[uClinux-dev] do_global_dtors problem

David McCullough davidm at snapgear.com
Sun Nov 4 17:37:38 EST 2001


Jivin Matt Waddel lays it down ...
> Hi,
> 
> I have been getting the following results on an M5272C3
> system when I use the "cat" command from the Fileutils
> userland package.
> 
> # cat /etc/inittab
> inet:unknown:/bin/inetd
> boa:unknown:/bin/boa
> Illegal instruction
> #
> 
> Using gdbserver I have traced it to the do_global_dtors
> routine.  It loads in a very large value then tries to jump
> there and this causes cat crash and return to a shell
> prompt.  (Sometimes cat will return a SIGSEGV.)
> 
> Since everything seems to run fine after that and
> the busybox cat functions properly, I am wondering
> if there is a subtle problem with the toolchains.
> 
> Has anyone else seen this behavior?  Isn't do_global_dtors
> a c++ function?  Any enlightenment on the constructor/
> destructor functions would be appreciated.

It is unlikely that the problem is with do_global_dtors.   What happens
is in the later version of gcc (ie 2.95) main calls __main which processes
do_global_[cd]tors lists.  So you get them with 'C' as well.  Actually,  I
think that atexit is processed this way,  but I may be wrong.

The fact that the program has run to completion and then crashed would make
me look at the stack size and see if it could be overflowing into the top
of the BSS/data sections.

Actually,  one look at the code shows a 4096 byte buffer,  and the Makefile
doesn't specify a stack size,  so you get the default 4096 stack.  Not a
very good combo ;-)  Make the char array a static:

	diff -r1.1 cat.c
	24c24
	<       char readbuf[CAT_BUF_SIZE];
	---
	>       static char readbuf[CAT_BUF_SIZE];

and you should be on your way ;-)

Cheers,
Davidm

-- 
David McCullough:    Ph: +61 7 3435 2815  http://www.SnapGear.com
davidm at snapgear.com  Fx: +61 7 3891 3630  825 Stanley St., W'gabba QLD 4102, Oz
This message resent by the uclinux-dev at uclinux.org list server http://www.uClinux.org/



More information about the uClinux-dev mailing list