landisk / Plextor Panicing problem

[John Klos <john%ziaspace.com@localhost>]

Hi,

I have several Plextor SH4 machines I used to use for NAT and pkgsrc 
building, but something's happened.

One in particular was stable for more than a year with no problems 
whatsoever. I upgraded the OS to 5.1 in the end of December, 2010, and it 
became unstable. It was finally returned to me and I pulled the disk and 
put it in another Plextor which has a serial console to see what was up.

I upgraded everything to netbsd-5 compiled from a few days ago, saw that 
there were UDMA errors, compiled a kernel to only use non-UDMA modes, but 
that made no difference. I ran and re-ran fsck on the filesystem a number 
of times, but after half a day of doing things I ran into a panic like 
this:

free inode //28750646 had 4 blocks
panic: buf mem pool index 23
Begin traceback...
calling through null pointer?
End traceback...

I'm certain it's not the hard drive (which is a 250 gig Hitachi 
which came with the Plextor) because another Plextor with a 16 gig CF card 
became unstable after a year when updating around the same time (December, 
2010).

The issue now is that I can run the system and do lots of I/O (cvs trees, 
et cetera), but after a while if I reboot and force an fsck, it always 
comes up with an unhappy filesystem which requires repair.

This is sort of reminiscent of an issue with VAX machines with larger 
partitions - stable for years, then all of a sudden the filesystem would 
become corrupt after a few days.

Does anyone have any clues as to what might have changed to cause this 
problem?

Thanks,
John Klos