Re: PREEMPT kernel with nvidia

On Fri, Jan 04, 2008 at 00:31:26 +1000, Paul Andreassen wrote:
On Wednesday 02 January 2008 02:44, Florian Kulzer wrote:
On Tue, Jan 01, 2008 at 23:25:47 +1000, Paul Andreassen wrote:
Is it possible to have a PREEMPT kernel while using the closed source
nVidia driver?

I've compiled the stable kernel 2.6.18 and the backported kernel 2.6.22,
and both kill random programs. Its strange because it some times kills
the game I'm running, sometime faults on boot up and once gcc actually
died.

Such sporadic problems can be caused by faulty hardware. (bad RAM, power
supply instability, insufficient cooling, ...) GCC for example can
sometimes function as a sort of CPU-and-memory-stress-tester.

Can you run your system (same load) on the "nv" driver without problems?
Did you test this for a sufficiently long time so that you can be sure
that the crashes only happen with the nvidia driver?

Hi Florian,

That's what I can't understand. The default Debian kernel 2.6.18 without
PREEMPT works fine.

Is it possible the PREEMPT option would bring to the surface faulty hardware
that non PREEMPT could handle?

I would not rule out something like that, but I do not dare to make a
guess how likely it is. GCC crashing the system seems suspicious to me
since that should not involve any fancy graphics stuff.

[...]

On Wednesday 02 January 2008 00:50, Hugo Vanwoerkom wrote:
hugo@debian:/sdb1$ uname -a
Linux debian 2.6.23-1-686 #1 SMP PREEMPT Sun Dec 9 10:02:52 CST 2007
i686 GNU/Linux

So I can conclude it is most likely a hardware problem on my end.

You can try memtest and lm-sensors if your hardware is supported.
Unfortunately memtest is only conclusive if it reports errors; if it
succeeds it still does not guarantee that your RAM is completely OK.

It might also be worthwhile to play around with certain nvidia options,
e.g. turning off RenderAccel or changing the AGP configuration.

--
Regards, | http://users.icfo.es/Florian.Kulzer
Florian |


--
To UNSUBSCRIBE, email to debian-user-REQUEST@xxxxxxxxxxxxxxxx
with a subject of "unsubscribe". Trouble? Contact listmaster@xxxxxxxxxxxxxxxx