[PATCH 10/10] IB/rdmavt: Post receive for QP in ERR state
DORMANTno replies
REVIEWED: 2 (0M)
From: Dennis Dalessandro <hidden>
Date: 2016-03-07 19:35:51
Subsystem:
infiniband subsystem, the rest · Maintainers:
Jason Gunthorpe, Leon Romanovsky, Linus Torvalds
2 review trailers.
From: Alex Estrin <redacted> Accordingly IB Spec post WR to receive queue must complete with error if QP is in Error state. Please refer to C10-42, C10-97.2.1 Reviewed-by: Mike Marciniszyn <redacted> Reviewed-by: Dennis Dalessandro <redacted> Signed-off-by: Alex Estrin <redacted> --- drivers/infiniband/sw/rdmavt/qp.c | 33 ++++++++++++++++++++++++--------- 1 files changed, 24 insertions(+), 9 deletions(-)
diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c
index de34474..bd82a69 100644
--- a/drivers/infiniband/sw/rdmavt/qp.c
+++ b/drivers/infiniband/sw/rdmavt/qp.c@@ -1364,6 +1364,8 @@ int rvt_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, struct rvt_qp *qp = ibqp_to_rvtqp(ibqp); struct rvt_rwq *wq = qp->r_rq.wq; unsigned long flags; + int qp_err_flush = (ib_rvt_state_ops[qp->state] & RVT_FLUSH_RECV) && + !qp->ibqp.srq; /* Check that state is OK to post receive. */ if (!(ib_rvt_state_ops[qp->state] & RVT_POST_RECV_OK) || !wq) {
@@ -1390,15 +1392,28 @@ int rvt_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, *bad_wr = wr; return -ENOMEM; } - - wqe = rvt_get_rwqe_ptr(&qp->r_rq, wq->head); - wqe->wr_id = wr->wr_id; - wqe->num_sge = wr->num_sge; - for (i = 0; i < wr->num_sge; i++) - wqe->sg_list[i] = wr->sg_list[i]; - /* Make sure queue entry is written before the head index. */ - smp_wmb(); - wq->head = next; + if (unlikely(qp_err_flush)) { + struct ib_wc wc; + + memset(&wc, 0, sizeof(wc)); + wc.qp = &qp->ibqp; + wc.opcode = IB_WC_RECV; + wc.wr_id = wr->wr_id; + wc.status = IB_WC_WR_FLUSH_ERR; + rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1); + } else { + wqe = rvt_get_rwqe_ptr(&qp->r_rq, wq->head); + wqe->wr_id = wr->wr_id; + wqe->num_sge = wr->num_sge; + for (i = 0; i < wr->num_sge; i++) + wqe->sg_list[i] = wr->sg_list[i]; + /* + * Make sure queue entry is written + * before the head index. + */ + smp_wmb(); + wq->head = next; + } spin_unlock_irqrestore(&qp->r_rq.lock, flags); } return 0; --
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html