On Thu, 2025-02-20 at 11:18 +0000, Hangbin Liu wrote: > > Thanks for the feedback and confirmation. Let me try it first. Hope > unregistering bond doesn't affect the gc works. I think this should be handled naturally as part of the bond device tear down. A quick peek shows: bond_uninit -> (for each slave) __bond_release_one -> bond_change_active_slave (_, NULL) -> bond_ipsec_del_sa_all deletes all matching xfrm_state entries from bond->ipsec_list, with the mutex held. Presumably, a new step after all slaves are deleted should call drain_workqueue on the new workqueue to wait for any scheduled work to be done before finally destroying the workqueue. Cosmin.
