Bug 748

Could this be a null pointer problem that has existed for a long time, but is just now exposed by this kernel change:

* Wed Jul 29 2009 Chuck Ebbert <cebbert@redhat.com> 2.6.29.6-217.2.3
- Don't optimize away NULL pointer tests where pointer is used before the test.
  (CVE-2009-1897)

You need to report the bug to nvidia using nvidia-bug-report.sh
Alternatively could you go back to the previous kernel using the new driver ?

Others have seems to have reported the nvidia driver to work with 185.18.31.
It could be a problem specifics from your hardware...

Can you confirm that :
- 185.18.31 works with the previous kernel ?
- 185.18.18 can be built using the current fedora kernel ? (if not what is the error log).


You could have experienced a driver version missmatch because of the use of akmod-nvidia over kmod-nvidia for standard Fedora kernel. Maybe you could test that the version is right for the nvidia module to match the xorg module.

I seem to be effected by this bug as well. When I boot into any kernel with the 185.18.31 NVIDIA driver using either kmod or akmod I get the same thing. The system loads up just fine but when it starts X my screen turns off. I can not change to another terminal to debug which makes things a little difficult. I am however able to ssh in so the machine is not dead. I get the same kernel opp's as above and my Xorg.log says no screen's found, obviously because of the nvidia kernel driver oppsing.

It seems that this is only effects Quadro users. I asked a friend who has a Gefore 8600 and everything works fine for them. Below is the PCI ID for my video card.

01:00.0 VGA compatible controller: nVidia Corporation Quadro FX 570M (rev a1)

The only way I was able to fix this was by disabling the NVIDIA driver. I did this by removing the NVIDIA process from startup and running nvidia-config-display disable. I then get the vesa driver which isn't what I want but at least I have X.

A quick search on nvnews.net shows that an upgrading to the latest NVIDIA beta driver fixes this issue. I haven't tried that yet.

http://www.nvnews.net/vbulletin/showthread.php?t=136959

I will re-introduce either beta and/or newest series for quadro users until nvidia update the stable drivers. 

(In reply to comment #5)
> I will re-introduce either beta and/or newest series for quadro users until
> nvidia update the stable drivers. 
> 

01:00.0 VGA compatible controller: nVidia Corporation Quadro FX 360M (rev a1)

I am having the same issue.  Glad to see a fix is coming.

I have the same Problem on different Hardware. 

I am using a Sony Vaio FE41M Notebook, 32 Bit F11, 

my graphic card is

01:00.0 VGA compatible controller: nVidia Corporation G72M [GeForce Go 7400] (rev a1)

The Problem exists with the previous kernel, too.

After downgrade to Nvidia 185.18.14 everything is ok.

To confirm what others have already found, I systematically:

0) downloaded the 185.18.14 and 185.18.31 nvidia-kmod SRPMs, and built both without detecting  any issues using "akmodsbuild -v -v -v -v".

then:

1) yum remove \*nvidia\*
2) yum localinstall  akmod-nvidia-<N.N.N-M> xorg-x11-drv-nvidia-<N.N.N-N> xorg-x11-drv-nvidia-libs--<N.N.N-N>
3) reboot

for N.N.N-M,N.N.N-N = 185.18.14-1,185.18.14-3 and 185.18.31-1,185.18.31-1 respectively, under kernels 2.6.29.6-213.fc11.x86_64 and 2.6.29.6-217.2.3.fc11.x86_64

I found that the 185.18.14 nvidia packages worked with both kernels and the 185.18.31 packages failed in identical fashions, consistent with all prior comments, with both kernels.

Unfortunately, just as I was finishing the tests, but before reporting on the results, my family sent out a search party to bring me back.

Thanks, Nicolas, for offering to package the latest nvidia drivers; I look forward eagerly to trying the 190.xx packages when you do. Until then I'm happy using the 185.18.14 drivers.

Is there any point now to submiting a bug report directly to Nvidia, as they seem to be aware of, and have fixed, the problem?

OK, so here is the choosen solution:
I have reverted the kmod-nvidia as 185.18.14 and built it for the lastest (current) Fedora kernel. The 185.18.31 driver will be moved back to the rpmfusion-nonfree-updates-testing repository.
If you have problem with the lastest driver, you will have to remove it with:
yum remove xorg-x11-drv-nvidia
then you reinstall (that will be the 185.18.14 in few hours) using
yum install kmod-nvidia(-PAE)

Thoses who did not experienced problem with the last driver will stay with 185.18.31.

On next fedora kernel, the kmod will be built for 185.18.31 or what would be next nvidia driver (from stable serie).

Hopefully early next week, I will re-introduce the nvidia-beta driver serie with a new packaging scheme. That will be versioned as 190.19


>Is there any point now to submiting a bug report directly to Nvidia, as they
>seem to be aware of, and have fixed, the problem?
You can learn how to report a problem to nvidia here:
http://www.nvnews.net/vbulletin/showthread.php?t=46678

Nicolas

WRT the chosen solution in Comment #9, are there any projections about *which* next week we will see the nvidia-beta driver serie in ;)?

I'm fine w/ the akmod rolling ne kmods w/ the newer kernels; I'm just curious.

(In reply to comment #10)
> Nicolas
> 
> WRT the chosen solution in Comment #9, are there any projections about *which*
> next week we will see the nvidia-beta driver serie in ;)?
I will be in holiday without inet access tommorrow and for two weeks
Today will be very short to receive the first feedbacks if ever the new packaging scheme was ready.

For now, you will have to update the rpm on you own, (not many things to change).

See also 

https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-180/+bug/417373


Basically 
185.18.31 is bad for some machines
185.18.36 is ok

You can test 185.18.36 that have be moved to rpmfusion-nonfree-updates-testing either testing a 2.6.30 kernel that is in Fedora updates-testing repository, or using a module built for 2.6.29 from the current updates-stable.

I will close the bug once 185.18.36 hit stable.

185.18.36 is pushed to stable, this bug should be fixed.