1 files changed, 106 insertions, 55 deletions

diff --git a/drivers/infiniband/sw/rxe/rxe_comp.c b/drivers/infiniband/sw/rxe/rxe_comp.c
index 20737fec392b..db18ace74d2b 100644
--- a/drivers/infiniband/sw/rxe/rxe_comp.c
+++ b/drivers/infiniband/sw/rxe/rxe_comp.c
@@ -118,10 +118,12 @@ void retransmit_timer(struct timer_list *t)
 
 	rxe_dbg_qp(qp, "retransmit timer fired\n");
 
+	spin_lock_bh(&qp->state_lock);
 	if (qp->valid) {
 		qp->comp.timeout = 1;
 		rxe_sched_task(&qp->comp.task);
 	}
+	spin_unlock_bh(&qp->state_lock);
 }
 
 void rxe_comp_queue_pkt(struct rxe_qp *qp, struct sk_buff *skb)
@@ -322,7 +324,7 @@ static inline enum comp_state check_ack(struct rxe_qp *qp,
 					qp->comp.psn = pkt->psn;
 					if (qp->req.wait_psn) {
 						qp->req.wait_psn = 0;
-						rxe_run_task(&qp->req.task);
+						rxe_sched_task(&qp->req.task);
 					}
 				}
 				return COMPST_ERROR_RETRY;
@@ -428,6 +430,10 @@ static void make_send_cqe(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
 				uwc->wc_flags = IB_WC_WITH_IMM;
 			uwc->byte_len = wqe->dma.length;
 		}
+	} else {
+		if (wqe->status != IB_WC_WR_FLUSH_ERR)
+			rxe_err_qp(qp, "non-flush error status = %d",
+				wqe->status);
 	}
 }
 
@@ -469,30 +475,16 @@ static void do_complete(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
 	 */
 	if (qp->req.wait_fence) {
 		qp->req.wait_fence = 0;
-		rxe_run_task(&qp->req.task);
+		rxe_sched_task(&qp->req.task);
 	}
 }
 
-static inline enum comp_state complete_ack(struct rxe_qp *qp,
-					   struct rxe_pkt_info *pkt,
-					   struct rxe_send_wqe *wqe)
+static void comp_check_sq_drain_done(struct rxe_qp *qp)
 {
-	if (wqe->has_rd_atomic) {
-		wqe->has_rd_atomic = 0;
-		atomic_inc(&qp->req.rd_atomic);
-		if (qp->req.need_rd_atomic) {
-			qp->comp.timeout_retry = 0;
-			qp->req.need_rd_atomic = 0;
-			rxe_run_task(&qp->req.task);
-		}
-	}
-
-	if (unlikely(qp->req.state == QP_STATE_DRAIN)) {
-		/* state_lock used by requester & completer */
-		spin_lock_bh(&qp->state_lock);
-		if ((qp->req.state == QP_STATE_DRAIN) &&
-		    (qp->comp.psn == qp->req.psn)) {
-			qp->req.state = QP_STATE_DRAINED;
+	spin_lock_bh(&qp->state_lock);
+	if (unlikely(qp_state(qp) == IB_QPS_SQD)) {
+		if (qp->attr.sq_draining && qp->comp.psn == qp->req.psn) {
+			qp->attr.sq_draining = 0;
 			spin_unlock_bh(&qp->state_lock);
 
 			if (qp->ibqp.event_handler) {
@@ -504,10 +496,27 @@ static inline enum comp_state complete_ack(struct rxe_qp *qp,
 				qp->ibqp.event_handler(&ev,
 					qp->ibqp.qp_context);
 			}
-		} else {
-			spin_unlock_bh(&qp->state_lock);
+			return;
 		}
 	}
+	spin_unlock_bh(&qp->state_lock);
+}
+
+static inline enum comp_state complete_ack(struct rxe_qp *qp,
+					   struct rxe_pkt_info *pkt,
+					   struct rxe_send_wqe *wqe)
+{
+	if (wqe->has_rd_atomic) {
+		wqe->has_rd_atomic = 0;
+		atomic_inc(&qp->req.rd_atomic);
+		if (qp->req.need_rd_atomic) {
+			qp->comp.timeout_retry = 0;
+			qp->req.need_rd_atomic = 0;
+			rxe_sched_task(&qp->req.task);
+		}
+	}
+
+	comp_check_sq_drain_done(qp);
 
 	do_complete(qp, wqe);
 
@@ -538,25 +547,60 @@ static inline enum comp_state complete_wqe(struct rxe_qp *qp,
 	return COMPST_GET_WQE;
 }
 
-static void rxe_drain_resp_pkts(struct rxe_qp *qp, bool notify)
+/* drain incoming response packet queue */
+static void drain_resp_pkts(struct rxe_qp *qp)
 {
 	struct sk_buff *skb;
-	struct rxe_send_wqe *wqe;
-	struct rxe_queue *q = qp->sq.queue;
 
 	while ((skb = skb_dequeue(&qp->resp_pkts))) {
 		rxe_put(qp);
 		kfree_skb(skb);
 		ib_device_put(qp->ibqp.device);
 	}
+}
+
+/* complete send wqe with flush error */
+static int flush_send_wqe(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
+{
+	struct rxe_cqe cqe = {};
+	struct ib_wc *wc = &cqe.ibwc;
+	struct ib_uverbs_wc *uwc = &cqe.uibwc;
+	int err;
+
+	if (qp->is_user) {
+		uwc->wr_id = wqe->wr.wr_id;
+		uwc->status = IB_WC_WR_FLUSH_ERR;
+		uwc->qp_num = qp->ibqp.qp_num;
+	} else {
+		wc->wr_id = wqe->wr.wr_id;
+		wc->status = IB_WC_WR_FLUSH_ERR;
+		wc->qp = &qp->ibqp;
+	}
+
+	err = rxe_cq_post(qp->scq, &cqe, 0);
+	if (err)
+		rxe_dbg_cq(qp->scq, "post cq failed, err = %d", err);
+
+	return err;
+}
+
+/* drain and optionally complete the send queue
+ * if unable to complete a wqe, i.e. cq is full, stop
+ * completing and flush the remaining wqes
+ */
+static void flush_send_queue(struct rxe_qp *qp, bool notify)
+{
+	struct rxe_send_wqe *wqe;
+	struct rxe_queue *q = qp->sq.queue;
+	int err;
 
 	while ((wqe = queue_head(q, q->type))) {
 		if (notify) {
-			wqe->status = IB_WC_WR_FLUSH_ERR;
-			do_complete(qp, wqe);
-		} else {
-			queue_advance_consumer(q, q->type);
+			err = flush_send_wqe(qp, wqe);
+			if (err)
+				notify = 0;
 		}
+		queue_advance_consumer(q, q->type);
 	}
 }
 
@@ -571,9 +615,28 @@ static void free_pkt(struct rxe_pkt_info *pkt)
 	ib_device_put(dev);
 }
 
-int rxe_completer(void *arg)
+/* reset the retry timer if
+ * - QP is type RC
+ * - there is a packet sent by the requester that
+ *   might be acked (we still might get spurious
+ *   timeouts but try to keep them as few as possible)
+ * - the timeout parameter is set
+ * - the QP is alive
+ */
+static void reset_retry_timer(struct rxe_qp *qp)
+{
+	if (qp_type(qp) == IB_QPT_RC && qp->qp_timeout_jiffies) {
+		spin_lock_bh(&qp->state_lock);
+		if (qp_state(qp) >= IB_QPS_RTS &&
+		    psn_compare(qp->req.psn, qp->comp.psn) > 0)
+			mod_timer(&qp->retrans_timer,
+				  jiffies + qp->qp_timeout_jiffies);
+		spin_unlock_bh(&qp->state_lock);
+	}
+}
+
+int rxe_completer(struct rxe_qp *qp)
 {
-	struct rxe_qp *qp = (struct rxe_qp *)arg;
 	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
 	struct rxe_send_wqe *wqe = NULL;
 	struct sk_buff *skb = NULL;
@@ -581,15 +644,17 @@ int rxe_completer(void *arg)
 	enum comp_state state;
 	int ret;
 
-	if (!rxe_get(qp))
-		return -EAGAIN;
+	spin_lock_bh(&qp->state_lock);
+	if (!qp->valid || qp_state(qp) == IB_QPS_ERR ||
+			  qp_state(qp) == IB_QPS_RESET) {
+		bool notify = qp->valid && (qp_state(qp) == IB_QPS_ERR);
 
-	if (!qp->valid || qp->comp.state == QP_STATE_ERROR ||
-	    qp->comp.state == QP_STATE_RESET) {
-		rxe_drain_resp_pkts(qp, qp->valid &&
-				    qp->comp.state == QP_STATE_ERROR);
+		drain_resp_pkts(qp);
+		flush_send_queue(qp, notify);
+		spin_unlock_bh(&qp->state_lock);
 		goto exit;
 	}
+	spin_unlock_bh(&qp->state_lock);
 
 	if (qp->comp.timeout) {
 		qp->comp.timeout_retry = 1;
@@ -677,20 +742,7 @@ int rxe_completer(void *arg)
 				break;
 			}
 
-			/* re reset the timeout counter if
-			 * (1) QP is type RC
-			 * (2) the QP is alive
-			 * (3) there is a packet sent by the requester that
-			 *     might be acked (we still might get spurious
-			 *     timeouts but try to keep them as few as possible)
-			 * (4) the timeout parameter is set
-			 */
-			if ((qp_type(qp) == IB_QPT_RC) &&
-			    (qp->req.state == QP_STATE_READY) &&
-			    (psn_compare(qp->req.psn, qp->comp.psn) > 0) &&
-			    qp->qp_timeout_jiffies)
-				mod_timer(&qp->retrans_timer,
-					  jiffies + qp->qp_timeout_jiffies);
+			reset_retry_timer(qp);
 			goto exit;
 
 		case COMPST_ERROR_RETRY:
@@ -730,7 +782,7 @@ int rxe_completer(void *arg)
 							RXE_CNT_COMP_RETRY);
 					qp->req.need_retry = 1;
 					qp->comp.started_retry = 1;
-					rxe_run_task(&qp->req.task);
+					rxe_sched_task(&qp->req.task);
 				}
 				goto done;
 
@@ -752,6 +804,7 @@ int rxe_completer(void *arg)
 				 */
 				qp->req.wait_for_rnr_timer = 1;
 				rxe_dbg_qp(qp, "set rnr nak timer\n");
+				// TODO who protects from destroy_qp??
 				mod_timer(&qp->rnr_nak_timer,
 					  jiffies + rnrnak_jiffies(aeth_syn(pkt)
 						& ~AETH_TYPE_MASK));
@@ -784,7 +837,5 @@ exit:
 out:
 	if (pkt)
 		free_pkt(pkt);
-	rxe_put(qp);
-
 	return ret;
 }