I've seen a lot of mysterious problems since I recompiled on Jan. 31.
In addition to the "dk_busy < 0" panics (plural, I just got a second
one), I've had two mysterious hangs while NFS was running (I don't
know if it was related, though) where processes trying to exec (or
having already done so?)  hung on something, though other processes
seemed OK.  (In particular, "su" wedged apparently trying to invoke
the shell.  How embarrassing.)  The first time this happened, I tried
to shut the system down and reboot, but while actually executing the
reboot command, it panicked (I don't recall the precise string, but
I think it was a fatal page fault in kernel mode) with this backtrace:

	mi_switch()
	tsleep()
	lock_clear_recursive()
	lock_clear_recursive()
	vm_pager_get_pages()
	vm_pager_get()
	vm_fault()
	vm_fault_wire()
	vm_map_pageable()
	swapin()
	scheduler()
	main()

The second time I got the mysterious process hang, killing a process
that was actually indulging in NFS I/O appeared to make everything
unblock.  I don't think there *was* such a process the first time
around, though.

I also got another fatal page fault while the system was trying to
panic from the second dk_busy crash; that time, the backtrace was just

	mi_switch()
	tsleep()
	scheduler()
	main()

Has anyone else seen anything like this, and (more important) is
anyone actively investigating problems like this?  Can kernels be
compiled with -g to make gdb -k more useful?