Hard lockups under SMP

[vince]

Apparently I'm not alone in this.

My Asus V8420 DT ( = GeForce4 Ti 4200) will hard lock my system
(not just X, the entire system) when I attempt to start X while running under an SMP kernel and have my Iiyama 4831D connected to the DVI port. It will work when the monitor is connected to VGA, and also _with_ DVI when running a UP kernel.

The system runs Redhat 8.0 with the latest stock Redhat 8.0 SMP kernel: 2.4.18-27.8.0smp.

$ cat /proc/driver/nvidia/cards/0
Model: GeForce4 Ti 4200
IRQ: 10
Video BIOS: 04.25.00.30.00
Card Type: AGP

$ cat /proc/driver/nvidia/version
NVRM version: NVIDIA Linux x86 nvidia.o Kernel Module 1.0-4349 Thu Mar 27 19:00:02 PST 2003

(the above two were collected while running X under a UP kernel).

The motherboard is a MSI 694D Pro; with dual Pentium III (coppermine) 800Mhz processors.

I've also noticed that the card appears to run at 2xAGP, where the motherboard

Attached are my XF86Config and XFree86.0.log of the last attempt I made.

[vince]

err. I'm a little new at this forums thing. Apparently, my files didn't get attached....

[vince]

And the XF86Config.

[eric42]

I can't help but I can confirm the problem.

I've had very bad lockup problem with both 4XXX drivers. No log files record the lockup, but it happens very often under SMP kernel (usually every day, X crashes once an hour).

I'm using
Dual 2.2 Xeons
Dell Precision 530
Quadro 4 750
2.4.18-3smp (up fixes the problem)
Using 3123 drivers fixes the problem. (Uptime of two weeks without a crash/lockup, either X or system, no 3D though.....)

Whatever the NVidia guys did with the 4XXX drivers, they broke the SMP compatibility.

I've tried diabling AGP and all the power managment stuff, but the system won't stabilize under 4XXX drivers.

NVidia? Are you listening? People have been reporting SMP lockups since 4191 was released........ Everything works fine until the system locks up.

Eric

[bwkaz]

OK... let me play dumb here for a minute.

Why don't you have 3D with 3123?

[kahuna0k]

Running 4349 fixes the hard lockups in SMP for me, now runnning with a problem (other than TV out overscan not working, see the overscan thread). Uptime of 7 days (last reset to check TVout in Windows).

[Andy Mecham]

This is a known problem - SMP+DVI hangs. See the errata on ftp://download.nvidia.com/XFree86/Linux-x86/1.0-4349

Current workarounds are to use UP+DVI, or SMP+Analog.

--andy

[eric42]

Under 4191 I had a working OpenGL install. (TuxRacer ran beautifully, GLX gears of 6000 fps.) I installed 3123 and glx gears won't even run. (Missing/bad glx, but it's still in the XF86Config.

kahuna: Did you have the lockups under 4191? What motherboard/system/graphics card do you have?

Eric

PS I'm not using DVI or TV out, just plain old vanilla VGA. (Cathode Ray Tube and all!)

[kahuna0k]

I have the lockups ONLY under 4191, under 3123 it works perfectly, and now under 4349 it also works perfectly. 4191 also gave corrupted 2D after some use. I have a Tyan Tiger MP S2460 and a Abit Siluro GeForce4 Ti 4400. I can use both CRT and TV, but TV overscan doesn't work (any clue Andy?)

[Andy Mecham]

From the 1.0-4349 errata:

o Overscan is currently only available on GeForce4 or newer GPUs
with either NVIDIA or Conexant TV encoders.

You can find the errata at ftp://download.nvidia.com/XFree86/Linux-x86/1.0-4349/errata-1.0-4349.txt

--andy

[donuts]

Andy,

Please see the "overscan" thread, where it becomes clear that overscan does not work for _anyone_, regardless of what type of GPU/TV encoder they have. For example I have GeForce4 MX440 with the NVidia (NV17) TV Encoder, and the "TVOverScan" option does nothing in any resolution. I have written about it to NVidia support, but ofcourse got no reply.

[Andy Mecham]

I have written about it to NVidia support, but ofcourse got no reply.
When did you write?

--andy

[donuts]

> When did you write?
Sun, 6 Apr 2003 18:42:45 -0700 (PDT) to "linux-bugs" (at nvidia.com). Maybe I should have used a different e-mail address?

[donuts]

The "TVOverscan" now works on my card with the new (4363) driver. See the "Overscan" thread for more details.
Thanks Andy/NVIDIA!

[jsolomon]

Hi,

Unfortunately, I am also seeing lockups with my smp/nvidia system. Here are the specs:

Dell Precision Workstation 450
2 x 2.8 GHz Xeon
2 GB DDR
Intel 7505 chipset
Nvidia 900 XGL Quadro 4 graphics
Redhat 8.0 - 2.4.18-14smp
Dell 2000FP DVI

I'm using the latest 4363 version of the drivers.

Now, the 4191 drivers locked up the system right from the get go, while I don't get lockups until some number of days with the current 4363 drivers.

For me, a lockup means:

- keyboard unresponsive
- telnet, ftp, rlogin, unresponsive
- ping is responsive
- screen just locks, but isn't garbled in any way

I've disabled AGP but the problem persists.

Some have suggested that I disable DMA on the IDE bus.

Some have suggested that I upgrade to the latest RH8.0 kernel or that I upgrade to RH9.0.

Andy - any idea on what is going on?

Thanks,

Jeff

[eric42]

Here is the only fix for this problem I have found:

Downgrade to 3123 drivers.

That simple, I've had 3123 up for over a month with no lockups or crashes of any type. I don't know what changed in 4XXX, but SMP compatibility was broken.

Eric