Fosdem impressions, frysk, packaging, tracepoints, kprobes and gdb

Fosdem impressions, frysk, packaging, tracepoints, kprobes and gdb

[Mark Wielaard]

Hi,

I was away for the last 4 days and also attended Fosdem this weekend in
Brussel http://fosdem.org/2007/. That was a blast. I was mainly there to
see how Sun and the GNU communities can work together on the future of
Java now that Sun is in the process of releasing all parts of their
implementation under the GPL. This was all very promising. Even though
there are still some holes and details to work out. I do think that in a
year choosing java as an implementation language on GNU/Linux platforms
will be much more accepted.

But I also got a chance to talk with various random people about Frysk.
There were people interested in packaging Frysk for other distros (like
Mandrake), but apparently they have some trouble with how we depend on
things we import into our source tree like junit. I will try to get them
to post to the Frysk list since I could immediately figure out what
their exact problems were.

While talking about Frysk with other people I did notice that I am still
missing some of the bigger picture. While I can clearly explain how our
tracing and monitoring is different for things like systemtap, valgrind
and oprofile, it is harder to explain how we will inter-operate with
things like that. And there is a huge interest to see how we are
different and complementary to gdb. This came a bit as a surprise to me
since I hadn't seen Frysk as that related to gdb or full-blown debuggers
before.

But now that I played a bit with the fhpd tool it is clear we should
have a much better story how we are tackling the same kind of problems
gdb is trying to solve. It looks like we could actually attract some
people to the project if we just had a page explaining what kind of
problems you might want to do with gdb but which would be much easier to
solve with Frysk, and the other way around. We do have some high-level
overview in our FAQ http://sourceware.org/frysk/questions/ about
comparisons with adb or sdb (both of which I don't really know how they
compare with gdb), but nothing about the techniques and libraries gdb
uses versus those used by frysk to get at the same kinds of information.

Driving home the point that we might actually share much more ideas with
something like gdb was the talk by Jim Blandy on gdb, tracepoints and
kprobes for debugging the linux kernel you are running
on your machine. This was really fascinating and he gave a very
impressive demo of collecting live trace data on the machine he was
giving the presentation on and then using the captured information to
drive gdb to show the call-stack, registers, function arguments and
contents of various structs passed around at various times during the
trace. http://www.red-bean.com/trac/tracepoints

I forgot to ask him how gdb can do this in user space. In kernel
space they are able to use kprobe handlers to collect the information
and then export them through the /proc interface for gdb to get at (to
be replaced by sysfs and kobjects). There is an old article on
tracepoints in gdb for embedded systems:
http://sourceware.org/gdb/talks/esc-west-1999/
But as far as I can tell that makes use of the fact that on the embedded
device gdb and the inferior both run in a shared memory space.

The tracepoint infrastructure is so useful that I think we should
definitely look into sharing ideas and implementation techniques with
gdb.

Cheers,

Mark

Re: Fosdem impressions, frysk, packaging, tracepoints, kprobes and gdb

[Mark Wielaard]

On Tue, 2007-02-27 at 14:55 +0100, Mark Wielaard wrote:
> Driving home the point that we might actually share much more ideas with
> something like gdb was the talk by Jim Blandy on gdb, tracepoints and
> kprobes for debugging the linux kernel you are running
> on your machine. This was really fascinating and he gave a very
> impressive demo of collecting live trace data on the machine he was
> giving the presentation on and then using the captured information to
> drive gdb to show the call-stack, registers, function arguments and
> contents of various structs passed around at various times during the
> trace. http://www.red-bean.com/trac/tracepoints

The video (Ogg/Theora) of his presentation is online at:
http://ftp.belnet.be/mirrors/FOSDEM/2007/FOSDEM2007-GDBTracePoints.ogg

Cheers,

Mark

Re: Fosdem impressions, frysk, packaging, tracepoints, kprobes and gdb

[Sami Wagiaalla]

Mark Wielaard wrote:
> On Tue, 2007-02-27 at 14:55 +0100, Mark Wielaard wrote:
>   
>> Driving home the point that we might actually share much more ideas with
>> something like gdb was the talk by Jim Blandy on gdb, tracepoints and
>> kprobes for debugging the linux kernel you are running
>> on your machine. This was really fascinating and he gave a very
>> impressive demo of collecting live trace data on the machine he was
>> giving the presentation on and then using the captured information to
>> drive gdb to show the call-stack, registers, function arguments and
>> contents of various structs passed around at various times during the
>> trace. http://www.red-bean.com/trac/tracepoints
>>     
>
> The video (Ogg/Theora) of his presentation is online at:
> http://ftp.belnet.be/mirrors/FOSDEM/2007/FOSDEM2007-GDBTracePoints.ogg
>   
Wow... way cool :)

    Sami Wagiaalla