.giraldeau at gmail.com) wrote:
> >> Le 2012-07-03 00:23, Mathieu Desnoyers a écrit :
> >>> * Francis Giraldeau (francis.giraldeau at gmail.com) wrote:
> >>>> By writing to the file /proc/lttng_uevent, a user-space application creates a
> >>>> kernel event. The event's payload is by default UTF-8 text, but any data can be
> >>>> written, up to 1024 bytes. Null-character is optional and is not enforced.
> >>>
> >>> The 1024 byte limit is removed now, right ?
> >>
> >> Exactly.
> > 
> > OK, please provide an updated version with the updated changelog.
> 
> Fixed.
> 
> > 
> >>
> >>>> + * Expose kernel tracer to user-space through /proc/lttng
> >>
> >> I just saw a typo here, we should read "through /proc/lttng_uevent"
> > 
> > Same here.
> 
> Fixed.
> 
> > 
> >>
> >>>> + * Copyright (C) 2009-2012 Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
> >>>
> >>> you might want to add your own copyright in here too.
> >>
> >> I transfer you the copyright.
> > 
> > Thanks!
> > 
> > One more thing:
> > 
> > In the testing you did, did you try running, in parallel:
> > 
> > for a in $(seq 1 10000); do
> >         modprobe nameofuevent_module
> >         rmmod nameofevent_module
> > done
> > 
> > and in 4 another terminals:
> > 
> > echo "blahblahblah" > /proc/lttng_uevent
> > 
> > ?
> > 
> > I understand that you tested only the write path in parallel, but I'm
> > not sure, from your changelog, that you stress-tested load/unload vs
> > write.
> 
> The test I did was to write in parallel to /proc/lttng_uevent and see
> the behavior. It takes about 15us on my i5 2.5Ghz to log an event this
> way in a tight loop, so yeah, it's quite slow.

Yep, that's benchmarking. I'm more concerned about correctness.

> It's not possible to remove the module while an active tracing session
> references the event lttng_uevent. I tested insertion/removal of the
> module while tracing with the previous version that was using RCU and it
> was working well.

But you don't need to have an active tracing session to perform the test
I point to above. Could you run it on this latest version ?

Thanks,

Mathieu

> 
> Cheers,
> 
> Francis
> 



> _______________________________________________
> lttng-dev mailing list
> lttng-dev at lists.lttng.org
> http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev


-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com