[plug] crashing server

[Peter F Bradshaw]

Hi;

On Tue, 14 Sep 2004, James Devenish wrote:

> In message <16710.57347.180888.888436 at pride.nsw.cmis.CSIRO.AU>
> on Tue, Sep 14, 2004 at 10:11:47PM +1000, Rob Dunne wrote:
> > the machine has to be rebooted.
>
> Good grief. One possibly difficult aspect is that capturing kernel
> coredumps after rebooting PC-type CPUs has apparently been unsupported
> in the past (according to previous discussions on this list). I don't
> know if Linux supports a crash-time debugger either. Not sure what
> kernel developers do. By the way: what version of Linux (i.e. kernel)
> are you using? How do you know the machine has to be rebooted? Does it:
> spit a panic message to the console? lock up completely? spontaneously
> reboot?

Try kgdb (see http://kgdb.sourceforge.net/). You need another machine,
connected to your server by serial cable, to run gdb. You might also
try linking your server program against Electric Fence (libefence.so)
to see if there is a double or invalid free().

>
> If you are using 64-bit hardware, perhaps the programme is triggering an
> obscure kernel bug that only manifests on 64-bit hardware.
>
> >  > > free((void *)b);
>
> The presence of a segfault on the above line strongly indicates that b
> is invalid. If b has also been used as part of the mutex code as you
> mentioned earlier, and if the kernel is failing to validate 'b' perhaps
> this leads to the crash (i.e. it's a combination of both bugs that is
> leading to the crash). Try fixing b and see if the crash stops. If the
> kernel is corrupting b, well, you may have to resort to tracking the
> value of b through the code (either with a debugger or with in-code
> methods).
>
>

Cheers

-- 
Peter F Bradshaw, pfb at users.sourceforge.net, ICQ 75431157 (exadios).
PGP public key at http://exadios.d2.net.au/public_key.html
"I love truth, and the way the government still uses it occasionally
 to keep us guessing." - Sam Kekovich.