I m confused with nVidia driver, so I'm opening a new thread for it. Here's my problem description: I'm running on AMD64 3000+, board Asus K8V-X SE, nVidia model is GeForce 6200. There are two gentoo-linux systems - one for amd64 architecture, one for generic x86. Both systems have 1.0.7676 driver installed, both systems are running same version of the kernel - 2.6.12.

On 64bit (amd64) system it works perfectly - no stability issues. On 32bit (x86) it works as well with one significant exception: If I boot to the x86 system FIRST AFTER POWERON, I get black screen and totally frozen box - probably only on user level, because I could hear sync from time to time when waiting for any more progress (10 min - nothing). I was even not able to obtain the bugreport from the frozen system state - script that should run in background hasn't even started.

Now the strangest thing: If I first boot to amd64 after poweron, do nothing more, just boot up to the level 3 WITHOUT starting X or inserting the module and reboot back to x86, everything works 100% and even after any reboot it works further. I just must not switch off the box otherwise, as I already wrote, I must first boot 64bit system to get it work in 32 bit..

My conclusion from this experience is that the 64bit kernel must do some initializations that the 32bit kernel doesn't. I tried to make those kernels as close as possible regarding configuration, but unfortunately nothing helped yet.

Maybe this problem is so well known that it wasn't worth mentioning anywhere, but in case of any more interest I can provide the bug reports and the kernel configs as well of course.

Thanks for any suggestions

It's hard to be sure what the issue is, but let me give it a try.
I am going to assume that both systems are running 64-bit processors, if this is not true, them you can probably stop reading and should post the correction.
The registers in a 64-bit processor must be initialized, as even though you only require 32-bit versions in x86 or 8086 modes, the OS and software know that the processor is 64-bit. This in turn implements stacks and buffers to suit such a setup. It's complicated, but based on the OS little/big endian uses 64-bit segments forwards or backwards.

What is likely happening is by starting the 64-bit version you initialize the registers/buffers/pipelines, etc for use, with their initial values (null, 0 or ignore values-known as dirty bits) once this has been done and the system is 'primed' for 64-bit use, it can easily implement 32-bit software and stack segments, the same way a 32-bit processor implements 16- and 8-bit software. backwards compatability usually works, the other way around from a cold boot into a 32-bit version.. not always.
*The problem may not be the kernel itself, but rather the implementation approach it uses in using memory, and it does not initialize correctly.
Hope this helps or gives some insight.

__________________

</Dream In Code>
-William. § (marvin_gohan)