[omniORB] Using NJAMD with Omniorb

[Robert Stent]

Sai-Lai,

I agree with you that the demonstration program will now function, however, my
original source still experiences the same symptom.  I am sure we just pushed off
the inevitable.  I will try to pare down what I have to something that is
demonstrable.
I am also willing to bet if you have any real object code of your own that you
attempt to leak check you will see the same problem.


Sai-Lai Lo wrote:

> Mike,
>
> omnithread is just a small C++ wrapper class that is implemented on top of
> pthread. I can reproduce your segv.
>
> I've done some investigation. The segv can be avoided by reordering the
> libraries when they are linked. Specifically, if one links -ltcpwrapGK
> before -lomnithread and -lpthread, the program runs fine with libnjamd.so
> preloaded.
>
> In other words, do something like this instead with your example:
>
> cat > bad.cpp << EOF
> #include <iostream.h>
> #include <omniORB3/CORBA.h>
>
> int main(int argc, char *argv[])
> {
>     CORBA::ORB_ptr orb;
>
>     orb = CORBA::ORB_init(argc, argv, "omniORB3");
>
>     return 0;
> }
> EOF
>
> g++ bad.cpp -o bad -D__x86__ -D__linux__ -D__OSVERSION__=2 -D_REENTRANT \
>   -g  -I/usr/src/omni/include \
>   -L/usr/src/omni/lib/i586_linux_2.0_glibc2.1
>   -lomniORB3 -lomniDynamic3 -ltcpwrapGK -lomnithread -lpthread
>
> Sai-Lai
>
> >>>>> Mike Perry writes:
>
> > Hi, I'm the author of a malloc debugger (http://fscked.org/proj/njamd.shtml),
> > and it was recently brought to my attention that when used with omniorb v3,
> > my debugger crashes in its init function.
>
> > I investigated this, and its actually crashing in glibc (2.1.3 on RH6.2),
> > during a READ of errno (one of the save_errno assignments). This crash happens
> > for just about any call that uses errno. Now, I'm guessing that since this
> > happens REALLY early, the memory address errno resides in isn't mapped yet.
> > (OmniORB calls malloc from one of its global constructors during its thread
> > init function). My init function is called from this malloc, and then uses a
> > number of system calls.
>
> > Doing a mmap on &errno stops the crash, but screws up things when debugging
> > programs that don't have this behavior (because I remap errno and probably
> > some important info in the page adjacent to it).
>
> > Ok, so I know the problem. My question now is: Is there anything that can
> > be done about it? Can you guys map errno in earlier in the thread init
> > proccess?  Do you even handle this? I'm surprised this doesn't happen in
> > the init function without my debugger. NJAMD works fine with pthreads, is
> > there any way to tell omniorb to use pthreads instead of omnithreads?
>
> > Any help you could give would be much appreciated by both me and at least one
> > of your users :)
>
> --
> Sai-Lai Lo                                   S.Lo@uk.research.att.com
> AT&T Laboratories Cambridge           WWW:   http://www.uk.research.att.com
> 24a Trumpington Street                Tel:   +44 1223 343000
> Cambridge CB2 1QA                     Fax:   +44 1223 313542
> ENGLAND