+static enum blk_eh_timer_return

quoted

quoted

+nvme_tcp_timeout(struct request *rq, bool reserved)
+{
+    struct nvme_tcp_request *req = blk_mq_rq_to_pdu(rq);
+    struct nvme_tcp_ctrl *ctrl = req->queue->ctrl;
+    struct nvme_tcp_cmd_pdu *pdu = req->pdu;
+
+    dev_dbg(ctrl->ctrl.device,
+        "queue %d: timeout request %#x type %d\n",
+        nvme_tcp_queue_id(req->queue), rq->tag,
+        pdu->hdr.type);
+
+    if (ctrl->ctrl.state != NVME_CTRL_LIVE) {
+        union nvme_result res = {};
+
+        nvme_req(rq)->flags |= NVME_REQ_CANCELLED;
+        nvme_end_request(rq, NVME_SC_ABORT_REQ, res);
+        return BLK_EH_DONE;

This looks odd.  It's not really the timeout handlers job to
call nvme_end_request here.

Well.. if we are not yet LIVE, we will not trigger error
recovery, which means nothing will complete this command so
something needs to do it...

I think that we need it for rdma too..

yes we do. and we've patches in our pipe.

I'm thinking on a wider change in the error/recovery flows but might 
send it "as is" meanwhile.

...

The rest of the comments will be addressed in the next submission..