GUI processes slip into uninterruptible sleep state

[nsoveiko]

i have a weird problem with GUI freezing (i run kde4). this results in load average skyrocketing (even though there are no tasks running) and non-killable X11.

usually i can switch to text console by ctrl-alt-f1, but can't login as myself, only as another user. i can log in as root, but can't restart xdm, nor can i do a clean shutdown. the only way out of this is to reboot, and even that only goes through if run as 'shutdown -rn now'. it happens seemingly randomly, but it only happens when i'm doing something GUI related. never on it's own. it can start with both native kde and no-kde applications.

i've traced it down to processes (e.g. browser) going into 'D' state (uninterruptible sleep) and never coming back. if i try to close the browser window, window manager follows the suit of going into D state. same with kill and 'shutdown -r now'.

how do i find the culprit?

(i'm not even sure if this should go into kernel and hardware or here, but as it's GUI related, let it be here)

[Ant P.]

May be the graphics card driver hanging, though if it was that it should have restarted automatically after 10 seconds.

[nsoveiko]

how do i find out? i'm running x11-drivers/nvidia-drivers.

although, i have a reason to believe it's not video driver. the only kernel process that locks into D state is [sync_supers]. although googling about it doesn't reveal much apart that it has been eliminated in recent kernels (which ones? - i'm on longterm 3.4 branch).

[Ant P.]

The hang auto-recovery is a feature only in KMS drivers. You may want to try reproducing the problem with nouveau.

[Section_8]

Something like this was happening to me a couple of years ago - it turned out that the video card fan was failing and so it would overheat and lock up at random times.

[nsoveiko]

[nsoveiko]