RDMA/nes: CM connection setup/teardown rework

Major rework of CM connection setup/teardown.  We had a number of issues
with MPI applications not starting/terminating properly over time.
With these changes we were able to run longer on larger clusters.

* Remove memory allocation from nes_connect() and nes_cm_connect().
* Fix mini_cm_dec_refcnt_listen() when destroying listener.
* Remove unnecessary code from schedule_nes_timer() and nes_cm_timer_tick().
* Functionalize mini_cm_recv_pkt() and process_packet().
* Clean up cm_node->ref_count usage.
* Reuse skbs if available.
Signed-off-by: Faisal Latif <flatif@neteffect.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/hw/nes/nes.c

@@ -276,6 +276,7 @@ static void nes_cqp_rem_ref_callback(struct nes_device *nesdev, struct nes_cqp_r
 	}
 	nes_free_resource(nesadapter, nesadapter->allocated_qps, nesqp->hwqp.qp_id);
 
+	nesadapter->qp_table[nesqp->hwqp.qp_id-NES_FIRST_QPN] = NULL;
 	kfree(nesqp->allocated_buffer);
 
 }
@@ -289,7 +290,6 @@ void nes_rem_ref(struct ib_qp *ibqp)
 	struct nes_qp *nesqp;
 	struct nes_vnic *nesvnic = to_nesvnic(ibqp->device);
 	struct nes_device *nesdev = nesvnic->nesdev;
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	struct nes_hw_cqp_wqe *cqp_wqe;
 	struct nes_cqp_request *cqp_request;
 	u32 opcode;
@@ -303,8 +303,6 @@ void nes_rem_ref(struct ib_qp *ibqp)
 	}
 
 	if (atomic_dec_and_test(&nesqp->refcount)) {
-		nesadapter->qp_table[nesqp->hwqp.qp_id-NES_FIRST_QPN] = NULL;
-
 		/* Destroy the QP */
 		cqp_request = nes_get_cqp_request(nesdev);
 		if (cqp_request == NULL) {

drivers/infiniband/hw/nes/nes_cm.h

@@ -83,6 +83,8 @@ enum nes_timer_type {
 #define SET_FIN 4
 #define SET_RST 8
 
+#define TCP_OPTIONS_PADDING	3
+
 struct option_base {
 	u8 optionnum;
 	u8 length;
@@ -177,6 +179,7 @@ enum nes_cm_node_state {
 	NES_CM_STATE_ESTABLISHED,
 	NES_CM_STATE_ACCEPTING,
 	NES_CM_STATE_MPAREQ_SENT,
+	NES_CM_STATE_MPAREQ_RCVD,
 	NES_CM_STATE_TSA,
 	NES_CM_STATE_FIN_WAIT1,
 	NES_CM_STATE_FIN_WAIT2,
@@ -187,6 +190,16 @@ enum nes_cm_node_state {
 	NES_CM_STATE_CLOSED
 };
 
+enum nes_tcpip_pkt_type {
+	NES_PKT_TYPE_UNKNOWN,
+	NES_PKT_TYPE_SYN,
+	NES_PKT_TYPE_SYNACK,
+	NES_PKT_TYPE_ACK,
+	NES_PKT_TYPE_FIN,
+	NES_PKT_TYPE_RST
+};
+
+
 /* type of nes connection */
 enum nes_cm_conn_type {
 	NES_CM_IWARP_CONN_TYPE,
@@ -257,7 +270,9 @@ struct nes_cm_node {
 	struct net_device         *netdev;
 
 	struct nes_cm_node        *loopbackpartner;
-	struct list_head          retrans_list;
+
+	struct nes_timer_entry	*send_entry;
+
 	spinlock_t                retrans_list_lock;
 	struct list_head          recv_list;
 	spinlock_t                recv_list_lock;
@@ -276,6 +291,8 @@ struct nes_cm_node {
 	struct nes_vnic           *nesvnic;
 	int                       apbvt_set;
 	int                       accept_pend;
+	int			freed;
+	struct nes_qp		*nesqp;
 };
 
 /* structure for client or CM to fill when making CM api calls. */
@@ -366,14 +383,14 @@ struct nes_cm_ops {
 			struct nes_cm_info *);
 	int (*stop_listener)(struct nes_cm_core *, struct nes_cm_listener *);
 	struct nes_cm_node * (*connect)(struct nes_cm_core *,
-			struct nes_vnic *, struct ietf_mpa_frame *,
+			struct nes_vnic *, u16, void *,
 			struct nes_cm_info *);
 	int (*close)(struct nes_cm_core *, struct nes_cm_node *);
 	int (*accept)(struct nes_cm_core *, struct ietf_mpa_frame *,
 			struct nes_cm_node *);
 	int (*reject)(struct nes_cm_core *, struct ietf_mpa_frame *,
 			struct nes_cm_node *);
-	int (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *,
+	void (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *,
 			struct sk_buff *);
 	int (*destroy_cm_core)(struct nes_cm_core *);
 	int (*get)(struct nes_cm_core *);

drivers/infiniband/hw/nes/nes_hw.c

@@ -2814,7 +2814,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			nesqp = *((struct nes_qp **)&context);
 			if (atomic_inc_return(&nesqp->close_timer_started) == 1) {
 				nesqp->cm_id->add_ref(nesqp->cm_id);
-				nes_add_ref(&nesqp->ibqp);
 				schedule_nes_timer(nesqp->cm_node, (struct sk_buff *)nesqp,
 						NES_TIMER_TYPE_CLOSE, 1, 0);
 				nes_debug(NES_DBG_AEQ, "QP%u Not decrementing QP refcount (%d),"
@@ -2838,7 +2837,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			if (async_event_id == NES_AEQE_AEID_RESET_SENT) {
 				tcp_state = NES_AEQE_TCP_STATE_CLOSED;
 			}
-			nes_add_ref(&nesqp->ibqp);
 			spin_lock_irqsave(&nesqp->lock, flags);
 			nesqp->hw_iwarp_state = iwarp_state;
 			nesqp->hw_tcp_state = tcp_state;
@@ -2876,7 +2874,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				}
 				spin_unlock_irqrestore(&nesqp->lock, flags);
 				if (next_iwarp_state) {
-					nes_add_ref(&nesqp->ibqp);
 					nes_debug(NES_DBG_AEQ, "issuing hw modifyqp for QP%u. next state = 0x%08X,"
 							" also added another reference\n",
 							nesqp->hwqp.qp_id, next_iwarp_state);
@@ -2888,7 +2885,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 					/* FIN Received but ib state not RTS,
 							close complete will be on its way */
 					spin_unlock_irqrestore(&nesqp->lock, flags);
-					nes_rem_ref(&nesqp->ibqp);
 					return;
 				}
 				spin_unlock_irqrestore(&nesqp->lock, flags);
@@ -2922,7 +2918,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			if ((tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
 					((nesqp->ibqp_state == IB_QPS_RTS)&&
 					(async_event_id == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
-				nes_add_ref(&nesqp->ibqp);
 				nes_cm_disconn(nesqp);
 			} else {
 				nesqp->in_disconnect = 0;
@@ -2931,7 +2926,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			break;
 		case NES_AEQE_AEID_LLP_TOO_MANY_RETRIES:
 			nesqp = *((struct nes_qp **)&context);
-			nes_add_ref(&nesqp->ibqp);
 			spin_lock_irqsave(&nesqp->lock, flags);
 			nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_ERROR;
 			nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
@@ -3042,7 +3036,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
 			}
 			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
 			nes_cm_disconn(nesqp);
 			break;
 		case NES_AEQE_AEID_DDP_UBE_INVALID_MSN_NO_BUFFER_AVAILABLE:
@@ -3062,7 +3055,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
 			}
 			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
 			nes_cm_disconn(nesqp);
 			break;
 		case NES_AEQE_AEID_LLP_RECEIVED_MPA_CRC_ERROR:
@@ -3082,7 +3074,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
 			}
 			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
 			nes_cm_disconn(nesqp);
 			break;
 			/* TODO: additional AEs need to be here */

drivers/infiniband/hw/nes/nes_verbs.c

@@ -2867,7 +2867,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			nesqp->hwqp.qp_id, attr->qp_state, nesqp->ibqp_state,
 			nesqp->iwarp_state, atomic_read(&nesqp->refcount));
 
-	nes_add_ref(&nesqp->ibqp);
 	spin_lock_irqsave(&nesqp->lock, qplockflags);
 
 	nes_debug(NES_DBG_MOD_QP, "QP%u: hw_iwarp_state=0x%X, hw_tcp_state=0x%X,"
@@ -2882,7 +2881,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state > (u32)NES_CQP_QP_IWARP_STATE_IDLE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_IDLE;
@@ -2893,7 +2891,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>(u32)NES_CQP_QP_IWARP_STATE_IDLE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_IDLE;
@@ -2904,14 +2901,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>(u32)NES_CQP_QP_IWARP_STATE_RTS) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				if (nesqp->cm_id == NULL) {
 					nes_debug(NES_DBG_MOD_QP, "QP%u: Failing attempt to move QP to RTS without a CM_ID. \n",
 							nesqp->hwqp.qp_id );
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_RTS;
@@ -2929,7 +2924,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id, nesqp->hwqp.sq_head, nesqp->hwqp.sq_tail);
 				if (nesqp->iwarp_state == (u32)NES_CQP_QP_IWARP_STATE_CLOSING) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return 0;
 				} else {
 					if (nesqp->iwarp_state > (u32)NES_CQP_QP_IWARP_STATE_CLOSING) {
@@ -2937,7 +2931,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 								" ignored due to current iWARP state\n",
 								nesqp->hwqp.qp_id);
 						spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-						nes_rem_ref(&nesqp->ibqp);
 						return -EINVAL;
 					}
 					if (nesqp->hw_iwarp_state != NES_AEQE_IWARP_STATE_RTS) {
@@ -2969,7 +2962,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>=(u32)NES_CQP_QP_IWARP_STATE_TERMINATE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				/* next_iwarp_state = (NES_CQP_QP_IWARP_STATE_TERMINATE | 0x02000000); */
@@ -2982,7 +2974,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			case IB_QPS_RESET:
 				if (nesqp->iwarp_state == (u32)NES_CQP_QP_IWARP_STATE_ERROR) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				nes_debug(NES_DBG_MOD_QP, "QP%u: new state = error\n",
@@ -3008,7 +2999,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 				break;
 			default:
 				spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-				nes_rem_ref(&nesqp->ibqp);
 				return -EINVAL;
 				break;
 		}
@@ -3088,7 +3078,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 							nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 							original_last_aeq, nesqp->last_aeq);
 					/* this one is for the cm_disconnect thread */
-					nes_add_ref(&nesqp->ibqp);
 					spin_lock_irqsave(&nesqp->lock, qplockflags);
 					nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
 					nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
@@ -3097,14 +3086,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 				} else {
 					nes_debug(NES_DBG_MOD_QP, "QP%u No fake disconnect, QP refcount=%d\n",
 							nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount));
-					nes_rem_ref(&nesqp->ibqp);
 				}
 			} else {
 				spin_lock_irqsave(&nesqp->lock, qplockflags);
 				if (nesqp->cm_id) {
 					/* These two are for the timer thread */
 					if (atomic_inc_return(&nesqp->close_timer_started) == 1) {
-						nes_add_ref(&nesqp->ibqp);
 						nesqp->cm_id->add_ref(nesqp->cm_id);
 						nes_debug(NES_DBG_MOD_QP, "QP%u Not decrementing QP refcount (%d),"
 								" need ae to finish up, original_last_aeq = 0x%04X."
@@ -3128,14 +3115,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 					" original_last_aeq = 0x%04X. last_aeq = 0x%04X.\n",
 					nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 					original_last_aeq, nesqp->last_aeq);
-			nes_rem_ref(&nesqp->ibqp);
 		}
 	} else {
 		nes_debug(NES_DBG_MOD_QP, "QP%u Decrementing QP refcount (%d), No ae to finish up,"
 				" original_last_aeq = 0x%04X. last_aeq = 0x%04X.\n",
 				nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 				original_last_aeq, nesqp->last_aeq);
-		nes_rem_ref(&nesqp->ibqp);
 	}
 
 	err = 0;