Baslinux mailing list

[sindi keesan <keesan AT sdf.lonestar.org>]

On Mon, 19 Dec 2005, David Moberg wrote:

> On 12/19/05, sindi keesan <keesan AT sdf.lonestar.org> wrote:
> > Does BL2 keep a log of error messages produced just before a crash?
> > Or do I need to start some program to keep the log?
>
> I think you need to start syslogd and klogd to record kernel
> messages. This probably will not survive a crash, so you
> will need to write down 'oops' messages.

Will either of these preserve the messages during boot that don't show up 
in dmesg?  I am getting them about hdb irq timeout waiting for DMA, and 
DMA disabled.  A friend thinks this is a primary controller problem, so I 
will move hdb to hdd as a test or a cure.  Would this DMA problem have any 
effect on running BL3 loop from hda DOS partition?

In another computer hdb was found only part of the time, but hda was fine.


> > After I upgraded one computer from 350 to 500MHz (AMD K6-2) and also
> > switched to a kernel that I compiled with added scsi (general and sd) and
> > framebuffer (vesafb) support, I have had three sudden crashes, including
> > two freezes, and just now when I switched vts:
> > unable to handle kernel NULL pointer dereference at virtual address:.....
> > .....
> > oops:0000
> > ....killing interrupt handler
> >
> > The keyboard also acted screwy a few times - either stopped responding at
> > all, or keys started typing double (some not at all).  Interrupt problem?

> Unlikely. A NULL pointer dereference is somewhat like a segfault,
> only it happens in kernel. Basically, the kernel is trying to access
> a portion of memory which does not exist (NULL). Since this
> happened with more than one compiler version, this is probably
> a bug in the source code. 2.4 is likely to have fixed this.
>
> David

This bug only occurs when using your or my two 2.2.16 framebuffer kernels 
(I tried with and without scsi support). It always crashes when I switch 
vts during a 'make bzImage' or 'make dep', sometimes with pages and pages 
of numbers rolling by, sometimes just one page and a freeze.

When I looked up some of the error messages there was much talk of kernel 
bugs, but why does it only happen with framebuffer, because it uses memory 
differently?

But it might also be related to my DMA hardware problem so I will test 
another computer with kernel source code on it.  The other one here could 
not be made to crash by simply switching vts.  What else could I do to 
try to crash things reproducibly?

I will try your 2.4.31 framebuffer kernel after this.