Re: LOR kqueue / sleep mtxpool for LOR list

On Tue, Dec 18, 2007 at 10:20:21AM -1000, Jeff Roberson wrote:

> On Tue, 18 Dec 2007, Ed Maste wrote:
> 
> >I just saw this LOR during startup of an application while testing
> >8-CURRENT on a dev box.  I've done no investigation yet.
> >
> >FreeBSD TPC-D13-08.phaedrus 8.0-CURRENT FreeBSD 8.0-CURRENT #0: Mon Dec 17 
> >16:27:54 EST 2007     
> >emaste_at_bsd-build3.phaedrus:/d2/emaste/HEAD/obj/d2/emaste/HEAD/src/sys/GENERIC  i386
> >
> >lock order reversal:
> >1st 0xc42b7580 kqueue (kqueue) _at_ 
> >/d2/emaste/HEAD/src/sys/kern/kern_event.c:1397
> >2nd 0xc3ece8e0 sleep mtxpool (sleep mtxpool) _at_ 
> >/d2/emaste/HEAD/src/sys/kern/sys_generic.c:1255
> >[...]
>
> Thanks, this is interesting.  The mtxpool should always be the leaf.  The 
> opposite order is the bug.  Can you add them to the static order list and 
> tell me where the reverse order is observed?

After adding it to the list this is what I got:

Mounting NFS file systems:lock order reversal:
 1st 0xc3ece280 sleep mtxpool (sleep mtxpool) _at_ /d2/emaste/HEAD/src/sys/kern/kern_event.c:993
 2nd 0xc4320d80 kqueue (kqueue) _at_ /d2/emaste/HEAD/src/sys/kern/kern_event.c:1001
KDB: stack backtrace:
db_trace_self_wrapper(c0b08831,e683daac,c079cabe,c0b0ada3,c4320d80,...) at db_trace_self_wrapper+0x26
kdb_backtrace(c0b0ada3,c4320d80,c0b01d5d,c0b01d5d,c0b01a4d,...) at kdb_backtrace+0x29
witness_checkorder(c4320d80,9,c0b01a4d,3e9,7db,...) at witness_checkorder+0x6de
_mtx_lock_flags(c4320d80,0,c0b01a4d,3e9,0,...) at _mtx_lock_flags+0xbc
kqueue_acquire(c445eaa0,4,e683db5c,c0b04436,f0,...) at kqueue_acquire+0x72
kern_kevent(c445eaa0,4,1,1,e683dc60,...) at kern_kevent+0x47
kevent(c445eaa0,e683dcfc,18,c0b0bacf,c0bb92e8,...) at kevent+0x19b
syscall(e683dd38) at syscall+0x2b3
Xint0x80_syscall() at Xint0x80_syscall+0x20
--- syscall (363, FreeBSD ELF32, kevent), eip = 0x2813dafb, esp = 0xbfbfd28c, ebp = 0xbfbfd388 ---

-Ed