[ltt-dev] MISSING: ltt-usertrace

[Gerlando Falauto]

Hi!

I'm new here so please forgive me if I'm being silly...

I am trying to use LTTng in order to pinpoint some some weird behaviours
in a complex multi threaded application (e.g. unexplicable delays in the
magnitude of hundreds of milliseconds) which appear under very
unpredictable situations.

I wish for an integraded user/kernel space tracing mechanism so to be
able to measure latencies between known userspace markers, while knowing
what the kernel is doing in the meantime.
I am using a 2.6.33 kernel so I guess LTTng 2.x is not an option for me.

I did something similar to
http://lttng.org/files/usecases/LTTng-examples-siemens.pdf
using custom unimplemented syscalls, which gives me interesting
information in a very simple way, although I can only see the syscall
number without its arguments (which would be very useful indeed).
I saw a recent patch which addresses this but it's for LTTng 2.x so I
don't think I can use it.

I see how in past versions of LTTng there was ltt-usertrace:

http://svn.lttng.org/svn/trunk/attic/ltt-usertrace/README

which sortof does what I want (allow for tracepoint messages from user
space by adding custom syscalls ltt_trace_generic and
ltt_register_generic), but this seems deprecated somehow.
*Could somebody please tell me what happened and why it was removed?*

I know I have two alternatives:

a) Writing messages to /mnt/debugfs/ltt/write_event,
but I'm afraid this would add more uncertainty to the behaviour of the
system - e.g., affect scheduling - due to the fact that I'm writing to a
file (but maybe I am wrong).

b) Use LTTng Userspace Tracer (UST),
but I see two drawbacks:
1) Need to link in liburcu and change the startup command which is
something I would rather avoid
2) The trace information would end up in a separate trace file and I
really don't want to take the risk of having the kernel and user traces
based on different timestamps.

So could someone please shed some light?

Thanks!
Gerlando