panic: witness_destroy: lock (sleep mutex) vnode interlock is not initialized

Running 6.0 on a SMP package machine with vfs.lookup_shared=1 and
WITNESS, I hit this panic after a few hours.  At boot time it hit the
vm page queue mutex/vnode interlock LOR, and it also hit a 'acquiring
duplicate lock of same type: "vnode interlock"' when mounting the
first nullfs instance, which may be related.

panic: witness_destroy: lock (sleep mutex) vnode interlock is not initialized
cpuid = 0
KDB: enter: panic
[thread pid 148 tid 100132 ]
Stopped at      kdb_enter+0x30: leave
db> wh
Tracing pid 148 tid 100132 td 0xc572d180
kdb_enter(c0700f44,0,c07054af,f5774c38,c572d180) at kdb_enter+0x30
panic(c07054af,c06dff2c,c0716859,c070abdc,c7f66440) at panic+0x13e
witness_destroy(c7f66440,c0700183,360,c0700545,c06df73d,c7f66440,c0716859,c070abdc,0,0) at witness_destroy+0x5a
mtx_destroy(c7f66440,c7f663a8,8,c7f663a8,f5774cb8) at mtx_destroy+0xf2
vdestroy(c7f663a8,0,c070aaaf,271,0) at vdestroy+0x192
vnlru_free(c,0,c070aaaf,293,3e8) at vnlru_free+0x188
vnlru_proc(0,f5774d38,c06fe130,30c,c572d180) at vnlru_proc+0xb1
fork_exit(c058b9ee,0,f5774d38) at fork_exit+0x116
fork_trampoline() at fork_trampoline+0x8
--- trap 0x1, eip = 0, esp = 0xf5774d6c, ebp = 0 ---
db>