Kernel freeze when initializing graphic card (during X startup)

Hi all helpful people,

I have a very serious problem with the 2.4.21 kernel (in SuSE 9.0).
The problem is that the kernel freezes when starting up X under certain
conditions, the background is the following:

The machine is a x86 PC with an Asustek A7V8X motherboard with a 2000Mhz
Celeron processor, a HighPoint IDE RAID controller (don't know the exact model),
and a Matrox Millennium G550 graphic card.

I installed the machine and tested it thouroughly (including some stress
testing) and it seemed completely stable.
The Matrox G550 that sits in the machine is a dual-head card, and until then I
had only used one monitor with the card, however when configuring the machine as
dual head, I started to get the kernel freeze problems.
What happens is that sometimes when X starts, the screen just goes blank, and
the system gets unresponsive to keyboard or mouse input. A few seconds later,
you can't ping the machine, and a hardware/norwegian reboot is necessary.
It can also happen, that if the kernel was accessing the disk at the time of
the freeze, the disk access diod on the chassi will stay alight indefinately,
which in my interpretation indicates a pretty bad kernel lockup in the middle
of some disk I/O.

Anyway, the only thing I could find in the logs was that the kernel reported
a problem with the mtrr. But after recompiling the kernel with mtrr left out
and still seeing this problem I assumed it wasn't cauing the crash.

Finally after some testing I realized this:
1) The kernel never freezes without the G550 card in the machine
2) The kernel never freezes with the card configured as single-head
3) The kernel never freezes with the card configured as dual-head, and monitors
plugged in to both VGA connectors.
4) The kernel almost always freezes with the card configured as dual-head, and
no monitor (or a turned off monitor) attached to the secondary VGA connector.
5) The freezes occur no matter what X settings are configured, (ie. DRI/no DRI,
bitdepth, resolution, type of dual-head config etc.)


(Note that the kernel only freezes when starting X and never when initalizing
the console framebuffer or similar)
In the conditions when it does freeze, that is also sporadic, happening every
third time or so that X is started.
The times X does in fact start properly with only one monitor attached to the
card, it runs stable from then on.

My conclusion is this: there's something wrong with the G550 drivers that
results in sporadic crashes when initalizing both VGA connectors under the
condition that there's nothing attached to the secondary connector.

I've tried different kernel versions (SuSE 2.4.21-99 and 2.4.21-166) and also
to download the latest version of the G550 drivers. I've changed X versions as
well and it doesn't help.

Anyway my questions are these:
1) How can X kill the kernel? I thought that using a dri module for the
graphic card was the only time when a device driver interfers with the internals
of the kernel (correct?) and can as such disrupt its functions.
2) How come the kernel doesn't even coredump, but just hangs, seemingly in the
middle of performing disk I/O and such?
3) How do X graphical device drivers work in respect to initializing hardware,
is there some general specification on this so that it's easy to guess what is
really cauing the freezes?
4) And most important, is there any patch or other solution available to fix
the problem?

Please help me, it's necessary that I solve this issue. The machine (actually
ten of them) is intended for a production environment and must run entirely
stable. Even more so as I've convinced the management of the company I work for,
that Linux is superior to Windows in these kind of environments, and that I've
finally got an approval to get rid of our Windows stations in favour of Linux :)

/Leon

On Fri, 06 Feb 2004 14:41:56 -0800, Leon Styhre wrote:

> Hi all helpful people,
>
> I have a very serious problem with the 2.4.21 kernel (in SuSE 9.0). The
> problem is that the kernel freezes when starting up X under certain
> conditions, the background is the following:
>
> The machine is a x86 PC with an Asustek A7V8X motherboard with a 2000Mhz
> Celeron processor, a HighPoint IDE RAID controller (don't know the exact
> model), and a Matrox Millennium G550 graphic card.
> (snip)
Tough problem. I would observe some things. First, G550 is supported
by X 4.3, not sure for older versions. As mandrake is 9.2 now I suppose
your X version should be 4.1 . G550 is a strange beast, following
the forum on matrox site I see they have a bug - dvi output delay - that
no one is capable of removing yet.
As you stated, in single head configuration works perfectly.
On double head output, X crashes and this somehow affects the kernel which
besides is quite old.
For the time being I have two suggestions, no one of them will appeal to
you.
First, try matrox module i.e. driver instead of Xfree one, on
www.matrox.com.
Second, try a more recent distribution, with a newer kernel and X.
Maybe you're triggering a bug, most likely.

As a suggestion, browse matrox forum and ask them for help also
http://www.matrox.com/mga/support/forum/agree.cfm

Leon Styhre wrote:

> The machine is a x86 PC with an Asustek A7V8X motherboard with a 2000Mhz
> Celeron processor, a HighPoint IDE RAID controller (don't know the exact model),
> and a Matrox Millennium G550 graphic card.

I'm quite surprised that you can at all boot up the machine, how did you get
that Celeron into that AMD CPU Socket?

Hi,

You're right, peculiar.
In YaST Hardware information it's listed as a A7V8X, but the BIOS reports it
as a P4GE-V. Of course A7V8X doesn't support Celeron.

/Leon

"J.O. Aho" <user@example.net> wrote in message news:<c0297e$sdr55$1@ID-130698.news.uni-berlin.de>...
> Leon Styhre wrote:
>
> > The machine is a x86 PC with an Asustek A7V8X motherboard with a 2000Mhz
> > Celeron processor, a HighPoint IDE RAID controller (don't know the exact model),
> > and a Matrox Millennium G550 graphic card.
>
> I'm quite surprised that you can at all boot up the machine, how did you get
> that Celeron into that AMD CPU Socket?

Hi, Thanks for the reply.

I tested already with the driver modules from Matrox and unfortunately the
result is the same.
And as for XFree86, the default version in SuSE is actually 4.3
A newer kernel will probably not solve the issue, but will hopefully just crash
X rather than freeze the whole system, so yes I may give this a try..
I don't know why SuSE keeps this older kernel in their 9.0 distribution, but
I guess it's because they find it more reliable (and besides it's highly
modified with much backported code included).

I'll take a look at the Matrox forums. What you said about the DVI output
delay bug sure sounds probable.

/Leon

U guess <bill@microsoft.com> wrote in message
news:<pan.2004.02.06.23.05.40.298157@microsoft.com >...
> On Fri, 06 Feb 2004 14:41:56 -0800, Leon Styhre wrote:
>
> > Hi all helpful people,
> >
> > I have a very serious problem with the 2.4.21 kernel (in SuSE 9.0). The
> > problem is that the kernel freezes when starting up X under certain
> > conditions, the background is the following:
> >
> > The machine is a x86 PC with an Asustek A7V8X motherboard with a 2000Mhz
> > Celeron processor, a HighPoint IDE RAID controller (don't know the exact
> > model), and a Matrox Millennium G550 graphic card.
> > (snip)
> Tough problem. I would observe some things. First, G550 is supported
> by X 4.3, not sure for older versions. As mandrake is 9.2 now I suppose
> your X version should be 4.1 . G550 is a strange beast, following
> the forum on matrox site I see they have a bug - dvi output delay - that
> no one is capable of removing yet.
> As you stated, in single head configuration works perfectly.
> On double head output, X crashes and this somehow affects the kernel which
> besides is quite old.
> For the time being I have two suggestions, no one of them will appeal to
> you.
> First, try matrox module i.e. driver instead of Xfree one, on
> www.matrox.com.
> Second, try a more recent distribution, with a newer kernel and X.
> Maybe you're triggering a bug, most likely.
>
> As a suggestion, browse matrox forum and ask them for help also
> http://www.matrox.com/mga/support/forum/agree.cfm

On Sat, 07 Feb 2004 07:28:59 -0800, Leon Styhre wrote:

> Hi, Thanks for the reply.
>
> I tested already with the driver modules from Matrox and unfortunately the
> result is the same.
> And as for XFree86, the default version in SuSE is actually 4.3 A newer
> kernel will probably not solve the issue, but will hopefully just crash X
> rather than freeze the whole system, so yes I may give this a try.. I
> don't know why SuSE keeps this older kernel in their 9.0 distribution, but
> I guess it's because they find it more reliable (and besides it's highly
> modified with much backported code included).
Aha, that would explain a lot of things. I hate this sort of modifications,
you'll never know what could happen. Maybe guys working on kernel have
a little more experience on that than a distribution firm.
That's the main reason open source works and proprietary software
cannot keep up.
>
> I'll take a look at the Matrox forums. What you said about the DVI output
> delay bug sure sounds probable.
>
> /Leon
Wait a moment. Do you use DVI output? What happens with VGA output instead?
This DVI stuff rings an alarm in my head.