Ok, I've got an example of deadlock.  slogin and ktruss -i the sshd.
ktruss waits for tty output and write of trace data to pipe blocks.
I guess I can drop a new entry if queue length is long.

> I seem to vaguely recall some similar-sounding changes so that ktrace
> could write out the trace file to NFS, but my memory is pretty hazy on 
> this, so I could be out in left field :-)

Ya, NFS will make hang due to ktrcsw() easier than local file system.

enami.