I have found a way to reliably repeat the endless loop of "dentry_iput"
messages. If the machine is busy compiling a good-sized application in
one session, and I start Bonnie against a server mounted over 100BaseTx,
it will almost always trigger the fault.
Let me know when patches are available, and I'll see if I can break them
<g>. On the bright side, since you identified and fixed the problem with
sparse files over NFS I'm able to build working kernels on my Alpha in a
mounted filesystem. All along I was blaming a kernel bug and it turned
out to be corrupted object files!
Steve