Re: lockdep: adding bonding device triggers warning

On Tue, 2006-08-15 at 14:11 -0400, Bill Nottingham wrote:

2.6.17-1.2564.fc6 is 2.6.18rc4+.

Happened with 'echo "+bond0" > /sys/class/net/bonding_masters'

bonding: bond0 is being created...

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.17-1.2564.fc6 #1
-------------------------------------------------------
bash/9497 is trying to acquire lock:
 (rtnl_mutex){--..}, at: [<c0612860>] mutex_lock+0x21/0x24

but task is already holding lock:
 (&bonding_rwsem){----}, at: [<f8c332e0>] bonding_store_bonds+0x28/0x1c6 [bonding]

which lock already depends on the new lock.

looks like a real deadlock:

SIOCSIFNAME ioctl takes  rtnl_lock() then calls dev_ifsioc which calls
dev_change_name which calls the netdev_chain notifier... which ends up
calling bond_event_changename() which does:
        down_write(&(bonding_rwsem));


on the other hand, bonding_store_bonds() does 
        down_write(&(bonding_rwsem));
then calls bond_create() which does:
int bond_create(char *name, struct bond_params *params, struct bonding
**newbond)
{
        struct net_device *bond_dev;
        int res;

        rtnl_lock();

since these both are global locks and not per device locks, this really
does look like an AB-BA deadlock to me....


-- 
if you want to mail me at work (you don't), use arjan (at) linux.intel.com