4496 on x86_64 w/8G RAM; IOMMU/AGP problems?
 

Hello all --

I've read this thread with some interest. We're running a somewhat bleeding
edge setup, and we're having problems with the nvidia drivers. Here's the
deal:

* The box is a dual opteron (246) w/8G RAM
* The card is a GeForce2 GTS/GeForce2 Pro
* Running RHEL 3.0 AS x86_64
* The disk is a SATA drive, which means we have to run kernel 2.6.0
* Kernel is patched with mm1 and Andi Kleen's x86_64 patchset

We're trying to use the 4496 drivers (yes, we applied the minion.de
patches), but they lock the system hard. People have recommended
using kernel parameters (such as noapic, acpi=off, etc.), but nothing
that we've tried has worked. This was all trying to use agpgart compiled
directly into the kernel.

The problem may be with IOMMU. Since the nvidia drivers lock hard with
the AGPGART kernel module compiled into the kernel, we decided to try
disabling it with NVAGP "1". No problem, except that AGPGART is
forced to be on if IOMMU is enabled. IOMMU should be enabled on
systems with more than 4G of RAM to make sure that certain older 32-bit
PCI cards work.

So, the problem is this: we have more than 4G of RAM, so we we have to
disable IOMMU to disable AGPGART, but even after we've done all of that,
the nvidia drivers hang hard.

Of course, disabling IOMMU could be what's causing the driver to hang.
The docs say that IOMMU is necessary for PCI cards that don't
support DAC.

So.

Um.

I'm way in over my head.

Any ideas? Does anyone know if this is an IOMMU problem? Is it
the fact that the card is old? Are there any more patchsets that
people would recommend? (I see that the mm2 patchset is out; I
may try that).

Any help would be appreciated.

Thanks!

-- David

More info
 

Sorry all; forgot some info:

The system is running a Tyan K8W board.

Contents of the XFree86 log are incomplete. I suspect it's because the system
crashes so hard it doesn't sync, so not everything is being written to disk. However,
not a lot happens when I start X -- the screen flashes a couple of times, and
it freezes.

A very similar X configuration works fine with the X 4.3.0 nv driver, so I
doubt that's the problem, but I will include both the log and the config
files.

-- David

XFree86.0.log:

XFree86 Version 4.3.0 (Red Hat Enterprise Linux release: 4.3.0-35.EL)
Release Date: 15 August 2003
X Protocol Version 11, Revision 0, Release 6.6
Build Operating System: Linux 2.4.21-2.ELsmp x86_64 [ELF]
Build Date: 03 October 2003
Build Host: dolly.devel.redhat.com

Before reporting any problems, please make sure you are using the most
recent XFree86 packages available from Red Hat by checking for updates
at http://rhn.redhat.com/errata or by using the Red Hat Network up2date
tool. If you still encounter problems, please file bug reports in the
XFree86.org bugzilla at http://bugs.xfree86.org and/or Red Hat
bugzilla at http://bugzilla.redhat.com

Module Loader present
OS Kernel: Linux version 2.6.0-mm1-x8664_1-dss (root@ruminate) (gcc version 3.2.3 20030502 (Red Hat Linux 3.2.3-20)) #17 S
MP Mon Jan 5 09:34:06 MST 2004 P
Markers: (--) probed, (**) from config file, (==) default setting,
(++) from command line, (!!) notice, (II) informational,
(WW) warning, (EE) error, (NI) not implemented, (??) unknown.
(==) Log file: "/var/log/XFree86.0.log", Time: Mon Jan 5 11:07:41 2004
(==) Using config file: "/etc/X11/XF86Config"
(==) ServerLayout "AGP"
(**) |-->Screen "Screen AGP" (0)
(**) | |-->Monitor "Monitor0"
(**) | |-->Device "NV AGP"
(**) |-->Input Device "Mouse0"
(**) |-->Input Device "Keyboard0"
(**) Option "AutoRepeat" "250 30"
(**) Option "XkbRules" "xfree86"
(**) XKB: rules: "xfree86"
(**) Option "XkbModel" "pc105"
(**) XKB: model: "pc105"
(**) Option "XkbLayout" "us"
(**) XKB: layout: "us"
(==) Keyboard: CustomKeycode disabled
(**) FontPath set to "unix/:7100"
(**) RgbPath set to "/usr/X11R6/lib/X11/rgb"
(==) ModulePath set to "/usr/X11R6/lib64/modules"
(**) Option "AllowMouseOpenFail"
(--) using VT number 7

Just curious, why are you running 4496 instead of 4499? I've had no
luck with the 4499 (search for my previous posts for more info). I've given
up and gone back to the lousy nv driver. Slow as heck, but at least I have
some 3-D.

My system is very similar to yours. We only have 4 Gig. We also are using SATA drives, but with the 2.4 kernel.

@wingated

You have to use 4499 (not 4496), IOMMU should be no problem, but K8 NUMA and a NPTL-enabled glibc cause problems... You should also patch the kernel with the patches from amd64.org, or use kernel 2.6.1-rc1.

@msc

There are problems with AGPGART on kernel 2.4.x, update to 2.6.0 and fetch the patches at amd64.org (or use kernel 2.6.1-rc1), and do not(!) use a NPTL-enabled glibc or K8 NUMA.