9 files changed, 185 insertions, 104 deletions

diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
index 80c4b6b401b8..46b64970058e 100644
--- a/drivers/infiniband/sw/rdmavt/mr.c
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -294,7 +294,7 @@ static void __rvt_free_mr(struct rvt_mr *mr)
 {
 	rvt_deinit_mregion(&mr->mr);
 	rvt_free_lkey(&mr->mr);
-	vfree(mr);
+	kfree(mr);
 }
 
 /**
diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c
index 870b4f212fbc..6500c3b5a89c 100644
--- a/drivers/infiniband/sw/rdmavt/qp.c
+++ b/drivers/infiniband/sw/rdmavt/qp.c
@@ -488,60 +488,23 @@ static void rvt_remove_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 	spin_unlock_irqrestore(&rdi->qp_dev->qpt_lock, flags);
 	if (removed) {
 		synchronize_rcu();
-		if (atomic_dec_and_test(&qp->refcount))
-			wake_up(&qp->wait);
+		rvt_put_qp(qp);
 	}
 }
 
 /**
- * reset_qp - initialize the QP state to the reset state
- * @qp: the QP to reset
+ * rvt_init_qp - initialize the QP state to the reset state
+ * @qp: the QP to init or reinit
  * @type: the QP type
- * r and s lock are required to be held by the caller
+ *
+ * This function is called from both rvt_create_qp() and
+ * rvt_reset_qp().   The difference is that the reset
+ * patch the necessary locks to protect against concurent
+ * access.
  */
-static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
-		  enum ib_qp_type type)
-	__releases(&qp->s_lock)
-	__releases(&qp->s_hlock)
-	__releases(&qp->r_lock)
-	__acquires(&qp->r_lock)
-	__acquires(&qp->s_hlock)
-	__acquires(&qp->s_lock)
+static void rvt_init_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+			enum ib_qp_type type)
 {
-	if (qp->state != IB_QPS_RESET) {
-		qp->state = IB_QPS_RESET;
-
-		/* Let drivers flush their waitlist */
-		rdi->driver_f.flush_qp_waiters(qp);
-		qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
-		spin_unlock(&qp->s_lock);
-		spin_unlock(&qp->s_hlock);
-		spin_unlock_irq(&qp->r_lock);
-
-		/* Stop the send queue and the retry timer */
-		rdi->driver_f.stop_send_queue(qp);
-
-		/* Wait for things to stop */
-		rdi->driver_f.quiesce_qp(qp);
-
-		/* take qp out the hash and wait for it to be unused */
-		rvt_remove_qp(rdi, qp);
-		wait_event(qp->wait, !atomic_read(&qp->refcount));
-
-		/* grab the lock b/c it was locked at call time */
-		spin_lock_irq(&qp->r_lock);
-		spin_lock(&qp->s_hlock);
-		spin_lock(&qp->s_lock);
-
-		rvt_clear_mr_refs(qp, 1);
-	}
-
-	/*
-	 * Let the driver do any tear down it needs to for a qp
-	 * that has been reset
-	 */
-	rdi->driver_f.notify_qp_reset(qp);
-
 	qp->remote_qpn = 0;
 	qp->qkey = 0;
 	qp->qp_access_flags = 0;
@@ -587,6 +550,60 @@ static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 }
 
 /**
+ * rvt_reset_qp - initialize the QP state to the reset state
+ * @qp: the QP to reset
+ * @type: the QP type
+ *
+ * r_lock, s_hlock, and s_lock are required to be held by the caller
+ */
+static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+			 enum ib_qp_type type)
+	__must_hold(&qp->s_lock)
+	__must_hold(&qp->s_hlock)
+	__must_hold(&qp->r_lock)
+{
+	lockdep_assert_held(&qp->r_lock);
+	lockdep_assert_held(&qp->s_hlock);
+	lockdep_assert_held(&qp->s_lock);
+	if (qp->state != IB_QPS_RESET) {
+		qp->state = IB_QPS_RESET;
+
+		/* Let drivers flush their waitlist */
+		rdi->driver_f.flush_qp_waiters(qp);
+		qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
+		spin_unlock(&qp->s_lock);
+		spin_unlock(&qp->s_hlock);
+		spin_unlock_irq(&qp->r_lock);
+
+		/* Stop the send queue and the retry timer */
+		rdi->driver_f.stop_send_queue(qp);
+
+		/* Wait for things to stop */
+		rdi->driver_f.quiesce_qp(qp);
+
+		/* take qp out the hash and wait for it to be unused */
+		rvt_remove_qp(rdi, qp);
+		wait_event(qp->wait, !atomic_read(&qp->refcount));
+
+		/* grab the lock b/c it was locked at call time */
+		spin_lock_irq(&qp->r_lock);
+		spin_lock(&qp->s_hlock);
+		spin_lock(&qp->s_lock);
+
+		rvt_clear_mr_refs(qp, 1);
+		/*
+		 * Let the driver do any tear down or re-init it needs to for
+		 * a qp that has been reset
+		 */
+		rdi->driver_f.notify_qp_reset(qp);
+	}
+	rvt_init_qp(rdi, qp, type);
+	lockdep_assert_held(&qp->r_lock);
+	lockdep_assert_held(&qp->s_hlock);
+	lockdep_assert_held(&qp->s_lock);
+}
+
+/**
  * rvt_create_qp - create a queue pair for a device
  * @ibpd: the protection domain who's device we create the queue pair for
  * @init_attr: the attributes of the queue pair
@@ -766,7 +783,7 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
 		}
 		qp->ibqp.qp_num = err;
 		qp->port_num = init_attr->port_num;
-		rvt_reset_qp(rdi, qp, init_attr->qp_type);
+		rvt_init_qp(rdi, qp, init_attr->qp_type);
 		break;
 
 	default:
@@ -906,6 +923,8 @@ int rvt_error_qp(struct rvt_qp *qp, enum ib_wc_status err)
 	int ret = 0;
 	struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
 
+	lockdep_assert_held(&qp->r_lock);
+	lockdep_assert_held(&qp->s_lock);
 	if (qp->state == IB_QPS_ERR || qp->state == IB_QPS_RESET)
 		goto bail;
 
@@ -980,7 +999,7 @@ static void rvt_insert_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 	struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
 	unsigned long flags;
 
-	atomic_inc(&qp->refcount);
+	rvt_get_qp(qp);
 	spin_lock_irqsave(&rdi->qp_dev->qpt_lock, flags);
 
 	if (qp->ibqp.qp_num <= 1) {
@@ -997,7 +1016,7 @@ static void rvt_insert_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 }
 
 /**
- * qib_modify_qp - modify the attributes of a queue pair
+ * rvt_modify_qp - modify the attributes of a queue pair
  * @ibqp: the queue pair who's attributes we're modifying
  * @attr: the new attributes
  * @attr_mask: the mask of attributes to modify
diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c
index 55f0e8f0ca79..ddd59270ff6d 100644
--- a/drivers/infiniband/sw/rxe/rxe.c
+++ b/drivers/infiniband/sw/rxe/rxe.c
@@ -362,15 +362,34 @@ static int __init rxe_module_init(void)
 		return err;
 	}
 
-	err = rxe_net_init();
+	err = rxe_net_ipv4_init();
 	if (err) {
-		pr_err("rxe: unable to init\n");
+		pr_err("rxe: unable to init ipv4 tunnel\n");
 		rxe_cache_exit();
-		return err;
+		goto exit;
+	}
+
+	err = rxe_net_ipv6_init();
+	if (err) {
+		pr_err("rxe: unable to init ipv6 tunnel\n");
+		rxe_cache_exit();
+		goto exit;
 	}
+
+	err = register_netdevice_notifier(&rxe_net_notifier);
+	if (err) {
+		pr_err("rxe: Failed to rigister netdev notifier\n");
+		goto exit;
+	}
+
 	pr_info("rxe: loaded\n");
 
 	return 0;
+
+exit:
+	rxe_release_udp_tunnel(recv_sockets.sk4);
+	rxe_release_udp_tunnel(recv_sockets.sk6);
+	return err;
 }
 
 static void __exit rxe_module_exit(void)
diff --git a/drivers/infiniband/sw/rxe/rxe_comp.c b/drivers/infiniband/sw/rxe/rxe_comp.c
index 36f67de44095..1c59ef2c67aa 100644
--- a/drivers/infiniband/sw/rxe/rxe_comp.c
+++ b/drivers/infiniband/sw/rxe/rxe_comp.c
@@ -689,7 +689,14 @@ int rxe_completer(void *arg)
 					qp->req.need_retry = 1;
 					rxe_run_task(&qp->req.task, 1);
 				}
+
+				if (pkt) {
+					rxe_drop_ref(pkt->qp);
+					kfree_skb(skb);
+				}
+
 				goto exit;
+
 			} else {
 				wqe->status = IB_WC_RETRY_EXC_ERR;
 				state = COMPST_ERROR;
@@ -716,6 +723,12 @@ int rxe_completer(void *arg)
 		case COMPST_ERROR:
 			do_complete(qp, wqe);
 			rxe_qp_error(qp);
+
+			if (pkt) {
+				rxe_drop_ref(pkt->qp);
+				kfree_skb(skb);
+			}
+
 			goto exit;
 		}
 	}
diff --git a/drivers/infiniband/sw/rxe/rxe_net.c b/drivers/infiniband/sw/rxe/rxe_net.c
index 0b8d2ea8b41d..eedf2f1cafdf 100644
--- a/drivers/infiniband/sw/rxe/rxe_net.c
+++ b/drivers/infiniband/sw/rxe/rxe_net.c
@@ -275,9 +275,10 @@ static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
 	return sock;
 }
 
-static void rxe_release_udp_tunnel(struct socket *sk)
+void rxe_release_udp_tunnel(struct socket *sk)
 {
-	udp_tunnel_sock_release(sk);
+	if (sk)
+		udp_tunnel_sock_release(sk);
 }
 
 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
@@ -658,51 +659,45 @@ out:
 	return NOTIFY_OK;
 }
 
-static struct notifier_block rxe_net_notifier = {
+struct notifier_block rxe_net_notifier = {
 	.notifier_call = rxe_notify,
 };
 
-int rxe_net_init(void)
+int rxe_net_ipv4_init(void)
 {
-	int err;
-
 	spin_lock_init(&dev_list_lock);
 
-	recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
-			htons(ROCE_V2_UDP_DPORT), true);
-	if (IS_ERR(recv_sockets.sk6)) {
-		recv_sockets.sk6 = NULL;
-		pr_err("rxe: Failed to create IPv6 UDP tunnel\n");
-		return -1;
-	}
-
 	recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
-			htons(ROCE_V2_UDP_DPORT), false);
+				htons(ROCE_V2_UDP_DPORT), false);
 	if (IS_ERR(recv_sockets.sk4)) {
-		rxe_release_udp_tunnel(recv_sockets.sk6);
 		recv_sockets.sk4 = NULL;
-		recv_sockets.sk6 = NULL;
 		pr_err("rxe: Failed to create IPv4 UDP tunnel\n");
 		return -1;
 	}
 
-	err = register_netdevice_notifier(&rxe_net_notifier);
-	if (err) {
-		rxe_release_udp_tunnel(recv_sockets.sk6);
-		rxe_release_udp_tunnel(recv_sockets.sk4);
-		pr_err("rxe: Failed to rigister netdev notifier\n");
-	}
-
-	return err;
+	return 0;
 }
 
-void rxe_net_exit(void)
+int rxe_net_ipv6_init(void)
 {
-	if (recv_sockets.sk6)
-		rxe_release_udp_tunnel(recv_sockets.sk6);
+#if IS_ENABLED(CONFIG_IPV6)
 
-	if (recv_sockets.sk4)
-		rxe_release_udp_tunnel(recv_sockets.sk4);
+	spin_lock_init(&dev_list_lock);
 
+	recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
+						htons(ROCE_V2_UDP_DPORT), true);
+	if (IS_ERR(recv_sockets.sk6)) {
+		recv_sockets.sk6 = NULL;
+		pr_err("rxe: Failed to create IPv6 UDP tunnel\n");
+		return -1;
+	}
+#endif
+	return 0;
+}
+
+void rxe_net_exit(void)
+{
+	rxe_release_udp_tunnel(recv_sockets.sk6);
+	rxe_release_udp_tunnel(recv_sockets.sk4);
 	unregister_netdevice_notifier(&rxe_net_notifier);
 }
diff --git a/drivers/infiniband/sw/rxe/rxe_net.h b/drivers/infiniband/sw/rxe/rxe_net.h
index 7b06f76d16cc..0daf7f09e5b5 100644
--- a/drivers/infiniband/sw/rxe/rxe_net.h
+++ b/drivers/infiniband/sw/rxe/rxe_net.h
@@ -44,10 +44,13 @@ struct rxe_recv_sockets {
 };
 
 extern struct rxe_recv_sockets recv_sockets;
+extern struct notifier_block rxe_net_notifier;
+void rxe_release_udp_tunnel(struct socket *sk);
 
 struct rxe_dev *rxe_net_add(struct net_device *ndev);
 
-int rxe_net_init(void);
+int rxe_net_ipv4_init(void);
+int rxe_net_ipv6_init(void);
 void rxe_net_exit(void);
 
 #endif /* RXE_NET_H */
diff --git a/drivers/infiniband/sw/rxe/rxe_recv.c b/drivers/infiniband/sw/rxe/rxe_recv.c
index 3d464c23e08b..144d2f129fcd 100644
--- a/drivers/infiniband/sw/rxe/rxe_recv.c
+++ b/drivers/infiniband/sw/rxe/rxe_recv.c
@@ -312,7 +312,7 @@ static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
 		 * make a copy of the skb to post to the next qp
 		 */
 		skb_copy = (mce->qp_list.next != &mcg->qp_list) ?
-				skb_clone(skb, GFP_KERNEL) : NULL;
+				skb_clone(skb, GFP_ATOMIC) : NULL;
 
 		pkt->qp = qp;
 		rxe_add_ref(qp);
diff --git a/drivers/infiniband/sw/rxe/rxe_req.c b/drivers/infiniband/sw/rxe/rxe_req.c
index 33b2d9d77021..13a848a518e8 100644
--- a/drivers/infiniband/sw/rxe/rxe_req.c
+++ b/drivers/infiniband/sw/rxe/rxe_req.c
@@ -511,24 +511,21 @@ static int fill_packet(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
 }
 
 static void update_wqe_state(struct rxe_qp *qp,
-			     struct rxe_send_wqe *wqe,
-			     struct rxe_pkt_info *pkt,
-			     enum wqe_state *prev_state)
+		struct rxe_send_wqe *wqe,
+		struct rxe_pkt_info *pkt)
 {
-	enum wqe_state prev_state_ = wqe->state;
-
 	if (pkt->mask & RXE_END_MASK) {
 		if (qp_type(qp) == IB_QPT_RC)
 			wqe->state = wqe_state_pending;
 	} else {
 		wqe->state = wqe_state_processing;
 	}
-
-	*prev_state = prev_state_;
 }
 
-static void update_state(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
-			 struct rxe_pkt_info *pkt, int payload)
+static void update_wqe_psn(struct rxe_qp *qp,
+			   struct rxe_send_wqe *wqe,
+			   struct rxe_pkt_info *pkt,
+			   int payload)
 {
 	/* number of packets left to send including current one */
 	int num_pkt = (wqe->dma.resid + payload + qp->mtu - 1) / qp->mtu;
@@ -546,9 +543,34 @@ static void update_state(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
 		qp->req.psn = (wqe->first_psn + num_pkt) & BTH_PSN_MASK;
 	else
 		qp->req.psn = (qp->req.psn + 1) & BTH_PSN_MASK;
+}
 
-	qp->req.opcode = pkt->opcode;
+static void save_state(struct rxe_send_wqe *wqe,
+		       struct rxe_qp *qp,
+		       struct rxe_send_wqe *rollback_wqe,
+		       struct rxe_qp *rollback_qp)
+{
+	rollback_wqe->state     = wqe->state;
+	rollback_wqe->first_psn = wqe->first_psn;
+	rollback_wqe->last_psn  = wqe->last_psn;
+	rollback_qp->req.psn    = qp->req.psn;
+}
 
+static void rollback_state(struct rxe_send_wqe *wqe,
+			   struct rxe_qp *qp,
+			   struct rxe_send_wqe *rollback_wqe,
+			   struct rxe_qp *rollback_qp)
+{
+	wqe->state     = rollback_wqe->state;
+	wqe->first_psn = rollback_wqe->first_psn;
+	wqe->last_psn  = rollback_wqe->last_psn;
+	qp->req.psn    = rollback_qp->req.psn;
+}
+
+static void update_state(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
+			 struct rxe_pkt_info *pkt, int payload)
+{
+	qp->req.opcode = pkt->opcode;
 
 	if (pkt->mask & RXE_END_MASK)
 		qp->req.wqe_index = next_index(qp->sq.queue, qp->req.wqe_index);
@@ -571,7 +593,8 @@ int rxe_requester(void *arg)
 	int mtu;
 	int opcode;
 	int ret;
-	enum wqe_state prev_state;
+	struct rxe_qp rollback_qp;
+	struct rxe_send_wqe rollback_wqe;
 
 next_wqe:
 	if (unlikely(!qp->valid || qp->req.state == QP_STATE_ERROR))
@@ -688,13 +711,21 @@ next_wqe:
 		goto err;
 	}
 
-	update_wqe_state(qp, wqe, &pkt, &prev_state);
+	/*
+	 * To prevent a race on wqe access between requester and completer,
+	 * wqe members state and psn need to be set before calling
+	 * rxe_xmit_packet().
+	 * Otherwise, completer might initiate an unjustified retry flow.
+	 */
+	save_state(wqe, qp, &rollback_wqe, &rollback_qp);
+	update_wqe_state(qp, wqe, &pkt);
+	update_wqe_psn(qp, wqe, &pkt, payload);
 	ret = rxe_xmit_packet(to_rdev(qp->ibqp.device), qp, &pkt, skb);
 	if (ret) {
 		qp->need_req_skb = 1;
 		kfree_skb(skb);
 
-		wqe->state = prev_state;
+		rollback_state(wqe, qp, &rollback_wqe, &rollback_qp);
 
 		if (ret == -EAGAIN) {
 			rxe_run_task(&qp->req.task, 1);
diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c
index ebb03b46e2ad..3e0f0f2baace 100644
--- a/drivers/infiniband/sw/rxe/rxe_resp.c
+++ b/drivers/infiniband/sw/rxe/rxe_resp.c
@@ -972,11 +972,13 @@ static int send_atomic_ack(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
 	free_rd_atomic_resource(qp, res);
 	rxe_advance_resp_resource(qp);
 
+	memcpy(SKB_TO_PKT(skb), &ack_pkt, sizeof(skb->cb));
+
 	res->type = RXE_ATOMIC_MASK;
 	res->atomic.skb = skb;
-	res->first_psn = qp->resp.psn;
-	res->last_psn = qp->resp.psn;
-	res->cur_psn = qp->resp.psn;
+	res->first_psn = ack_pkt.psn;
+	res->last_psn  = ack_pkt.psn;
+	res->cur_psn   = ack_pkt.psn;
 
 	rc = rxe_xmit_packet(rxe, qp, &ack_pkt, skb_copy);
 	if (rc) {
@@ -1116,8 +1118,7 @@ static enum resp_states duplicate_request(struct rxe_qp *qp,
 				rc = RESPST_CLEANUP;
 				goto out;
 			}
-			bth_set_psn(SKB_TO_PKT(skb_copy),
-				    qp->resp.psn - 1);
+
 			/* Resend the result. */
 			rc = rxe_xmit_packet(to_rdev(qp->ibqp.device), qp,
 					     pkt, skb_copy);