lockmgr_args, lock order reversal: (sleepable after non-sleepable)

[David Xu]

I saw so many LORs like following, I think this is a false reporting
because the interlock is released by lockmgr whether the thread is
blocked or not, shouldn't WITNESS_CHECKORDER bypass this interlock ?


lock order reversal: (sleepable after non-sleepable)
  1st 0xc48d728c vnode interlock (vnode interlock) @ 
fs/devfs/devfs_vnops.c:286
  2nd 0xc48d7270 devfs (devfs) @ kern/vfs_subr.c:2051
KDB: stack backtrace:
db_trace_self_wrapper(c0b6d1e1,c4199a2c,c07f1d35,4,c0b68bf7,...) at 
db_trace_self_wrapper+0x26
kdb_backtrace(4,c0b68bf7,c0de8a58,c44768d0,c4199b64,...) at 
kdb_backtrace+0x29
_witness_debugger(c0b6fa7b,c48d6000,c0b761c4,c44768d0,c0b7679d,...) at 
_witness_debugger+0x25
witness_checkorder(c48d6000,1,c0b76794,174,c0b68bf7,...) at 
witness_checkorder+0x7c9
__lockmgr_args(c48d6000,200100,c48d601c,0,0,...) at __lockmgr_args+0x230
vfs_busy(c48d6000,200,0,1,c44c6d20,...) at vfs_busy+0x1bc
vfs_mount_alloc(0,c0c3bae0,c0b7653a,c44b1400,c0831580,...) at 
vfs_mount_alloc+0x74
vfs_mountroot(c0caef30,4,c0b64bc9,264,0,...) at vfs_mountroot+0x272
start_init(0,c4199d38,c0b66578,322,c44c4d0c,...) at start_init+0x65
fork_exit(c077a040,0,c4199d38) at fork_exit+0xb8


David Xu