KDE randomly freezes, NVIDIA driver at fault?

[joober]

I'm on FC16, using KDE 4.8.3 and for the past several weeks, I have been having problems with KDE randomly freezing or logging me out without warning. Needless to say that this is very annoying and I have lost some work because of it. I have been having these problems on my desktop machine, which has an NVIDIA GPU. My laptop is installed the same way but does not have problems. It has an Intel GPU. Further, I believe my problems started when moving to version 295.49-1 of the NVIDIA proprietary driver.

First, the relevant information of my system:

rpm -qa *\nvidia\*  *\kernel\*|sort;uname -r;lsmod |grep -e nvidia -e nouveau;cat /etc/X11/xorg.conf
abrt-addon-kerneloops-2.0.7-2.fc16.i686
akmod-nvidia-295.49-1.fc16.i686
kernel-3.3.4-1.fc16.i686
kernel-3.3.4-3.fc16.i686
kernel-3.3.5-2.fc16.i686
kernel-devel-3.3.4-1.fc16.i686
kernel-devel-3.3.4-3.fc16.i686
kernel-devel-3.3.5-2.fc16.i686
kernel-headers-3.3.5-2.fc16.i686
kernel-tools-3.3.5-2.fc16.i686
kmod-nvidia-295.49-1.fc16.2.i686
kmod-nvidia-3.3.5-2.fc16.i686-295.49-1.fc16.2.i686
libreport-plugin-kerneloops-2.0.8-4.fc16.i686
nvidia-settings-1.0-17.fc16.i686
nvidia-xconfig-1.0-15.fc16.i686
xorg-x11-drv-nvidia-295.49-1.fc16.i686
xorg-x11-drv-nvidia-libs-295.49-1.fc16.i686
3.3.5-2.fc16.i686
nvidia              10971248  84 
i2c_core               28151  3 videodev,nvidia,i2c_i801
# nvidia-settings: X configuration file generated by nvidia-settings
# nvidia-settings:  version 295.49  (mockbuild@builder1.ovh.rpmfusion.lan)  Thu May  3 20:09:20 CEST 2012

# RPM Fusion - nvidia-xorg.conf
# 

Section "ServerLayout"
    Identifier     "Default Layout"
    Screen      0  "Screen0" 0 0
    InputDevice    "Keyboard0" "CoreKeyboard"
    InputDevice    "Mouse0" "CorePointer"
    Option         "Xinerama" "0"
EndSection

Section "InputDevice"
    # generated from data in "/etc/sysconfig/keyboard"
    Identifier     "Keyboard0"
    Driver         "kbd"
    Option         "XkbLayout" "us"
    Option         "XkbModel" "pc105"
EndSection

Section "InputDevice"
    # generated from default
    Identifier     "Mouse0"
    Driver         "mouse"
    Option         "Protocol" "auto"
    Option         "Device" "/dev/input/mice"
    Option         "Emulate3Buttons" "no"
    Option         "ZAxisMapping" "4 5"
EndSection

Section "Monitor"
    Identifier     "Monitor0"
    VendorName     "Unknown"
    ModelName      "Samsung SyncMaster"
    HorizSync       30.0 - 81.0
    VertRefresh     56.0 - 75.0
EndSection

Section "Device"
    Identifier     "Videocard0"
    Driver         "nvidia"
EndSection

Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "GeForce 8600 GT"
EndSection

Section "Screen"
    Identifier     "Default Screen"
    Device         "Videocard0"
EndSection

Section "Screen"
    Identifier     "Screen0"
    Device         "Device0"
    Monitor        "Monitor0"
    DefaultDepth    24
    Option         "TwinView" "1"
    Option         "TwinViewXineramaInfoOrder" "CRT-0"
    Option         "metamodes" "CRT-0: nvidia-auto-select +1280+0, CRT-1: nvidia-auto-select +0+0"
    SubSection     "Display"
        Depth       24
    EndSubSection
EndSection

I've been monitoring /var/log/messages, /var/log/Xorg.0.log and ~/.xsession-errors for clues but so far I have failed to come up with anything conclusive. I did find messages such as these in /var/log/messages which seem to coincide with X crapping out:

May 12 17:10:27 deepthought kernel: [172119.249950] NVRM: GPU at 0000:01:00.0 has fallen off the bus.
May 12 17:10:27 deepthought kernel: [172119.249956] NVRM: GPU at 0000:01:00.0 has fallen off the bus.

Right after these messages, KDE freezes completely, except for the mouse pointer which continues to move. The system itself is still running though. I can ssh in from another machine and top tells me that X is putting one CPU on 100%. A reboot is the only way to regain a working desktop.
Google tells me that the above error can be resolved by enabling "Persistence Mode" in the NVIDIA driver like so:

# /usr/bin/nvidia-smi -pm 1

This seems to have stopped KDE from freezing completely but it still boots me to the login screen every now and then.

I am also seeing this in /var/log/messages at times, which coincides with me getting booted out of my session:

May 13 14:33:35 deepthought kernel: [76223.746031] NVRM: os_schedule: Attempted to yield the CPU while in atomic or interrupt context
May 13 14:33:37 deepthought kernel: [76225.746120] NVRM: os_schedule: Attempted to yield the CPU while in atomic or interrupt context

Google has been unable to enlighten me as to what these messages mean, just as these messages:

May 12 17:18:36 deepthought kernel: [172607.922716] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:39 deepthought kernel: [172610.923376] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.924098] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.924156] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.924212] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.924264] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.924315] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:42 deepthought kernel: [172613.962104] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:45 deepthought kernel: [172616.963780] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:56 deepthought kernel: [172628.007898] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007
May 12 17:18:56 deepthought kernel: [172628.007966] NVRM: Xid (0000:01:00): 56, CMDre 00000000 00000080 00000000 00000005 00000007

Switching my system to the Nouveau driver, yields a stable desktop at the expense of not being able to play 3d accelerated games, of course. Gnome 3 also seems to be unaffected but I haven't used that extensively so I can't be certain.

I have found this thread http://forums.fedoraforum.org/showthread.php?p=1574003? but the problem with KDE happen without a video player running. It even happens on switching a tab in Firefox.

Any ideas on how I can get my stable desktop back?

[Anjo92]

I 'm adding a me too here to keep the thread alive.
My setting and problem is basically the same as described above.

Edit:
After trying different versions of nvidias driver I found out that
302.07 is not working
295.49 is not working
295.40 is not working
295.33 _is_ working

So there you have it. The changes between 33 and 40 are rather few so it should be easy for some intelligent nvidia developer to mend the broken fix.

/A

[joober]

Originally Posted by Anjo92

I 'm adding a me too here to keep the thread alive.
My setting and problem is basically the same as described above.

I'm glad somebody else is seeing this problem too. I was starting to get worried if my GPU was on the fritz. It still could be, but if other people have the same issue, that becomes less likely.

Thanks for commenting!

[ah7013]

Yeah had the same problem too back when I had F15 KDE on my desktop using the proprietary nvidia driver. I remember switching to nouveau and after that everything was OK. Luckily I didn't do all that much 3D stuff though so I just stayed with nouveau.

[george_toolan]

Can you try to disable all cores except one and see if this changes anything? Either disable SMP in you bios setup or boot with maxcpus=1

[joober]

Originally Posted by george_toolan

Can you try to disable all cores except one and see if this changes anything? Either disable SMP in you bios setup or boot with maxcpus=1

I'll give it a go when I am back home and report the results, though the machine has worked perfectly on 4 cores until this mess began.

[DBelton]

possibly you are hitting the bug with the nvidia drivers and prelink?

try running:

su -
(root password)

prelink -ua

If that fixes your freezing issue, then look at blacklisting the nvidia stuff from prelink

[joober]

Originally Posted by DBelton

possibly you are hitting the bug with the nvidia drivers and prelink?

try running:

prelink -ua

I've run the prelink -ua and I am now waiting to see what will happen. Uptime is now at 1 day, 23 hours and some odd minutes without freezes. Fingers crossed!
Mind, I haven't done any intense 3D stuff yet...

[george_toolan]

So there you have it. The changes between 33 and 40 are rather few so it should be easy for some intelligent nvidia developer to mend the broken fix.

Only if you supply a detailed bug report to nvidia.

Contrary to popular belief it is really difficult to figure out what changes broke something escpecially if the problem only occurs intermittently.

The only "easy" way would be to revert all the changes so you might as well us the old working driver ;-)

[joober]

Originally Posted by Anjo92

I 'm adding a me too here to keep the thread alive.
My setting and problem is basically the same as described above.

Edit:
After trying different versions of nvidias driver I found out that
302.07 is not working
295.49 is not working
295.40 is not working
295.33 _is_ working

So there you have it. The changes between 33 and 40 are rather few so it should be easy for some intelligent nvidia developer to mend the broken fix.

I'm currently running 295.49-1 and my system's been stable for 2 days, 11 hours which is longer than was possible before the freezes began. Mind, I haven't done much heavy 3D stuff yet but it looks promising. I'll keep you posted.
Otherwise, I'll try a downgrade to 295.33 and see how that goes.

[joober]

After 4 days running without a problem, my system just froze again. The freeze occurred while playing Civilization 4 through wine so during relatively heavy load on the GPU. Last messages in /var/log/messages:

May 17 23:17:41 deepthought kernel: [346763.813871] NVRM: GPU at 0000:01:00.0 has fallen off the bus.
May 17 23:17:41 deepthought kernel: [346763.813897] NVRM: os_pci_init_handle: invalid context!
May 17 23:17:41 deepthought kernel: [346763.813901] NVRM: os_pci_init_handle: invalid context!

The first one I have seen before, and this should have been avoided by enabling persistence mode for the NVidia driver. However, I have already done so, so that is not an answer in this case.

I'm going to see if I can downgrade to 295.33 and see if that alleviates matters...if I can convince yum to install that version from rpmfusion. Anyone? Yum insists on downloading the latest binary drivers only, even if I tell it to:

yum install akmod-nvidia-295.33

[jbuckley2004]

I can't quite tell, but this seems almost like the problem I've been having.

I've been running KDE on F16 with an AMD Phenom(tm) 9600 Quad-Core Processor, a GeForce 8400 and special effects with the nVidiia drivers. Several times a day (maybe once every half hour at worst), ksensors tells me one of the cores is at 100% utilization and it stays there for about 30 seconds. I usually don't notice unless that's the same core on which Firefox is running, and then it seems completely frozen (except that I still can switch desktops). This has been going on since F14, I think.

I'm asking if these are the same symptoms, because it's not clear that what you're describing is a temporary effect like I'm seeing.

[joober]

Originally Posted by joober

Right after these messages, KDE freezes completely, except for the mouse pointer which continues to move. A reboot is the only way to regain a working desktop.

Quoted from my first post, because maybe you missed that bit.

To clarify further: what I am seeing is a permanent freeze. At least, I have waited for up to five minutes to see if the desktop becomes responsive again. It doesn't. Only the mouse pointer still moves but the PC itself becomes completely unresponsive. I can't switch windows, desktops or even to a console. The only way to get a working desktop back (gracefully) is to log in from another machine and initiate a reboot from there.