kern/116886: Lock Order Reversal on Current for kbd and zyd

[Weongyo Jeong]

The following reply was made to PR kern/116886; it has been noted by GNATS.

From: Weongyo Jeong <weongyo.jeong at gmail.com>
To: John Merryweather Cooper <j.m.cooper at borgsdemons.com>
Cc: FreeBSD-gnats-submit at freebsd.org
Subject: Re: kern/116886: Lock Order Reversal on Current for kbd and zyd
Date: Mon, 8 Oct 2007 10:55:29 +0900

 On Wed, Oct 03, 2007 at 07:56:46PM -0700, John Merryweather Cooper wrote:
 > 
 > >Number:         116886
 > >Category:       kern
 > >Synopsis:       Lock Order Reversal on Current for kbd and zyd
 > >Confidential:   no
 > >Severity:       non-critical
 > >Priority:       low
 > >Responsible:    freebsd-bugs
 > >State:          open
 > >Quarter:        
 > >Keywords:       
 > >Date-Required:
 > >Class:          sw-bug
 > >Submitter-Id:   current-users
 > >Arrival-Date:   Thu Oct 04 05:20:03 GMT 2007
 > >Closed-Date:
 > >Last-Modified:
 > >Originator:     John Merryweather Cooper
 > >Release:        FreeBSD 7.0-CURRENT amd64
 > >Organization:
 > >Environment:
 > System: FreeBSD borgdemon3.temp.wsu.edu 7.0-CURRENT FreeBSD 7.0-CURRENT #8: Mon Oct 1 12:32:21 PDT 2007 root at borgdemon3.temp.wsu.edu:/usr/obj/usr/src/sys/TURION amd64
 > 
 > 
 > 	
 > >Description:
 > 	When the zyd0 interface first takes a DHCP IP (or when it roams),
 > 	it usually (but not always) throws the attach lock order reversal.
 > 	Rarely, the machine will lock up as a result.
 > >How-To-Repeat:
 > 	When interface takes a DHCP IP address or roams.
 > >Fix:
 > 
 > 	unknown
 > 
 > --- lock-order-reversal begins here ---
 > Oct  3 19:27:26 borgdemon3 kernel: lock order reversal: (Giant after non-sleepable)
 > Oct  3 19:27:26 borgdemon3 kernel: 1st 0xffffff0001674cf0 inp (udpinp) @ /usr/src/sys/netinet/in_mcast.c:670
 > Oct  3 19:27:26 borgdemon3 kernel: 2nd 0xffffffff8078e3c0 Giant (Giant) @ /usr/src/sys/netinet/in_mcast.c:317
 > Oct  3 19:27:26 borgdemon3 kernel: KDB: stack backtrace:
 > Oct  3 19:27:26 borgdemon3 kernel: db_trace_self_wrapper() at db_trace_self_wrapper+0x2a
 > Oct  3 19:27:26 borgdemon3 kernel: witness_checkorder() at witness_checkorder+0x64b
 > Oct  3 19:27:26 borgdemon3 kernel: _mtx_lock_flags() at _mtx_lock_flags+0x75
 > Oct  3 19:27:26 borgdemon3 kernel: in_addmulti() at in_addmulti+0x1e3
 > Oct  3 19:27:26 borgdemon3 kernel: inp_setmoptions() at inp_setmoptions+0x111c
 > Oct  3 19:27:26 borgdemon3 kernel: ip_ctloutput() at ip_ctloutput+0x16f
 > Oct  3 19:27:26 borgdemon3 kernel: sosetopt() at sosetopt+0x3b
 > Oct  3 19:27:26 borgdemon3 kernel: kern_setsockopt() at kern_setsockopt+0xa2
 > Oct  3 19:27:26 borgdemon3 kernel: setsockopt() at setsockopt+0x22
 > Oct  3 19:27:26 borgdemon3 kernel: syscall() at syscall+0x1ca
 > Oct  3 19:27:26 borgdemon3 kernel: Xfast_syscall() at 
 > Oct  3 19:27:26 borgdemon3 kernel: Xfast_syscall+0xab
 > Oct  3 19:27:26 borgdemon3 kernel: --- syscall (105, FreeBSD ELF64, setsockopt), rip = 0x800eb8ebc, rsp = 0x7fffffffeb18, rbp = 0x7fffffffeb90 ---
 > --- lock-order-reversal ends here ---
 
 It seems that this LOR isn't occured in the zyd module only when I review
 the LOR message you sent.
 
 LOR is happened in in_addmulti().
 
 Regards,
 Weongyo Jeong