A netfinity 4500R worked ok with 512 MB for two months with the 2.4.9-13 SMP kernel.

Today we added another 512 MB (again IBM ECC memory) because the server was
getting a little busy.
It booted the default SMP kernel alright but apparently the "fsck -T -a
$fsckoptions /"
failed since we were dropped into a shell. Which we found strange because
normally ext3
would handle these situations in the background ...

Trying to run fsck.ext3 from this shell on any filesystem (I tried the one where 
/opt was mounted) resulted in an immediate segmentation violation.
Booting the SMP kernel second time over it hung completely on `Checking root
filesystem'

Because of the segv I thought the memory was the culprit but:
o We booted the 2.4.9-13 UP kernel and it has been running ok for six+ hours now
  and it doesn't exhibit any problems so far. It went some 100MB into swap.

o The extra memory was one IBM ECC 512MB DIMM from another netfinity 4500R
  to which we added it six months ago and which has been running fine with it for
  six months or so, lately with kernel-2.2.19-6.2.10.i686.rpm ...

Unfortunately this is a production system so we cannot just take it down
and do all sorts of tests ..
  
I'll attach UP (with 1024MB) kernel boot messages and SMP (with 512MB) kernel boot 
messages below.  I do not have a log of the boot messages during the segv
mentioned above
ie SMP with 1024 MB, but AFAICT they where the normal messages.

Do you have any idea ?