>OK, the smaller patch below seems to fix things to, so that's what >I'll merge. I think the real issue was that I left the code as > > ret = kobject_init_and_add(&cm_dev->dev_obj, &cm_dev_obj_type, > &cm_class.subsys.kobj, "%s", device->name); > >without doing kobject_get() on subsys.kobj, which meant on unload, we >did an extra put of that subsys.kobj, which left the infiniband_cm >class in a messed-up state. But there's no reason to do an extra get >of the parents (kobject_init_and_add() already gets the parent once >for us), so we can just simplify everything as below:
Thanks for clarifying this to me. - Sean _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
