Hi, If a vhost-user port is deleted (vsctl del-port) while the vhost device is still attached to a VM, the port is ending up in some "semi-deleted" state. Even though we indicate a reject of the command in netdev_dpdk_vhost_destruct() by an ERROR log, we still run netdev_close()/netdev_unref() meaning the name is removed from netdev_shash and the memory is also freed (!), given back to dpdk. However, the device is NOT linked out from dpdk_list, meaning it will be handled in the dpdk_watchdog thread. Since the name is removed from netdev_shash, it is treated as not existing from an operators view. Whenever this newly freed memory is allocated again, a lot of strange things may happen to the "ghost device" present in the dpdk_list.
I think the check for an attached vhost device should be done a lot earlier, maybe ofproto_port_delete() is a good place? I have only checked the vhost-user port type, may the same problem exists for other types as well? Anyone else seen the problem? BR /jaw _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
