On 11/16/2016 09:39 AM, Leon Romanovsky wrote: > @@ -745,13 +746,17 @@ int rxe_requester(void *arg) > wqe->status = IB_WC_LOC_PROT_ERR; > wqe->state = wqe_state_error; > > -complete: > - if (qp_type(qp) != IB_QPT_RC) { > - while (rxe_completer(qp) == 0) > - ; > - } > - > - return 0; > + /* > + * IBA Spec. Section 10.7.3.1 SIGNALED COMPLETIONS > + * ---------8<---------8<------------- > + * ...Note that if a completion error occurs, a Work Completion > + * will always be generated, even if the signaling > + * indicator requests an Unsignaled Completion. > + * ---------8<---------8<------------- > + */ > + wqe->wr.send_flags |= IB_SEND_SIGNALED; > + __rxe_do_task(&qp->comp.task); > + return -EAGAIN; Hello Leon and Yonatan, Sorry for the late reply but I think setting IB_SEND_SIGNALED for WQE's reporting a completion error is wrong. I noticed this because this change conflicts with Doug's rxe branch on github. Have you considered to apply the following (untested) patch instead of setting the IB_SEND_SIGNALED flag? --- a/drivers/infiniband/sw/rxe/rxe_comp.c +++ b/drivers/infiniband/sw/rxe/rxe_comp.c @@ -418,7 +418,7 @@ static void do_complete(struct rxe_qp *qp, struct rxe_send_w qe *wqe) if ((qp->sq_sig_type == IB_SIGNAL_ALL_WR) || (wqe->wr.send_flags & IB_SEND_SIGNALED) || - (qp->req.state == QP_STATE_ERROR)) { + wqe->status != IB_WC_SUCCESS) { make_send_cqe(qp, wqe, &cqe); advance_consumer(qp->sq.queue); rxe_cq_post(qp->scq, &cqe, 0); Bart. -- 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