Why is this not needed for others as well?Well, I can only consider nvme case, why it is not needed there, but in other ULPs that can be probably a bug or they have other protections, I do not know (seems net/rds/ib.c does nice things). *_remove_one() callback will be firstly called if device is dying and only then cm_handler() will get RDMA_CM_EVENT_DEVICE_REMOVAL. Order matters. NVMe host and target do flush_scheduled_work() in nvme[t]_rdma_remove_one(), thus it is guaranteed that *_remove_one() is the last one who puts the reference on the device, i.e.: nvmet_rdma_remove_one(...) { ... flush_scheduled_work(); WARN_ON(!nvmet_rdma_dev_put(ndev)); }
I think it would be a good idea to do the same for iser. If you don't have time, keep it as is, I'll try to find some time to get it incrementally. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
