Re: 2.6.17-rc5-mm3

[Dave Jones <davej@redhat.com>]

On Mon, Jun 05, 2006 at 01:53:54PM -0700, Andrew Morton wrote:
 
 > >  >  > Try reverting debug-shared-irqs.patch, or disable the sound driver?
 > >  > Will turn off the sound driver, and see what happens.
 > > Win! It now boots.
 > So does Windows 95.

Hey, it's my turn to play "optimist" today. :)

 > >   I blew it up really easy with a socket-fuzzer though.
 > > (http://people.redhat.com/davej/sfuzz.c)
 > 
 > But it kept running OK, yes?

Yep, still ticking along (for now).

 > > [  874.865028] ======================================
 > > [  874.943738] [ BUG: bad unlock ordering detected! ]
 > > [  875.002919] --------------------------------------
 > > [  875.062134] sfuzz/23915 is trying to release lock (&sctp_port_alloc_lock) at:
 > > [  875.149619]  [<d128ed4e>] sctp_get_port_local+0xd0/0x285 [sctp]
 > > [  875.222636] but the next lock to release is:
 > > [  875.276019]  (&sctp_port_hashtable[i].lock){-...}, at: [<d128ed0e>] sctp_get_port_local+0x90/0x285 [sctp]
 > > [  875.393031]
 > > [  875.393032] other info that might help us debug this:
 > > [  875.476583] 1 locks held by sfuzz/23915:
 > > [  875.526247]  #0:  (&sctp_port_alloc_lock){-...}, at: [<d128ecd9>] sctp_get_port_local+0x5b/0x285 [sctp]
 > > [  875.641621]
 > > [  875.641623] stack backtrace:
 > > [  875.699891]  [<c0104966>] show_trace_log_lvl+0x54/0xfd
 > > [  875.764425]  [<c0104f1a>] show_trace+0xd/0x10
 > > [  875.819622]  [<c010502f>] dump_stack+0x19/0x1b
 > > [  875.875924]  [<c013b4af>] lockdep_release+0x150/0x2d1
 > > [  875.939610]  [<c032341e>] _spin_unlock+0x16/0x20
 > > [  875.998171]  [<d128ed4e>] sctp_get_port_local+0xd0/0x285 [sctp]
 > > [  876.072345]  [<d128efd4>] sctp_do_bind+0x9a/0x158 [sctp]
 > > [  876.139315]  [<d128f0ce>] sctp_autobind+0x3c/0x44 [sctp]
 > > [  876.206310]  [<d129253d>] sctp_inet_listen+0xe9/0x139 [sctp]
 > > [  876.277539]  [<c02c20af>] sys_listen+0x4a/0x65
 > > [  876.334730]  [<c02c308d>] sys_socketcall+0x98/0x186
 > > [  876.397175]  [<c03239cb>] syscall_call+0x7/0xb
 > 
 > This is a really really fussy "BUG", IMO.  So we undid the locks in an
 > inappropriate order - big deal.
 > 
 > But often these _are_ things which we should tune up, as an efficiency
 > thing, so it is interesting to hear about them.  But calling it a "BUG" is
 > a bit alarmist.

Maybe so, but it's still pretty grotty though.

		Dave

-- 
http://www.codemonkey.org.uk
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/