[ltt-dev] LTTng. Strings with missing characters (ARM)

[mathieu.desnoyers@polymtl.ca (Mathieu Desnoyers)]

(adding the ltt-dev list in CC, so other developers can have a look at
it and help reproducing this ARM problem)

Hrm, weird,

So /proc/ltt outputs correct strings, but ltt-armall output is messed
up.

Can you do the following ?

which ltt-armall

and

cat `which ltt-armall`

I guess it has something to do with the IFS=$'\n' line in the script,
which works on x86, but maybe not on arm for some reason.

Also, can you give me the output after this sequence of commands ?
ltt-armall
lttctl -C -w /tmp/trace trace
find /mnt/debugfs/ltt  (where /mnt/debugfs is the path to your debugfs
mount)

Thanks

Mathieu

* Akyurek, Ali (EXT) (ali.akyurek.ext at siemens.com) wrote:
> 
> Hello,
> 
> Strings are seem to be cutted off at 'n' character.
> 
> -----Urspr?ngliche Nachricht-----
> Von: Mathieu Desnoyers [mailto:mathieu.desnoyers at polymtl.ca] 
> Gesendet: Montag, 22. Dezember 2008 19:51
> An: Akyurek, Ali (EXT)
> Betreff: Re: LTTng. Strings with missing characters
> 
> 
> Hrm,
> 
> What do you get if you do :
> 
> cat /proc/ltt ?
> 
> And what is your kernel config ?
> 
> Mathieu
> 
> * Akyurek, Ali (EXT) (ali.akyurek.ext at siemens.com) wrote:
> > 
> > 
> > 
> > Hello again,
> > 
> > I've installed latest versions LTTng 0.70, lttcontrol 0.61.
> > 
> > When i run armall, in the output, i see strings with missing characters like:
> > 
> > ...
> > Connecting el irq_exit
> > ker
> > Connecting el kthread_create
> > ker
> > Connecting el kthread_stop
> > ker
> > Connecting el kthread_stop_ret
> > ker
> > ...
> > 
> >  
> > Also after i'm done with tracing, the trace output directory looks like:
> > 
> > drwxr-xr-x    2 root     0               0 Jan  1 02:12 .
> > drwxr-xr-x    3 root     0               0 Jan  1 01:55 ..
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 _page_e_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 d_sig_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 dle_fault_e_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 el_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 etif_state_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 etwork_ip_i_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 etwork_ipv4_i_0
> > -rwxr-xr-x    1 root     0           65536 Jan  1 01:55 fd_state_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 metadata_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 mm_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 put_0
> > -rwxr-xr-x    1 root     0          262144 Jan  1 01:55 put_eve_0
> > 
> > Do you have an idea about it?
> >  
> > Thanks.
> > 
> > 
> > -----Urspr?ngliche Nachricht-----
> > Von: Mathieu Desnoyers [mailto:mathieu.desnoyers at polymtl.ca]
> > Gesendet: Freitag, 19. Dezember 2008 18:17
> > An: Akyurek, Ali (EXT)
> > Betreff: Re: arm userspace tracing
> > 
> > Hello !
> > 
> > Well, it's your lucky day I guess. I just added a neat feature to LTTng
> > which will help a great deal sending a subset of the data types
> > supported (a simple string) into the trace buffers from userspace _very_
> > easily. It's not integrated with the event management, so I would not
> > add this to application in production (because it would add overhead
> > even when tracing is disabled). With the latest LTTng, all you have to
> > do is :
> > 
> > modprobe ltt-userspace-event
> > lttctl -C -w /tmp/trace trace (note : the arguments recently changed)
> > run stuff...
> > echo "Some event happened!" > /mnt/debugfs/ltt/write_event
> > lttctl -D trace
> > looking specifically for the event :
> > lttv -e userspace.event -m textDump -t /tmp/trace
> > 
> > So basically, you can look for your own string into the trace stream as
> > a starting point. This is portable, so should work on ARM without any
> > problem.
> > 
> > There is still a lot of work to do on ARM to have per-site activation of
> > tracepoints like we currently have in the kernel, but at least this
> > solution gives us _something_. It's just that it's always enabled.
> > 
> > * Akyurek, Ali (EXT) (ali.akyurek.ext at siemens.com) wrote:
> > > Hello Mr.Desnoyers,
> > > 
> > > First of all, thanks for the great work you've done.
> > > What i wonder is that is it possible to trace user space in arm architecture?
> > > In the quick start document it writes :
> > > 
> > > "Get the latest markers-userspace-*.tar.bz2 and see the Makefile and examples. It
> > > allows inserting markers in executables and libraries, currently only on x86_32
> > > and x86_64."
> > > 
> > > But ,then i've seen a patch in http://ltt.polymtl.ca/packages/markers-user/ called lttng-userspace-tracing-arm.patch <http://ltt.polymtl.ca/packages/markers-user/lttng-userspace-tracing-arm.patch> .
> > >
> > 
> > That was a complete syscall implementation at some point, which could be
> > ported to the new infrastructure, but note that the current
> > markers-userspace-*.tar.bz2 code also needs linker script modification.
> > 
> > Mathieu
> > 
> > > Thanks.
> > > 
> > > Mit freundlichem Gruss / Best Regards
> > > Ali G?nhan Aky?rek
> > > evoline TR
> > > 
> > 
> > --
> > Mathieu Desnoyers
> > OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
> > 
> 
> -- 
> Mathieu Desnoyers
> OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68