x86_64 lock-up problems - any ideas?

[chunkey]

is there already a workaround for this "known bug", since I guess he is not the only one with these _known_ problems, or?

[netllama]

No, I'm afraid there are no known workarounds right now. This bug is currently with development.

[TwinFinity]

Thanks netllama,

Yes, the machine boots with normal output on the serial console.
Would you be interested in seeing the boot details?

I was hoping it was the known bug! I'd noticed the lack of obvious errors anywhere so I was holding off on reporting my problems until the known bug had been fixed... Then yesterday I saw bms20's message and thought it would be a good time to mention it!

But it only crashes as root? That's just weird!


J

[gilboa]

TwinFinity,

Do you really need Cool&Quiet?
It's known to cause problems with SMP kernels.

It is very possible to assume that C&Q is crashing something inside the kernel once you load the drivers; and long after, the same badness causes the kernel to crash.

Cheers,

[TwinFinity]

hi gilboa,

No I don't need cool'n'quiet, but I've pretty much ruled it out as a cause of the problem. I recompiled the kernel without any frequency control/cnq support and it still locks up when running as root.

Tried it a second ago, it will run two lots of glxgears at the same time but shut them both down and try to start another and it locks up. It opens the window to draw into and there is the black background but that's it. Mouse can be moved for a few seconds (and then not smoothly), and then stops totally.

As SMP will be the norm by this time next year we shouldn't have to wait long for a fix ;-)

Just for info, in case it's not quite right-
with one window glxgears runs around 15,000fps and one core at 100%
with two windows glxgears runs approx 1,400fps and two cores at 100%...
So twice as much processor usage and one tenth the frame rate.
Might have to try that test on my athlon mp box for comparison!

[gilboa]

First, did you try running as user. I doubt that it'll help... but who knows.
Second, can you try an older card? (GF6xxx)

As for the CPU usage, glxgears is a bad benchmark. Weird never the less.
You should have seen ~6000-7000fps on each instance.

Cheers,

[gilboa]

Double post...

[TwinFinity]

gilboa,

I'm not having any issues when running as user.
Results from top with four instances of glxgears running...
Cpu0 : 17.6% us, 82.4% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu1 : 17.9% us, 82.1% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu2 : 17.6% us, 82.4% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si
Cpu3 : 19.0% us, 81.0% sy, 0.0% ni, 0.0% id, 0.0% wa, 0.0% hi, 0.0% si

Each window is getting about 850fps.

I'm now running the latest test kernel, going to log out of this user account and in as root to try my luck... and it still locks as root, it does manage multiple windows but close them all and it freezes after opening a new instance of glxgears.

[gilboa]

The first Linux/Unix rule is:
Don't work as root.

Maybe the nVidia driver it trying to remind you of that

Never the less, if all works well as you, does it *need* to work as root?

[netllama]

As much as I'd like to think that NVIDIA encourages good security practices, there isn't any such enforcement in the driver that would cause applications to fail as root yet work as other users.

TwinFinity, the bug that I referenced only exists with G70 based cards (7800GTX, etc). So if you do happen to have an older different card that you can test with that would help to confirm whether you are experiencing the same bug.

thanks,
Lonni

[TwinFinity]

Thanks Lonni,
yeah, the root of the problem shouldn't be that root is the problem ;-)

I'd try another card if I had a suitable one here at home,
Oh for an AGP slot... I have FX5200, Geforce3, Geforce256 and TNT cards available but just the one PCI-e, the 7800GTX.
I will dig out a PCI card Geforce2MX when I'm back in work next week. I know that card works in this machine but don't remember trying OpenGL on it, should be interesting to see the frames per second on that one ;-)

Jason

[bms20]

Hi Folks,

Thanks for the glxgears tip. In two days of trying I couldn't reproduce the problem. Using the glxgears tip I can definitely crash the system. I've got a debug trace from the serial port and the nvidia-bug-report.log attached.

It seems that I'm getting a slightly different problem rather than a null pointer dereference, and my machine definitely does hard lock. Furhtermore, I'm not running SMP - I couldn't afford the second Opteron 252 yet.

Any ideas?

-BMS20