Random Hard Lockups on Compile

[energyman76b]

@uxbod

first of all:

NEVER USE THE NVIDIA INSTALLER!

NEVER!

NEVER EVER!

After we have setz that straight - use the latest gentoo kernel (13-r5) or the latest vanilla kernel (2.6.14) and the bug is fixed.
The latest nvidia-kernel ebuild (7676-r1) should also incorporate some of the patches.

Again: use one of the kernels or the ebuilds and never the installer. This will bork your system in a very unpleasant and hard to fix way. Others running into their doom is no reason to follow them.
_________________

[coco-loco]

I encounter that problem since more than one year... with all the different kernels and nvidia drivers that I had installed during all this time.
What I have discovered, having the system monitor open, is that a few seconds before the lockup, the use of memory turns 100% (sys, nice and user). As someone told before, it happens always with the same packages (mySQL, KDE, kdedevelop, etc.).
I does not seem to be 64bit related since I run a dual pentium system.

My resolution for the problem:
- stopping window manager with /etc/init.d/xdm stop
- emerge whatever huge packages
- starting window manager again /etc/init.d/xdm start

That way, I never crashed the system again.
IMHO it must have something to do with the memory management while running the X-server, probably related to the glx drivers - my old PIII machine I use for testing purposes with an i810 built in graphics chip doesn't lockup at all.

[coco-loco]

Seems like if I got the solution!

My thought was basically that there must be a problem with the memory management (as I wrote in my last post). I tried to find out what differences are on these two machines... and I found one that got my attention: on the old PIII test box, I have 256 MB ram and a 512 MB swap, while on my main box I have 512 MB ram and a 124 MB swap. I did this in order to avoid too much swapping, to speed up the machine. Usually, the memory used in 'normal mode' is around 200 MB, so this solution works fine.

I grabbed an old 2 GB scsi disk that I had left from an old server that is out of order and put it into the box. I mounted it as a 2 GB swap drive (insane, I know!) and compiled mysql in graphic mode while working on it (surfed on firefox, Evolution was open and writing some text in Ooo.writer). It used to lock-up the machine, but this time everything worked fine. At several stages of the compilation, the machine seemed to be hardly tortured, and the old scsi disk worked a lot. Checking the use of the swap space showed that sometimes there are over 400 MB swap-space used!!!

This explains somehow the lockups I think. Anyways... this should not happen IMHO, I always thought that there might be some function that stops such processes before the memory overflow... have I missed something?

[dom_cyrus]

[energyman76b]