Linux Format forums

[davecs]

My only real frustration when using Linux is a fault that seems only to happen on some boxes and not others, and only when using GLX/3D acceleration, especially on nVidia but also occasionally on ATi.

What happens is that everything freezes except the mouse pointer which moves around the screen but its cursor stays fixed. Apparently if the computer is on a network, you can SSH into it (whatever that means) and find that X has taken over all your memory, you can close X and the computer is still running. Otherwise you are forced to press ALT-SYSRQ-S to save all open files, ALT-SYSRQ-U to unmount them and reopen them as read-only, and ALT-SYSRQ-B to reboot. There are other combinations that restart the keyboard, etc, but ultimately you never get up and running properly unless you reboot.

There are several theories about the origin of this fault. One is the nVidia drivers. I'm not sure about this, because I have read that it can happen in any 3D-enabled X setup. Another is the kernel. Another is somewhere in X.

Everyone blames the other, and you only read about it on forums. It never seems to be discussed in magazines, or acknowledged by any developers. At the nvidia help site, it is one of the biggest topics, but no "official" answers at all.

It also seems to strike randomly, I didn't have this problem for ages, and then it got so bad that I had to switch to the nv (2D) driver which is slow even in 2D.

It can strike in any distro, and there is no rhyme or reason. Two people with identical computers and identical distros can find that one persion never sees this bug and another sees it all the time. A little tweak of xorg.conf can result in it going away for a while only for it to come back with a vengeance later.

If developers can't work out what's causing it, can they at least add a means of escaping from it without having to reboot? Has anyone else had this problem?

[towy71]

It happened to me yesterday but I did not know the keyboard sequence to reboot so had to resort to reset button which then set off whole sequence of fsck stuff but the system did eventually get going again!
so thank davecs for the tip will bear it mind if it happens again
Dick
_________________
still looking for that door into summer

[Rhakios]

One thing to be aware of with the alt+sysrq sequence, some distros disable this function by default for security reasons. SuSE, for example: one of my first tasks after a fresh installation of a SuSE distribution is to re-enable it.

There is a full explanantion of all the magic sysrq key sequnces in the kernel docs, such as
/usr/src/linux/Documentation/sysrq.txt
on SuSE. IIRC, it has also been dealt with briefly in the magazine.
_________________
Bye, Rhakios

[davecs]

If you don't actually need 3D/GLX, the "nv" solution is OK, it just slows the machine down so much. Oh and if you have an analogue connection to your monitor, it shifts the picture over, about 6-7%, so you have to adjust it, which is annoying if you need to change back and forth.

But still, is there a way to force this issue right out into the open so that the developers will get their heads together and deal with it? Maybe we need a "bribery" fund for them.
_________________

Asus Asus M2N32 WS Pro+Athlon AM2/4200+ — GeForce 7600GT — 2Gb Cosair VS RAM — 500Gb WD5000AAKS SATA Drive — PCLinuxOS

[towy71]

Bribery? What are you implying?

Thats my tuppence!
_________________
still looking for that door into summer

[fingers99]

So,

Alt + backspace won't do it?

That would seem to point to something more fundamental than X.

[linuxgirlie]

I must admit I always resort to alt + backspace if I have a problem, unless I have a winex problem when I do kill wine -9 I use that, but i'm lazy

Jo
_________________
My knowledge comes with no warranty...........

Server operating system designed for schools:www.linuxschools.com

[davecs]

[nelz]

It doesn't sound like a kernel issue. If, as you said earlier, you can still SSH into the computer from another, the OS is running fine. The problem is that X has locked up and, as you say, X is grabbing almost all input so you cannot get out of it. Using SSH bypasses all of that since it sets up a new console login that has nothing to do with X. From there you can kill X.

A kernel level lockup would also kill networking, so SSH wouldn't work.

[fingers99]

OK, so what effect does

Alt + SysRq + R have?

Is it true for all the nVidia drivers or only the latest one(s).

[towy71]

I don't like you fingers99

I had to try it and then had to Alt+SysRq+B grrrr
_________________
still looking for that door into summer

[Rhakios]

Why? You should have been able to switch to another terminal and shutdown gracefully. That is the purpose of alt+sysrq+r, to free the keyboard from a locked X session and allow you to switch to a terminal where you can do things in a more controlled manner.
_________________
Bye, Rhakios

[systemx]

The current state of nvidia is pretty unstable and it is my experience that anything over 1.0-6629 can and will cause lockups on most systems...

As for ATI, I have no idea because I don't use it

If yr already using that nvidia version, maybe you should apply there patch set: http://www.nvnews.net/vbulletin/showpost.php?p=553970&postcount=1

It fixes most, if not all the known bugs in that release...

[davecs]

Unfortunately, Nvidia 6629 won't install on my PCLinuxOS system, only 7174 will. There is a fix in the full version of the above post, but that doesn't work for me either!

I've put back "nvidia" in my xorg.conf and am running in 3D at present. Just to see how long it lasts!
_________________

Asus Asus M2N32 WS Pro+Athlon AM2/4200+ — GeForce 7600GT — 2Gb Cosair VS RAM — 500Gb WD5000AAKS SATA Drive — PCLinuxOS

[doctorflange]