Skip to contentSkip to conversation

"lttng enable-event" fails

DISCUSSION David Goulet 2 participants 2 messages Jun 24, 2014 – Jun 26, 2014 2 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 28 lines · Thierry Bultel

Hi,

I am using lttng (2.4.1), target is ARM Cortex A9; compiler is gcc-4.7.3, but I have the same issue with gcc-4.8.2
The kernel is Freescale's 3.0.35-4.1.0

lttng fails to start:

[root@buildroot ~]# lttng create
Session auto-19700319-135920 created.
Traces will be written in /root/lttng-traces/auto-19700319-135920

[root@buildroot~]# lttng enable-event --kernel --all
PERROR [5831/5915]: wait: Success (in run_as_clone() at runas.c:243)
Error: Trace directory creation error
Error: Events: Kernel create session failed (channel channel0, session
auto-19700319-135920)

My analysis show that lttng attempts to create the trace directory, when "lttng enable-event" is called.
To do so, it uses a slave thread that performs a kind of "mkdir -p" , that thread is spawned via a call to "clone"
and the father waits for it via "waitpid".

When the slave returns, waitpid says it is not completed. Added a little code to shows that the thread
is killed with SIGSEGV at the time it calls "return"

Quoted 5 lines

Changing it to call "exit" instead, makes it a little better, but other numerous errors come when
calling "lttng start"

Did anyone encounter such an issue ?

Hrm, tbh I never tried lttng-tools on ARM but if a SIGSEGV is raised you should be able to get a core dump and with gdb we can understand which callsite failed badly.

Thanks!
David

Quoted 11 lines

Best regards
Thierry Bultel

_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

A message that is not in the archive (sent off-list or not yet received): 20140624135319.GB11698@thessa

#3 lore raw

Some other useful information probably:
The bug does not happen with the (old) Linaro 2012.10 compiler:

# ./host/usr/bin/arm-linux-gnueabihf-gcc --version
arm-linux-gnueabihf-gcc (crosstool-NG linaro-1.13.1-4.7-2012.10-20121022 - Linaro GCC 2012.10) 4.7.3 20121001 (prerelease)
Copyright © 2012 Free Software Foundation, Inc.

Whereas it happens with the one provided by buildroot:

../buildroot_agv_v15_amos820/host/usr/bin/arm-linux-gcc --version arm-linux-gcc (Buildroot 2014.02-rc3-00323-g2f12017) 4.7.3 Copyright © 2012 Free Software Foundation, Inc.

Thus I am submitting that bug report to buildroot as well.

Regards
Thierry

Quoted 53 lines · David Goulet
Quoted 28 lines · Thierry Bultel

Hi,

I am using lttng (2.4.1), target is ARM Cortex A9; compiler is gcc-4.7.3, but I have the same issue with gcc-4.8.2
The kernel is Freescale's 3.0.35-4.1.0

lttng fails to start:

[root@buildroot ~]# lttng create
Session auto-19700319-135920 created.
Traces will be written in /root/lttng-traces/auto-19700319-135920

[root@buildroot~]# lttng enable-event --kernel --all
PERROR [5831/5915]: wait: Success (in run_as_clone() at runas.c:243)
Error: Trace directory creation error
Error: Events: Kernel create session failed (channel channel0, session
auto-19700319-135920)

My analysis show that lttng attempts to create the trace directory, when "lttng enable-event" is called.
To do so, it uses a slave thread that performs a kind of "mkdir -p" , that thread is spawned via a call to "clone"
and the father waits for it via "waitpid".

When the slave returns, waitpid says it is not completed. Added a little code to shows that the thread
is killed with SIGSEGV at the time it calls "return"

Quoted 5 lines

Changing it to call "exit" instead, makes it a little better, but other numerous errors come when
calling "lttng start"

Did anyone encounter such an issue ?

Hrm, tbh I never tried lttng-tools on ARM but if a SIGSEGV is raised you should be able to get a core dump and with gdb we can understand which callsite failed badly.

Thanks!
David

Quoted 11 lines

Best regards
Thierry Bultel

_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev