Re: [lkcd-devel] forcing a dump in deadlocksituations

[Troy H. <tro...@hp...>]

On 10/14/05 12:45, Golbach, Klaus wrote:

> SUSE SLES9 +SP1 is installed on our Intel machine (with
> 4 processors).

I'm not positive but I believe that SuSE uses a 6 version of LKCD which means
that LKCD turns interrupts back on in order to complete the dump.
 
> Can you give me some hints, what to do to get a dump in such case of
> deadly embrace?

Using interrupt based dumping dumping there are several pitfalls in which the
machine will hang and you will not be able to get a dump. I would strongly
recommend trying to move to the code in our CVS repository in the 7.X.X
branch. This code has the ability to do non-interrupt base polling dump (like
diskdump) which makes crash dumping much more reliable in a wide variety of
nasty situations. 

I have no experience trying to get the 7.X.X branch to work with a SuSE
kernel. Maybe someone else on the list can give some pointers. 
 
> Is there something additional  to do?

No, sounds like you have done everything correctly but are hitting a
limitation of the 6.x.x. version of LKCD. If you absolutely can not move to
the 7.X.X version you might try the software NMI watchdog. You said you are
running on Intel but not which architecture: if it's x86 or x86_64 try
enabling the software NMI Watchdog, it might detect the hang and gives you a
very small chance of getting a dump, most likely you will just hang again when
LKCD re-enables interrupts but its worth a try. 

If you are really trying to debug a driver you really need to get to the
7.X.X branch of LKCD with polling mode!! Even if this requires grabbing a
kernel.org kernel and patching it with 7.X.X LKCD, do the debugging there,
get the drivers fixed then move back to a SuSE kernel to finalize the work. 

I'm leaving now and won't have access to a computer till Wednesday, so if you
respond I won't be able to answer till then. Hopefully, others on the list
will. 

Troy