gdb and too many threads causes SIGTRAP

gdb and too many threads causes SIGTRAP

[Kaarthik Sivakumar]

Hi

We are developing a multi-threaded application on RH9 which creates and 
deletes a large number of threads in a short time. After this program runs 
for sometime, gdb fails with a SIGTRAP in a thread that is at 
__ntpl_death_event() (when the thread wants to quit). Following some of the 
mails at gdb-patches, I noticed a thread (subject: [RFC]: fix for recycled 
thread ids) that seems to address this issue.

According to that mail thread, and this specific mail 
http://sources.redhat.com/ml/gdb-patches/2004-03/msg00672.html, a fix was 
issued to check for TD_DEATH when possible and to attach to threads sooner 
when thread ids are being reused. This fix seems to be available in gdb 
6.2.1, so we used that to debug the application. But this still does not seem 
to have fixed the problem. I can confirm that the code to test for TD_DEATH 
is activated and that glibc version is 2.3.2, so this fix should apply and 
work. But it doesnt.

Is there anything else I should be checking? If you need more information, 
please let me know. The OS is RH 9, with just the gdb 6.2.1 tar.bz2 compiled 
and being used. Thanks.

kaarthik

Re: gdb and too many threads causes SIGTRAP

[Daniel Jacobowitz]

On Wed, May 25, 2005 at 03:51:06PM -0700, Kaarthik Sivakumar wrote:
> Hi
> 
> We are developing a multi-threaded application on RH9 which creates and 
> deletes a large number of threads in a short time. After this program runs 
> for sometime, gdb fails with a SIGTRAP in a thread that is at 
> __ntpl_death_event() (when the thread wants to quit). Following some of the 
> mails at gdb-patches, I noticed a thread (subject: [RFC]: fix for recycled 
> thread ids) that seems to address this issue.
> 
> According to that mail thread, and this specific mail 
> http://sources.redhat.com/ml/gdb-patches/2004-03/msg00672.html, a fix was 
> issued to check for TD_DEATH when possible and to attach to threads sooner 
> when thread ids are being reused. This fix seems to be available in gdb 
> 6.2.1, so we used that to debug the application. But this still does not seem 
> to have fixed the problem. I can confirm that the code to test for TD_DEATH 
> is activated and that glibc version is 2.3.2, so this fix should apply and 
> work. But it doesnt.
> 
> Is there anything else I should be checking? If you need more information, 
> please let me know. The OS is RH 9, with just the gdb 6.2.1 tar.bz2 compiled 
> and being used. Thanks.

If you are compiling a new version of GDB anyway, please try a more
current release - if possible, a current CVS snapshot.  After that, if
you still have trouble, it may be a kernel problem.  Current versions
of GDB should not have this problem.

-- 
Daniel Jacobowitz
CodeSourcery, LLC