I had problems with the ide disk recently on my MiniITX machine.
After some days or even several hours only, the machine was issuing tons of errors on the FS, the access to the IDE disk, etc. No way to see the first errors reported, since the console had scrolled too much :(
Everything already in RAM was still working but the rest became unavailable. As the machine acts as my firewall and gateway, I didn't notice as far as the network connection, but the wiki installed on it was not operational any longer :(
I have had no time to reroute the syslog logs to another of my machines to have a hint on what was happening... but this may be something relating to DMA problems, as lots of users of similar systems seem to have experienced this kind of behaviour.
But I can't confirm until I can see the first errors happening...
Using Debian stable had caused no real pain before... but recently the problem got worse and worse... maybe I'm using this machine much more now, or something has been upgraded so that it's become more demanding to the kernel ?
I learned that there are specific patches in the kernel 2.6 (and probably a bunch also hanging around). See http://epialinux.org/files/epia_howto/index.html for an excellent reference.
I've then decided to recompile the kernel with the C3 processor type option, using the Debian stable's stock 386 kernel's
.config file. Although the processor emulates the 386 or 686, a specifically designed kernel may be much preferable for such an uncommon processor.
-initrd commands and a reboot later, my machine seems to be much more stable...
Dunno if this is the end of the problems, but can't do much more harm I think.
Anyway, this tends to convince me more and more that every new machine that I touch needs real customisation (i.e. recompilation of some parts), even in a world of standards (libre software, packaged distros, x86 hardware) ;)
Update 30/10/2005 : Looks like there are still problems... however less often... yesterday, it has had problems again, and this time, I had redirected syslog to another machine, so I've ben able to get a hand on the error messages... Hope it will help me solve it. In the mean time, I'll stick to this new kernel, since it lasted one full week at least.