How to avoid a bonding instance in mode 1 to send GARP for slave NIC when going down

Issue

When a machine is going down, a bond instance running in mode 1 is sending GARP for a
a not active interface.
This is happening when the active interface is removed from the bond and the slave interface is becoming a new master before removing it from the bond instance completely.

That is happening because the bonding instance is making "ifdown" first before removing it completely.

Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: Removing slave xxx
Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: Releasing backup interface xxx
Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: the permanent HWaddr of xxx - xx:xx:xx:xx:xx:xx - is still in use by bond0 - set the HWaddr of xxx to a different address to avoid conflicts
Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: making interface yyy the new active one <<<<< here is the GARP sent
Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: Removing slave yyy
Dec 24 22:23:41 TBCNBSRV03 kernel: bond0: Releasing backup interface yyy
Dec 24 22:23:41 TBCNBSRV03 network: Shutting down interface bond0:  [  OK  ]