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

Pull rdma fixes from Doug Ledford:
"This is the second batch of queued up rdma patches for this rc cycle.

There isn't anything really major in here. It's passed 0day,
linux-next, and local testing across a wide variety of hardware.
There are still a few known issues to be tracked down, but this should
amount to the vast majority of the rdma RC fixes.

Round two of 4.7 rc fixes:

- A couple minor fixes to the rdma core
- Multiple minor fixes to hfi1
- Multiple minor fixes to mlx4/mlx4
- A few minor fixes to i40iw"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (31 commits)
IB/srpt: Reduce QP buffer size
i40iw: Enable level-1 PBL for fast memory registration
i40iw: Return correct max_fast_reg_page_list_len
i40iw: Correct status check on i40iw_get_pble
i40iw: Correct CQ arming
IB/rdmavt: Correct qp_priv_alloc() return value test
IB/hfi1: Don't zero out qp->s_ack_queue in rvt_reset_qp
IB/hfi1: Fix deadlock with txreq allocation slow path
IB/mlx4: Prevent cross page boundary allocation
IB/mlx4: Fix memory leak if QP creation failed
IB/mlx4: Verify port number in flow steering create flow
IB/mlx4: Fix error flow when sending mads under SRIOV
IB/mlx4: Fix the SQ size of an RC QP
IB/mlx5: Fix wrong naming of port_rcv_data counter
IB/mlx5: Fix post send fence logic
IB/uverbs: Initialize ib_qp_init_attr with zeros
IB/core: Fix false search of the IB_SA_WELL_KNOWN_GUID
IB/core: Fix RoCE v1 multicast join logic issue
IB/core: Fix no default GIDs when netdevice reregisters
IB/hfi1: Send a pkey change event on driver pkey update
...

drivers/infiniband/core/cache.c

···
 
 	for (ix = 0; ix < table->sz; ix++)
 		if (table->data_vec[ix].attr.ndev == ndev)
-			if (!del_gid(ib_dev, port, table, ix, false))
+			if (!del_gid(ib_dev, port, table, ix,
+				     !!(table->data_vec[ix].props &
+					GID_TABLE_ENTRY_DEFAULT)))
 				deleted = true;
 
 	write_unlock_irq(&table->rwlock);

drivers/infiniband/core/cma.c

···
 		complete(&id_priv->comp);
 }
 
-static int cma_disable_callback(struct rdma_id_private *id_priv,
-				enum rdma_cm_state state)
-{
-	mutex_lock(&id_priv->handler_mutex);
-	if (id_priv->state != state) {
-		mutex_unlock(&id_priv->handler_mutex);
-		return -EINVAL;
-	}
-	return 0;
-}
-
 struct rdma_cm_id *rdma_create_id(struct net *net,
 				  rdma_cm_event_handler event_handler,
 				  void *context, enum rdma_port_space ps,
···
 	struct rdma_cm_event event;
 	int ret = 0;
 
+	mutex_lock(&id_priv->handler_mutex);
 	if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
-		cma_disable_callback(id_priv, RDMA_CM_CONNECT)) ||
+	     id_priv->state != RDMA_CM_CONNECT) ||
 	    (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
-		cma_disable_callback(id_priv, RDMA_CM_DISCONNECT)))
-		return 0;
+	     id_priv->state != RDMA_CM_DISCONNECT))
+		goto out;
 
 	memset(&event, 0, sizeof event);
 	switch (ib_event->event) {
···
 
 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
 {
-	struct rdma_id_private *listen_id, *conn_id;
+	struct rdma_id_private *listen_id, *conn_id = NULL;
 	struct rdma_cm_event event;
 	struct net_device *net_dev;
 	int offset, ret;
···
 		goto net_dev_put;
 	}
 
-	if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) {
+	mutex_lock(&listen_id->handler_mutex);
+	if (listen_id->state != RDMA_CM_LISTEN) {
 		ret = -ECONNABORTED;
-		goto net_dev_put;
+		goto err1;
 	}
 
 	memset(&event, 0, sizeof event);
···
 	struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
 	struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
 
-	if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
-		return 0;
+	mutex_lock(&id_priv->handler_mutex);
+	if (id_priv->state != RDMA_CM_CONNECT)
+		goto out;
 
 	memset(&event, 0, sizeof event);
 	switch (iw_event->event) {
···
 		return ret;
 	}
 
+out:
 	mutex_unlock(&id_priv->handler_mutex);
 	return ret;
 }
···
 	struct rdma_cm_id *new_cm_id;
 	struct rdma_id_private *listen_id, *conn_id;
 	struct rdma_cm_event event;
-	int ret;
+	int ret = -ECONNABORTED;
 	struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
 	struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
 
 	listen_id = cm_id->context;
-	if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
-		return -ECONNABORTED;
+
+	mutex_lock(&listen_id->handler_mutex);
+	if (listen_id->state != RDMA_CM_LISTEN)
+		goto out;
 
 	/* Create a new RDMA id for the new IW CM ID */
 	new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
···
 	struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
 	int ret = 0;
 
-	if (cma_disable_callback(id_priv, RDMA_CM_CONNECT))
-		return 0;
+	mutex_lock(&id_priv->handler_mutex);
+	if (id_priv->state != RDMA_CM_CONNECT)
+		goto out;
 
 	memset(&event, 0, sizeof event);
 	switch (ib_event->event) {
···
 	struct rdma_id_private *id_priv;
 	struct cma_multicast *mc = multicast->context;
 	struct rdma_cm_event event;
-	int ret;
+	int ret = 0;
 
 	id_priv = mc->id_priv;
-	if (cma_disable_callback(id_priv, RDMA_CM_ADDR_BOUND) &&
-	    cma_disable_callback(id_priv, RDMA_CM_ADDR_RESOLVED))
-		return 0;
+	mutex_lock(&id_priv->handler_mutex);
+	if (id_priv->state != RDMA_CM_ADDR_BOUND &&
+	    id_priv->state != RDMA_CM_ADDR_RESOLVED)
+		goto out;
 
 	if (!status)
 		status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
···
 		return 0;
 	}
 
+out:
 	mutex_unlock(&id_priv->handler_mutex);
 	return 0;
 }
···
 	gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
 		   rdma_start_port(id_priv->cma_dev->device)];
 	if (addr->sa_family == AF_INET) {
-		if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
+		if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
+			mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
 			err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
 					    true);
-		if (!err) {
-			mc->igmp_joined = true;
-			mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
+			if (!err)
+				mc->igmp_joined = true;
 		}
 	} else {
 		if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)

drivers/infiniband/core/uverbs_cmd.c

···
 	struct ib_srq			*srq = NULL;
 	struct ib_qp			*qp;
 	char				*buf;
-	struct ib_qp_init_attr		attr;
+	struct ib_qp_init_attr		attr = {};
 	struct ib_uverbs_ex_create_qp_resp resp;
 	int				ret;
 

drivers/infiniband/core/verbs.c

···
 		ah_attr->grh.dgid = sgid;
 
 		if (!rdma_cap_eth_ah(device, port_num)) {
-			ret = ib_find_cached_gid_by_port(device, &dgid,
-							 IB_GID_TYPE_IB,
-							 port_num, NULL,
-							 &gid_index);
-			if (ret)
-				return ret;
+			if (dgid.global.interface_id != cpu_to_be64(IB_SA_WELL_KNOWN_GUID)) {
+				ret = ib_find_cached_gid_by_port(device, &dgid,
+								 IB_GID_TYPE_IB,
+								 port_num, NULL,
+								 &gid_index);
+				if (ret)
+					return ret;
+			} else {
+				gid_index = 0;
+			}
 		}
 
 		ah_attr->grh.sgid_index = (u8) gid_index;

drivers/infiniband/hw/hfi1/chip.c

···
 static void dc_start(struct hfi1_devdata *);
 static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
 			   unsigned int *np);
-static void remove_full_mgmt_pkey(struct hfi1_pportdata *ppd);
+static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd);
 
 /*
  * Error interrupt table entry.  This is used as input to the interrupt
···
 	}
 
 	reset_neighbor_info(ppd);
-	if (ppd->mgmt_allowed)
-		remove_full_mgmt_pkey(ppd);
 
 	/* disable the port */
 	clear_rcvctrl(ppd->dd, RCV_CTRL_RCV_PORT_ENABLE_SMASK);
···
 			    __func__, ppd->pkeys[2], FULL_MGMT_P_KEY);
 	ppd->pkeys[2] = FULL_MGMT_P_KEY;
 	(void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
+	hfi1_event_pkey_change(ppd->dd, ppd->port);
 }
 
-static void remove_full_mgmt_pkey(struct hfi1_pportdata *ppd)
+static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd)
 {
-	ppd->pkeys[2] = 0;
-	(void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
+	if (ppd->pkeys[2] != 0) {
+		ppd->pkeys[2] = 0;
+		(void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
+		hfi1_event_pkey_change(ppd->dd, ppd->port);
+	}
 }
 
 /*
···
 		return 0;
 	}
 
+	/*
+	 * FULL_MGMT_P_KEY is cleared from the pkey table, so that the
+	 * pkey table can be configured properly if the HFI unit is connected
+	 * to switch port with MgmtAllowed=NO
+	 */
+	clear_full_mgmt_pkey(ppd);
+
 	return set_link_state(ppd, HLS_DN_POLL);
 }
 
···
 	u64 len1 = 0, len2 = (((dd->vld[15].mtu + max_hb) >> 2)
 			      & SEND_LEN_CHECK1_LEN_VL15_MASK) <<
 		SEND_LEN_CHECK1_LEN_VL15_SHIFT;
-	int i;
+	int i, j;
 	u32 thres;
 
 	for (i = 0; i < ppd->vls_supported; i++) {
···
 			    sc_mtu_to_threshold(dd->vld[i].sc,
 						dd->vld[i].mtu,
 						dd->rcd[0]->rcvhdrqentsize));
-		sc_set_cr_threshold(dd->vld[i].sc, thres);
+		for (j = 0; j < INIT_SC_PER_VL; j++)
+			sc_set_cr_threshold(
+					pio_select_send_context_vl(dd, j, i),
+					    thres);
 	}
 	thres = min(sc_percent_to_threshold(dd->vld[15].sc, 50),
 		    sc_mtu_to_threshold(dd->vld[15].sc,

drivers/infiniband/hw/hfi1/file_ops.c

···
 
 	switch (cmd) {
 	case HFI1_IOCTL_ASSIGN_CTXT:
+		if (uctxt)
+			return -EINVAL;
+
 		if (copy_from_user(&uinfo,
 				   (struct hfi1_user_info __user *)arg,
 				   sizeof(uinfo)))

drivers/infiniband/hw/hfi1/init.c

···
 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	int ret = 0, j, pidx, initfail;
-	struct hfi1_devdata *dd = NULL;
+	struct hfi1_devdata *dd = ERR_PTR(-EINVAL);
 	struct hfi1_pportdata *ppd;
 
 	/* First, lock the non-writable module parameters */

drivers/infiniband/hw/hfi1/mad.c

···
 	memset(data, 0, size);
 }
 
+void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port)
+{
+	struct ib_event event;
+
+	event.event = IB_EVENT_PKEY_CHANGE;
+	event.device = &dd->verbs_dev.rdi.ibdev;
+	event.element.port_num = port;
+	ib_dispatch_event(&event);
+}
+
 static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len)
 {
 	struct ib_mad_send_buf *send_buf;
···
 	}
 
 	if (changed) {
-		struct ib_event event;
-
 		(void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
-
-		event.event = IB_EVENT_PKEY_CHANGE;
-		event.device = &dd->verbs_dev.rdi.ibdev;
-		event.element.port_num = port;
-		ib_dispatch_event(&event);
+		hfi1_event_pkey_change(dd, port);
 	}
+
 	return 0;
 }
 

drivers/infiniband/hw/hfi1/mad.h

···
 		    COUNTER_MASK(1, 3) | \
 		    COUNTER_MASK(1, 4))
 
+void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port);
+
 #endif				/* _HFI1_MAD_H */

drivers/infiniband/hw/hfi1/pio.c

···
 		/* counter is reset if occupancy count changes */
 		if (reg != reg_prev)
 			loop = 0;
-		if (loop > 500) {
+		if (loop > 50000) {
 			/* timed out - bounce the link */
 			dd_dev_err(dd,
 				   "%s: context %u(%u) timeout waiting for packets to egress, remaining count %u, bouncing link\n",
···
 }
 
 /*
+ * Set credit return threshold for the kernel send context
+ */
+static void set_threshold(struct hfi1_devdata *dd, int scontext, int i)
+{
+	u32 thres;
+
+	thres = min(sc_percent_to_threshold(dd->kernel_send_context[scontext],
+					    50),
+		    sc_mtu_to_threshold(dd->kernel_send_context[scontext],
+					dd->vld[i].mtu,
+					dd->rcd[0]->rcvhdrqentsize));
+	sc_set_cr_threshold(dd->kernel_send_context[scontext], thres);
+}
+
+/*
  * pio_map_init - called when #vls change
  * @dd: hfi1_devdata
  * @port: port number
···
 			if (!newmap->map[i])
 				goto bail;
 			newmap->map[i]->mask = (1 << ilog2(sz)) - 1;
-			/* assign send contexts */
+			/*
+			 * assign send contexts and
+			 * adjust credit return threshold
+			 */
 			for (j = 0; j < sz; j++) {
-				if (dd->kernel_send_context[scontext])
+				if (dd->kernel_send_context[scontext]) {
 					newmap->map[i]->ksc[j] =
 					dd->kernel_send_context[scontext];
+					set_threshold(dd, scontext, i);
+				}
 				if (++scontext >= first_scontext +
 						  vl_scontexts[i])
 					/* wrap back to first send context */

drivers/infiniband/hw/hfi1/qsfp.c

···
 
 	if (ppd->qsfp_info.cache_valid) {
 		if (QSFP_IS_CU(cache[QSFP_MOD_TECH_OFFS]))
-			sprintf(lenstr, "%dM ", cache[QSFP_MOD_LEN_OFFS]);
+			snprintf(lenstr, sizeof(lenstr), "%dM ",
+				 cache[QSFP_MOD_LEN_OFFS]);
 
 		power_byte = cache[QSFP_MOD_PWR_OFFS];
 		sofar += scnprintf(buf + sofar, len - sofar, "PWR:%.3sW\n",

drivers/infiniband/hw/hfi1/verbs_txreq.c

···
 
 struct verbs_txreq *__get_txreq(struct hfi1_ibdev *dev,
 				struct rvt_qp *qp)
+	__must_hold(&qp->s_lock)
 {
 	struct verbs_txreq *tx = ERR_PTR(-EBUSY);
-	unsigned long flags;
 
-	spin_lock_irqsave(&qp->s_lock, flags);
 	write_seqlock(&dev->iowait_lock);
 	if (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) {
 		struct hfi1_qp_priv *priv;
···
 	}
 out:
 	write_sequnlock(&dev->iowait_lock);
-	spin_unlock_irqrestore(&qp->s_lock, flags);
 	return tx;
 }
 

drivers/infiniband/hw/hfi1/verbs_txreq.h

···
 
 static inline struct verbs_txreq *get_txreq(struct hfi1_ibdev *dev,
 					    struct rvt_qp *qp)
+	__must_hold(&qp->slock)
 {
 	struct verbs_txreq *tx;
 	struct hfi1_qp_priv *priv = qp->priv;

drivers/infiniband/hw/i40iw/i40iw.h

···
 
 #define IW_HMC_OBJ_TYPE_NUM ARRAY_SIZE(iw_hmc_obj_types)
 #define IW_CFG_FPM_QP_COUNT		32768
+#define I40IW_MAX_PAGES_PER_FMR		512
+#define I40IW_MIN_PAGES_PER_FMR		1
 
 #define I40IW_MTU_TO_MSS		40
 #define I40IW_DEFAULT_MSS		1460

drivers/infiniband/hw/i40iw/i40iw_verbs.c

···
 	props->max_qp_init_rd_atom = props->max_qp_rd_atom;
 	props->atomic_cap = IB_ATOMIC_NONE;
 	props->max_map_per_fmr = 1;
+	props->max_fast_reg_page_list_len = I40IW_MAX_PAGES_PER_FMR;
 	return 0;
 }
 
···
 	mutex_lock(&iwdev->pbl_mutex);
 	status = i40iw_get_pble(&iwdev->sc_dev, iwdev->pble_rsrc, palloc, iwmr->page_cnt);
 	mutex_unlock(&iwdev->pbl_mutex);
-	if (!status)
+	if (status)
 		goto err1;
 
 	if (palloc->level != I40IW_LEVEL_1)
···
 			struct i40iw_sc_dev *dev = &iwqp->iwdev->sc_dev;
 			struct i40iw_fast_reg_stag_info info;
 
+			memset(&info, 0, sizeof(info));
 			info.access_rights = I40IW_ACCESS_FLAGS_LOCALREAD;
 			info.access_rights |= i40iw_get_user_access(flags);
 			info.stag_key = reg_wr(ib_wr)->key & 0xff;
···
 			info.addr_type = I40IW_ADDR_TYPE_VA_BASED;
 			info.va = (void *)(uintptr_t)iwmr->ibmr.iova;
 			info.total_len = iwmr->ibmr.length;
+			info.reg_addr_pa = *(u64 *)palloc->level1.addr;
 			info.first_pm_pbl_index = palloc->level1.idx;
 			info.local_fence = ib_wr->send_flags & IB_SEND_FENCE;
 			info.signaled = ib_wr->send_flags & IB_SEND_SIGNALED;
+
+			if (iwmr->npages > I40IW_MIN_PAGES_PER_FMR)
+				info.chunk_size = 1;
 
 			if (page_shift == 21)
 				info.page_size = 1; /* 2M page */
···
 {
 	struct i40iw_cq *iwcq;
 	struct i40iw_cq_uk *ukcq;
-	enum i40iw_completion_notify cq_notify = IW_CQ_COMPL_SOLICITED;
+	unsigned long flags;
+	enum i40iw_completion_notify cq_notify = IW_CQ_COMPL_EVENT;
 
 	iwcq = (struct i40iw_cq *)ibcq;
 	ukcq = &iwcq->sc_cq.cq_uk;
-	if (notify_flags == IB_CQ_NEXT_COMP)
-		cq_notify = IW_CQ_COMPL_EVENT;
+	if (notify_flags == IB_CQ_SOLICITED)
+		cq_notify = IW_CQ_COMPL_SOLICITED;
+	spin_lock_irqsave(&iwcq->lock, flags);
 	ukcq->ops.iw_cq_request_notification(ukcq, cq_notify);
+	spin_unlock_irqrestore(&iwcq->lock, flags);
 	return 0;
 }
 

drivers/infiniband/hw/mlx4/ah.c

···
 
 	ah->av.ib.port_pd = cpu_to_be32(to_mpd(pd)->pdn | (ah_attr->port_num << 24));
 	ah->av.ib.g_slid  = ah_attr->src_path_bits;
+	ah->av.ib.sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
 	if (ah_attr->ah_flags & IB_AH_GRH) {
 		ah->av.ib.g_slid   |= 0x80;
 		ah->av.ib.gid_index = ah_attr->grh.sgid_index;
···
 		       !(1 << ah->av.ib.stat_rate & dev->caps.stat_rate_support))
 			--ah->av.ib.stat_rate;
 	}
-	ah->av.ib.sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
 
 	return &ah->ibah;
 }

drivers/infiniband/hw/mlx4/mad.c

···
 		tun_tx_ix = (++tun_qp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
 	spin_unlock(&tun_qp->tx_lock);
 	if (ret)
-		goto out;
+		goto end;
 
 	tun_mad = (struct mlx4_rcv_tunnel_mad *) (tun_qp->tx_ring[tun_tx_ix].buf.addr);
 	if (tun_qp->tx_ring[tun_tx_ix].ah)
···
 	wr.wr.send_flags = IB_SEND_SIGNALED;
 
 	ret = ib_post_send(src_qp, &wr.wr, &bad_wr);
-out:
-	if (ret)
-		ib_destroy_ah(ah);
+	if (!ret)
+		return 0;
+ out:
+	spin_lock(&tun_qp->tx_lock);
+	tun_qp->tx_ix_tail++;
+	spin_unlock(&tun_qp->tx_lock);
+	tun_qp->tx_ring[tun_tx_ix].ah = NULL;
+end:
+	ib_destroy_ah(ah);
 	return ret;
 }
 
···
 
 
 	ret = ib_post_send(send_qp, &wr.wr, &bad_wr);
+	if (!ret)
+		return 0;
+
+	spin_lock(&sqp->tx_lock);
+	sqp->tx_ix_tail++;
+	spin_unlock(&sqp->tx_lock);
+	sqp->tx_ring[wire_tx_ix].ah = NULL;
 out:
-	if (ret)
-		ib_destroy_ah(ah);
+	ib_destroy_ah(ah);
 	return ret;
 }
 

drivers/infiniband/hw/mlx4/main.c

···
 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
 	int is_bonded = mlx4_is_bonded(dev);
 
+	if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
+		return ERR_PTR(-EINVAL);
+
 	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
 	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
 		return ERR_PTR(-EOPNOTSUPP);

drivers/infiniband/hw/mlx4/mlx4_ib.h

···
 	u32			max_pages;
 	struct mlx4_mr		mmr;
 	struct ib_umem	       *umem;
-	void			*pages_alloc;
+	size_t			page_map_size;
 };
 
 struct mlx4_ib_mw {

drivers/infiniband/hw/mlx4/mr.c

···
 		      struct mlx4_ib_mr *mr,
 		      int max_pages)
 {
-	int size = max_pages * sizeof(u64);
-	int add_size;
 	int ret;
 
-	add_size = max_t(int, MLX4_MR_PAGES_ALIGN - ARCH_KMALLOC_MINALIGN, 0);
+	/* Ensure that size is aligned to DMA cacheline
+	 * requirements.
+	 * max_pages is limited to MLX4_MAX_FAST_REG_PAGES
+	 * so page_map_size will never cross PAGE_SIZE.
+	 */
+	mr->page_map_size = roundup(max_pages * sizeof(u64),
+				    MLX4_MR_PAGES_ALIGN);
 
-	mr->pages_alloc = kzalloc(size + add_size, GFP_KERNEL);
-	if (!mr->pages_alloc)
+	/* Prevent cross page boundary allocation. */
+	mr->pages = (__be64 *)get_zeroed_page(GFP_KERNEL);
+	if (!mr->pages)
 		return -ENOMEM;
 
-	mr->pages = PTR_ALIGN(mr->pages_alloc, MLX4_MR_PAGES_ALIGN);
-
 	mr->page_map = dma_map_single(device->dma_device, mr->pages,
-				      size, DMA_TO_DEVICE);
+				      mr->page_map_size, DMA_TO_DEVICE);
 
 	if (dma_mapping_error(device->dma_device, mr->page_map)) {
 		ret = -ENOMEM;
···
 	}
 
 	return 0;
-err:
-	kfree(mr->pages_alloc);
 
+err:
+	free_page((unsigned long)mr->pages);
 	return ret;
 }
 
···
 {
 	if (mr->pages) {
 		struct ib_device *device = mr->ibmr.device;
-		int size = mr->max_pages * sizeof(u64);
 
 		dma_unmap_single(device->dma_device, mr->page_map,
-				 size, DMA_TO_DEVICE);
-		kfree(mr->pages_alloc);
+				 mr->page_map_size, DMA_TO_DEVICE);
+		free_page((unsigned long)mr->pages);
 		mr->pages = NULL;
 	}
 }
···
 	mr->npages = 0;
 
 	ib_dma_sync_single_for_cpu(ibmr->device, mr->page_map,
-				   sizeof(u64) * mr->max_pages,
-				   DMA_TO_DEVICE);
+				   mr->page_map_size, DMA_TO_DEVICE);
 
 	rc = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, mlx4_set_page);
 
 	ib_dma_sync_single_for_device(ibmr->device, mr->page_map,
-				      sizeof(u64) * mr->max_pages,
-				      DMA_TO_DEVICE);
+				      mr->page_map_size, DMA_TO_DEVICE);
 
 	return rc;
 }

drivers/infiniband/hw/mlx4/qp.c

···
 			sizeof (struct mlx4_wqe_raddr_seg);
 	case MLX4_IB_QPT_RC:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
-			sizeof (struct mlx4_wqe_atomic_seg) +
+			sizeof (struct mlx4_wqe_masked_atomic_seg) +
 			sizeof (struct mlx4_wqe_raddr_seg);
 	case MLX4_IB_QPT_SMI:
 	case MLX4_IB_QPT_GSI:
···
 	{
 		err = create_qp_common(to_mdev(pd->device), pd, init_attr,
 				       udata, 0, &qp, gfp);
-		if (err)
+		if (err) {
+			kfree(qp);
 			return ERR_PTR(err);
+		}
 
 		qp->ibqp.qp_num = qp->mqp.qpn;
 		qp->xrcdn = xrcdn;

drivers/infiniband/hw/mlx5/mad.c

···
 	pma_cnt_ext->port_xmit_data =
 		cpu_to_be64(MLX5_SUM_CNT(out, transmitted_ib_unicast.octets,
 					 transmitted_ib_multicast.octets) >> 2);
-	pma_cnt_ext->port_xmit_data =
+	pma_cnt_ext->port_rcv_data =
 		cpu_to_be64(MLX5_SUM_CNT(out, received_ib_unicast.octets,
 					 received_ib_multicast.octets) >> 2);
 	pma_cnt_ext->port_xmit_packets =

drivers/infiniband/hw/mlx5/qp.c

···
 			return MLX5_FENCE_MODE_SMALL_AND_FENCE;
 		else
 			return fence;
-
-	} else {
-		return 0;
+	} else if (unlikely(wr->send_flags & IB_SEND_FENCE)) {
+		return MLX5_FENCE_MODE_FENCE;
 	}
+
+	return 0;
 }
 
 static int begin_wqe(struct mlx5_ib_qp *qp, void **seg,

drivers/infiniband/hw/qib/qib_file_ops.c

···
 
 	switch (cmd.type) {
 	case QIB_CMD_ASSIGN_CTXT:
+		if (rcd) {
+			ret = -EINVAL;
+			goto bail;
+		}
+
 		ret = qib_assign_ctxt(fp, &cmd.cmd.user_info);
 		if (ret)
 			goto bail;

drivers/infiniband/sw/rdmavt/qp.c

···
 			/* wrap to first map page, invert bit 0 */
 			offset = qpt->incr | ((offset & 1) ^ 1);
 		}
-		/* there can be no bits at shift and below */
-		WARN_ON(offset & (rdi->dparms.qos_shift - 1));
+		/* there can be no set bits in low-order QoS bits */
+		WARN_ON(offset & (BIT(rdi->dparms.qos_shift) - 1));
 		qpn = mk_qpn(qpt, map, offset);
 	}
 
···
 	qp->s_ssn = 1;
 	qp->s_lsn = 0;
 	qp->s_mig_state = IB_MIG_MIGRATED;
-	if (qp->s_ack_queue)
-		memset(
-			qp->s_ack_queue,
-			0,
-			rvt_max_atomic(rdi) *
-				sizeof(*qp->s_ack_queue));
 	qp->r_head_ack_queue = 0;
 	qp->s_tail_ack_queue = 0;
 	qp->s_num_rd_atomic = 0;
···
 		 * initialization that is needed.
 		 */
 		priv = rdi->driver_f.qp_priv_alloc(rdi, qp, gfp);
-		if (!priv)
+		if (IS_ERR(priv)) {
+			ret = priv;
 			goto bail_qp;
+		}
 		qp->priv = priv;
 		qp->timeout_jiffies =
 			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /

drivers/infiniband/sw/rdmavt/vt.c

···
 			    !rdi->driver_f.quiesce_qp ||
 			    !rdi->driver_f.notify_error_qp ||
 			    !rdi->driver_f.mtu_from_qp ||
-			    !rdi->driver_f.mtu_to_path_mtu ||
-			    !rdi->driver_f.shut_down_port ||
-			    !rdi->driver_f.cap_mask_chg)
+			    !rdi->driver_f.mtu_to_path_mtu)
 				return -EINVAL;
 		break;
 

drivers/infiniband/ulp/srpt/ib_srpt.c

···
 	 */
 	qp_init->cap.max_send_wr = srp_sq_size / 2;
 	qp_init->cap.max_rdma_ctxs = srp_sq_size / 2;
-	qp_init->cap.max_send_sge = max(sdev->device->attrs.max_sge_rd,
-					sdev->device->attrs.max_sge);
+	qp_init->cap.max_send_sge = SRPT_DEF_SG_PER_WQE;
 	qp_init->port_num = ch->sport->port;
 
 	ch->qp = ib_create_qp(sdev->pd, qp_init);

drivers/infiniband/ulp/srpt/ib_srpt.h

···
 	SRP_LOGIN_RSP_MULTICHAN_MAINTAINED = 0x2,
 
 	SRPT_DEF_SG_TABLESIZE = 128,
+	SRPT_DEF_SG_PER_WQE = 16,
 
 	MIN_SRPT_SQ_SIZE = 16,
 	DEF_SRPT_SQ_SIZE = 4096,

include/linux/mlx5/qp.h

···
 enum {
 	MLX5_FENCE_MODE_NONE			= 0 << 5,
 	MLX5_FENCE_MODE_INITIATOR_SMALL		= 1 << 5,
+	MLX5_FENCE_MODE_FENCE			= 2 << 5,
 	MLX5_FENCE_MODE_STRONG_ORDERING		= 3 << 5,
 	MLX5_FENCE_MODE_SMALL_AND_FENCE		= 4 << 5,
 };

include/rdma/rdma_vt.h

···
 
 	/*
 	 * Allocate a private queue pair data structure for driver specific
-	 * information which is opaque to rdmavt.
+	 * information which is opaque to rdmavt.  Errors are returned via
+	 * ERR_PTR(err).  The driver is free to return NULL or a valid
+	 * pointer.
 	 */
 	void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 				gfp_t gfp);