First v5.17 rc request

Some medium sized bugs in the various drivers. A couple are more recent
 regressions:
 
 - Fix two panics in hfi1 and two allocation problems
 
 - Send the IGMP to the correct address in cma
 
 - Squash a syzkaller bug related to races reading the multicast list
 
 - Memory leak in siw and cm
 
 - Fix a corner case spec compliance for HFI/QIB
 
 - Correct the implementation of fences in siw
 
 - Error unwind bug in mlx4
 -----BEGIN PGP SIGNATURE-----
 
 iQIzBAABCgAdFiEEfB7FMLh+8QxL+6i3OG33FX4gmxoFAmH9eHIACgkQOG33FX4g
 mxr6dBAAglSsJbV3Q0/x/idhj+5gBRKYg00DVj9hnwTme53dJioMDKO1r4x+3ium
 +GIvH7nOntoMXmgWBCZaEEvxZpQZ6QpuF3qgwqFDLn3eUSif66yiT67N8GthD277
 0RR/hHplX0s6meZn5e7bNvNcvh8Nq/hhpTgOPCLAfxWBHWUtVanp0L67UecYthdw
 tlRB/Cs31nB28xUY6fFyYA/pG1wBMNDeywOSve4iQvl3w46HCqveeUyNrIeULHau
 YzZM70yCzGkS5RQ+uMd7OQ7lutzIUlenx4PboP3A3ld5JqoAFdDgKqG0BovWrj3f
 aQRtoIf1F0y+LSDxZKwQ/CauqtBuS7HeJgsA6lbMNSK2O0ZkLDIPAexRhCZhnHT+
 Eabnc4Vm5T3jMqFbEJwpLUR+F1/VxmM3kNcfl+sBwBjK3JzjEjVNenRUaFSvMra0
 zQlqlqvvJKBQxfWaBkeE1C72IJySn73JDrBPGYeVhw86QvDAWWfOPbsDFknx49T6
 4R4S4aK6JZ4pHRuuKE8Xx87RnVAcwqGAmkrJzRJFK59Brlxc5vexo45Y9NEfpPzk
 3sM/RgQxJXY2IoTZNKdbCTcwxD71pgz8CYMQ0HZwoylC4Wp7LMXQRI2I5LBpYTmK
 pD1l5FVTYbcnUq2jqGjeZnwho2RwowSGSgn1A+TUIEBqQUXta78=
 =RrpA
 -----END PGP SIGNATURE-----

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "Some medium sized bugs in the various drivers. A couple are more
  recent regressions:

   - Fix two panics in hfi1 and two allocation problems

   - Send the IGMP to the correct address in cma

   - Squash a syzkaller bug related to races reading the multicast list

   - Memory leak in siw and cm

   - Fix a corner case spec compliance for HFI/QIB

   - Correct the implementation of fences in siw

   - Error unwind bug in mlx4"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/mlx4: Don't continue event handler after memory allocation failure
  RDMA/siw: Fix broken RDMA Read Fence/Resume logic.
  IB/rdmavt: Validate remote_addr during loopback atomic tests
  IB/cm: Release previously acquired reference counter in the cm_id_priv
  RDMA/siw: Fix refcounting leak in siw_create_qp()
  RDMA/ucma: Protect mc during concurrent multicast leaves
  RDMA/cma: Use correct address when leaving multicast group
  IB/hfi1: Fix tstats alloc and dealloc
  IB/hfi1: Fix AIP early init panic
  IB/hfi1: Fix alloc failure with larger txqueuelen
  IB/hfi1: Fix panic with larger ipoib send_queue_size

drivers/infiniband/core/cm.c

@@ -3322,7 +3322,7 @@ static int cm_lap_handler(struct cm_work *work)
 	ret = cm_init_av_by_path(param->alternate_path, NULL, &alt_av);
 	if (ret) {
 		rdma_destroy_ah_attr(&ah_attr);
-		return -EINVAL;
+		goto deref;
 	}
 
 	spin_lock_irq(&cm_id_priv->lock);

drivers/infiniband/core/cma.c

@@ -67,8 +67,8 @@ static const char * const cma_events[] = {
 	[RDMA_CM_EVENT_TIMEWAIT_EXIT]	 = "timewait exit",
 };
 
-static void cma_set_mgid(struct rdma_id_private *id_priv, struct sockaddr *addr,
-			 union ib_gid *mgid);
+static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
+			      enum ib_gid_type gid_type);
 
 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
 {
@@ -1846,17 +1846,19 @@ static void destroy_mc(struct rdma_id_private *id_priv,
 		if (dev_addr->bound_dev_if)
 			ndev = dev_get_by_index(dev_addr->net,
 						dev_addr->bound_dev_if);
-		if (ndev) {
+		if (ndev && !send_only) {
+			enum ib_gid_type gid_type;
 			union ib_gid mgid;
 
-			cma_set_mgid(id_priv, (struct sockaddr *)&mc->addr,
-				     &mgid);
-
-			if (!send_only)
-				cma_igmp_send(ndev, &mgid, false);
-
-			dev_put(ndev);
+			gid_type = id_priv->cma_dev->default_gid_type
+					   [id_priv->id.port_num -
+					    rdma_start_port(
+						    id_priv->cma_dev->device)];
+			cma_iboe_set_mgid((struct sockaddr *)&mc->addr, &mgid,
+					  gid_type);
+			cma_igmp_send(ndev, &mgid, false);
 		}
+		dev_put(ndev);
 
 		cancel_work_sync(&mc->iboe_join.work);
 	}

drivers/infiniband/core/ucma.c

@@ -95,6 +95,7 @@ struct ucma_context {
 	u64			uid;
 
 	struct list_head	list;
+	struct list_head	mc_list;
 	struct work_struct	close_work;
 };
 
@@ -105,6 +106,7 @@ struct ucma_multicast {
 
 	u64			uid;
 	u8			join_state;
+	struct list_head	list;
 	struct sockaddr_storage	addr;
 };
 
@@ -198,6 +200,7 @@ static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)
 
 	INIT_WORK(&ctx->close_work, ucma_close_id);
 	init_completion(&ctx->comp);
+	INIT_LIST_HEAD(&ctx->mc_list);
 	/* So list_del() will work if we don't do ucma_finish_ctx() */
 	INIT_LIST_HEAD(&ctx->list);
 	ctx->file = file;
@@ -484,19 +487,19 @@ err1:
 
 static void ucma_cleanup_multicast(struct ucma_context *ctx)
 {
-	struct ucma_multicast *mc;
-	unsigned long index;
+	struct ucma_multicast *mc, *tmp;
 
-	xa_for_each(&multicast_table, index, mc) {
-		if (mc->ctx != ctx)
-			continue;
+	xa_lock(&multicast_table);
+	list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) {
+		list_del(&mc->list);
 		/*
 		 * At this point mc->ctx->ref is 0 so the mc cannot leave the
 		 * lock on the reader and this is enough serialization
 		 */
-		xa_erase(&multicast_table, index);
+		__xa_erase(&multicast_table, mc->id);
 		kfree(mc);
 	}
+	xa_unlock(&multicast_table);
 }
 
 static void ucma_cleanup_mc_events(struct ucma_multicast *mc)
@@ -1469,12 +1472,16 @@ static ssize_t ucma_process_join(struct ucma_file *file,
 	mc->uid = cmd->uid;
 	memcpy(&mc->addr, addr, cmd->addr_size);
 
-	if (xa_alloc(&multicast_table, &mc->id, NULL, xa_limit_32b,
+	xa_lock(&multicast_table);
+	if (__xa_alloc(&multicast_table, &mc->id, NULL, xa_limit_32b,
 		     GFP_KERNEL)) {
 		ret = -ENOMEM;
 		goto err_free_mc;
 	}
 
+	list_add_tail(&mc->list, &ctx->mc_list);
+	xa_unlock(&multicast_table);
+
 	mutex_lock(&ctx->mutex);
 	ret = rdma_join_multicast(ctx->cm_id, (struct sockaddr *)&mc->addr,
 				  join_state, mc);
@@ -1500,8 +1507,11 @@ err_leave_multicast:
 	mutex_unlock(&ctx->mutex);
 	ucma_cleanup_mc_events(mc);
 err_xa_erase:
-	xa_erase(&multicast_table, mc->id);
+	xa_lock(&multicast_table);
+	list_del(&mc->list);
+	__xa_erase(&multicast_table, mc->id);
 err_free_mc:
+	xa_unlock(&multicast_table);
 	kfree(mc);
 err_put_ctx:
 	ucma_put_ctx(ctx);
@@ -1569,15 +1579,17 @@ static ssize_t ucma_leave_multicast(struct ucma_file *file,
 		mc = ERR_PTR(-EINVAL);
 	else if (!refcount_inc_not_zero(&mc->ctx->ref))
 		mc = ERR_PTR(-ENXIO);
-	else
-		__xa_erase(&multicast_table, mc->id);
-	xa_unlock(&multicast_table);
 
 	if (IS_ERR(mc)) {
+		xa_unlock(&multicast_table);
 		ret = PTR_ERR(mc);
 		goto out;
 	}
 
+	list_del(&mc->list);
+	__xa_erase(&multicast_table, mc->id);
+	xa_unlock(&multicast_table);
+
 	mutex_lock(&mc->ctx->mutex);
 	rdma_leave_multicast(mc->ctx->cm_id, (struct sockaddr *) &mc->addr);
 	mutex_unlock(&mc->ctx->mutex);

drivers/infiniband/hw/hfi1/ipoib.h

@@ -55,7 +55,7 @@ union hfi1_ipoib_flow {
  */
 struct ipoib_txreq {
 	struct sdma_txreq           txreq;
-	struct hfi1_sdma_header     sdma_hdr;
+	struct hfi1_sdma_header     *sdma_hdr;
 	int                         sdma_status;
 	int                         complete;
 	struct hfi1_ipoib_dev_priv *priv;

drivers/infiniband/hw/hfi1/ipoib_main.c

@@ -22,26 +22,35 @@ static int hfi1_ipoib_dev_init(struct net_device *dev)
 	int ret;
 
 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
+	if (!dev->tstats)
+		return -ENOMEM;
 
 	ret = priv->netdev_ops->ndo_init(dev);
 	if (ret)
-		return ret;
+		goto out_ret;
 
 	ret = hfi1_netdev_add_data(priv->dd,
 				   qpn_from_mac(priv->netdev->dev_addr),
 				   dev);
 	if (ret < 0) {
 		priv->netdev_ops->ndo_uninit(dev);
-		return ret;
+		goto out_ret;
 	}
 
 	return 0;
+out_ret:
+	free_percpu(dev->tstats);
+	dev->tstats = NULL;
+	return ret;
 }
 
 static void hfi1_ipoib_dev_uninit(struct net_device *dev)
 {
 	struct hfi1_ipoib_dev_priv *priv = hfi1_ipoib_priv(dev);
 
+	free_percpu(dev->tstats);
+	dev->tstats = NULL;
+
 	hfi1_netdev_remove_data(priv->dd, qpn_from_mac(priv->netdev->dev_addr));
 
 	priv->netdev_ops->ndo_uninit(dev);
@@ -166,12 +175,7 @@ static void hfi1_ipoib_netdev_dtor(struct net_device *dev)
 	hfi1_ipoib_rxq_deinit(priv->netdev);
 
 	free_percpu(dev->tstats);
-}
-
-static void hfi1_ipoib_free_rdma_netdev(struct net_device *dev)
-{
-	hfi1_ipoib_netdev_dtor(dev);
-	free_netdev(dev);
+	dev->tstats = NULL;
 }
 
 static void hfi1_ipoib_set_id(struct net_device *dev, int id)
@@ -211,24 +215,23 @@ static int hfi1_ipoib_setup_rn(struct ib_device *device,
 	priv->port_num = port_num;
 	priv->netdev_ops = netdev->netdev_ops;
 
-	netdev->netdev_ops = &hfi1_ipoib_netdev_ops;
-
 	ib_query_pkey(device, port_num, priv->pkey_index, &priv->pkey);
 
 	rc = hfi1_ipoib_txreq_init(priv);
 	if (rc) {
 		dd_dev_err(dd, "IPoIB netdev TX init - failed(%d)\n", rc);
-		hfi1_ipoib_free_rdma_netdev(netdev);
 		return rc;
 	}
 
 	rc = hfi1_ipoib_rxq_init(netdev);
 	if (rc) {
 		dd_dev_err(dd, "IPoIB netdev RX init - failed(%d)\n", rc);
-		hfi1_ipoib_free_rdma_netdev(netdev);
+		hfi1_ipoib_txreq_deinit(priv);
 		return rc;
 	}
 
+	netdev->netdev_ops = &hfi1_ipoib_netdev_ops;
+
 	netdev->priv_destructor = hfi1_ipoib_netdev_dtor;
 	netdev->needs_free_netdev = true;
 

drivers/infiniband/hw/hfi1/ipoib_tx.c

@@ -122,7 +122,7 @@ static void hfi1_ipoib_free_tx(struct ipoib_txreq *tx, int budget)
 		dd_dev_warn(priv->dd,
 			    "%s: Status = 0x%x pbc 0x%llx txq = %d sde = %d\n",
 			    __func__, tx->sdma_status,
-			    le64_to_cpu(tx->sdma_hdr.pbc), tx->txq->q_idx,
+			    le64_to_cpu(tx->sdma_hdr->pbc), tx->txq->q_idx,
 			    tx->txq->sde->this_idx);
 	}
 
@@ -231,7 +231,7 @@ static int hfi1_ipoib_build_tx_desc(struct ipoib_txreq *tx,
 {
 	struct hfi1_devdata *dd = txp->dd;
 	struct sdma_txreq *txreq = &tx->txreq;
-	struct hfi1_sdma_header *sdma_hdr = &tx->sdma_hdr;
+	struct hfi1_sdma_header *sdma_hdr = tx->sdma_hdr;
 	u16 pkt_bytes =
 		sizeof(sdma_hdr->pbc) + (txp->hdr_dwords << 2) + tx->skb->len;
 	int ret;
@@ -256,7 +256,7 @@ static void hfi1_ipoib_build_ib_tx_headers(struct ipoib_txreq *tx,
 					   struct ipoib_txparms *txp)
 {
 	struct hfi1_ipoib_dev_priv *priv = tx->txq->priv;
-	struct hfi1_sdma_header *sdma_hdr = &tx->sdma_hdr;
+	struct hfi1_sdma_header *sdma_hdr = tx->sdma_hdr;
 	struct sk_buff *skb = tx->skb;
 	struct hfi1_pportdata *ppd = ppd_from_ibp(txp->ibp);
 	struct rdma_ah_attr *ah_attr = txp->ah_attr;
@@ -483,7 +483,7 @@ static int hfi1_ipoib_send_dma_single(struct net_device *dev,
 	if (likely(!ret)) {
 tx_ok:
 		trace_sdma_output_ibhdr(txq->priv->dd,
-					&tx->sdma_hdr.hdr,
+					&tx->sdma_hdr->hdr,
 					ib_is_sc5(txp->flow.sc5));
 		hfi1_ipoib_check_queue_depth(txq);
 		return NETDEV_TX_OK;
@@ -547,7 +547,7 @@ static int hfi1_ipoib_send_dma_list(struct net_device *dev,
 	hfi1_ipoib_check_queue_depth(txq);
 
 	trace_sdma_output_ibhdr(txq->priv->dd,
-				&tx->sdma_hdr.hdr,
+				&tx->sdma_hdr->hdr,
 				ib_is_sc5(txp->flow.sc5));
 
 	if (!netdev_xmit_more())
@@ -683,7 +683,8 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv)
 {
 	struct net_device *dev = priv->netdev;
 	u32 tx_ring_size, tx_item_size;
-	int i;
+	struct hfi1_ipoib_circ_buf *tx_ring;
+	int i, j;
 
 	/*
 	 * Ring holds 1 less than tx_ring_size
@@ -701,7 +702,9 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv)
 
 	for (i = 0; i < dev->num_tx_queues; i++) {
 		struct hfi1_ipoib_txq *txq = &priv->txqs[i];
+		struct ipoib_txreq *tx;
 
+		tx_ring = &txq->tx_ring;
 		iowait_init(&txq->wait,
 			    0,
 			    hfi1_ipoib_flush_txq,
@@ -725,14 +728,19 @@ int hfi1_ipoib_txreq_init(struct hfi1_ipoib_dev_priv *priv)
 					     priv->dd->node);
 
 		txq->tx_ring.items =
-			kcalloc_node(tx_ring_size, tx_item_size,
-				     GFP_KERNEL, priv->dd->node);
+			kvzalloc_node(array_size(tx_ring_size, tx_item_size),
+				      GFP_KERNEL, priv->dd->node);
 		if (!txq->tx_ring.items)
 			goto free_txqs;
 
 		txq->tx_ring.max_items = tx_ring_size;
-		txq->tx_ring.shift = ilog2(tx_ring_size);
+		txq->tx_ring.shift = ilog2(tx_item_size);
 		txq->tx_ring.avail = hfi1_ipoib_ring_hwat(txq);
+		tx_ring = &txq->tx_ring;
+		for (j = 0; j < tx_ring_size; j++)
+			hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr =
+				kzalloc_node(sizeof(*tx->sdma_hdr),
+					     GFP_KERNEL, priv->dd->node);
 
 		netif_tx_napi_add(dev, &txq->napi,
 				  hfi1_ipoib_poll_tx_ring,
@@ -746,7 +754,10 @@ free_txqs:
 		struct hfi1_ipoib_txq *txq = &priv->txqs[i];
 
 		netif_napi_del(&txq->napi);
-		kfree(txq->tx_ring.items);
+		tx_ring = &txq->tx_ring;
+		for (j = 0; j < tx_ring_size; j++)
+			kfree(hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr);
+		kvfree(tx_ring->items);
 	}
 
 	kfree(priv->txqs);
@@ -780,17 +791,20 @@ static void hfi1_ipoib_drain_tx_list(struct hfi1_ipoib_txq *txq)
 
 void hfi1_ipoib_txreq_deinit(struct hfi1_ipoib_dev_priv *priv)
 {
-	int i;
+	int i, j;
 
 	for (i = 0; i < priv->netdev->num_tx_queues; i++) {
 		struct hfi1_ipoib_txq *txq = &priv->txqs[i];
+		struct hfi1_ipoib_circ_buf *tx_ring = &txq->tx_ring;
 
 		iowait_cancel_work(&txq->wait);
 		iowait_sdma_drain(&txq->wait);
 		hfi1_ipoib_drain_tx_list(txq);
 		netif_napi_del(&txq->napi);
 		hfi1_ipoib_drain_tx_ring(txq);
-		kfree(txq->tx_ring.items);
+		for (j = 0; j < tx_ring->max_items; j++)
+			kfree(hfi1_txreq_from_idx(tx_ring, j)->sdma_hdr);
+		kvfree(tx_ring->items);
 	}
 
 	kfree(priv->txqs);

drivers/infiniband/hw/mlx4/main.c

@@ -3237,7 +3237,7 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
 		if (!ew)
-			break;
+			return;
 
 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);

drivers/infiniband/sw/rdmavt/qp.c

@@ -3073,6 +3073,8 @@ do_write:
 	case IB_WR_ATOMIC_FETCH_AND_ADD:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
 			goto inv_err;
+		if (unlikely(wqe->atomic_wr.remote_addr & (sizeof(u64) - 1)))
+			goto inv_err;
 		if (unlikely(!rvt_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
 					  wqe->atomic_wr.remote_addr,
 					  wqe->atomic_wr.rkey,

drivers/infiniband/sw/siw/siw.h

@@ -644,14 +644,9 @@ static inline struct siw_sqe *orq_get_current(struct siw_qp *qp)
 	return &qp->orq[qp->orq_get % qp->attrs.orq_size];
 }
 
-static inline struct siw_sqe *orq_get_tail(struct siw_qp *qp)
-{
-	return &qp->orq[qp->orq_put % qp->attrs.orq_size];
-}
-
 static inline struct siw_sqe *orq_get_free(struct siw_qp *qp)
 {
-	struct siw_sqe *orq_e = orq_get_tail(qp);
+	struct siw_sqe *orq_e = &qp->orq[qp->orq_put % qp->attrs.orq_size];
 
 	if (READ_ONCE(orq_e->flags) == 0)
 		return orq_e;

drivers/infiniband/sw/siw/siw_qp_rx.c

@@ -1153,11 +1153,12 @@ static int siw_check_tx_fence(struct siw_qp *qp)
 
 	spin_lock_irqsave(&qp->orq_lock, flags);
 
-	rreq = orq_get_current(qp);
-
 	/* free current orq entry */
+	rreq = orq_get_current(qp);
 	WRITE_ONCE(rreq->flags, 0);
 
+	qp->orq_get++;
+
 	if (qp->tx_ctx.orq_fence) {
 		if (unlikely(tx_waiting->wr_status != SIW_WR_QUEUED)) {
 			pr_warn("siw: [QP %u]: fence resume: bad status %d\n",
@@ -1165,10 +1166,12 @@ static int siw_check_tx_fence(struct siw_qp *qp)
 			rv = -EPROTO;
 			goto out;
 		}
-		/* resume SQ processing */
+		/* resume SQ processing, if possible */
 		if (tx_waiting->sqe.opcode == SIW_OP_READ ||
 		    tx_waiting->sqe.opcode == SIW_OP_READ_LOCAL_INV) {
-			rreq = orq_get_tail(qp);
+
+			/* SQ processing was stopped because of a full ORQ */
+			rreq = orq_get_free(qp);
 			if (unlikely(!rreq)) {
 				pr_warn("siw: [QP %u]: no ORQE\n", qp_id(qp));
 				rv = -EPROTO;
@@ -1181,15 +1184,14 @@ static int siw_check_tx_fence(struct siw_qp *qp)
 			resume_tx = 1;
 
 		} else if (siw_orq_empty(qp)) {
+			/*
+			 * SQ processing was stopped by fenced work request.
+			 * Resume since all previous Read's are now completed.
+			 */
 			qp->tx_ctx.orq_fence = 0;
 			resume_tx = 1;
-		} else {
-			pr_warn("siw: [QP %u]: fence resume: orq idx: %d:%d\n",
-				qp_id(qp), qp->orq_get, qp->orq_put);
-			rv = -EPROTO;
 		}
 	}
-	qp->orq_get++;
 out:
 	spin_unlock_irqrestore(&qp->orq_lock, flags);
 

drivers/infiniband/sw/siw/siw_verbs.c

@@ -313,7 +313,8 @@ int siw_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attrs,
 
 	if (atomic_inc_return(&sdev->num_qp) > SIW_MAX_QP) {
 		siw_dbg(base_dev, "too many QP's\n");
-		return -ENOMEM;
+		rv = -ENOMEM;
+		goto err_atomic;
 	}
 	if (attrs->qp_type != IB_QPT_RC) {
 		siw_dbg(base_dev, "only RC QP's supported\n");