Linux Trace Toolkit
@) Preliminary Draft @) under construction
- 1 Introduction
- 2 Downloads
- 3 How To Use
- 4 References
- 5 Sample Results
- 6 Future Work
- 7 LKML Reaction to tracing
The Linux Trace Toolkit is used to examine the flow of execution (between processes, kernel threads, and interrupts) in a Linux system. This is useful for analyzing where delays occur in the system, and to see how processes interact (especially with regard to scheduling, interrupts, synchronization primitives, etc.)
* Note that as of 2.6.14, LTT is being replaced with LTTng. See LTTng
See also the announcement here: Patch: Linux Trace Toolkit Viewer/Next Generation announcement (LTTV/LTTng)
Tracing is useful for analyzing stuff.
See the QUICKSTART guide at: Quickstart
Sample i386 trace for LTTng
Here is a zip file containing a sample trace on an i386 machine: i386-ltt-trace.zip
0.9.6 Release (Nov 21, 2004)
I found it a bit difficult to get the appropriate set of patches for LTT. After a bit of frustration, I built my own release for Nov. 23, 2004. The release is 0.9.6 (no 'pre') and it works with the Linux kernel 2.6.9. The tar file for the release is in the PatchArchive.
The latest stable release is 0.9.5a, but this release is over 2 years old.
Here is the patch set I finally used, for kernel 188.8.131.52:
- message with ltt-184.108.40.206-relayfs.patch.bz2 by Mathieu Desnoyers
- message with ltt-220.127.116.11-ltt.patch.bz2 by Mathieu Desnoyers
getting attachments from ltt-dev archive
I had to use 'munpack' to extract the compressed patches from the mail messages in the archive. 'munpack' is part of the package 'mpack', which is available at: munpack
Here are some patches used by Tim Bird at Sony, targeted at MIPS, but known to also work on other platforms. This tarball contains both an all-in-one patch and a series file with discreet sub-patches.
- Patches for Linux kernel version 18.104.22.168, and !TraceToolkit tarfile are here:
These are bundles with multiple sub-patches and sub-tars. To install them, do the following:
- untar kernel patches:
tar -xzvf ltt-22.214.171.124.tar.gz
- apply patches:
./tpm -t linux-126.96.36.199.tar.bz2 -f ltt-188.8.131.52.pl -o linux-184.108.40.206-ltt
- untar !TraceToolkit stuff
tar -xzvf TraceToolkit-0.9.6pre3-plus.tar.gz
- unpack and apply patch:
./tpm -f TraceToolkit-0.9.6pre3-plus.pl -o TraceToolkit
Now, follow the build and usage instructions for the software in the include docs (!TraceToolkit/Help/index.html). For cross-compiling, use the instructions on this wiki page.
How To Use
Building the software
Apply the ltt and relayfs patches to your kernel:
- configure kernel with LTT
- turn on RelayFS and LTT
File systems --->
Pseudo filesystems --->
- Relayfs file system support
General Setup --->
- Linux Trace Toolkit support
You can leave klog debugging support turned off.
Compile the user-space tracedaemon program:
I couldn't figure out if the tracevisualizer tools handle cross-compilation correctly (by which I mean that you can natively compile the tracevisualizer but cross-compile the tracedaemon). Instead of blindly trying configure tricks, I instead used the instructions from Karim's book "Building Embedded Linux Systems". In a nutshell, the instructions go something like this:
* install source for user-space stuff * configure for native build * hand-build tracedaemon specifying a cross-compiler * build rest of user-space suite using native compiler * install programs as appropriate
Detail (in my case):
* install !TraceToolkit-0.9.6pre3, apply "plus" patch ** tpm -f user-space.pl -o !TraceToolkit-0.9.6pre3 * cd !TraceToolkit-0.9.6pre3 * configure --prefix=/home/tbird/work/ltt/tools * make -C !LibUserTrace CC=<cross-compiler-gcc> !UserTrace.o * make -C !LibUserTrace CC=<cross-compiler-gcc> LDFLAGS="-static" * make -C Daemon CC=<cross-compiler-gcc> LDFLAGS="-static" * cp Daemon/tracedaemon Daemon/Scripts/trace Daemon/Scripts/tracecore
Daemon/Scripts/traceu <target file syste>/usr/sbin
* make * make install
Using the software
In April, 2004, Karim wrote: lkml thread
The documentation is out of date. Basically, the createdev.sh script isn't needed anymore because of relayfs. You need to mount relayfs to use LTT. See the classic dox on filesystem mounting for this kind of thing. It's
going to be something like: # mount -t relayfs nodev /mnt/relay
There's no insmod for LTT. It isn't a device driver module, following LKML recommendations.
- LTT home page is at: LTT
- Presentation from OLS 2006:
- Online documentation is at: LTT Documentation
- This documentation is a bit old (2002), and has parts that are out of date.
Here is a list of things that could be worked on for this feature:
- online reference needs to be updated
- e.g. no mention of relayfs
- project seems fairly quiet, and unmaintained
- see if LTT patch should be refactored to take into account new kprobe support in the kernel
LKML Reaction to tracing
In Sep 2002, there was a thread about tracing, where some major kernel developers expressed their concerns about tracing infrastructure in the kernel.
my problem with this stuff is conceptual: it introduces a constant drag on the kernel sourcecode, while 99% of development will not want to trace, ever. When i do need tracing occasionally, then i take those 30 minutes to write up a tracer from pre-existing tracing patches, tailored to specific problems.</source>
so use the power of the GPL-ed kernel and keep your patches separate, releasing them for specific stable kernel ranches (or even development kernels).
> To summarize: You find tracing useful, but software tracing is only of limited value in areas you're working at. What about other developers, which only want to develop a simple driver, without having to understand the whole kernel? Traces still work where printk() or kgdb don't work. I think it's reasonable to ask an user to enable tracing and reproduce the problem, which you can't reproduce yourself.
That makes adding source bloat ok? I've debugged some drivers with dprintk() style tracing, and it often makes the code harder to follow, even if it ends up being compiled away.
From what I've seen from the LTT thing, it's too heavy-weight to be good for many things (taking SMP-global locks for trace events is _not_ a good idea if the trace is for doing things like doing performance tracing, where a tracer that adds synchronization fundamentally _changes_ what is going on in ways that have nothing to do with timing).
I suspect we'll want to have some form of event tracing eventually, but I'm personally pretty convinced that it needs to be a per-CPU thing, and the core mechanism would need to be very lightweight. It's easier to build up complexity on top of a lightweight interface than it is to make a lightweight interface out of a heavy one.