Add the rt linux 4.1.3-rt3 as base

Import the rt linux 4.1.3-rt3 as OPNFV kvm base.

It's from git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git linux-4.1.y-rt and
the base is:

commit 0917f823c59692d751951bf5ea699a2d1e2f26a2
Author: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date:   Sat Jul 25 12:13:34 2015 +0200

    Prepare v4.1.3-rt3

    Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

We lose all the git history this way and it's not good. We
should apply another opnfv project repo in future.

Change-Id: I87543d81c9df70d99c5001fbdf646b202c19f423
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>

14 files changed, 9892 insertions, 0 deletions

diff --git a/kernel/drivers/infiniband/hw/mlx5/Kconfig b/kernel/drivers/infiniband/hw/mlx5/Kconfig
new file mode 100644
index 000000000..10df386c6
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/Kconfig
@@ -0,0 +1,10 @@
+config MLX5_INFINIBAND
+	tristate "Mellanox Connect-IB HCA support"
+	depends on NETDEVICES && ETHERNET && PCI
+	select NET_VENDOR_MELLANOX
+	select MLX5_CORE
+	---help---
+	  This driver provides low-level InfiniBand support for
+	  Mellanox Connect-IB PCI Express host channel adapters (HCAs).
+	  This is required to use InfiniBand protocols such as
+	  IP-over-IB or SRP with these devices.
diff --git a/kernel/drivers/infiniband/hw/mlx5/Makefile b/kernel/drivers/infiniband/hw/mlx5/Makefile
new file mode 100644
index 000000000..27a70159e
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/Makefile
@@ -0,0 +1,4 @@
+obj-$(CONFIG_MLX5_INFINIBAND)	+= mlx5_ib.o
+
+mlx5_ib-y :=	main.o cq.o doorbell.o qp.o mem.o srq.o mr.o ah.o mad.o
+mlx5_ib-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += odp.o
diff --git a/kernel/drivers/infiniband/hw/mlx5/ah.c b/kernel/drivers/infiniband/hw/mlx5/ah.c
new file mode 100644
index 000000000..66080580e
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/ah.c
@@ -0,0 +1,92 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "mlx5_ib.h"
+
+struct ib_ah *create_ib_ah(struct ib_ah_attr *ah_attr,
+			   struct mlx5_ib_ah *ah)
+{
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		memcpy(ah->av.rgid, &ah_attr->grh.dgid, 16);
+		ah->av.grh_gid_fl = cpu_to_be32(ah_attr->grh.flow_label |
+						(1 << 30) |
+						ah_attr->grh.sgid_index << 20);
+		ah->av.hop_limit = ah_attr->grh.hop_limit;
+		ah->av.tclass = ah_attr->grh.traffic_class;
+	}
+
+	ah->av.rlid = cpu_to_be16(ah_attr->dlid);
+	ah->av.fl_mlid = ah_attr->src_path_bits & 0x7f;
+	ah->av.stat_rate_sl = (ah_attr->static_rate << 4) | (ah_attr->sl & 0xf);
+
+	return &ah->ibah;
+}
+
+struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
+{
+	struct mlx5_ib_ah *ah;
+
+	ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	return create_ib_ah(ah_attr, ah); /* never fails */
+}
+
+int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
+{
+	struct mlx5_ib_ah *ah = to_mah(ibah);
+	u32 tmp;
+
+	memset(ah_attr, 0, sizeof(*ah_attr));
+
+	tmp = be32_to_cpu(ah->av.grh_gid_fl);
+	if (tmp & (1 << 30)) {
+		ah_attr->ah_flags = IB_AH_GRH;
+		ah_attr->grh.sgid_index = (tmp >> 20) & 0xff;
+		ah_attr->grh.flow_label = tmp & 0xfffff;
+		memcpy(&ah_attr->grh.dgid, ah->av.rgid, 16);
+		ah_attr->grh.hop_limit = ah->av.hop_limit;
+		ah_attr->grh.traffic_class = ah->av.tclass;
+	}
+	ah_attr->dlid = be16_to_cpu(ah->av.rlid);
+	ah_attr->static_rate = ah->av.stat_rate_sl >> 4;
+	ah_attr->sl = ah->av.stat_rate_sl & 0xf;
+
+	return 0;
+}
+
+int mlx5_ib_destroy_ah(struct ib_ah *ah)
+{
+	kfree(to_mah(ah));
+	return 0;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/cq.c b/kernel/drivers/infiniband/hw/mlx5/cq.c
new file mode 100644
index 000000000..2ee6b1051
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/cq.c
@@ -0,0 +1,1189 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/kref.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_user_verbs.h>
+#include "mlx5_ib.h"
+#include "user.h"
+
+static void mlx5_ib_cq_comp(struct mlx5_core_cq *cq)
+{
+	struct ib_cq *ibcq = &to_mibcq(cq)->ibcq;
+
+	ibcq->comp_handler(ibcq, ibcq->cq_context);
+}
+
+static void mlx5_ib_cq_event(struct mlx5_core_cq *mcq, enum mlx5_event type)
+{
+	struct mlx5_ib_cq *cq = container_of(mcq, struct mlx5_ib_cq, mcq);
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct ib_cq *ibcq = &cq->ibcq;
+	struct ib_event event;
+
+	if (type != MLX5_EVENT_TYPE_CQ_ERROR) {
+		mlx5_ib_warn(dev, "Unexpected event type %d on CQ %06x\n",
+			     type, mcq->cqn);
+		return;
+	}
+
+	if (ibcq->event_handler) {
+		event.device     = &dev->ib_dev;
+		event.event      = IB_EVENT_CQ_ERR;
+		event.element.cq = ibcq;
+		ibcq->event_handler(&event, ibcq->cq_context);
+	}
+}
+
+static void *get_cqe_from_buf(struct mlx5_ib_cq_buf *buf, int n, int size)
+{
+	return mlx5_buf_offset(&buf->buf, n * size);
+}
+
+static void *get_cqe(struct mlx5_ib_cq *cq, int n)
+{
+	return get_cqe_from_buf(&cq->buf, n, cq->mcq.cqe_sz);
+}
+
+static u8 sw_ownership_bit(int n, int nent)
+{
+	return (n & nent) ? 1 : 0;
+}
+
+static void *get_sw_cqe(struct mlx5_ib_cq *cq, int n)
+{
+	void *cqe = get_cqe(cq, n & cq->ibcq.cqe);
+	struct mlx5_cqe64 *cqe64;
+
+	cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+
+	if (likely((cqe64->op_own) >> 4 != MLX5_CQE_INVALID) &&
+	    !((cqe64->op_own & MLX5_CQE_OWNER_MASK) ^ !!(n & (cq->ibcq.cqe + 1)))) {
+		return cqe;
+	} else {
+		return NULL;
+	}
+}
+
+static void *next_cqe_sw(struct mlx5_ib_cq *cq)
+{
+	return get_sw_cqe(cq, cq->mcq.cons_index);
+}
+
+static enum ib_wc_opcode get_umr_comp(struct mlx5_ib_wq *wq, int idx)
+{
+	switch (wq->wr_data[idx]) {
+	case MLX5_IB_WR_UMR:
+		return 0;
+
+	case IB_WR_LOCAL_INV:
+		return IB_WC_LOCAL_INV;
+
+	case IB_WR_FAST_REG_MR:
+		return IB_WC_FAST_REG_MR;
+
+	default:
+		pr_warn("unknown completion status\n");
+		return 0;
+	}
+}
+
+static void handle_good_req(struct ib_wc *wc, struct mlx5_cqe64 *cqe,
+			    struct mlx5_ib_wq *wq, int idx)
+{
+	wc->wc_flags = 0;
+	switch (be32_to_cpu(cqe->sop_drop_qpn) >> 24) {
+	case MLX5_OPCODE_RDMA_WRITE_IMM:
+		wc->wc_flags |= IB_WC_WITH_IMM;
+	case MLX5_OPCODE_RDMA_WRITE:
+		wc->opcode    = IB_WC_RDMA_WRITE;
+		break;
+	case MLX5_OPCODE_SEND_IMM:
+		wc->wc_flags |= IB_WC_WITH_IMM;
+	case MLX5_OPCODE_SEND:
+	case MLX5_OPCODE_SEND_INVAL:
+		wc->opcode    = IB_WC_SEND;
+		break;
+	case MLX5_OPCODE_RDMA_READ:
+		wc->opcode    = IB_WC_RDMA_READ;
+		wc->byte_len  = be32_to_cpu(cqe->byte_cnt);
+		break;
+	case MLX5_OPCODE_ATOMIC_CS:
+		wc->opcode    = IB_WC_COMP_SWAP;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_FA:
+		wc->opcode    = IB_WC_FETCH_ADD;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_MASKED_CS:
+		wc->opcode    = IB_WC_MASKED_COMP_SWAP;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_MASKED_FA:
+		wc->opcode    = IB_WC_MASKED_FETCH_ADD;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_BIND_MW:
+		wc->opcode    = IB_WC_BIND_MW;
+		break;
+	case MLX5_OPCODE_UMR:
+		wc->opcode = get_umr_comp(wq, idx);
+		break;
+	}
+}
+
+enum {
+	MLX5_GRH_IN_BUFFER = 1,
+	MLX5_GRH_IN_CQE	   = 2,
+};
+
+static void handle_responder(struct ib_wc *wc, struct mlx5_cqe64 *cqe,
+			     struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.device);
+	struct mlx5_ib_srq *srq;
+	struct mlx5_ib_wq *wq;
+	u16 wqe_ctr;
+	u8 g;
+
+	if (qp->ibqp.srq || qp->ibqp.xrcd) {
+		struct mlx5_core_srq *msrq = NULL;
+
+		if (qp->ibqp.xrcd) {
+			msrq = mlx5_core_get_srq(dev->mdev,
+						 be32_to_cpu(cqe->srqn));
+			srq = to_mibsrq(msrq);
+		} else {
+			srq = to_msrq(qp->ibqp.srq);
+		}
+		if (srq) {
+			wqe_ctr = be16_to_cpu(cqe->wqe_counter);
+			wc->wr_id = srq->wrid[wqe_ctr];
+			mlx5_ib_free_srq_wqe(srq, wqe_ctr);
+			if (msrq && atomic_dec_and_test(&msrq->refcount))
+				complete(&msrq->free);
+		}
+	} else {
+		wq	  = &qp->rq;
+		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		++wq->tail;
+	}
+	wc->byte_len = be32_to_cpu(cqe->byte_cnt);
+
+	switch (cqe->op_own >> 4) {
+	case MLX5_CQE_RESP_WR_IMM:
+		wc->opcode	= IB_WC_RECV_RDMA_WITH_IMM;
+		wc->wc_flags	= IB_WC_WITH_IMM;
+		wc->ex.imm_data = cqe->imm_inval_pkey;
+		break;
+	case MLX5_CQE_RESP_SEND:
+		wc->opcode   = IB_WC_RECV;
+		wc->wc_flags = 0;
+		break;
+	case MLX5_CQE_RESP_SEND_IMM:
+		wc->opcode	= IB_WC_RECV;
+		wc->wc_flags	= IB_WC_WITH_IMM;
+		wc->ex.imm_data = cqe->imm_inval_pkey;
+		break;
+	case MLX5_CQE_RESP_SEND_INV:
+		wc->opcode	= IB_WC_RECV;
+		wc->wc_flags	= IB_WC_WITH_INVALIDATE;
+		wc->ex.invalidate_rkey = be32_to_cpu(cqe->imm_inval_pkey);
+		break;
+	}
+	wc->slid	   = be16_to_cpu(cqe->slid);
+	wc->sl		   = (be32_to_cpu(cqe->flags_rqpn) >> 24) & 0xf;
+	wc->src_qp	   = be32_to_cpu(cqe->flags_rqpn) & 0xffffff;
+	wc->dlid_path_bits = cqe->ml_path;
+	g = (be32_to_cpu(cqe->flags_rqpn) >> 28) & 3;
+	wc->wc_flags |= g ? IB_WC_GRH : 0;
+	wc->pkey_index     = be32_to_cpu(cqe->imm_inval_pkey) & 0xffff;
+}
+
+static void dump_cqe(struct mlx5_ib_dev *dev, struct mlx5_err_cqe *cqe)
+{
+	__be32 *p = (__be32 *)cqe;
+	int i;
+
+	mlx5_ib_warn(dev, "dump error cqe\n");
+	for (i = 0; i < sizeof(*cqe) / 16; i++, p += 4)
+		pr_info("%08x %08x %08x %08x\n", be32_to_cpu(p[0]),
+			be32_to_cpu(p[1]), be32_to_cpu(p[2]),
+			be32_to_cpu(p[3]));
+}
+
+static void mlx5_handle_error_cqe(struct mlx5_ib_dev *dev,
+				  struct mlx5_err_cqe *cqe,
+				  struct ib_wc *wc)
+{
+	int dump = 1;
+
+	switch (cqe->syndrome) {
+	case MLX5_CQE_SYNDROME_LOCAL_LENGTH_ERR:
+		wc->status = IB_WC_LOC_LEN_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_QP_OP_ERR:
+		wc->status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_PROT_ERR:
+		wc->status = IB_WC_LOC_PROT_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_WR_FLUSH_ERR:
+		dump = 0;
+		wc->status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_MW_BIND_ERR:
+		wc->status = IB_WC_MW_BIND_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_BAD_RESP_ERR:
+		wc->status = IB_WC_BAD_RESP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_ACCESS_ERR:
+		wc->status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_INVAL_REQ_ERR:
+		wc->status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_ACCESS_ERR:
+		wc->status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_OP_ERR:
+		wc->status = IB_WC_REM_OP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_TRANSPORT_RETRY_EXC_ERR:
+		wc->status = IB_WC_RETRY_EXC_ERR;
+		dump = 0;
+		break;
+	case MLX5_CQE_SYNDROME_RNR_RETRY_EXC_ERR:
+		wc->status = IB_WC_RNR_RETRY_EXC_ERR;
+		dump = 0;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_ABORTED_ERR:
+		wc->status = IB_WC_REM_ABORT_ERR;
+		break;
+	default:
+		wc->status = IB_WC_GENERAL_ERR;
+		break;
+	}
+
+	wc->vendor_err = cqe->vendor_err_synd;
+	if (dump)
+		dump_cqe(dev, cqe);
+}
+
+static int is_atomic_response(struct mlx5_ib_qp *qp, uint16_t idx)
+{
+	/* TBD: waiting decision
+	*/
+	return 0;
+}
+
+static void *mlx5_get_atomic_laddr(struct mlx5_ib_qp *qp, uint16_t idx)
+{
+	struct mlx5_wqe_data_seg *dpseg;
+	void *addr;
+
+	dpseg = mlx5_get_send_wqe(qp, idx) + sizeof(struct mlx5_wqe_ctrl_seg) +
+		sizeof(struct mlx5_wqe_raddr_seg) +
+		sizeof(struct mlx5_wqe_atomic_seg);
+	addr = (void *)(unsigned long)be64_to_cpu(dpseg->addr);
+	return addr;
+}
+
+static void handle_atomic(struct mlx5_ib_qp *qp, struct mlx5_cqe64 *cqe64,
+			  uint16_t idx)
+{
+	void *addr;
+	int byte_count;
+	int i;
+
+	if (!is_atomic_response(qp, idx))
+		return;
+
+	byte_count = be32_to_cpu(cqe64->byte_cnt);
+	addr = mlx5_get_atomic_laddr(qp, idx);
+
+	if (byte_count == 4) {
+		*(uint32_t *)addr = be32_to_cpu(*((__be32 *)addr));
+	} else {
+		for (i = 0; i < byte_count; i += 8) {
+			*(uint64_t *)addr = be64_to_cpu(*((__be64 *)addr));
+			addr += 8;
+		}
+	}
+
+	return;
+}
+
+static void handle_atomics(struct mlx5_ib_qp *qp, struct mlx5_cqe64 *cqe64,
+			   u16 tail, u16 head)
+{
+	u16 idx;
+
+	do {
+		idx = tail & (qp->sq.wqe_cnt - 1);
+		handle_atomic(qp, cqe64, idx);
+		if (idx == head)
+			break;
+
+		tail = qp->sq.w_list[idx].next;
+	} while (1);
+	tail = qp->sq.w_list[idx].next;
+	qp->sq.last_poll = tail;
+}
+
+static void free_cq_buf(struct mlx5_ib_dev *dev, struct mlx5_ib_cq_buf *buf)
+{
+	mlx5_buf_free(dev->mdev, &buf->buf);
+}
+
+static void get_sig_err_item(struct mlx5_sig_err_cqe *cqe,
+			     struct ib_sig_err *item)
+{
+	u16 syndrome = be16_to_cpu(cqe->syndrome);
+
+#define GUARD_ERR   (1 << 13)
+#define APPTAG_ERR  (1 << 12)
+#define REFTAG_ERR  (1 << 11)
+
+	if (syndrome & GUARD_ERR) {
+		item->err_type = IB_SIG_BAD_GUARD;
+		item->expected = be32_to_cpu(cqe->expected_trans_sig) >> 16;
+		item->actual = be32_to_cpu(cqe->actual_trans_sig) >> 16;
+	} else
+	if (syndrome & REFTAG_ERR) {
+		item->err_type = IB_SIG_BAD_REFTAG;
+		item->expected = be32_to_cpu(cqe->expected_reftag);
+		item->actual = be32_to_cpu(cqe->actual_reftag);
+	} else
+	if (syndrome & APPTAG_ERR) {
+		item->err_type = IB_SIG_BAD_APPTAG;
+		item->expected = be32_to_cpu(cqe->expected_trans_sig) & 0xffff;
+		item->actual = be32_to_cpu(cqe->actual_trans_sig) & 0xffff;
+	} else {
+		pr_err("Got signature completion error with bad syndrome %04x\n",
+		       syndrome);
+	}
+
+	item->sig_err_offset = be64_to_cpu(cqe->err_offset);
+	item->key = be32_to_cpu(cqe->mkey);
+}
+
+static int mlx5_poll_one(struct mlx5_ib_cq *cq,
+			 struct mlx5_ib_qp **cur_qp,
+			 struct ib_wc *wc)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct mlx5_err_cqe *err_cqe;
+	struct mlx5_cqe64 *cqe64;
+	struct mlx5_core_qp *mqp;
+	struct mlx5_ib_wq *wq;
+	struct mlx5_sig_err_cqe *sig_err_cqe;
+	struct mlx5_core_mr *mmr;
+	struct mlx5_ib_mr *mr;
+	uint8_t opcode;
+	uint32_t qpn;
+	u16 wqe_ctr;
+	void *cqe;
+	int idx;
+
+repoll:
+	cqe = next_cqe_sw(cq);
+	if (!cqe)
+		return -EAGAIN;
+
+	cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+
+	++cq->mcq.cons_index;
+
+	/* Make sure we read CQ entry contents after we've checked the
+	 * ownership bit.
+	 */
+	rmb();
+
+	opcode = cqe64->op_own >> 4;
+	if (unlikely(opcode == MLX5_CQE_RESIZE_CQ)) {
+		if (likely(cq->resize_buf)) {
+			free_cq_buf(dev, &cq->buf);
+			cq->buf = *cq->resize_buf;
+			kfree(cq->resize_buf);
+			cq->resize_buf = NULL;
+			goto repoll;
+		} else {
+			mlx5_ib_warn(dev, "unexpected resize cqe\n");
+		}
+	}
+
+	qpn = ntohl(cqe64->sop_drop_qpn) & 0xffffff;
+	if (!*cur_qp || (qpn != (*cur_qp)->ibqp.qp_num)) {
+		/* We do not have to take the QP table lock here,
+		 * because CQs will be locked while QPs are removed
+		 * from the table.
+		 */
+		mqp = __mlx5_qp_lookup(dev->mdev, qpn);
+		if (unlikely(!mqp)) {
+			mlx5_ib_warn(dev, "CQE@CQ %06x for unknown QPN %6x\n",
+				     cq->mcq.cqn, qpn);
+			return -EINVAL;
+		}
+
+		*cur_qp = to_mibqp(mqp);
+	}
+
+	wc->qp  = &(*cur_qp)->ibqp;
+	switch (opcode) {
+	case MLX5_CQE_REQ:
+		wq = &(*cur_qp)->sq;
+		wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+		idx = wqe_ctr & (wq->wqe_cnt - 1);
+		handle_good_req(wc, cqe64, wq, idx);
+		handle_atomics(*cur_qp, cqe64, wq->last_poll, idx);
+		wc->wr_id = wq->wrid[idx];
+		wq->tail = wq->wqe_head[idx] + 1;
+		wc->status = IB_WC_SUCCESS;
+		break;
+	case MLX5_CQE_RESP_WR_IMM:
+	case MLX5_CQE_RESP_SEND:
+	case MLX5_CQE_RESP_SEND_IMM:
+	case MLX5_CQE_RESP_SEND_INV:
+		handle_responder(wc, cqe64, *cur_qp);
+		wc->status = IB_WC_SUCCESS;
+		break;
+	case MLX5_CQE_RESIZE_CQ:
+		break;
+	case MLX5_CQE_REQ_ERR:
+	case MLX5_CQE_RESP_ERR:
+		err_cqe = (struct mlx5_err_cqe *)cqe64;
+		mlx5_handle_error_cqe(dev, err_cqe, wc);
+		mlx5_ib_dbg(dev, "%s error cqe on cqn 0x%x:\n",
+			    opcode == MLX5_CQE_REQ_ERR ?
+			    "Requestor" : "Responder", cq->mcq.cqn);
+		mlx5_ib_dbg(dev, "syndrome 0x%x, vendor syndrome 0x%x\n",
+			    err_cqe->syndrome, err_cqe->vendor_err_synd);
+		if (opcode == MLX5_CQE_REQ_ERR) {
+			wq = &(*cur_qp)->sq;
+			wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+			idx = wqe_ctr & (wq->wqe_cnt - 1);
+			wc->wr_id = wq->wrid[idx];
+			wq->tail = wq->wqe_head[idx] + 1;
+		} else {
+			struct mlx5_ib_srq *srq;
+
+			if ((*cur_qp)->ibqp.srq) {
+				srq = to_msrq((*cur_qp)->ibqp.srq);
+				wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+				wc->wr_id = srq->wrid[wqe_ctr];
+				mlx5_ib_free_srq_wqe(srq, wqe_ctr);
+			} else {
+				wq = &(*cur_qp)->rq;
+				wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+				++wq->tail;
+			}
+		}
+		break;
+	case MLX5_CQE_SIG_ERR:
+		sig_err_cqe = (struct mlx5_sig_err_cqe *)cqe64;
+
+		read_lock(&dev->mdev->priv.mr_table.lock);
+		mmr = __mlx5_mr_lookup(dev->mdev,
+				       mlx5_base_mkey(be32_to_cpu(sig_err_cqe->mkey)));
+		if (unlikely(!mmr)) {
+			read_unlock(&dev->mdev->priv.mr_table.lock);
+			mlx5_ib_warn(dev, "CQE@CQ %06x for unknown MR %6x\n",
+				     cq->mcq.cqn, be32_to_cpu(sig_err_cqe->mkey));
+			return -EINVAL;
+		}
+
+		mr = to_mibmr(mmr);
+		get_sig_err_item(sig_err_cqe, &mr->sig->err_item);
+		mr->sig->sig_err_exists = true;
+		mr->sig->sigerr_count++;
+
+		mlx5_ib_warn(dev, "CQN: 0x%x Got SIGERR on key: 0x%x err_type %x err_offset %llx expected %x actual %x\n",
+			     cq->mcq.cqn, mr->sig->err_item.key,
+			     mr->sig->err_item.err_type,
+			     mr->sig->err_item.sig_err_offset,
+			     mr->sig->err_item.expected,
+			     mr->sig->err_item.actual);
+
+		read_unlock(&dev->mdev->priv.mr_table.lock);
+		goto repoll;
+	}
+
+	return 0;
+}
+
+int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	struct mlx5_ib_cq *cq = to_mcq(ibcq);
+	struct mlx5_ib_qp *cur_qp = NULL;
+	unsigned long flags;
+	int npolled;
+	int err = 0;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	for (npolled = 0; npolled < num_entries; npolled++) {
+		err = mlx5_poll_one(cq, &cur_qp, wc + npolled);
+		if (err)
+			break;
+	}
+
+	if (npolled)
+		mlx5_cq_set_ci(&cq->mcq);
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	if (err == 0 || err == -EAGAIN)
+		return npolled;
+	else
+		return err;
+}
+
+int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
+{
+	struct mlx5_core_dev *mdev = to_mdev(ibcq->device)->mdev;
+	void __iomem *uar_page = mdev->priv.uuari.uars[0].map;
+
+	mlx5_cq_arm(&to_mcq(ibcq)->mcq,
+		    (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
+		    MLX5_CQ_DB_REQ_NOT_SOL : MLX5_CQ_DB_REQ_NOT,
+		    uar_page,
+		    MLX5_GET_DOORBELL_LOCK(&mdev->priv.cq_uar_lock),
+		    to_mcq(ibcq)->mcq.cons_index);
+
+	return 0;
+}
+
+static int alloc_cq_buf(struct mlx5_ib_dev *dev, struct mlx5_ib_cq_buf *buf,
+			int nent, int cqe_size)
+{
+	int err;
+
+	err = mlx5_buf_alloc(dev->mdev, nent * cqe_size,
+			     PAGE_SIZE * 2, &buf->buf);
+	if (err)
+		return err;
+
+	buf->cqe_size = cqe_size;
+	buf->nent = nent;
+
+	return 0;
+}
+
+static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
+			  struct ib_ucontext *context, struct mlx5_ib_cq *cq,
+			  int entries, struct mlx5_create_cq_mbox_in **cqb,
+			  int *cqe_size, int *index, int *inlen)
+{
+	struct mlx5_ib_create_cq ucmd;
+	size_t ucmdlen;
+	int page_shift;
+	int npages;
+	int ncont;
+	int err;
+
+	ucmdlen =
+		(udata->inlen - sizeof(struct ib_uverbs_cmd_hdr) <
+		 sizeof(ucmd)) ? (sizeof(ucmd) -
+				  sizeof(ucmd.reserved)) : sizeof(ucmd);
+
+	if (ib_copy_from_udata(&ucmd, udata, ucmdlen))
+		return -EFAULT;
+
+	if (ucmdlen == sizeof(ucmd) &&
+	    ucmd.reserved != 0)
+		return -EINVAL;
+
+	if (ucmd.cqe_size != 64 && ucmd.cqe_size != 128)
+		return -EINVAL;
+
+	*cqe_size = ucmd.cqe_size;
+
+	cq->buf.umem = ib_umem_get(context, ucmd.buf_addr,
+				   entries * ucmd.cqe_size,
+				   IB_ACCESS_LOCAL_WRITE, 1);
+	if (IS_ERR(cq->buf.umem)) {
+		err = PTR_ERR(cq->buf.umem);
+		return err;
+	}
+
+	err = mlx5_ib_db_map_user(to_mucontext(context), ucmd.db_addr,
+				  &cq->db);
+	if (err)
+		goto err_umem;
+
+	mlx5_ib_cont_pages(cq->buf.umem, ucmd.buf_addr, &npages, &page_shift,
+			   &ncont, NULL);
+	mlx5_ib_dbg(dev, "addr 0x%llx, size %u, npages %d, page_shift %d, ncont %d\n",
+		    ucmd.buf_addr, entries * ucmd.cqe_size, npages, page_shift, ncont);
+
+	*inlen = sizeof(**cqb) + sizeof(*(*cqb)->pas) * ncont;
+	*cqb = mlx5_vzalloc(*inlen);
+	if (!*cqb) {
+		err = -ENOMEM;
+		goto err_db;
+	}
+	mlx5_ib_populate_pas(dev, cq->buf.umem, page_shift, (*cqb)->pas, 0);
+	(*cqb)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+
+	*index = to_mucontext(context)->uuari.uars[0].index;
+
+	return 0;
+
+err_db:
+	mlx5_ib_db_unmap_user(to_mucontext(context), &cq->db);
+
+err_umem:
+	ib_umem_release(cq->buf.umem);
+	return err;
+}
+
+static void destroy_cq_user(struct mlx5_ib_cq *cq, struct ib_ucontext *context)
+{
+	mlx5_ib_db_unmap_user(to_mucontext(context), &cq->db);
+	ib_umem_release(cq->buf.umem);
+}
+
+static void init_cq_buf(struct mlx5_ib_cq *cq, struct mlx5_ib_cq_buf *buf)
+{
+	int i;
+	void *cqe;
+	struct mlx5_cqe64 *cqe64;
+
+	for (i = 0; i < buf->nent; i++) {
+		cqe = get_cqe_from_buf(buf, i, buf->cqe_size);
+		cqe64 = buf->cqe_size == 64 ? cqe : cqe + 64;
+		cqe64->op_own = MLX5_CQE_INVALID << 4;
+	}
+}
+
+static int create_cq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+			    int entries, int cqe_size,
+			    struct mlx5_create_cq_mbox_in **cqb,
+			    int *index, int *inlen)
+{
+	int err;
+
+	err = mlx5_db_alloc(dev->mdev, &cq->db);
+	if (err)
+		return err;
+
+	cq->mcq.set_ci_db  = cq->db.db;
+	cq->mcq.arm_db     = cq->db.db + 1;
+	cq->mcq.cqe_sz = cqe_size;
+
+	err = alloc_cq_buf(dev, &cq->buf, entries, cqe_size);
+	if (err)
+		goto err_db;
+
+	init_cq_buf(cq, &cq->buf);
+
+	*inlen = sizeof(**cqb) + sizeof(*(*cqb)->pas) * cq->buf.buf.npages;
+	*cqb = mlx5_vzalloc(*inlen);
+	if (!*cqb) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	mlx5_fill_page_array(&cq->buf.buf, (*cqb)->pas);
+
+	(*cqb)->ctx.log_pg_sz = cq->buf.buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	*index = dev->mdev->priv.uuari.uars[0].index;
+
+	return 0;
+
+err_buf:
+	free_cq_buf(dev, &cq->buf);
+
+err_db:
+	mlx5_db_free(dev->mdev, &cq->db);
+	return err;
+}
+
+static void destroy_cq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq)
+{
+	free_cq_buf(dev, &cq->buf);
+	mlx5_db_free(dev->mdev, &cq->db);
+}
+
+struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, int entries,
+				int vector, struct ib_ucontext *context,
+				struct ib_udata *udata)
+{
+	struct mlx5_create_cq_mbox_in *cqb = NULL;
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_cq *cq;
+	int uninitialized_var(index);
+	int uninitialized_var(inlen);
+	int cqe_size;
+	int irqn;
+	int eqn;
+	int err;
+
+	if (entries < 0)
+		return ERR_PTR(-EINVAL);
+
+	entries = roundup_pow_of_two(entries + 1);
+	if (entries > dev->mdev->caps.gen.max_cqes)
+		return ERR_PTR(-EINVAL);
+
+	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-ENOMEM);
+
+	cq->ibcq.cqe = entries - 1;
+	mutex_init(&cq->resize_mutex);
+	spin_lock_init(&cq->lock);
+	cq->resize_buf = NULL;
+	cq->resize_umem = NULL;
+
+	if (context) {
+		err = create_cq_user(dev, udata, context, cq, entries,
+				     &cqb, &cqe_size, &index, &inlen);
+		if (err)
+			goto err_create;
+	} else {
+		/* for now choose 64 bytes till we have a proper interface */
+		cqe_size = 64;
+		err = create_cq_kernel(dev, cq, entries, cqe_size, &cqb,
+				       &index, &inlen);
+		if (err)
+			goto err_create;
+	}
+
+	cq->cqe_size = cqe_size;
+	cqb->ctx.cqe_sz_flags = cqe_sz_to_mlx_sz(cqe_size) << 5;
+	cqb->ctx.log_sz_usr_page = cpu_to_be32((ilog2(entries) << 24) | index);
+	err = mlx5_vector2eqn(dev->mdev, vector, &eqn, &irqn);
+	if (err)
+		goto err_cqb;
+
+	cqb->ctx.c_eqn = cpu_to_be16(eqn);
+	cqb->ctx.db_record_addr = cpu_to_be64(cq->db.dma);
+
+	err = mlx5_core_create_cq(dev->mdev, &cq->mcq, cqb, inlen);
+	if (err)
+		goto err_cqb;
+
+	mlx5_ib_dbg(dev, "cqn 0x%x\n", cq->mcq.cqn);
+	cq->mcq.irqn = irqn;
+	cq->mcq.comp  = mlx5_ib_cq_comp;
+	cq->mcq.event = mlx5_ib_cq_event;
+
+	if (context)
+		if (ib_copy_to_udata(udata, &cq->mcq.cqn, sizeof(__u32))) {
+			err = -EFAULT;
+			goto err_cmd;
+		}
+
+
+	kvfree(cqb);
+	return &cq->ibcq;
+
+err_cmd:
+	mlx5_core_destroy_cq(dev->mdev, &cq->mcq);
+
+err_cqb:
+	kvfree(cqb);
+	if (context)
+		destroy_cq_user(cq, context);
+	else
+		destroy_cq_kernel(dev, cq);
+
+err_create:
+	kfree(cq);
+
+	return ERR_PTR(err);
+}
+
+
+int mlx5_ib_destroy_cq(struct ib_cq *cq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->device);
+	struct mlx5_ib_cq *mcq = to_mcq(cq);
+	struct ib_ucontext *context = NULL;
+
+	if (cq->uobject)
+		context = cq->uobject->context;
+
+	mlx5_core_destroy_cq(dev->mdev, &mcq->mcq);
+	if (context)
+		destroy_cq_user(mcq, context);
+	else
+		destroy_cq_kernel(dev, mcq);
+
+	kfree(mcq);
+
+	return 0;
+}
+
+static int is_equal_rsn(struct mlx5_cqe64 *cqe64, u32 rsn)
+{
+	return rsn == (ntohl(cqe64->sop_drop_qpn) & 0xffffff);
+}
+
+void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 rsn, struct mlx5_ib_srq *srq)
+{
+	struct mlx5_cqe64 *cqe64, *dest64;
+	void *cqe, *dest;
+	u32 prod_index;
+	int nfreed = 0;
+	u8 owner_bit;
+
+	if (!cq)
+		return;
+
+	/* First we need to find the current producer index, so we
+	 * know where to start cleaning from.  It doesn't matter if HW
+	 * adds new entries after this loop -- the QP we're worried
+	 * about is already in RESET, so the new entries won't come
+	 * from our QP and therefore don't need to be checked.
+	 */
+	for (prod_index = cq->mcq.cons_index; get_sw_cqe(cq, prod_index); prod_index++)
+		if (prod_index == cq->mcq.cons_index + cq->ibcq.cqe)
+			break;
+
+	/* Now sweep backwards through the CQ, removing CQ entries
+	 * that match our QP by copying older entries on top of them.
+	 */
+	while ((int) --prod_index - (int) cq->mcq.cons_index >= 0) {
+		cqe = get_cqe(cq, prod_index & cq->ibcq.cqe);
+		cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+		if (is_equal_rsn(cqe64, rsn)) {
+			if (srq && (ntohl(cqe64->srqn) & 0xffffff))
+				mlx5_ib_free_srq_wqe(srq, be16_to_cpu(cqe64->wqe_counter));
+			++nfreed;
+		} else if (nfreed) {
+			dest = get_cqe(cq, (prod_index + nfreed) & cq->ibcq.cqe);
+			dest64 = (cq->mcq.cqe_sz == 64) ? dest : dest + 64;
+			owner_bit = dest64->op_own & MLX5_CQE_OWNER_MASK;
+			memcpy(dest, cqe, cq->mcq.cqe_sz);
+			dest64->op_own = owner_bit |
+				(dest64->op_own & ~MLX5_CQE_OWNER_MASK);
+		}
+	}
+
+	if (nfreed) {
+		cq->mcq.cons_index += nfreed;
+		/* Make sure update of buffer contents is done before
+		 * updating consumer index.
+		 */
+		wmb();
+		mlx5_cq_set_ci(&cq->mcq);
+	}
+}
+
+void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq)
+{
+	if (!cq)
+		return;
+
+	spin_lock_irq(&cq->lock);
+	__mlx5_ib_cq_clean(cq, qpn, srq);
+	spin_unlock_irq(&cq->lock);
+}
+
+int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
+{
+	struct mlx5_modify_cq_mbox_in *in;
+	struct mlx5_ib_dev *dev = to_mdev(cq->device);
+	struct mlx5_ib_cq *mcq = to_mcq(cq);
+	int err;
+	u32 fsel;
+
+	if (!(dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_CQ_MODER))
+		return -ENOSYS;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	in->cqn = cpu_to_be32(mcq->mcq.cqn);
+	fsel = (MLX5_CQ_MODIFY_PERIOD | MLX5_CQ_MODIFY_COUNT);
+	in->ctx.cq_period = cpu_to_be16(cq_period);
+	in->ctx.cq_max_count = cpu_to_be16(cq_count);
+	in->field_select = cpu_to_be32(fsel);
+	err = mlx5_core_modify_cq(dev->mdev, &mcq->mcq, in, sizeof(*in));
+	kfree(in);
+
+	if (err)
+		mlx5_ib_warn(dev, "modify cq 0x%x failed\n", mcq->mcq.cqn);
+
+	return err;
+}
+
+static int resize_user(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+		       int entries, struct ib_udata *udata, int *npas,
+		       int *page_shift, int *cqe_size)
+{
+	struct mlx5_ib_resize_cq ucmd;
+	struct ib_umem *umem;
+	int err;
+	int npages;
+	struct ib_ucontext *context = cq->buf.umem->context;
+
+	err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
+	if (err)
+		return err;
+
+	if (ucmd.reserved0 || ucmd.reserved1)
+		return -EINVAL;
+
+	umem = ib_umem_get(context, ucmd.buf_addr, entries * ucmd.cqe_size,
+			   IB_ACCESS_LOCAL_WRITE, 1);
+	if (IS_ERR(umem)) {
+		err = PTR_ERR(umem);
+		return err;
+	}
+
+	mlx5_ib_cont_pages(umem, ucmd.buf_addr, &npages, page_shift,
+			   npas, NULL);
+
+	cq->resize_umem = umem;
+	*cqe_size = ucmd.cqe_size;
+
+	return 0;
+}
+
+static void un_resize_user(struct mlx5_ib_cq *cq)
+{
+	ib_umem_release(cq->resize_umem);
+}
+
+static int resize_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+			 int entries, int cqe_size)
+{
+	int err;
+
+	cq->resize_buf = kzalloc(sizeof(*cq->resize_buf), GFP_KERNEL);
+	if (!cq->resize_buf)
+		return -ENOMEM;
+
+	err = alloc_cq_buf(dev, cq->resize_buf, entries, cqe_size);
+	if (err)
+		goto ex;
+
+	init_cq_buf(cq, cq->resize_buf);
+
+	return 0;
+
+ex:
+	kfree(cq->resize_buf);
+	return err;
+}
+
+static void un_resize_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq)
+{
+	free_cq_buf(dev, cq->resize_buf);
+	cq->resize_buf = NULL;
+}
+
+static int copy_resize_cqes(struct mlx5_ib_cq *cq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct mlx5_cqe64 *scqe64;
+	struct mlx5_cqe64 *dcqe64;
+	void *start_cqe;
+	void *scqe;
+	void *dcqe;
+	int ssize;
+	int dsize;
+	int i;
+	u8 sw_own;
+
+	ssize = cq->buf.cqe_size;
+	dsize = cq->resize_buf->cqe_size;
+	if (ssize != dsize) {
+		mlx5_ib_warn(dev, "resize from different cqe size is not supported\n");
+		return -EINVAL;
+	}
+
+	i = cq->mcq.cons_index;
+	scqe = get_sw_cqe(cq, i);
+	scqe64 = ssize == 64 ? scqe : scqe + 64;
+	start_cqe = scqe;
+	if (!scqe) {
+		mlx5_ib_warn(dev, "expected cqe in sw ownership\n");
+		return -EINVAL;
+	}
+
+	while ((scqe64->op_own >> 4) != MLX5_CQE_RESIZE_CQ) {
+		dcqe = get_cqe_from_buf(cq->resize_buf,
+					(i + 1) & (cq->resize_buf->nent),
+					dsize);
+		dcqe64 = dsize == 64 ? dcqe : dcqe + 64;
+		sw_own = sw_ownership_bit(i + 1, cq->resize_buf->nent);
+		memcpy(dcqe, scqe, dsize);
+		dcqe64->op_own = (dcqe64->op_own & ~MLX5_CQE_OWNER_MASK) | sw_own;
+
+		++i;
+		scqe = get_sw_cqe(cq, i);
+		scqe64 = ssize == 64 ? scqe : scqe + 64;
+		if (!scqe) {
+			mlx5_ib_warn(dev, "expected cqe in sw ownership\n");
+			return -EINVAL;
+		}
+
+		if (scqe == start_cqe) {
+			pr_warn("resize CQ failed to get resize CQE, CQN 0x%x\n",
+				cq->mcq.cqn);
+			return -ENOMEM;
+		}
+	}
+	++cq->mcq.cons_index;
+	return 0;
+}
+
+int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibcq->device);
+	struct mlx5_ib_cq *cq = to_mcq(ibcq);
+	struct mlx5_modify_cq_mbox_in *in;
+	int err;
+	int npas;
+	int page_shift;
+	int inlen;
+	int uninitialized_var(cqe_size);
+	unsigned long flags;
+
+	if (!(dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_RESIZE_CQ)) {
+		pr_info("Firmware does not support resize CQ\n");
+		return -ENOSYS;
+	}
+
+	if (entries < 1)
+		return -EINVAL;
+
+	entries = roundup_pow_of_two(entries + 1);
+	if (entries > dev->mdev->caps.gen.max_cqes + 1)
+		return -EINVAL;
+
+	if (entries == ibcq->cqe + 1)
+		return 0;
+
+	mutex_lock(&cq->resize_mutex);
+	if (udata) {
+		err = resize_user(dev, cq, entries, udata, &npas, &page_shift,
+				  &cqe_size);
+	} else {
+		cqe_size = 64;
+		err = resize_kernel(dev, cq, entries, cqe_size);
+		if (!err) {
+			npas = cq->resize_buf->buf.npages;
+			page_shift = cq->resize_buf->buf.page_shift;
+		}
+	}
+
+	if (err)
+		goto ex;
+
+	inlen = sizeof(*in) + npas * sizeof(in->pas[0]);
+	in = mlx5_vzalloc(inlen);
+	if (!in) {
+		err = -ENOMEM;
+		goto ex_resize;
+	}
+
+	if (udata)
+		mlx5_ib_populate_pas(dev, cq->resize_umem, page_shift,
+				     in->pas, 0);
+	else
+		mlx5_fill_page_array(&cq->resize_buf->buf, in->pas);
+
+	in->field_select = cpu_to_be32(MLX5_MODIFY_CQ_MASK_LOG_SIZE  |
+				       MLX5_MODIFY_CQ_MASK_PG_OFFSET |
+				       MLX5_MODIFY_CQ_MASK_PG_SIZE);
+	in->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	in->ctx.cqe_sz_flags = cqe_sz_to_mlx_sz(cqe_size) << 5;
+	in->ctx.page_offset = 0;
+	in->ctx.log_sz_usr_page = cpu_to_be32(ilog2(entries) << 24);
+	in->hdr.opmod = cpu_to_be16(MLX5_CQ_OPMOD_RESIZE);
+	in->cqn = cpu_to_be32(cq->mcq.cqn);
+
+	err = mlx5_core_modify_cq(dev->mdev, &cq->mcq, in, inlen);
+	if (err)
+		goto ex_alloc;
+
+	if (udata) {
+		cq->ibcq.cqe = entries - 1;
+		ib_umem_release(cq->buf.umem);
+		cq->buf.umem = cq->resize_umem;
+		cq->resize_umem = NULL;
+	} else {
+		struct mlx5_ib_cq_buf tbuf;
+		int resized = 0;
+
+		spin_lock_irqsave(&cq->lock, flags);
+		if (cq->resize_buf) {
+			err = copy_resize_cqes(cq);
+			if (!err) {
+				tbuf = cq->buf;
+				cq->buf = *cq->resize_buf;
+				kfree(cq->resize_buf);
+				cq->resize_buf = NULL;
+				resized = 1;
+			}
+		}
+		cq->ibcq.cqe = entries - 1;
+		spin_unlock_irqrestore(&cq->lock, flags);
+		if (resized)
+			free_cq_buf(dev, &tbuf);
+	}
+	mutex_unlock(&cq->resize_mutex);
+
+	kvfree(in);
+	return 0;
+
+ex_alloc:
+	kvfree(in);
+
+ex_resize:
+	if (udata)
+		un_resize_user(cq);
+	else
+		un_resize_kernel(dev, cq);
+ex:
+	mutex_unlock(&cq->resize_mutex);
+	return err;
+}
+
+int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq)
+{
+	struct mlx5_ib_cq *cq;
+
+	if (!ibcq)
+		return 128;
+
+	cq = to_mcq(ibcq);
+	return cq->cqe_size;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/doorbell.c b/kernel/drivers/infiniband/hw/mlx5/doorbell.c
new file mode 100644
index 000000000..a0e4e6ddb
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/doorbell.c
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/kref.h>
+#include <linux/slab.h>
+#include <rdma/ib_umem.h>
+
+#include "mlx5_ib.h"
+
+struct mlx5_ib_user_db_page {
+	struct list_head	list;
+	struct ib_umem	       *umem;
+	unsigned long		user_virt;
+	int			refcnt;
+};
+
+int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
+			struct mlx5_db *db)
+{
+	struct mlx5_ib_user_db_page *page;
+	int err = 0;
+
+	mutex_lock(&context->db_page_mutex);
+
+	list_for_each_entry(page, &context->db_page_list, list)
+		if (page->user_virt == (virt & PAGE_MASK))
+			goto found;
+
+	page = kmalloc(sizeof(*page), GFP_KERNEL);
+	if (!page) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	page->user_virt = (virt & PAGE_MASK);
+	page->refcnt    = 0;
+	page->umem      = ib_umem_get(&context->ibucontext, virt & PAGE_MASK,
+				      PAGE_SIZE, 0, 0);
+	if (IS_ERR(page->umem)) {
+		err = PTR_ERR(page->umem);
+		kfree(page);
+		goto out;
+	}
+
+	list_add(&page->list, &context->db_page_list);
+
+found:
+	db->dma = sg_dma_address(page->umem->sg_head.sgl) + (virt & ~PAGE_MASK);
+	db->u.user_page = page;
+	++page->refcnt;
+
+out:
+	mutex_unlock(&context->db_page_mutex);
+
+	return err;
+}
+
+void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db)
+{
+	mutex_lock(&context->db_page_mutex);
+
+	if (!--db->u.user_page->refcnt) {
+		list_del(&db->u.user_page->list);
+		ib_umem_release(db->u.user_page->umem);
+		kfree(db->u.user_page);
+	}
+
+	mutex_unlock(&context->db_page_mutex);
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/mad.c b/kernel/drivers/infiniband/hw/mlx5/mad.c
new file mode 100644
index 000000000..9cf9a37bb
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/mad.c
@@ -0,0 +1,139 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mlx5/cmd.h>
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include "mlx5_ib.h"
+
+enum {
+	MLX5_IB_VENDOR_CLASS1 = 0x9,
+	MLX5_IB_VENDOR_CLASS2 = 0xa
+};
+
+int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+		 u8 port, struct ib_wc *in_wc, struct ib_grh *in_grh,
+		 void *in_mad, void *response_mad)
+{
+	u8 op_modifier = 0;
+
+	/* Key check traps can't be generated unless we have in_wc to
+	 * tell us where to send the trap.
+	 */
+	if (ignore_mkey || !in_wc)
+		op_modifier |= 0x1;
+	if (ignore_bkey || !in_wc)
+		op_modifier |= 0x2;
+
+	return mlx5_core_mad_ifc(dev->mdev, in_mad, response_mad, op_modifier, port);
+}
+
+int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	u16 slid;
+	int err;
+
+	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0)
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
+			return IB_MAD_RESULT_SUCCESS;
+
+		/* Don't process SMInfo queries -- the SMA can't handle them.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO)
+			return IB_MAD_RESULT_SUCCESS;
+	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
+		   in_mad->mad_hdr.mgmt_class == MLX5_IB_VENDOR_CLASS1   ||
+		   in_mad->mad_hdr.mgmt_class == MLX5_IB_VENDOR_CLASS2   ||
+		   in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CONG_MGMT) {
+		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
+			return IB_MAD_RESULT_SUCCESS;
+	} else {
+		return IB_MAD_RESULT_SUCCESS;
+	}
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev),
+			   mad_flags & IB_MAD_IGNORE_MKEY,
+			   mad_flags & IB_MAD_IGNORE_BKEY,
+			   port_num, in_wc, in_grh, in_mad, out_mad);
+	if (err)
+		return IB_MAD_RESULT_FAILURE;
+
+	/* set return bit in status of directed route responses */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
+		/* no response for trap repress */
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u16 packet_error;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+
+	packet_error = be16_to_cpu(out_mad->status);
+
+	dev->mdev->caps.gen.ext_port_cap[port - 1] = (!err && !packet_error) ?
+		MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO : 0;
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/main.c b/kernel/drivers/infiniband/hw/mlx5/main.c
new file mode 100644
index 000000000..57c9809e8
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/main.c
@@ -0,0 +1,1397 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <asm-generic/kmap_types.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/io-mapping.h>
+#include <linux/sched.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_umem.h>
+#include "user.h"
+#include "mlx5_ib.h"
+
+#define DRIVER_NAME "mlx5_ib"
+#define DRIVER_VERSION "2.2-1"
+#define DRIVER_RELDATE	"Feb 2014"
+
+MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
+MODULE_DESCRIPTION("Mellanox Connect-IB HCA IB driver");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRIVER_VERSION);
+
+static int deprecated_prof_sel = 2;
+module_param_named(prof_sel, deprecated_prof_sel, int, 0444);
+MODULE_PARM_DESC(prof_sel, "profile selector. Deprecated here. Moved to module mlx5_core");
+
+static char mlx5_version[] =
+	DRIVER_NAME ": Mellanox Connect-IB Infiniband driver v"
+	DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
+
+static int mlx5_ib_query_device(struct ib_device *ibdev,
+				struct ib_device_attr *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	struct mlx5_general_caps *gen;
+	int err = -ENOMEM;
+	int max_rq_sg;
+	int max_sq_sg;
+	u64 flags;
+
+	gen = &dev->mdev->caps.gen;
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memset(props, 0, sizeof(*props));
+
+	props->fw_ver = ((u64)fw_rev_maj(dev->mdev) << 32) |
+		(fw_rev_min(dev->mdev) << 16) |
+		fw_rev_sub(dev->mdev);
+	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
+		IB_DEVICE_PORT_ACTIVE_EVENT		|
+		IB_DEVICE_SYS_IMAGE_GUID		|
+		IB_DEVICE_RC_RNR_NAK_GEN;
+	flags = gen->flags;
+	if (flags & MLX5_DEV_CAP_FLAG_BAD_PKEY_CNTR)
+		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
+	if (flags & MLX5_DEV_CAP_FLAG_BAD_QKEY_CNTR)
+		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
+	if (flags & MLX5_DEV_CAP_FLAG_APM)
+		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
+	props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
+	if (flags & MLX5_DEV_CAP_FLAG_XRC)
+		props->device_cap_flags |= IB_DEVICE_XRC;
+	props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	if (flags & MLX5_DEV_CAP_FLAG_SIG_HAND_OVER) {
+		props->device_cap_flags |= IB_DEVICE_SIGNATURE_HANDOVER;
+		/* At this stage no support for signature handover */
+		props->sig_prot_cap = IB_PROT_T10DIF_TYPE_1 |
+				      IB_PROT_T10DIF_TYPE_2 |
+				      IB_PROT_T10DIF_TYPE_3;
+		props->sig_guard_cap = IB_GUARD_T10DIF_CRC |
+				       IB_GUARD_T10DIF_CSUM;
+	}
+	if (flags & MLX5_DEV_CAP_FLAG_BLOCK_MCAST)
+		props->device_cap_flags |= IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
+
+	props->vendor_id	   = be32_to_cpup((__be32 *)(out_mad->data + 36)) &
+		0xffffff;
+	props->vendor_part_id	   = be16_to_cpup((__be16 *)(out_mad->data + 30));
+	props->hw_ver		   = be32_to_cpup((__be32 *)(out_mad->data + 32));
+	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
+
+	props->max_mr_size	   = ~0ull;
+	props->page_size_cap	   = gen->min_page_sz;
+	props->max_qp		   = 1 << gen->log_max_qp;
+	props->max_qp_wr	   = gen->max_wqes;
+	max_rq_sg = gen->max_rq_desc_sz / sizeof(struct mlx5_wqe_data_seg);
+	max_sq_sg = (gen->max_sq_desc_sz - sizeof(struct mlx5_wqe_ctrl_seg)) /
+		sizeof(struct mlx5_wqe_data_seg);
+	props->max_sge = min(max_rq_sg, max_sq_sg);
+	props->max_cq		   = 1 << gen->log_max_cq;
+	props->max_cqe		   = gen->max_cqes - 1;
+	props->max_mr		   = 1 << gen->log_max_mkey;
+	props->max_pd		   = 1 << gen->log_max_pd;
+	props->max_qp_rd_atom	   = 1 << gen->log_max_ra_req_qp;
+	props->max_qp_init_rd_atom = 1 << gen->log_max_ra_res_qp;
+	props->max_srq		   = 1 << gen->log_max_srq;
+	props->max_srq_wr	   = gen->max_srq_wqes - 1;
+	props->local_ca_ack_delay  = gen->local_ca_ack_delay;
+	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
+	props->max_srq_sge	   = max_rq_sg - 1;
+	props->max_fast_reg_page_list_len = (unsigned int)-1;
+	props->local_ca_ack_delay  = gen->local_ca_ack_delay;
+	props->atomic_cap	   = IB_ATOMIC_NONE;
+	props->masked_atomic_cap   = IB_ATOMIC_NONE;
+	props->max_pkeys	   = be16_to_cpup((__be16 *)(out_mad->data + 28));
+	props->max_mcast_grp	   = 1 << gen->log_max_mcg;
+	props->max_mcast_qp_attach = gen->max_qp_mcg;
+	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
+					   props->max_mcast_grp;
+	props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	if (dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_ON_DMND_PG)
+		props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING;
+	props->odp_caps = dev->odp_caps;
+#endif
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+
+	return err;
+}
+
+int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
+		       struct ib_port_attr *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	struct mlx5_general_caps *gen;
+	int ext_active_speed;
+	int err = -ENOMEM;
+
+	gen = &dev->mdev->caps.gen;
+	if (port < 1 || port > gen->num_ports) {
+		mlx5_ib_warn(dev, "invalid port number %d\n", port);
+		return -EINVAL;
+	}
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(props, 0, sizeof(*props));
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(dev, 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err) {
+		mlx5_ib_warn(dev, "err %d\n", err);
+		goto out;
+	}
+
+
+	props->lid		= be16_to_cpup((__be16 *)(out_mad->data + 16));
+	props->lmc		= out_mad->data[34] & 0x7;
+	props->sm_lid		= be16_to_cpup((__be16 *)(out_mad->data + 18));
+	props->sm_sl		= out_mad->data[36] & 0xf;
+	props->state		= out_mad->data[32] & 0xf;
+	props->phys_state	= out_mad->data[33] >> 4;
+	props->port_cap_flags	= be32_to_cpup((__be32 *)(out_mad->data + 20));
+	props->gid_tbl_len	= out_mad->data[50];
+	props->max_msg_sz	= 1 << gen->log_max_msg;
+	props->pkey_tbl_len	= gen->port[port - 1].pkey_table_len;
+	props->bad_pkey_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 46));
+	props->qkey_viol_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 48));
+	props->active_width	= out_mad->data[31] & 0xf;
+	props->active_speed	= out_mad->data[35] >> 4;
+	props->max_mtu		= out_mad->data[41] & 0xf;
+	props->active_mtu	= out_mad->data[36] >> 4;
+	props->subnet_timeout	= out_mad->data[51] & 0x1f;
+	props->max_vl_num	= out_mad->data[37] >> 4;
+	props->init_type_reply	= out_mad->data[41] >> 4;
+
+	/* Check if extended speeds (EDR/FDR/...) are supported */
+	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
+		ext_active_speed = out_mad->data[62] >> 4;
+
+		switch (ext_active_speed) {
+		case 1:
+			props->active_speed = 16; /* FDR */
+			break;
+		case 2:
+			props->active_speed = 32; /* EDR */
+			break;
+		}
+	}
+
+	/* If reported active speed is QDR, check if is FDR-10 */
+	if (props->active_speed == 4) {
+		if (gen->ext_port_cap[port - 1] &
+		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
+			init_query_mad(in_mad);
+			in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
+			in_mad->attr_mod = cpu_to_be32(port);
+
+			err = mlx5_MAD_IFC(dev, 1, 1, port,
+					   NULL, NULL, in_mad, out_mad);
+			if (err)
+				goto out;
+
+			/* Checking LinkSpeedActive for FDR-10 */
+			if (out_mad->data[15] & 0x1)
+				props->active_speed = 8;
+		}
+	}
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+
+	return err;
+}
+
+static int mlx5_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+			     union ib_gid *gid)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(gid->raw, out_mad->data + 8, 8);
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
+	in_mad->attr_mod = cpu_to_be32(index / 8);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mlx5_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			      u16 *pkey)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
+	in_mad->attr_mod = cpu_to_be32(index / 32);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	*pkey = be16_to_cpu(((__be16 *)out_mad->data)[index % 32]);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+struct mlx5_reg_node_desc {
+	u8	desc[64];
+};
+
+static int mlx5_ib_modify_device(struct ib_device *ibdev, int mask,
+				 struct ib_device_modify *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_reg_node_desc in;
+	struct mlx5_reg_node_desc out;
+	int err;
+
+	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
+		return -EOPNOTSUPP;
+
+	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
+		return 0;
+
+	/*
+	 * If possible, pass node desc to FW, so it can generate
+	 * a 144 trap.  If cmd fails, just ignore.
+	 */
+	memcpy(&in, props->node_desc, 64);
+	err = mlx5_core_access_reg(dev->mdev, &in, sizeof(in), &out,
+				   sizeof(out), MLX5_REG_NODE_DESC, 0, 1);
+	if (err)
+		return err;
+
+	memcpy(ibdev->node_desc, props->node_desc, 64);
+
+	return err;
+}
+
+static int mlx5_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
+			       struct ib_port_modify *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_port_attr attr;
+	u32 tmp;
+	int err;
+
+	mutex_lock(&dev->cap_mask_mutex);
+
+	err = mlx5_ib_query_port(ibdev, port, &attr);
+	if (err)
+		goto out;
+
+	tmp = (attr.port_cap_flags | props->set_port_cap_mask) &
+		~props->clr_port_cap_mask;
+
+	err = mlx5_set_port_caps(dev->mdev, port, tmp);
+
+out:
+	mutex_unlock(&dev->cap_mask_mutex);
+	return err;
+}
+
+static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
+						  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_alloc_ucontext_req_v2 req;
+	struct mlx5_ib_alloc_ucontext_resp resp;
+	struct mlx5_ib_ucontext *context;
+	struct mlx5_general_caps *gen;
+	struct mlx5_uuar_info *uuari;
+	struct mlx5_uar *uars;
+	int gross_uuars;
+	int num_uars;
+	int ver;
+	int uuarn;
+	int err;
+	int i;
+	size_t reqlen;
+
+	gen = &dev->mdev->caps.gen;
+	if (!dev->ib_active)
+		return ERR_PTR(-EAGAIN);
+
+	memset(&req, 0, sizeof(req));
+	reqlen = udata->inlen - sizeof(struct ib_uverbs_cmd_hdr);
+	if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req))
+		ver = 0;
+	else if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req_v2))
+		ver = 2;
+	else
+		return ERR_PTR(-EINVAL);
+
+	err = ib_copy_from_udata(&req, udata, reqlen);
+	if (err)
+		return ERR_PTR(err);
+
+	if (req.flags || req.reserved)
+		return ERR_PTR(-EINVAL);
+
+	if (req.total_num_uuars > MLX5_MAX_UUARS)
+		return ERR_PTR(-ENOMEM);
+
+	if (req.total_num_uuars == 0)
+		return ERR_PTR(-EINVAL);
+
+	req.total_num_uuars = ALIGN(req.total_num_uuars,
+				    MLX5_NON_FP_BF_REGS_PER_PAGE);
+	if (req.num_low_latency_uuars > req.total_num_uuars - 1)
+		return ERR_PTR(-EINVAL);
+
+	num_uars = req.total_num_uuars / MLX5_NON_FP_BF_REGS_PER_PAGE;
+	gross_uuars = num_uars * MLX5_BF_REGS_PER_PAGE;
+	resp.qp_tab_size      = 1 << gen->log_max_qp;
+	resp.bf_reg_size      = gen->bf_reg_size;
+	resp.cache_line_size  = L1_CACHE_BYTES;
+	resp.max_sq_desc_sz = gen->max_sq_desc_sz;
+	resp.max_rq_desc_sz = gen->max_rq_desc_sz;
+	resp.max_send_wqebb = gen->max_wqes;
+	resp.max_recv_wr = gen->max_wqes;
+	resp.max_srq_recv_wr = gen->max_srq_wqes;
+
+	context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return ERR_PTR(-ENOMEM);
+
+	uuari = &context->uuari;
+	mutex_init(&uuari->lock);
+	uars = kcalloc(num_uars, sizeof(*uars), GFP_KERNEL);
+	if (!uars) {
+		err = -ENOMEM;
+		goto out_ctx;
+	}
+
+	uuari->bitmap = kcalloc(BITS_TO_LONGS(gross_uuars),
+				sizeof(*uuari->bitmap),
+				GFP_KERNEL);
+	if (!uuari->bitmap) {
+		err = -ENOMEM;
+		goto out_uar_ctx;
+	}
+	/*
+	 * clear all fast path uuars
+	 */
+	for (i = 0; i < gross_uuars; i++) {
+		uuarn = i & 3;
+		if (uuarn == 2 || uuarn == 3)
+			set_bit(i, uuari->bitmap);
+	}
+
+	uuari->count = kcalloc(gross_uuars, sizeof(*uuari->count), GFP_KERNEL);
+	if (!uuari->count) {
+		err = -ENOMEM;
+		goto out_bitmap;
+	}
+
+	for (i = 0; i < num_uars; i++) {
+		err = mlx5_cmd_alloc_uar(dev->mdev, &uars[i].index);
+		if (err)
+			goto out_count;
+	}
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range;
+#endif
+
+	INIT_LIST_HEAD(&context->db_page_list);
+	mutex_init(&context->db_page_mutex);
+
+	resp.tot_uuars = req.total_num_uuars;
+	resp.num_ports = gen->num_ports;
+	err = ib_copy_to_udata(udata, &resp,
+			       sizeof(resp) - sizeof(resp.reserved));
+	if (err)
+		goto out_uars;
+
+	uuari->ver = ver;
+	uuari->num_low_latency_uuars = req.num_low_latency_uuars;
+	uuari->uars = uars;
+	uuari->num_uars = num_uars;
+	return &context->ibucontext;
+
+out_uars:
+	for (i--; i >= 0; i--)
+		mlx5_cmd_free_uar(dev->mdev, uars[i].index);
+out_count:
+	kfree(uuari->count);
+
+out_bitmap:
+	kfree(uuari->bitmap);
+
+out_uar_ctx:
+	kfree(uars);
+
+out_ctx:
+	kfree(context);
+	return ERR_PTR(err);
+}
+
+static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
+{
+	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
+	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
+	struct mlx5_uuar_info *uuari = &context->uuari;
+	int i;
+
+	for (i = 0; i < uuari->num_uars; i++) {
+		if (mlx5_cmd_free_uar(dev->mdev, uuari->uars[i].index))
+			mlx5_ib_warn(dev, "failed to free UAR 0x%x\n", uuari->uars[i].index);
+	}
+
+	kfree(uuari->count);
+	kfree(uuari->bitmap);
+	kfree(uuari->uars);
+	kfree(context);
+
+	return 0;
+}
+
+static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, int index)
+{
+	return (pci_resource_start(dev->mdev->pdev, 0) >> PAGE_SHIFT) + index;
+}
+
+static int get_command(unsigned long offset)
+{
+	return (offset >> MLX5_IB_MMAP_CMD_SHIFT) & MLX5_IB_MMAP_CMD_MASK;
+}
+
+static int get_arg(unsigned long offset)
+{
+	return offset & ((1 << MLX5_IB_MMAP_CMD_SHIFT) - 1);
+}
+
+static int get_index(unsigned long offset)
+{
+	return get_arg(offset);
+}
+
+static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
+{
+	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
+	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
+	struct mlx5_uuar_info *uuari = &context->uuari;
+	unsigned long command;
+	unsigned long idx;
+	phys_addr_t pfn;
+
+	command = get_command(vma->vm_pgoff);
+	switch (command) {
+	case MLX5_IB_MMAP_REGULAR_PAGE:
+		if (vma->vm_end - vma->vm_start != PAGE_SIZE)
+			return -EINVAL;
+
+		idx = get_index(vma->vm_pgoff);
+		if (idx >= uuari->num_uars)
+			return -EINVAL;
+
+		pfn = uar_index2pfn(dev, uuari->uars[idx].index);
+		mlx5_ib_dbg(dev, "uar idx 0x%lx, pfn 0x%llx\n", idx,
+			    (unsigned long long)pfn);
+
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+		if (io_remap_pfn_range(vma, vma->vm_start, pfn,
+				       PAGE_SIZE, vma->vm_page_prot))
+			return -EAGAIN;
+
+		mlx5_ib_dbg(dev, "mapped WC at 0x%lx, PA 0x%llx\n",
+			    vma->vm_start,
+			    (unsigned long long)pfn << PAGE_SHIFT);
+		break;
+
+	case MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES:
+		return -ENOSYS;
+
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int alloc_pa_mkey(struct mlx5_ib_dev *dev, u32 *key, u32 pdn)
+{
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_mkey_seg *seg;
+	struct mlx5_core_mr mr;
+	int err;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	seg = &in->seg;
+	seg->flags = MLX5_PERM_LOCAL_READ | MLX5_ACCESS_MODE_PA;
+	seg->flags_pd = cpu_to_be32(pdn | MLX5_MKEY_LEN64);
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	seg->start_addr = 0;
+
+	err = mlx5_core_create_mkey(dev->mdev, &mr, in, sizeof(*in),
+				    NULL, NULL, NULL);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to create mkey, %d\n", err);
+		goto err_in;
+	}
+
+	kfree(in);
+	*key = mr.key;
+
+	return 0;
+
+err_in:
+	kfree(in);
+
+	return err;
+}
+
+static void free_pa_mkey(struct mlx5_ib_dev *dev, u32 key)
+{
+	struct mlx5_core_mr mr;
+	int err;
+
+	memset(&mr, 0, sizeof(mr));
+	mr.key = key;
+	err = mlx5_core_destroy_mkey(dev->mdev, &mr);
+	if (err)
+		mlx5_ib_warn(dev, "failed to destroy mkey 0x%x\n", key);
+}
+
+static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev,
+				      struct ib_ucontext *context,
+				      struct ib_udata *udata)
+{
+	struct mlx5_ib_alloc_pd_resp resp;
+	struct mlx5_ib_pd *pd;
+	int err;
+
+	pd = kmalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx5_core_alloc_pd(to_mdev(ibdev)->mdev, &pd->pdn);
+	if (err) {
+		kfree(pd);
+		return ERR_PTR(err);
+	}
+
+	if (context) {
+		resp.pdn = pd->pdn;
+		if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
+			mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
+			kfree(pd);
+			return ERR_PTR(-EFAULT);
+		}
+	} else {
+		err = alloc_pa_mkey(to_mdev(ibdev), &pd->pa_lkey, pd->pdn);
+		if (err) {
+			mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
+			kfree(pd);
+			return ERR_PTR(err);
+		}
+	}
+
+	return &pd->ibpd;
+}
+
+static int mlx5_ib_dealloc_pd(struct ib_pd *pd)
+{
+	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
+	struct mlx5_ib_pd *mpd = to_mpd(pd);
+
+	if (!pd->uobject)
+		free_pa_mkey(mdev, mpd->pa_lkey);
+
+	mlx5_core_dealloc_pd(mdev->mdev, mpd->pdn);
+	kfree(mpd);
+
+	return 0;
+}
+
+static int mlx5_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	int err;
+
+	err = mlx5_core_attach_mcg(dev->mdev, gid, ibqp->qp_num);
+	if (err)
+		mlx5_ib_warn(dev, "failed attaching QPN 0x%x, MGID %pI6\n",
+			     ibqp->qp_num, gid->raw);
+
+	return err;
+}
+
+static int mlx5_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	int err;
+
+	err = mlx5_core_detach_mcg(dev->mdev, gid, ibqp->qp_num);
+	if (err)
+		mlx5_ib_warn(dev, "failed detaching QPN 0x%x, MGID %pI6\n",
+			     ibqp->qp_num, gid->raw);
+
+	return err;
+}
+
+static int init_node_data(struct mlx5_ib_dev *dev)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
+
+	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	dev->mdev->rev_id = be32_to_cpup((__be32 *)(out_mad->data + 32));
+	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static ssize_t show_fw_pages(struct device *device, struct device_attribute *attr,
+			     char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+
+	return sprintf(buf, "%d\n", dev->mdev->priv.fw_pages);
+}
+
+static ssize_t show_reg_pages(struct device *device,
+			      struct device_attribute *attr, char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+
+	return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages));
+}
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "MT%d\n", dev->mdev->pdev->device);
+}
+
+static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
+			   char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%d.%d.%d\n", fw_rev_maj(dev->mdev),
+		       fw_rev_min(dev->mdev), fw_rev_sub(dev->mdev));
+}
+
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%x\n", dev->mdev->rev_id);
+}
+
+static ssize_t show_board(struct device *device, struct device_attribute *attr,
+			  char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN,
+		       dev->mdev->board_id);
+}
+
+static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
+static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
+static DEVICE_ATTR(fw_pages, S_IRUGO, show_fw_pages, NULL);
+static DEVICE_ATTR(reg_pages, S_IRUGO, show_reg_pages, NULL);
+
+static struct device_attribute *mlx5_class_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_fw_ver,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+	&dev_attr_fw_pages,
+	&dev_attr_reg_pages,
+};
+
+static void mlx5_ib_event(struct mlx5_core_dev *dev, void *context,
+			  enum mlx5_dev_event event, unsigned long param)
+{
+	struct mlx5_ib_dev *ibdev = (struct mlx5_ib_dev *)context;
+	struct ib_event ibev;
+
+	u8 port = 0;
+
+	switch (event) {
+	case MLX5_DEV_EVENT_SYS_ERROR:
+		ibdev->ib_active = false;
+		ibev.event = IB_EVENT_DEVICE_FATAL;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_UP:
+		ibev.event = IB_EVENT_PORT_ACTIVE;
+		port = (u8)param;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_DOWN:
+		ibev.event = IB_EVENT_PORT_ERR;
+		port = (u8)param;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_INITIALIZED:
+		/* not used by ULPs */
+		return;
+
+	case MLX5_DEV_EVENT_LID_CHANGE:
+		ibev.event = IB_EVENT_LID_CHANGE;
+		port = (u8)param;
+		break;
+
+	case MLX5_DEV_EVENT_PKEY_CHANGE:
+		ibev.event = IB_EVENT_PKEY_CHANGE;
+		port = (u8)param;
+		break;
+
+	case MLX5_DEV_EVENT_GUID_CHANGE:
+		ibev.event = IB_EVENT_GID_CHANGE;
+		port = (u8)param;
+		break;
+
+	case MLX5_DEV_EVENT_CLIENT_REREG:
+		ibev.event = IB_EVENT_CLIENT_REREGISTER;
+		port = (u8)param;
+		break;
+	}
+
+	ibev.device	      = &ibdev->ib_dev;
+	ibev.element.port_num = port;
+
+	if (port < 1 || port > ibdev->num_ports) {
+		mlx5_ib_warn(ibdev, "warning: event on port %d\n", port);
+		return;
+	}
+
+	if (ibdev->ib_active)
+		ib_dispatch_event(&ibev);
+}
+
+static void get_ext_port_caps(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_general_caps *gen;
+	int port;
+
+	gen = &dev->mdev->caps.gen;
+	for (port = 1; port <= gen->num_ports; port++)
+		mlx5_query_ext_port_caps(dev, port);
+}
+
+static int get_port_caps(struct mlx5_ib_dev *dev)
+{
+	struct ib_device_attr *dprops = NULL;
+	struct ib_port_attr *pprops = NULL;
+	struct mlx5_general_caps *gen;
+	int err = -ENOMEM;
+	int port;
+
+	gen = &dev->mdev->caps.gen;
+	pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
+	if (!pprops)
+		goto out;
+
+	dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
+	if (!dprops)
+		goto out;
+
+	err = mlx5_ib_query_device(&dev->ib_dev, dprops);
+	if (err) {
+		mlx5_ib_warn(dev, "query_device failed %d\n", err);
+		goto out;
+	}
+
+	for (port = 1; port <= gen->num_ports; port++) {
+		err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
+		if (err) {
+			mlx5_ib_warn(dev, "query_port %d failed %d\n", port, err);
+			break;
+		}
+		gen->port[port - 1].pkey_table_len = dprops->max_pkeys;
+		gen->port[port - 1].gid_table_len = pprops->gid_tbl_len;
+		mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
+			    dprops->max_pkeys, pprops->gid_tbl_len);
+	}
+
+out:
+	kfree(pprops);
+	kfree(dprops);
+
+	return err;
+}
+
+static void destroy_umrc_res(struct mlx5_ib_dev *dev)
+{
+	int err;
+
+	err = mlx5_mr_cache_cleanup(dev);
+	if (err)
+		mlx5_ib_warn(dev, "mr cache cleanup failed\n");
+
+	mlx5_ib_destroy_qp(dev->umrc.qp);
+	ib_destroy_cq(dev->umrc.cq);
+	ib_dereg_mr(dev->umrc.mr);
+	ib_dealloc_pd(dev->umrc.pd);
+}
+
+enum {
+	MAX_UMR_WR = 128,
+};
+
+static int create_umr_res(struct mlx5_ib_dev *dev)
+{
+	struct ib_qp_init_attr *init_attr = NULL;
+	struct ib_qp_attr *attr = NULL;
+	struct ib_pd *pd;
+	struct ib_cq *cq;
+	struct ib_qp *qp;
+	struct ib_mr *mr;
+	int ret;
+
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
+	init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
+	if (!attr || !init_attr) {
+		ret = -ENOMEM;
+		goto error_0;
+	}
+
+	pd = ib_alloc_pd(&dev->ib_dev);
+	if (IS_ERR(pd)) {
+		mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
+		ret = PTR_ERR(pd);
+		goto error_0;
+	}
+
+	mr = ib_get_dma_mr(pd,  IB_ACCESS_LOCAL_WRITE);
+	if (IS_ERR(mr)) {
+		mlx5_ib_dbg(dev, "Couldn't create DMA MR for sync UMR QP\n");
+		ret = PTR_ERR(mr);
+		goto error_1;
+	}
+
+	cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL, 128,
+			  0);
+	if (IS_ERR(cq)) {
+		mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
+		ret = PTR_ERR(cq);
+		goto error_2;
+	}
+	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+
+	init_attr->send_cq = cq;
+	init_attr->recv_cq = cq;
+	init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
+	init_attr->cap.max_send_wr = MAX_UMR_WR;
+	init_attr->cap.max_send_sge = 1;
+	init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
+	init_attr->port_num = 1;
+	qp = mlx5_ib_create_qp(pd, init_attr, NULL);
+	if (IS_ERR(qp)) {
+		mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
+		ret = PTR_ERR(qp);
+		goto error_3;
+	}
+	qp->device     = &dev->ib_dev;
+	qp->real_qp    = qp;
+	qp->uobject    = NULL;
+	qp->qp_type    = MLX5_IB_QPT_REG_UMR;
+
+	attr->qp_state = IB_QPS_INIT;
+	attr->port_num = 1;
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
+				IB_QP_PORT, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
+		goto error_4;
+	}
+
+	memset(attr, 0, sizeof(*attr));
+	attr->qp_state = IB_QPS_RTR;
+	attr->path_mtu = IB_MTU_256;
+
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
+		goto error_4;
+	}
+
+	memset(attr, 0, sizeof(*attr));
+	attr->qp_state = IB_QPS_RTS;
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
+		goto error_4;
+	}
+
+	dev->umrc.qp = qp;
+	dev->umrc.cq = cq;
+	dev->umrc.mr = mr;
+	dev->umrc.pd = pd;
+
+	sema_init(&dev->umrc.sem, MAX_UMR_WR);
+	ret = mlx5_mr_cache_init(dev);
+	if (ret) {
+		mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
+		goto error_4;
+	}
+
+	kfree(attr);
+	kfree(init_attr);
+
+	return 0;
+
+error_4:
+	mlx5_ib_destroy_qp(qp);
+
+error_3:
+	ib_destroy_cq(cq);
+
+error_2:
+	ib_dereg_mr(mr);
+
+error_1:
+	ib_dealloc_pd(pd);
+
+error_0:
+	kfree(attr);
+	kfree(init_attr);
+	return ret;
+}
+
+static int create_dev_resources(struct mlx5_ib_resources *devr)
+{
+	struct ib_srq_init_attr attr;
+	struct mlx5_ib_dev *dev;
+	int ret = 0;
+
+	dev = container_of(devr, struct mlx5_ib_dev, devr);
+
+	devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->p0)) {
+		ret = PTR_ERR(devr->p0);
+		goto error0;
+	}
+	devr->p0->device  = &dev->ib_dev;
+	devr->p0->uobject = NULL;
+	atomic_set(&devr->p0->usecnt, 0);
+
+	devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, 1, 0, NULL, NULL);
+	if (IS_ERR(devr->c0)) {
+		ret = PTR_ERR(devr->c0);
+		goto error1;
+	}
+	devr->c0->device        = &dev->ib_dev;
+	devr->c0->uobject       = NULL;
+	devr->c0->comp_handler  = NULL;
+	devr->c0->event_handler = NULL;
+	devr->c0->cq_context    = NULL;
+	atomic_set(&devr->c0->usecnt, 0);
+
+	devr->x0 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->x0)) {
+		ret = PTR_ERR(devr->x0);
+		goto error2;
+	}
+	devr->x0->device = &dev->ib_dev;
+	devr->x0->inode = NULL;
+	atomic_set(&devr->x0->usecnt, 0);
+	mutex_init(&devr->x0->tgt_qp_mutex);
+	INIT_LIST_HEAD(&devr->x0->tgt_qp_list);
+
+	devr->x1 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->x1)) {
+		ret = PTR_ERR(devr->x1);
+		goto error3;
+	}
+	devr->x1->device = &dev->ib_dev;
+	devr->x1->inode = NULL;
+	atomic_set(&devr->x1->usecnt, 0);
+	mutex_init(&devr->x1->tgt_qp_mutex);
+	INIT_LIST_HEAD(&devr->x1->tgt_qp_list);
+
+	memset(&attr, 0, sizeof(attr));
+	attr.attr.max_sge = 1;
+	attr.attr.max_wr = 1;
+	attr.srq_type = IB_SRQT_XRC;
+	attr.ext.xrc.cq = devr->c0;
+	attr.ext.xrc.xrcd = devr->x0;
+
+	devr->s0 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
+	if (IS_ERR(devr->s0)) {
+		ret = PTR_ERR(devr->s0);
+		goto error4;
+	}
+	devr->s0->device	= &dev->ib_dev;
+	devr->s0->pd		= devr->p0;
+	devr->s0->uobject       = NULL;
+	devr->s0->event_handler = NULL;
+	devr->s0->srq_context   = NULL;
+	devr->s0->srq_type      = IB_SRQT_XRC;
+	devr->s0->ext.xrc.xrcd	= devr->x0;
+	devr->s0->ext.xrc.cq	= devr->c0;
+	atomic_inc(&devr->s0->ext.xrc.xrcd->usecnt);
+	atomic_inc(&devr->s0->ext.xrc.cq->usecnt);
+	atomic_inc(&devr->p0->usecnt);
+	atomic_set(&devr->s0->usecnt, 0);
+
+	return 0;
+
+error4:
+	mlx5_ib_dealloc_xrcd(devr->x1);
+error3:
+	mlx5_ib_dealloc_xrcd(devr->x0);
+error2:
+	mlx5_ib_destroy_cq(devr->c0);
+error1:
+	mlx5_ib_dealloc_pd(devr->p0);
+error0:
+	return ret;
+}
+
+static void destroy_dev_resources(struct mlx5_ib_resources *devr)
+{
+	mlx5_ib_destroy_srq(devr->s0);
+	mlx5_ib_dealloc_xrcd(devr->x0);
+	mlx5_ib_dealloc_xrcd(devr->x1);
+	mlx5_ib_destroy_cq(devr->c0);
+	mlx5_ib_dealloc_pd(devr->p0);
+}
+
+static void *mlx5_ib_add(struct mlx5_core_dev *mdev)
+{
+	struct mlx5_ib_dev *dev;
+	int err;
+	int i;
+
+	printk_once(KERN_INFO "%s", mlx5_version);
+
+	dev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*dev));
+	if (!dev)
+		return NULL;
+
+	dev->mdev = mdev;
+
+	err = get_port_caps(dev);
+	if (err)
+		goto err_dealloc;
+
+	get_ext_port_caps(dev);
+
+	MLX5_INIT_DOORBELL_LOCK(&dev->uar_lock);
+
+	strlcpy(dev->ib_dev.name, "mlx5_%d", IB_DEVICE_NAME_MAX);
+	dev->ib_dev.owner		= THIS_MODULE;
+	dev->ib_dev.node_type		= RDMA_NODE_IB_CA;
+	dev->ib_dev.local_dma_lkey	= mdev->caps.gen.reserved_lkey;
+	dev->num_ports		= mdev->caps.gen.num_ports;
+	dev->ib_dev.phys_port_cnt     = dev->num_ports;
+	dev->ib_dev.num_comp_vectors    =
+		dev->mdev->priv.eq_table.num_comp_vectors;
+	dev->ib_dev.dma_device	= &mdev->pdev->dev;
+
+	dev->ib_dev.uverbs_abi_ver	= MLX5_IB_UVERBS_ABI_VERSION;
+	dev->ib_dev.uverbs_cmd_mask	=
+		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
+		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
+		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
+		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
+		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
+	dev->ib_dev.uverbs_ex_cmd_mask =
+		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE);
+
+	dev->ib_dev.query_device	= mlx5_ib_query_device;
+	dev->ib_dev.query_port		= mlx5_ib_query_port;
+	dev->ib_dev.query_gid		= mlx5_ib_query_gid;
+	dev->ib_dev.query_pkey		= mlx5_ib_query_pkey;
+	dev->ib_dev.modify_device	= mlx5_ib_modify_device;
+	dev->ib_dev.modify_port		= mlx5_ib_modify_port;
+	dev->ib_dev.alloc_ucontext	= mlx5_ib_alloc_ucontext;
+	dev->ib_dev.dealloc_ucontext	= mlx5_ib_dealloc_ucontext;
+	dev->ib_dev.mmap		= mlx5_ib_mmap;
+	dev->ib_dev.alloc_pd		= mlx5_ib_alloc_pd;
+	dev->ib_dev.dealloc_pd		= mlx5_ib_dealloc_pd;
+	dev->ib_dev.create_ah		= mlx5_ib_create_ah;
+	dev->ib_dev.query_ah		= mlx5_ib_query_ah;
+	dev->ib_dev.destroy_ah		= mlx5_ib_destroy_ah;
+	dev->ib_dev.create_srq		= mlx5_ib_create_srq;
+	dev->ib_dev.modify_srq		= mlx5_ib_modify_srq;
+	dev->ib_dev.query_srq		= mlx5_ib_query_srq;
+	dev->ib_dev.destroy_srq		= mlx5_ib_destroy_srq;
+	dev->ib_dev.post_srq_recv	= mlx5_ib_post_srq_recv;
+	dev->ib_dev.create_qp		= mlx5_ib_create_qp;
+	dev->ib_dev.modify_qp		= mlx5_ib_modify_qp;
+	dev->ib_dev.query_qp		= mlx5_ib_query_qp;
+	dev->ib_dev.destroy_qp		= mlx5_ib_destroy_qp;
+	dev->ib_dev.post_send		= mlx5_ib_post_send;
+	dev->ib_dev.post_recv		= mlx5_ib_post_recv;
+	dev->ib_dev.create_cq		= mlx5_ib_create_cq;
+	dev->ib_dev.modify_cq		= mlx5_ib_modify_cq;
+	dev->ib_dev.resize_cq		= mlx5_ib_resize_cq;
+	dev->ib_dev.destroy_cq		= mlx5_ib_destroy_cq;
+	dev->ib_dev.poll_cq		= mlx5_ib_poll_cq;
+	dev->ib_dev.req_notify_cq	= mlx5_ib_arm_cq;
+	dev->ib_dev.get_dma_mr		= mlx5_ib_get_dma_mr;
+	dev->ib_dev.reg_user_mr		= mlx5_ib_reg_user_mr;
+	dev->ib_dev.dereg_mr		= mlx5_ib_dereg_mr;
+	dev->ib_dev.destroy_mr		= mlx5_ib_destroy_mr;
+	dev->ib_dev.attach_mcast	= mlx5_ib_mcg_attach;
+	dev->ib_dev.detach_mcast	= mlx5_ib_mcg_detach;
+	dev->ib_dev.process_mad		= mlx5_ib_process_mad;
+	dev->ib_dev.create_mr		= mlx5_ib_create_mr;
+	dev->ib_dev.alloc_fast_reg_mr	= mlx5_ib_alloc_fast_reg_mr;
+	dev->ib_dev.alloc_fast_reg_page_list = mlx5_ib_alloc_fast_reg_page_list;
+	dev->ib_dev.free_fast_reg_page_list  = mlx5_ib_free_fast_reg_page_list;
+	dev->ib_dev.check_mr_status	= mlx5_ib_check_mr_status;
+
+	mlx5_ib_internal_query_odp_caps(dev);
+
+	if (mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_XRC) {
+		dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd;
+		dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd;
+		dev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
+			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+	}
+
+	err = init_node_data(dev);
+	if (err)
+		goto err_dealloc;
+
+	mutex_init(&dev->cap_mask_mutex);
+
+	err = create_dev_resources(&dev->devr);
+	if (err)
+		goto err_dealloc;
+
+	err = mlx5_ib_odp_init_one(dev);
+	if (err)
+		goto err_rsrc;
+
+	err = ib_register_device(&dev->ib_dev, NULL);
+	if (err)
+		goto err_odp;
+
+	err = create_umr_res(dev);
+	if (err)
+		goto err_dev;
+
+	for (i = 0; i < ARRAY_SIZE(mlx5_class_attributes); i++) {
+		err = device_create_file(&dev->ib_dev.dev,
+					 mlx5_class_attributes[i]);
+		if (err)
+			goto err_umrc;
+	}
+
+	dev->ib_active = true;
+
+	return dev;
+
+err_umrc:
+	destroy_umrc_res(dev);
+
+err_dev:
+	ib_unregister_device(&dev->ib_dev);
+
+err_odp:
+	mlx5_ib_odp_remove_one(dev);
+
+err_rsrc:
+	destroy_dev_resources(&dev->devr);
+
+err_dealloc:
+	ib_dealloc_device((struct ib_device *)dev);
+
+	return NULL;
+}
+
+static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context)
+{
+	struct mlx5_ib_dev *dev = context;
+
+	ib_unregister_device(&dev->ib_dev);
+	destroy_umrc_res(dev);
+	mlx5_ib_odp_remove_one(dev);
+	destroy_dev_resources(&dev->devr);
+	ib_dealloc_device(&dev->ib_dev);
+}
+
+static struct mlx5_interface mlx5_ib_interface = {
+	.add            = mlx5_ib_add,
+	.remove         = mlx5_ib_remove,
+	.event          = mlx5_ib_event,
+	.protocol	= MLX5_INTERFACE_PROTOCOL_IB,
+};
+
+static int __init mlx5_ib_init(void)
+{
+	int err;
+
+	if (deprecated_prof_sel != 2)
+		pr_warn("prof_sel is deprecated for mlx5_ib, set it for mlx5_core\n");
+
+	err = mlx5_ib_odp_init();
+	if (err)
+		return err;
+
+	err = mlx5_register_interface(&mlx5_ib_interface);
+	if (err)
+		goto clean_odp;
+
+	return err;
+
+clean_odp:
+	mlx5_ib_odp_cleanup();
+	return err;
+}
+
+static void __exit mlx5_ib_cleanup(void)
+{
+	mlx5_unregister_interface(&mlx5_ib_interface);
+	mlx5_ib_odp_cleanup();
+}
+
+module_init(mlx5_ib_init);
+module_exit(mlx5_ib_cleanup);
diff --git a/kernel/drivers/infiniband/hw/mlx5/mem.c b/kernel/drivers/infiniband/hw/mlx5/mem.c
new file mode 100644
index 000000000..40df2cca0
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/mem.c
@@ -0,0 +1,225 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_umem_odp.h>
+#include "mlx5_ib.h"
+
+/* @umem: umem object to scan
+ * @addr: ib virtual address requested by the user
+ * @count: number of PAGE_SIZE pages covered by umem
+ * @shift: page shift for the compound pages found in the region
+ * @ncont: number of compund pages
+ * @order: log2 of the number of compound pages
+ */
+void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
+			int *ncont, int *order)
+{
+	unsigned long tmp;
+	unsigned long m;
+	int i, k;
+	u64 base = 0;
+	int p = 0;
+	int skip;
+	int mask;
+	u64 len;
+	u64 pfn;
+	struct scatterlist *sg;
+	int entry;
+	unsigned long page_shift = ilog2(umem->page_size);
+
+	/* With ODP we must always match OS page size. */
+	if (umem->odp_data) {
+		*count = ib_umem_page_count(umem);
+		*shift = PAGE_SHIFT;
+		*ncont = *count;
+		if (order)
+			*order = ilog2(roundup_pow_of_two(*count));
+
+		return;
+	}
+
+	addr = addr >> page_shift;
+	tmp = (unsigned long)addr;
+	m = find_first_bit(&tmp, sizeof(tmp));
+	skip = 1 << m;
+	mask = skip - 1;
+	i = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> page_shift;
+		pfn = sg_dma_address(sg) >> page_shift;
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				tmp = (unsigned long)pfn;
+				m = min_t(unsigned long, m, find_first_bit(&tmp, sizeof(tmp)));
+				skip = 1 << m;
+				mask = skip - 1;
+				base = pfn;
+				p = 0;
+			} else {
+				if (base + p != pfn) {
+					tmp = (unsigned long)p;
+					m = find_first_bit(&tmp, sizeof(tmp));
+					skip = 1 << m;
+					mask = skip - 1;
+					base = pfn;
+					p = 0;
+				}
+			}
+			p++;
+			i++;
+		}
+	}
+
+	if (i) {
+		m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m);
+
+		if (order)
+			*order = ilog2(roundup_pow_of_two(i) >> m);
+
+		*ncont = DIV_ROUND_UP(i, (1 << m));
+	} else {
+		m  = 0;
+
+		if (order)
+			*order = 0;
+
+		*ncont = 0;
+	}
+	*shift = page_shift + m;
+	*count = i;
+}
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+static u64 umem_dma_to_mtt(dma_addr_t umem_dma)
+{
+	u64 mtt_entry = umem_dma & ODP_DMA_ADDR_MASK;
+
+	if (umem_dma & ODP_READ_ALLOWED_BIT)
+		mtt_entry |= MLX5_IB_MTT_READ;
+	if (umem_dma & ODP_WRITE_ALLOWED_BIT)
+		mtt_entry |= MLX5_IB_MTT_WRITE;
+
+	return mtt_entry;
+}
+#endif
+
+/*
+ * Populate the given array with bus addresses from the umem.
+ *
+ * dev - mlx5_ib device
+ * umem - umem to use to fill the pages
+ * page_shift - determines the page size used in the resulting array
+ * offset - offset into the umem to start from,
+ *          only implemented for ODP umems
+ * num_pages - total number of pages to fill
+ * pas - bus addresses array to fill
+ * access_flags - access flags to set on all present pages.
+		  use enum mlx5_ib_mtt_access_flags for this.
+ */
+void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			    int page_shift, size_t offset, size_t num_pages,
+			    __be64 *pas, int access_flags)
+{
+	unsigned long umem_page_shift = ilog2(umem->page_size);
+	int shift = page_shift - umem_page_shift;
+	int mask = (1 << shift) - 1;
+	int i, k;
+	u64 cur = 0;
+	u64 base;
+	int len;
+	struct scatterlist *sg;
+	int entry;
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	const bool odp = umem->odp_data != NULL;
+
+	if (odp) {
+		WARN_ON(shift != 0);
+		WARN_ON(access_flags != (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE));
+
+		for (i = 0; i < num_pages; ++i) {
+			dma_addr_t pa = umem->odp_data->dma_list[offset + i];
+
+			pas[i] = cpu_to_be64(umem_dma_to_mtt(pa));
+		}
+		return;
+	}
+#endif
+
+	i = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> umem_page_shift;
+		base = sg_dma_address(sg);
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				cur = base + (k << umem_page_shift);
+				cur |= access_flags;
+
+				pas[i >> shift] = cpu_to_be64(cur);
+				mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
+					    i >> shift, be64_to_cpu(pas[i >> shift]));
+			}  else
+				mlx5_ib_dbg(dev, "=====> 0x%llx\n",
+					    base + (k << umem_page_shift));
+			i++;
+		}
+	}
+}
+
+void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			  int page_shift, __be64 *pas, int access_flags)
+{
+	return __mlx5_ib_populate_pas(dev, umem, page_shift, 0,
+				      ib_umem_num_pages(umem), pas,
+				      access_flags);
+}
+int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset)
+{
+	u64 page_size;
+	u64 page_mask;
+	u64 off_size;
+	u64 off_mask;
+	u64 buf_off;
+
+	page_size = (u64)1 << page_shift;
+	page_mask = page_size - 1;
+	buf_off = addr & page_mask;
+	off_size = page_size >> 6;
+	off_mask = off_size - 1;
+
+	if (buf_off & off_mask)
+		return -EINVAL;
+
+	*offset = buf_off >> ilog2(off_size);
+	return 0;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/mlx5_ib.h b/kernel/drivers/infiniband/hw/mlx5/mlx5_ib.h
new file mode 100644
index 000000000..dff1cfcdf
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -0,0 +1,669 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef MLX5_IB_H
+#define MLX5_IB_H
+
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_smi.h>
+#include <linux/mlx5/driver.h>
+#include <linux/mlx5/cq.h>
+#include <linux/mlx5/qp.h>
+#include <linux/mlx5/srq.h>
+#include <linux/types.h>
+
+#define mlx5_ib_dbg(dev, format, arg...)				\
+pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	 __LINE__, current->pid, ##arg)
+
+#define mlx5_ib_err(dev, format, arg...)				\
+pr_err("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	__LINE__, current->pid, ##arg)
+
+#define mlx5_ib_warn(dev, format, arg...)				\
+pr_warn("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	__LINE__, current->pid, ##arg)
+
+enum {
+	MLX5_IB_MMAP_CMD_SHIFT	= 8,
+	MLX5_IB_MMAP_CMD_MASK	= 0xff,
+};
+
+enum mlx5_ib_mmap_cmd {
+	MLX5_IB_MMAP_REGULAR_PAGE		= 0,
+	MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES	= 1, /* always last */
+};
+
+enum {
+	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
+	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
+	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
+	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
+};
+
+enum mlx5_ib_latency_class {
+	MLX5_IB_LATENCY_CLASS_LOW,
+	MLX5_IB_LATENCY_CLASS_MEDIUM,
+	MLX5_IB_LATENCY_CLASS_HIGH,
+	MLX5_IB_LATENCY_CLASS_FAST_PATH
+};
+
+enum mlx5_ib_mad_ifc_flags {
+	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
+	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
+	MLX5_MAD_IFC_NET_VIEW		= 4,
+};
+
+struct mlx5_ib_ucontext {
+	struct ib_ucontext	ibucontext;
+	struct list_head	db_page_list;
+
+	/* protect doorbell record alloc/free
+	 */
+	struct mutex		db_page_mutex;
+	struct mlx5_uuar_info	uuari;
+};
+
+static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
+{
+	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
+}
+
+struct mlx5_ib_pd {
+	struct ib_pd		ibpd;
+	u32			pdn;
+	u32			pa_lkey;
+};
+
+/* Use macros here so that don't have to duplicate
+ * enum ib_send_flags and enum ib_qp_type for low-level driver
+ */
+
+#define MLX5_IB_SEND_UMR_UNREG	IB_SEND_RESERVED_START
+#define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1)
+#define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2)
+#define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
+#define MLX5_IB_WR_UMR		IB_WR_RESERVED1
+
+struct wr_list {
+	u16	opcode;
+	u16	next;
+};
+
+struct mlx5_ib_wq {
+	u64		       *wrid;
+	u32		       *wr_data;
+	struct wr_list	       *w_list;
+	unsigned	       *wqe_head;
+	u16		        unsig_count;
+
+	/* serialize post to the work queue
+	 */
+	spinlock_t		lock;
+	int			wqe_cnt;
+	int			max_post;
+	int			max_gs;
+	int			offset;
+	int			wqe_shift;
+	unsigned		head;
+	unsigned		tail;
+	u16			cur_post;
+	u16			last_poll;
+	void		       *qend;
+};
+
+enum {
+	MLX5_QP_USER,
+	MLX5_QP_KERNEL,
+	MLX5_QP_EMPTY
+};
+
+/*
+ * Connect-IB can trigger up to four concurrent pagefaults
+ * per-QP.
+ */
+enum mlx5_ib_pagefault_context {
+	MLX5_IB_PAGEFAULT_RESPONDER_READ,
+	MLX5_IB_PAGEFAULT_REQUESTOR_READ,
+	MLX5_IB_PAGEFAULT_RESPONDER_WRITE,
+	MLX5_IB_PAGEFAULT_REQUESTOR_WRITE,
+	MLX5_IB_PAGEFAULT_CONTEXTS
+};
+
+static inline enum mlx5_ib_pagefault_context
+	mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault)
+{
+	return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE);
+}
+
+struct mlx5_ib_pfault {
+	struct work_struct	work;
+	struct mlx5_pagefault	mpfault;
+};
+
+struct mlx5_ib_qp {
+	struct ib_qp		ibqp;
+	struct mlx5_core_qp	mqp;
+	struct mlx5_buf		buf;
+
+	struct mlx5_db		db;
+	struct mlx5_ib_wq	rq;
+
+	u32			doorbell_qpn;
+	u8			sq_signal_bits;
+	u8			fm_cache;
+	int			sq_max_wqes_per_wr;
+	int			sq_spare_wqes;
+	struct mlx5_ib_wq	sq;
+
+	struct ib_umem	       *umem;
+	int			buf_size;
+
+	/* serialize qp state modifications
+	 */
+	struct mutex		mutex;
+	u16			xrcdn;
+	u32			flags;
+	u8			port;
+	u8			alt_port;
+	u8			atomic_rd_en;
+	u8			resp_depth;
+	u8			state;
+	int			mlx_type;
+	int			wq_sig;
+	int			scat_cqe;
+	int			max_inline_data;
+	struct mlx5_bf	       *bf;
+	int			has_rq;
+
+	/* only for user space QPs. For kernel
+	 * we have it from the bf object
+	 */
+	int			uuarn;
+
+	int			create_type;
+	u32			pa_lkey;
+
+	/* Store signature errors */
+	bool			signature_en;
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	/*
+	 * A flag that is true for QP's that are in a state that doesn't
+	 * allow page faults, and shouldn't schedule any more faults.
+	 */
+	int                     disable_page_faults;
+	/*
+	 * The disable_page_faults_lock protects a QP's disable_page_faults
+	 * field, allowing for a thread to atomically check whether the QP
+	 * allows page faults, and if so schedule a page fault.
+	 */
+	spinlock_t              disable_page_faults_lock;
+	struct mlx5_ib_pfault	pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS];
+#endif
+};
+
+struct mlx5_ib_cq_buf {
+	struct mlx5_buf		buf;
+	struct ib_umem		*umem;
+	int			cqe_size;
+	int			nent;
+};
+
+enum mlx5_ib_qp_flags {
+	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = 1 << 0,
+	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 1,
+};
+
+struct mlx5_umr_wr {
+	union {
+		u64			virt_addr;
+		u64			offset;
+	} target;
+	struct ib_pd		       *pd;
+	unsigned int			page_shift;
+	unsigned int			npages;
+	u32				length;
+	int				access_flags;
+	u32				mkey;
+};
+
+struct mlx5_shared_mr_info {
+	int mr_id;
+	struct ib_umem		*umem;
+};
+
+struct mlx5_ib_cq {
+	struct ib_cq		ibcq;
+	struct mlx5_core_cq	mcq;
+	struct mlx5_ib_cq_buf	buf;
+	struct mlx5_db		db;
+
+	/* serialize access to the CQ
+	 */
+	spinlock_t		lock;
+
+	/* protect resize cq
+	 */
+	struct mutex		resize_mutex;
+	struct mlx5_ib_cq_buf  *resize_buf;
+	struct ib_umem	       *resize_umem;
+	int			cqe_size;
+};
+
+struct mlx5_ib_srq {
+	struct ib_srq		ibsrq;
+	struct mlx5_core_srq	msrq;
+	struct mlx5_buf		buf;
+	struct mlx5_db		db;
+	u64		       *wrid;
+	/* protect SRQ hanlding
+	 */
+	spinlock_t		lock;
+	int			head;
+	int			tail;
+	u16			wqe_ctr;
+	struct ib_umem	       *umem;
+	/* serialize arming a SRQ
+	 */
+	struct mutex		mutex;
+	int			wq_sig;
+};
+
+struct mlx5_ib_xrcd {
+	struct ib_xrcd		ibxrcd;
+	u32			xrcdn;
+};
+
+enum mlx5_ib_mtt_access_flags {
+	MLX5_IB_MTT_READ  = (1 << 0),
+	MLX5_IB_MTT_WRITE = (1 << 1),
+};
+
+#define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
+
+struct mlx5_ib_mr {
+	struct ib_mr		ibmr;
+	struct mlx5_core_mr	mmr;
+	struct ib_umem	       *umem;
+	struct mlx5_shared_mr_info	*smr_info;
+	struct list_head	list;
+	int			order;
+	int			umred;
+	int			npages;
+	struct mlx5_ib_dev     *dev;
+	struct mlx5_create_mkey_mbox_out out;
+	struct mlx5_core_sig_ctx    *sig;
+	int			live;
+};
+
+struct mlx5_ib_fast_reg_page_list {
+	struct ib_fast_reg_page_list	ibfrpl;
+	__be64			       *mapped_page_list;
+	dma_addr_t			map;
+};
+
+struct mlx5_ib_umr_context {
+	enum ib_wc_status	status;
+	struct completion	done;
+};
+
+static inline void mlx5_ib_init_umr_context(struct mlx5_ib_umr_context *context)
+{
+	context->status = -1;
+	init_completion(&context->done);
+}
+
+struct umr_common {
+	struct ib_pd	*pd;
+	struct ib_cq	*cq;
+	struct ib_qp	*qp;
+	struct ib_mr	*mr;
+	/* control access to UMR QP
+	 */
+	struct semaphore	sem;
+};
+
+enum {
+	MLX5_FMR_INVALID,
+	MLX5_FMR_VALID,
+	MLX5_FMR_BUSY,
+};
+
+struct mlx5_ib_fmr {
+	struct ib_fmr			ibfmr;
+	struct mlx5_core_mr		mr;
+	int				access_flags;
+	int				state;
+	/* protect fmr state
+	 */
+	spinlock_t			lock;
+	u64				wrid;
+	struct ib_send_wr		wr[2];
+	u8				page_shift;
+	struct ib_fast_reg_page_list	page_list;
+};
+
+struct mlx5_cache_ent {
+	struct list_head	head;
+	/* sync access to the cahce entry
+	 */
+	spinlock_t		lock;
+
+
+	struct dentry	       *dir;
+	char                    name[4];
+	u32                     order;
+	u32			size;
+	u32                     cur;
+	u32                     miss;
+	u32			limit;
+
+	struct dentry          *fsize;
+	struct dentry          *fcur;
+	struct dentry          *fmiss;
+	struct dentry          *flimit;
+
+	struct mlx5_ib_dev     *dev;
+	struct work_struct	work;
+	struct delayed_work	dwork;
+	int			pending;
+};
+
+struct mlx5_mr_cache {
+	struct workqueue_struct *wq;
+	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
+	int			stopped;
+	struct dentry		*root;
+	unsigned long		last_add;
+};
+
+struct mlx5_ib_resources {
+	struct ib_cq	*c0;
+	struct ib_xrcd	*x0;
+	struct ib_xrcd	*x1;
+	struct ib_pd	*p0;
+	struct ib_srq	*s0;
+};
+
+struct mlx5_ib_dev {
+	struct ib_device		ib_dev;
+	struct mlx5_core_dev		*mdev;
+	MLX5_DECLARE_DOORBELL_LOCK(uar_lock);
+	int				num_ports;
+	/* serialize update of capability mask
+	 */
+	struct mutex			cap_mask_mutex;
+	bool				ib_active;
+	struct umr_common		umrc;
+	/* sync used page count stats
+	 */
+	struct mlx5_ib_resources	devr;
+	struct mlx5_mr_cache		cache;
+	struct timer_list		delay_timer;
+	int				fill_delay;
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	struct ib_odp_caps	odp_caps;
+	/*
+	 * Sleepable RCU that prevents destruction of MRs while they are still
+	 * being used by a page fault handler.
+	 */
+	struct srcu_struct      mr_srcu;
+#endif
+};
+
+static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
+{
+	return container_of(mcq, struct mlx5_ib_cq, mcq);
+}
+
+static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
+{
+	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
+}
+
+static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
+}
+
+static inline struct mlx5_ib_fmr *to_mfmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct mlx5_ib_fmr, ibfmr);
+}
+
+static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
+}
+
+static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
+{
+	return container_of(mqp, struct mlx5_ib_qp, mqp);
+}
+
+static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mr *mmr)
+{
+	return container_of(mmr, struct mlx5_ib_mr, mmr);
+}
+
+static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
+}
+
+static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
+{
+	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
+}
+
+static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
+}
+
+static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
+{
+	return container_of(msrq, struct mlx5_ib_srq, msrq);
+}
+
+static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
+}
+
+static inline struct mlx5_ib_fast_reg_page_list *to_mfrpl(struct ib_fast_reg_page_list *ibfrpl)
+{
+	return container_of(ibfrpl, struct mlx5_ib_fast_reg_page_list, ibfrpl);
+}
+
+struct mlx5_ib_ah {
+	struct ib_ah		ibah;
+	struct mlx5_av		av;
+};
+
+static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct mlx5_ib_ah, ibah);
+}
+
+int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
+			struct mlx5_db *db);
+void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
+void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
+void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
+void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
+int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+		 u8 port, struct ib_wc *in_wc, struct ib_grh *in_grh,
+		 void *in_mad, void *response_mad);
+struct ib_ah *create_ib_ah(struct ib_ah_attr *ah_attr,
+			   struct mlx5_ib_ah *ah);
+struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
+int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr);
+int mlx5_ib_destroy_ah(struct ib_ah *ah);
+struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
+				  struct ib_srq_init_attr *init_attr,
+				  struct ib_udata *udata);
+int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
+int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
+int mlx5_ib_destroy_srq(struct ib_srq *srq);
+int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			  struct ib_recv_wr **bad_wr);
+struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
+				struct ib_qp_init_attr *init_attr,
+				struct ib_udata *udata);
+int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask, struct ib_udata *udata);
+int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr);
+int mlx5_ib_destroy_qp(struct ib_qp *qp);
+int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr);
+int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr);
+void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n);
+int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index,
+			  void *buffer, u32 length);
+struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, int entries,
+				int vector, struct ib_ucontext *context,
+				struct ib_udata *udata);
+int mlx5_ib_destroy_cq(struct ib_cq *cq);
+int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
+int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
+struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
+struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				  u64 virt_addr, int access_flags,
+				  struct ib_udata *udata);
+int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index,
+		       int npages, int zap);
+int mlx5_ib_dereg_mr(struct ib_mr *ibmr);
+int mlx5_ib_destroy_mr(struct ib_mr *ibmr);
+struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd,
+				struct ib_mr_init_attr *mr_init_attr);
+struct ib_mr *mlx5_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len);
+struct ib_fast_reg_page_list *mlx5_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len);
+void mlx5_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
+struct ib_fmr *mlx5_ib_fmr_alloc(struct ib_pd *pd, int acc,
+				 struct ib_fmr_attr *fmr_attr);
+int mlx5_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		      int npages, u64 iova);
+int mlx5_ib_unmap_fmr(struct list_head *fmr_list);
+int mlx5_ib_fmr_dealloc(struct ib_fmr *ibfmr);
+int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad);
+struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata);
+int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd);
+int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset);
+int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
+int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
+		       struct ib_port_attr *props);
+int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev);
+void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev);
+void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
+			int *ncont, int *order);
+void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			    int page_shift, size_t offset, size_t num_pages,
+			    __be64 *pas, int access_flags);
+void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			  int page_shift, __be64 *pas, int access_flags);
+void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
+int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq);
+int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
+int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
+int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift);
+void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context);
+int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
+			    struct ib_mr_status *mr_status);
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+extern struct workqueue_struct *mlx5_ib_page_fault_wq;
+
+int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev);
+void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
+			       struct mlx5_ib_pfault *pfault);
+void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp);
+int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
+void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev);
+int __init mlx5_ib_odp_init(void);
+void mlx5_ib_odp_cleanup(void);
+void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp);
+void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp);
+void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start,
+			      unsigned long end);
+
+#else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
+static inline int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev)
+{
+	return 0;
+}
+
+static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)		{}
+static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
+static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)	{}
+static inline int mlx5_ib_odp_init(void) { return 0; }
+static inline void mlx5_ib_odp_cleanup(void)				{}
+static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {}
+static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)  {}
+
+#endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
+
+static inline void init_query_mad(struct ib_smp *mad)
+{
+	mad->base_version  = 1;
+	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	mad->class_version = 1;
+	mad->method	   = IB_MGMT_METHOD_GET;
+}
+
+static inline u8 convert_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
+	       MLX5_PERM_LOCAL_READ;
+}
+
+#define MLX5_MAX_UMR_SHIFT 16
+#define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
+
+#endif /* MLX5_IB_H */
diff --git a/kernel/drivers/infiniband/hw/mlx5/mr.c b/kernel/drivers/infiniband/hw/mlx5/mr.c
new file mode 100644
index 000000000..71c593583
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/mr.c
@@ -0,0 +1,1479 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+
+#include <linux/kref.h>
+#include <linux/random.h>
+#include <linux/debugfs.h>
+#include <linux/export.h>
+#include <linux/delay.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_umem_odp.h>
+#include <rdma/ib_verbs.h>
+#include "mlx5_ib.h"
+
+enum {
+	MAX_PENDING_REG_MR = 8,
+};
+
+#define MLX5_UMR_ALIGN 2048
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+static __be64 mlx5_ib_update_mtt_emergency_buffer[
+		MLX5_UMR_MTT_MIN_CHUNK_SIZE/sizeof(__be64)]
+	__aligned(MLX5_UMR_ALIGN);
+static DEFINE_MUTEX(mlx5_ib_update_mtt_emergency_buffer_mutex);
+#endif
+
+static int clean_mr(struct mlx5_ib_mr *mr);
+
+static int destroy_mkey(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
+{
+	int err = mlx5_core_destroy_mkey(dev->mdev, &mr->mmr);
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	/* Wait until all page fault handlers using the mr complete. */
+	synchronize_srcu(&dev->mr_srcu);
+#endif
+
+	return err;
+}
+
+static int order2idx(struct mlx5_ib_dev *dev, int order)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+
+	if (order < cache->ent[0].order)
+		return 0;
+	else
+		return order - cache->ent[0].order;
+}
+
+static void reg_mr_callback(int status, void *context)
+{
+	struct mlx5_ib_mr *mr = context;
+	struct mlx5_ib_dev *dev = mr->dev;
+	struct mlx5_mr_cache *cache = &dev->cache;
+	int c = order2idx(dev, mr->order);
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	u8 key;
+	unsigned long flags;
+	struct mlx5_mr_table *table = &dev->mdev->priv.mr_table;
+	int err;
+
+	spin_lock_irqsave(&ent->lock, flags);
+	ent->pending--;
+	spin_unlock_irqrestore(&ent->lock, flags);
+	if (status) {
+		mlx5_ib_warn(dev, "async reg mr failed. status %d\n", status);
+		kfree(mr);
+		dev->fill_delay = 1;
+		mod_timer(&dev->delay_timer, jiffies + HZ);
+		return;
+	}
+
+	if (mr->out.hdr.status) {
+		mlx5_ib_warn(dev, "failed - status %d, syndorme 0x%x\n",
+			     mr->out.hdr.status,
+			     be32_to_cpu(mr->out.hdr.syndrome));
+		kfree(mr);
+		dev->fill_delay = 1;
+		mod_timer(&dev->delay_timer, jiffies + HZ);
+		return;
+	}
+
+	spin_lock_irqsave(&dev->mdev->priv.mkey_lock, flags);
+	key = dev->mdev->priv.mkey_key++;
+	spin_unlock_irqrestore(&dev->mdev->priv.mkey_lock, flags);
+	mr->mmr.key = mlx5_idx_to_mkey(be32_to_cpu(mr->out.mkey) & 0xffffff) | key;
+
+	cache->last_add = jiffies;
+
+	spin_lock_irqsave(&ent->lock, flags);
+	list_add_tail(&mr->list, &ent->head);
+	ent->cur++;
+	ent->size++;
+	spin_unlock_irqrestore(&ent->lock, flags);
+
+	write_lock_irqsave(&table->lock, flags);
+	err = radix_tree_insert(&table->tree, mlx5_base_mkey(mr->mmr.key),
+				&mr->mmr);
+	if (err)
+		pr_err("Error inserting to mr tree. 0x%x\n", -err);
+	write_unlock_irqrestore(&table->lock, flags);
+}
+
+static int add_keys(struct mlx5_ib_dev *dev, int c, int num)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int npages = 1 << ent->order;
+	int err = 0;
+	int i;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	for (i = 0; i < num; i++) {
+		if (ent->pending >= MAX_PENDING_REG_MR) {
+			err = -EAGAIN;
+			break;
+		}
+
+		mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+		if (!mr) {
+			err = -ENOMEM;
+			break;
+		}
+		mr->order = ent->order;
+		mr->umred = 1;
+		mr->dev = dev;
+		in->seg.status = MLX5_MKEY_STATUS_FREE;
+		in->seg.xlt_oct_size = cpu_to_be32((npages + 1) / 2);
+		in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+		in->seg.flags = MLX5_ACCESS_MODE_MTT | MLX5_PERM_UMR_EN;
+		in->seg.log2_page_size = 12;
+
+		spin_lock_irq(&ent->lock);
+		ent->pending++;
+		spin_unlock_irq(&ent->lock);
+		err = mlx5_core_create_mkey(dev->mdev, &mr->mmr, in,
+					    sizeof(*in), reg_mr_callback,
+					    mr, &mr->out);
+		if (err) {
+			spin_lock_irq(&ent->lock);
+			ent->pending--;
+			spin_unlock_irq(&ent->lock);
+			mlx5_ib_warn(dev, "create mkey failed %d\n", err);
+			kfree(mr);
+			break;
+		}
+	}
+
+	kfree(in);
+	return err;
+}
+
+static void remove_keys(struct mlx5_ib_dev *dev, int c, int num)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_ib_mr *mr;
+	int err;
+	int i;
+
+	for (i = 0; i < num; i++) {
+		spin_lock_irq(&ent->lock);
+		if (list_empty(&ent->head)) {
+			spin_unlock_irq(&ent->lock);
+			return;
+		}
+		mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list);
+		list_del(&mr->list);
+		ent->cur--;
+		ent->size--;
+		spin_unlock_irq(&ent->lock);
+		err = destroy_mkey(dev, mr);
+		if (err)
+			mlx5_ib_warn(dev, "failed destroy mkey\n");
+		else
+			kfree(mr);
+	}
+}
+
+static ssize_t size_write(struct file *filp, const char __user *buf,
+			  size_t count, loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	struct mlx5_ib_dev *dev = ent->dev;
+	char lbuf[20];
+	u32 var;
+	int err;
+	int c;
+
+	if (copy_from_user(lbuf, buf, sizeof(lbuf)))
+		return -EFAULT;
+
+	c = order2idx(dev, ent->order);
+	lbuf[sizeof(lbuf) - 1] = 0;
+
+	if (sscanf(lbuf, "%u", &var) != 1)
+		return -EINVAL;
+
+	if (var < ent->limit)
+		return -EINVAL;
+
+	if (var > ent->size) {
+		do {
+			err = add_keys(dev, c, var - ent->size);
+			if (err && err != -EAGAIN)
+				return err;
+
+			usleep_range(3000, 5000);
+		} while (err);
+	} else if (var < ent->size) {
+		remove_keys(dev, c, ent->size - var);
+	}
+
+	return count;
+}
+
+static ssize_t size_read(struct file *filp, char __user *buf, size_t count,
+			 loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	char lbuf[20];
+	int err;
+
+	if (*pos)
+		return 0;
+
+	err = snprintf(lbuf, sizeof(lbuf), "%d\n", ent->size);
+	if (err < 0)
+		return err;
+
+	if (copy_to_user(buf, lbuf, err))
+		return -EFAULT;
+
+	*pos += err;
+
+	return err;
+}
+
+static const struct file_operations size_fops = {
+	.owner	= THIS_MODULE,
+	.open	= simple_open,
+	.write	= size_write,
+	.read	= size_read,
+};
+
+static ssize_t limit_write(struct file *filp, const char __user *buf,
+			   size_t count, loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	struct mlx5_ib_dev *dev = ent->dev;
+	char lbuf[20];
+	u32 var;
+	int err;
+	int c;
+
+	if (copy_from_user(lbuf, buf, sizeof(lbuf)))
+		return -EFAULT;
+
+	c = order2idx(dev, ent->order);
+	lbuf[sizeof(lbuf) - 1] = 0;
+
+	if (sscanf(lbuf, "%u", &var) != 1)
+		return -EINVAL;
+
+	if (var > ent->size)
+		return -EINVAL;
+
+	ent->limit = var;
+
+	if (ent->cur < ent->limit) {
+		err = add_keys(dev, c, 2 * ent->limit - ent->cur);
+		if (err)
+			return err;
+	}
+
+	return count;
+}
+
+static ssize_t limit_read(struct file *filp, char __user *buf, size_t count,
+			  loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	char lbuf[20];
+	int err;
+
+	if (*pos)
+		return 0;
+
+	err = snprintf(lbuf, sizeof(lbuf), "%d\n", ent->limit);
+	if (err < 0)
+		return err;
+
+	if (copy_to_user(buf, lbuf, err))
+		return -EFAULT;
+
+	*pos += err;
+
+	return err;
+}
+
+static const struct file_operations limit_fops = {
+	.owner	= THIS_MODULE,
+	.open	= simple_open,
+	.write	= limit_write,
+	.read	= limit_read,
+};
+
+static int someone_adding(struct mlx5_mr_cache *cache)
+{
+	int i;
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		if (cache->ent[i].cur < cache->ent[i].limit)
+			return 1;
+	}
+
+	return 0;
+}
+
+static void __cache_work_func(struct mlx5_cache_ent *ent)
+{
+	struct mlx5_ib_dev *dev = ent->dev;
+	struct mlx5_mr_cache *cache = &dev->cache;
+	int i = order2idx(dev, ent->order);
+	int err;
+
+	if (cache->stopped)
+		return;
+
+	ent = &dev->cache.ent[i];
+	if (ent->cur < 2 * ent->limit && !dev->fill_delay) {
+		err = add_keys(dev, i, 1);
+		if (ent->cur < 2 * ent->limit) {
+			if (err == -EAGAIN) {
+				mlx5_ib_dbg(dev, "returned eagain, order %d\n",
+					    i + 2);
+				queue_delayed_work(cache->wq, &ent->dwork,
+						   msecs_to_jiffies(3));
+			} else if (err) {
+				mlx5_ib_warn(dev, "command failed order %d, err %d\n",
+					     i + 2, err);
+				queue_delayed_work(cache->wq, &ent->dwork,
+						   msecs_to_jiffies(1000));
+			} else {
+				queue_work(cache->wq, &ent->work);
+			}
+		}
+	} else if (ent->cur > 2 * ent->limit) {
+		if (!someone_adding(cache) &&
+		    time_after(jiffies, cache->last_add + 300 * HZ)) {
+			remove_keys(dev, i, 1);
+			if (ent->cur > ent->limit)
+				queue_work(cache->wq, &ent->work);
+		} else {
+			queue_delayed_work(cache->wq, &ent->dwork, 300 * HZ);
+		}
+	}
+}
+
+static void delayed_cache_work_func(struct work_struct *work)
+{
+	struct mlx5_cache_ent *ent;
+
+	ent = container_of(work, struct mlx5_cache_ent, dwork.work);
+	__cache_work_func(ent);
+}
+
+static void cache_work_func(struct work_struct *work)
+{
+	struct mlx5_cache_ent *ent;
+
+	ent = container_of(work, struct mlx5_cache_ent, work);
+	__cache_work_func(ent);
+}
+
+static struct mlx5_ib_mr *alloc_cached_mr(struct mlx5_ib_dev *dev, int order)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_ib_mr *mr = NULL;
+	struct mlx5_cache_ent *ent;
+	int c;
+	int i;
+
+	c = order2idx(dev, order);
+	if (c < 0 || c >= MAX_MR_CACHE_ENTRIES) {
+		mlx5_ib_warn(dev, "order %d, cache index %d\n", order, c);
+		return NULL;
+	}
+
+	for (i = c; i < MAX_MR_CACHE_ENTRIES; i++) {
+		ent = &cache->ent[i];
+
+		mlx5_ib_dbg(dev, "order %d, cache index %d\n", ent->order, i);
+
+		spin_lock_irq(&ent->lock);
+		if (!list_empty(&ent->head)) {
+			mr = list_first_entry(&ent->head, struct mlx5_ib_mr,
+					      list);
+			list_del(&mr->list);
+			ent->cur--;
+			spin_unlock_irq(&ent->lock);
+			if (ent->cur < ent->limit)
+				queue_work(cache->wq, &ent->work);
+			break;
+		}
+		spin_unlock_irq(&ent->lock);
+
+		queue_work(cache->wq, &ent->work);
+
+		if (mr)
+			break;
+	}
+
+	if (!mr)
+		cache->ent[c].miss++;
+
+	return mr;
+}
+
+static void free_cached_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int shrink = 0;
+	int c;
+
+	c = order2idx(dev, mr->order);
+	if (c < 0 || c >= MAX_MR_CACHE_ENTRIES) {
+		mlx5_ib_warn(dev, "order %d, cache index %d\n", mr->order, c);
+		return;
+	}
+	ent = &cache->ent[c];
+	spin_lock_irq(&ent->lock);
+	list_add_tail(&mr->list, &ent->head);
+	ent->cur++;
+	if (ent->cur > 2 * ent->limit)
+		shrink = 1;
+	spin_unlock_irq(&ent->lock);
+
+	if (shrink)
+		queue_work(cache->wq, &ent->work);
+}
+
+static void clean_keys(struct mlx5_ib_dev *dev, int c)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	cancel_delayed_work(&ent->dwork);
+	while (1) {
+		spin_lock_irq(&ent->lock);
+		if (list_empty(&ent->head)) {
+			spin_unlock_irq(&ent->lock);
+			return;
+		}
+		mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list);
+		list_del(&mr->list);
+		ent->cur--;
+		ent->size--;
+		spin_unlock_irq(&ent->lock);
+		err = destroy_mkey(dev, mr);
+		if (err)
+			mlx5_ib_warn(dev, "failed destroy mkey\n");
+		else
+			kfree(mr);
+	}
+}
+
+static int mlx5_mr_cache_debugfs_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int i;
+
+	if (!mlx5_debugfs_root)
+		return 0;
+
+	cache->root = debugfs_create_dir("mr_cache", dev->mdev->priv.dbg_root);
+	if (!cache->root)
+		return -ENOMEM;
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		ent = &cache->ent[i];
+		sprintf(ent->name, "%d", ent->order);
+		ent->dir = debugfs_create_dir(ent->name,  cache->root);
+		if (!ent->dir)
+			return -ENOMEM;
+
+		ent->fsize = debugfs_create_file("size", 0600, ent->dir, ent,
+						 &size_fops);
+		if (!ent->fsize)
+			return -ENOMEM;
+
+		ent->flimit = debugfs_create_file("limit", 0600, ent->dir, ent,
+						  &limit_fops);
+		if (!ent->flimit)
+			return -ENOMEM;
+
+		ent->fcur = debugfs_create_u32("cur", 0400, ent->dir,
+					       &ent->cur);
+		if (!ent->fcur)
+			return -ENOMEM;
+
+		ent->fmiss = debugfs_create_u32("miss", 0600, ent->dir,
+						&ent->miss);
+		if (!ent->fmiss)
+			return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static void mlx5_mr_cache_debugfs_cleanup(struct mlx5_ib_dev *dev)
+{
+	if (!mlx5_debugfs_root)
+		return;
+
+	debugfs_remove_recursive(dev->cache.root);
+}
+
+static void delay_time_func(unsigned long ctx)
+{
+	struct mlx5_ib_dev *dev = (struct mlx5_ib_dev *)ctx;
+
+	dev->fill_delay = 0;
+}
+
+int mlx5_mr_cache_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int limit;
+	int err;
+	int i;
+
+	cache->wq = create_singlethread_workqueue("mkey_cache");
+	if (!cache->wq) {
+		mlx5_ib_warn(dev, "failed to create work queue\n");
+		return -ENOMEM;
+	}
+
+	setup_timer(&dev->delay_timer, delay_time_func, (unsigned long)dev);
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		INIT_LIST_HEAD(&cache->ent[i].head);
+		spin_lock_init(&cache->ent[i].lock);
+
+		ent = &cache->ent[i];
+		INIT_LIST_HEAD(&ent->head);
+		spin_lock_init(&ent->lock);
+		ent->order = i + 2;
+		ent->dev = dev;
+
+		if (dev->mdev->profile->mask & MLX5_PROF_MASK_MR_CACHE)
+			limit = dev->mdev->profile->mr_cache[i].limit;
+		else
+			limit = 0;
+
+		INIT_WORK(&ent->work, cache_work_func);
+		INIT_DELAYED_WORK(&ent->dwork, delayed_cache_work_func);
+		ent->limit = limit;
+		queue_work(cache->wq, &ent->work);
+	}
+
+	err = mlx5_mr_cache_debugfs_init(dev);
+	if (err)
+		mlx5_ib_warn(dev, "cache debugfs failure\n");
+
+	return 0;
+}
+
+int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev)
+{
+	int i;
+
+	dev->cache.stopped = 1;
+	flush_workqueue(dev->cache.wq);
+
+	mlx5_mr_cache_debugfs_cleanup(dev);
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++)
+		clean_keys(dev, i);
+
+	destroy_workqueue(dev->cache.wq);
+	del_timer_sync(&dev->delay_timer);
+
+	return 0;
+}
+
+struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_core_dev *mdev = dev->mdev;
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_mkey_seg *seg;
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	seg = &in->seg;
+	seg->flags = convert_access(acc) | MLX5_ACCESS_MODE_PA;
+	seg->flags_pd = cpu_to_be32(to_mpd(pd)->pdn | MLX5_MKEY_LEN64);
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	seg->start_addr = 0;
+
+	err = mlx5_core_create_mkey(mdev, &mr->mmr, in, sizeof(*in), NULL, NULL,
+				    NULL);
+	if (err)
+		goto err_in;
+
+	kfree(in);
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+err_in:
+	kfree(in);
+
+err_free:
+	kfree(mr);
+
+	return ERR_PTR(err);
+}
+
+static int get_octo_len(u64 addr, u64 len, int page_size)
+{
+	u64 offset;
+	int npages;
+
+	offset = addr & (page_size - 1);
+	npages = ALIGN(len + offset, page_size) >> ilog2(page_size);
+	return (npages + 1) / 2;
+}
+
+static int use_umr(int order)
+{
+	return order <= MLX5_MAX_UMR_SHIFT;
+}
+
+static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr,
+			     struct ib_sge *sg, u64 dma, int n, u32 key,
+			     int page_shift, u64 virt_addr, u64 len,
+			     int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct ib_mr *mr = dev->umrc.mr;
+	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+
+	sg->addr = dma;
+	sg->length = ALIGN(sizeof(u64) * n, 64);
+	sg->lkey = mr->lkey;
+
+	wr->next = NULL;
+	wr->send_flags = 0;
+	wr->sg_list = sg;
+	if (n)
+		wr->num_sge = 1;
+	else
+		wr->num_sge = 0;
+
+	wr->opcode = MLX5_IB_WR_UMR;
+
+	umrwr->npages = n;
+	umrwr->page_shift = page_shift;
+	umrwr->mkey = key;
+	umrwr->target.virt_addr = virt_addr;
+	umrwr->length = len;
+	umrwr->access_flags = access_flags;
+	umrwr->pd = pd;
+}
+
+static void prep_umr_unreg_wqe(struct mlx5_ib_dev *dev,
+			       struct ib_send_wr *wr, u32 key)
+{
+	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+
+	wr->send_flags = MLX5_IB_SEND_UMR_UNREG | MLX5_IB_SEND_UMR_FAIL_IF_FREE;
+	wr->opcode = MLX5_IB_WR_UMR;
+	umrwr->mkey = key;
+}
+
+void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context)
+{
+	struct mlx5_ib_umr_context *context;
+	struct ib_wc wc;
+	int err;
+
+	while (1) {
+		err = ib_poll_cq(cq, 1, &wc);
+		if (err < 0) {
+			pr_warn("poll cq error %d\n", err);
+			return;
+		}
+		if (err == 0)
+			break;
+
+		context = (struct mlx5_ib_umr_context *) (unsigned long) wc.wr_id;
+		context->status = wc.status;
+		complete(&context->done);
+	}
+	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+}
+
+static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem,
+				  u64 virt_addr, u64 len, int npages,
+				  int page_shift, int order, int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct device *ddev = dev->ib_dev.dma_device;
+	struct umr_common *umrc = &dev->umrc;
+	struct mlx5_ib_umr_context umr_context;
+	struct ib_send_wr wr, *bad;
+	struct mlx5_ib_mr *mr;
+	struct ib_sge sg;
+	int size;
+	__be64 *mr_pas;
+	__be64 *pas;
+	dma_addr_t dma;
+	int err = 0;
+	int i;
+
+	for (i = 0; i < 1; i++) {
+		mr = alloc_cached_mr(dev, order);
+		if (mr)
+			break;
+
+		err = add_keys(dev, order2idx(dev, order), 1);
+		if (err && err != -EAGAIN) {
+			mlx5_ib_warn(dev, "add_keys failed, err %d\n", err);
+			break;
+		}
+	}
+
+	if (!mr)
+		return ERR_PTR(-EAGAIN);
+
+	/* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
+	 * To avoid copying garbage after the pas array, we allocate
+	 * a little more. */
+	size = ALIGN(sizeof(u64) * npages, MLX5_UMR_MTT_ALIGNMENT);
+	mr_pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL);
+	if (!mr_pas) {
+		err = -ENOMEM;
+		goto free_mr;
+	}
+
+	pas = PTR_ALIGN(mr_pas, MLX5_UMR_ALIGN);
+	mlx5_ib_populate_pas(dev, umem, page_shift, pas, MLX5_IB_MTT_PRESENT);
+	/* Clear padding after the actual pages. */
+	memset(pas + npages, 0, size - npages * sizeof(u64));
+
+	dma = dma_map_single(ddev, pas, size, DMA_TO_DEVICE);
+	if (dma_mapping_error(ddev, dma)) {
+		err = -ENOMEM;
+		goto free_pas;
+	}
+
+	memset(&wr, 0, sizeof(wr));
+	wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_reg_wqe(pd, &wr, &sg, dma, npages, mr->mmr.key, page_shift,
+			 virt_addr, len, access_flags);
+
+	mlx5_ib_init_umr_context(&umr_context);
+	down(&umrc->sem);
+	err = ib_post_send(umrc->qp, &wr, &bad);
+	if (err) {
+		mlx5_ib_warn(dev, "post send failed, err %d\n", err);
+		goto unmap_dma;
+	} else {
+		wait_for_completion(&umr_context.done);
+		if (umr_context.status != IB_WC_SUCCESS) {
+			mlx5_ib_warn(dev, "reg umr failed\n");
+			err = -EFAULT;
+		}
+	}
+
+	mr->mmr.iova = virt_addr;
+	mr->mmr.size = len;
+	mr->mmr.pd = to_mpd(pd)->pdn;
+
+	mr->live = 1;
+
+unmap_dma:
+	up(&umrc->sem);
+	dma_unmap_single(ddev, dma, size, DMA_TO_DEVICE);
+
+free_pas:
+	kfree(mr_pas);
+
+free_mr:
+	if (err) {
+		free_cached_mr(dev, mr);
+		return ERR_PTR(err);
+	}
+
+	return mr;
+}
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, int npages,
+		       int zap)
+{
+	struct mlx5_ib_dev *dev = mr->dev;
+	struct device *ddev = dev->ib_dev.dma_device;
+	struct umr_common *umrc = &dev->umrc;
+	struct mlx5_ib_umr_context umr_context;
+	struct ib_umem *umem = mr->umem;
+	int size;
+	__be64 *pas;
+	dma_addr_t dma;
+	struct ib_send_wr wr, *bad;
+	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr.wr.fast_reg;
+	struct ib_sge sg;
+	int err = 0;
+	const int page_index_alignment = MLX5_UMR_MTT_ALIGNMENT / sizeof(u64);
+	const int page_index_mask = page_index_alignment - 1;
+	size_t pages_mapped = 0;
+	size_t pages_to_map = 0;
+	size_t pages_iter = 0;
+	int use_emergency_buf = 0;
+
+	/* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes,
+	 * so we need to align the offset and length accordingly */
+	if (start_page_index & page_index_mask) {
+		npages += start_page_index & page_index_mask;
+		start_page_index &= ~page_index_mask;
+	}
+
+	pages_to_map = ALIGN(npages, page_index_alignment);
+
+	if (start_page_index + pages_to_map > MLX5_MAX_UMR_PAGES)
+		return -EINVAL;
+
+	size = sizeof(u64) * pages_to_map;
+	size = min_t(int, PAGE_SIZE, size);
+	/* We allocate with GFP_ATOMIC to avoid recursion into page-reclaim
+	 * code, when we are called from an invalidation. The pas buffer must
+	 * be 2k-aligned for Connect-IB. */
+	pas = (__be64 *)get_zeroed_page(GFP_ATOMIC);
+	if (!pas) {
+		mlx5_ib_warn(dev, "unable to allocate memory during MTT update, falling back to slower chunked mechanism.\n");
+		pas = mlx5_ib_update_mtt_emergency_buffer;
+		size = MLX5_UMR_MTT_MIN_CHUNK_SIZE;
+		use_emergency_buf = 1;
+		mutex_lock(&mlx5_ib_update_mtt_emergency_buffer_mutex);
+		memset(pas, 0, size);
+	}
+	pages_iter = size / sizeof(u64);
+	dma = dma_map_single(ddev, pas, size, DMA_TO_DEVICE);
+	if (dma_mapping_error(ddev, dma)) {
+		mlx5_ib_err(dev, "unable to map DMA during MTT update.\n");
+		err = -ENOMEM;
+		goto free_pas;
+	}
+
+	for (pages_mapped = 0;
+	     pages_mapped < pages_to_map && !err;
+	     pages_mapped += pages_iter, start_page_index += pages_iter) {
+		dma_sync_single_for_cpu(ddev, dma, size, DMA_TO_DEVICE);
+
+		npages = min_t(size_t,
+			       pages_iter,
+			       ib_umem_num_pages(umem) - start_page_index);
+
+		if (!zap) {
+			__mlx5_ib_populate_pas(dev, umem, PAGE_SHIFT,
+					       start_page_index, npages, pas,
+					       MLX5_IB_MTT_PRESENT);
+			/* Clear padding after the pages brought from the
+			 * umem. */
+			memset(pas + npages, 0, size - npages * sizeof(u64));
+		}
+
+		dma_sync_single_for_device(ddev, dma, size, DMA_TO_DEVICE);
+
+		memset(&wr, 0, sizeof(wr));
+		wr.wr_id = (u64)(unsigned long)&umr_context;
+
+		sg.addr = dma;
+		sg.length = ALIGN(npages * sizeof(u64),
+				MLX5_UMR_MTT_ALIGNMENT);
+		sg.lkey = dev->umrc.mr->lkey;
+
+		wr.send_flags = MLX5_IB_SEND_UMR_FAIL_IF_FREE |
+				MLX5_IB_SEND_UMR_UPDATE_MTT;
+		wr.sg_list = &sg;
+		wr.num_sge = 1;
+		wr.opcode = MLX5_IB_WR_UMR;
+		umrwr->npages = sg.length / sizeof(u64);
+		umrwr->page_shift = PAGE_SHIFT;
+		umrwr->mkey = mr->mmr.key;
+		umrwr->target.offset = start_page_index;
+
+		mlx5_ib_init_umr_context(&umr_context);
+		down(&umrc->sem);
+		err = ib_post_send(umrc->qp, &wr, &bad);
+		if (err) {
+			mlx5_ib_err(dev, "UMR post send failed, err %d\n", err);
+		} else {
+			wait_for_completion(&umr_context.done);
+			if (umr_context.status != IB_WC_SUCCESS) {
+				mlx5_ib_err(dev, "UMR completion failed, code %d\n",
+					    umr_context.status);
+				err = -EFAULT;
+			}
+		}
+		up(&umrc->sem);
+	}
+	dma_unmap_single(ddev, dma, size, DMA_TO_DEVICE);
+
+free_pas:
+	if (!use_emergency_buf)
+		free_page((unsigned long)pas);
+	else
+		mutex_unlock(&mlx5_ib_update_mtt_emergency_buffer_mutex);
+
+	return err;
+}
+#endif
+
+static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr,
+				     u64 length, struct ib_umem *umem,
+				     int npages, int page_shift,
+				     int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int inlen;
+	int err;
+	bool pg_cap = !!(dev->mdev->caps.gen.flags &
+			 MLX5_DEV_CAP_FLAG_ON_DMND_PG);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	inlen = sizeof(*in) + sizeof(*in->pas) * ((npages + 1) / 2) * 2;
+	in = mlx5_vzalloc(inlen);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_1;
+	}
+	mlx5_ib_populate_pas(dev, umem, page_shift, in->pas,
+			     pg_cap ? MLX5_IB_MTT_PRESENT : 0);
+
+	/* The MLX5_MKEY_INBOX_PG_ACCESS bit allows setting the access flags
+	 * in the page list submitted with the command. */
+	in->flags = pg_cap ? cpu_to_be32(MLX5_MKEY_INBOX_PG_ACCESS) : 0;
+	in->seg.flags = convert_access(access_flags) |
+		MLX5_ACCESS_MODE_MTT;
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	in->seg.start_addr = cpu_to_be64(virt_addr);
+	in->seg.len = cpu_to_be64(length);
+	in->seg.bsfs_octo_size = 0;
+	in->seg.xlt_oct_size = cpu_to_be32(get_octo_len(virt_addr, length, 1 << page_shift));
+	in->seg.log2_page_size = page_shift;
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->xlat_oct_act_size = cpu_to_be32(get_octo_len(virt_addr, length,
+							 1 << page_shift));
+	err = mlx5_core_create_mkey(dev->mdev, &mr->mmr, in, inlen, NULL,
+				    NULL, NULL);
+	if (err) {
+		mlx5_ib_warn(dev, "create mkey failed\n");
+		goto err_2;
+	}
+	mr->umem = umem;
+	mr->dev = dev;
+	mr->live = 1;
+	kvfree(in);
+
+	mlx5_ib_dbg(dev, "mkey = 0x%x\n", mr->mmr.key);
+
+	return mr;
+
+err_2:
+	kvfree(in);
+
+err_1:
+	kfree(mr);
+
+	return ERR_PTR(err);
+}
+
+struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				  u64 virt_addr, int access_flags,
+				  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_mr *mr = NULL;
+	struct ib_umem *umem;
+	int page_shift;
+	int npages;
+	int ncont;
+	int order;
+	int err;
+
+	mlx5_ib_dbg(dev, "start 0x%llx, virt_addr 0x%llx, length 0x%llx, access_flags 0x%x\n",
+		    start, virt_addr, length, access_flags);
+	umem = ib_umem_get(pd->uobject->context, start, length, access_flags,
+			   0);
+	if (IS_ERR(umem)) {
+		mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem));
+		return (void *)umem;
+	}
+
+	mlx5_ib_cont_pages(umem, start, &npages, &page_shift, &ncont, &order);
+	if (!npages) {
+		mlx5_ib_warn(dev, "avoid zero region\n");
+		err = -EINVAL;
+		goto error;
+	}
+
+	mlx5_ib_dbg(dev, "npages %d, ncont %d, order %d, page_shift %d\n",
+		    npages, ncont, order, page_shift);
+
+	if (use_umr(order)) {
+		mr = reg_umr(pd, umem, virt_addr, length, ncont, page_shift,
+			     order, access_flags);
+		if (PTR_ERR(mr) == -EAGAIN) {
+			mlx5_ib_dbg(dev, "cache empty for order %d", order);
+			mr = NULL;
+		}
+	} else if (access_flags & IB_ACCESS_ON_DEMAND) {
+		err = -EINVAL;
+		pr_err("Got MR registration for ODP MR > 512MB, not supported for Connect-IB");
+		goto error;
+	}
+
+	if (!mr)
+		mr = reg_create(pd, virt_addr, length, umem, ncont, page_shift,
+				access_flags);
+
+	if (IS_ERR(mr)) {
+		err = PTR_ERR(mr);
+		goto error;
+	}
+
+	mlx5_ib_dbg(dev, "mkey 0x%x\n", mr->mmr.key);
+
+	mr->umem = umem;
+	mr->npages = npages;
+	atomic_add(npages, &dev->mdev->priv.reg_pages);
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	if (umem->odp_data) {
+		/*
+		 * This barrier prevents the compiler from moving the
+		 * setting of umem->odp_data->private to point to our
+		 * MR, before reg_umr finished, to ensure that the MR
+		 * initialization have finished before starting to
+		 * handle invalidations.
+		 */
+		smp_wmb();
+		mr->umem->odp_data->private = mr;
+		/*
+		 * Make sure we will see the new
+		 * umem->odp_data->private value in the invalidation
+		 * routines, before we can get page faults on the
+		 * MR. Page faults can happen once we put the MR in
+		 * the tree, below this line. Without the barrier,
+		 * there can be a fault handling and an invalidation
+		 * before umem->odp_data->private == mr is visible to
+		 * the invalidation handler.
+		 */
+		smp_wmb();
+	}
+#endif
+
+	return &mr->ibmr;
+
+error:
+	/*
+	 * Destroy the umem *before* destroying the MR, to ensure we
+	 * will not have any in-flight notifiers when destroying the
+	 * MR.
+	 *
+	 * As the MR is completely invalid to begin with, and this
+	 * error path is only taken if we can't push the mr entry into
+	 * the pagefault tree, this is safe.
+	 */
+
+	ib_umem_release(umem);
+	/* Kill the MR, and return an error code. */
+	clean_mr(mr);
+	return ERR_PTR(err);
+}
+
+static int unreg_umr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
+{
+	struct umr_common *umrc = &dev->umrc;
+	struct mlx5_ib_umr_context umr_context;
+	struct ib_send_wr wr, *bad;
+	int err;
+
+	memset(&wr, 0, sizeof(wr));
+	wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_unreg_wqe(dev, &wr, mr->mmr.key);
+
+	mlx5_ib_init_umr_context(&umr_context);
+	down(&umrc->sem);
+	err = ib_post_send(umrc->qp, &wr, &bad);
+	if (err) {
+		up(&umrc->sem);
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto error;
+	} else {
+		wait_for_completion(&umr_context.done);
+		up(&umrc->sem);
+	}
+	if (umr_context.status != IB_WC_SUCCESS) {
+		mlx5_ib_warn(dev, "unreg umr failed\n");
+		err = -EFAULT;
+		goto error;
+	}
+	return 0;
+
+error:
+	return err;
+}
+
+static int clean_mr(struct mlx5_ib_mr *mr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device);
+	int umred = mr->umred;
+	int err;
+
+	if (!umred) {
+		err = destroy_mkey(dev, mr);
+		if (err) {
+			mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n",
+				     mr->mmr.key, err);
+			return err;
+		}
+	} else {
+		err = unreg_umr(dev, mr);
+		if (err) {
+			mlx5_ib_warn(dev, "failed unregister\n");
+			return err;
+		}
+		free_cached_mr(dev, mr);
+	}
+
+	if (!umred)
+		kfree(mr);
+
+	return 0;
+}
+
+int mlx5_ib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibmr->device);
+	struct mlx5_ib_mr *mr = to_mmr(ibmr);
+	int npages = mr->npages;
+	struct ib_umem *umem = mr->umem;
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	if (umem && umem->odp_data) {
+		/* Prevent new page faults from succeeding */
+		mr->live = 0;
+		/* Wait for all running page-fault handlers to finish. */
+		synchronize_srcu(&dev->mr_srcu);
+		/* Destroy all page mappings */
+		mlx5_ib_invalidate_range(umem, ib_umem_start(umem),
+					 ib_umem_end(umem));
+		/*
+		 * We kill the umem before the MR for ODP,
+		 * so that there will not be any invalidations in
+		 * flight, looking at the *mr struct.
+		 */
+		ib_umem_release(umem);
+		atomic_sub(npages, &dev->mdev->priv.reg_pages);
+
+		/* Avoid double-freeing the umem. */
+		umem = NULL;
+	}
+#endif
+
+	clean_mr(mr);
+
+	if (umem) {
+		ib_umem_release(umem);
+		atomic_sub(npages, &dev->mdev->priv.reg_pages);
+	}
+
+	return 0;
+}
+
+struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd,
+				struct ib_mr_init_attr *mr_init_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int access_mode, err;
+	int ndescs = roundup(mr_init_attr->max_reg_descriptors, 4);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	in->seg.status = MLX5_MKEY_STATUS_FREE;
+	in->seg.xlt_oct_size = cpu_to_be32(ndescs);
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	access_mode = MLX5_ACCESS_MODE_MTT;
+
+	if (mr_init_attr->flags & IB_MR_SIGNATURE_EN) {
+		u32 psv_index[2];
+
+		in->seg.flags_pd = cpu_to_be32(be32_to_cpu(in->seg.flags_pd) |
+							   MLX5_MKEY_BSF_EN);
+		in->seg.bsfs_octo_size = cpu_to_be32(MLX5_MKEY_BSF_OCTO_SIZE);
+		mr->sig = kzalloc(sizeof(*mr->sig), GFP_KERNEL);
+		if (!mr->sig) {
+			err = -ENOMEM;
+			goto err_free_in;
+		}
+
+		/* create mem & wire PSVs */
+		err = mlx5_core_create_psv(dev->mdev, to_mpd(pd)->pdn,
+					   2, psv_index);
+		if (err)
+			goto err_free_sig;
+
+		access_mode = MLX5_ACCESS_MODE_KLM;
+		mr->sig->psv_memory.psv_idx = psv_index[0];
+		mr->sig->psv_wire.psv_idx = psv_index[1];
+
+		mr->sig->sig_status_checked = true;
+		mr->sig->sig_err_exists = false;
+		/* Next UMR, Arm SIGERR */
+		++mr->sig->sigerr_count;
+	}
+
+	in->seg.flags = MLX5_PERM_UMR_EN | access_mode;
+	err = mlx5_core_create_mkey(dev->mdev, &mr->mmr, in, sizeof(*in),
+				    NULL, NULL, NULL);
+	if (err)
+		goto err_destroy_psv;
+
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+	kfree(in);
+
+	return &mr->ibmr;
+
+err_destroy_psv:
+	if (mr->sig) {
+		if (mlx5_core_destroy_psv(dev->mdev,
+					  mr->sig->psv_memory.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy mem psv %d\n",
+				     mr->sig->psv_memory.psv_idx);
+		if (mlx5_core_destroy_psv(dev->mdev,
+					  mr->sig->psv_wire.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy wire psv %d\n",
+				     mr->sig->psv_wire.psv_idx);
+	}
+err_free_sig:
+	kfree(mr->sig);
+err_free_in:
+	kfree(in);
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+int mlx5_ib_destroy_mr(struct ib_mr *ibmr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibmr->device);
+	struct mlx5_ib_mr *mr = to_mmr(ibmr);
+	int err;
+
+	if (mr->sig) {
+		if (mlx5_core_destroy_psv(dev->mdev,
+					  mr->sig->psv_memory.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy mem psv %d\n",
+				     mr->sig->psv_memory.psv_idx);
+		if (mlx5_core_destroy_psv(dev->mdev,
+					  mr->sig->psv_wire.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy wire psv %d\n",
+				     mr->sig->psv_wire.psv_idx);
+		kfree(mr->sig);
+	}
+
+	err = destroy_mkey(dev, mr);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n",
+			     mr->mmr.key, err);
+		return err;
+	}
+
+	kfree(mr);
+
+	return err;
+}
+
+struct ib_mr *mlx5_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	in->seg.status = MLX5_MKEY_STATUS_FREE;
+	in->seg.xlt_oct_size = cpu_to_be32((max_page_list_len + 1) / 2);
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->seg.flags = MLX5_PERM_UMR_EN | MLX5_ACCESS_MODE_MTT;
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	/*
+	 * TBD not needed - issue 197292 */
+	in->seg.log2_page_size = PAGE_SHIFT;
+
+	err = mlx5_core_create_mkey(dev->mdev, &mr->mmr, in, sizeof(*in), NULL,
+				    NULL, NULL);
+	kfree(in);
+	if (err)
+		goto err_free;
+
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+struct ib_fast_reg_page_list *mlx5_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl;
+	int size = page_list_len * sizeof(u64);
+
+	mfrpl = kmalloc(sizeof(*mfrpl), GFP_KERNEL);
+	if (!mfrpl)
+		return ERR_PTR(-ENOMEM);
+
+	mfrpl->ibfrpl.page_list = kmalloc(size, GFP_KERNEL);
+	if (!mfrpl->ibfrpl.page_list)
+		goto err_free;
+
+	mfrpl->mapped_page_list = dma_alloc_coherent(ibdev->dma_device,
+						     size, &mfrpl->map,
+						     GFP_KERNEL);
+	if (!mfrpl->mapped_page_list)
+		goto err_free;
+
+	WARN_ON(mfrpl->map & 0x3f);
+
+	return &mfrpl->ibfrpl;
+
+err_free:
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+	return ERR_PTR(-ENOMEM);
+}
+
+void mlx5_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list);
+	struct mlx5_ib_dev *dev = to_mdev(page_list->device);
+	int size = page_list->max_page_list_len * sizeof(u64);
+
+	dma_free_coherent(&dev->mdev->pdev->dev, size, mfrpl->mapped_page_list,
+			  mfrpl->map);
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+}
+
+int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
+			    struct ib_mr_status *mr_status)
+{
+	struct mlx5_ib_mr *mmr = to_mmr(ibmr);
+	int ret = 0;
+
+	if (check_mask & ~IB_MR_CHECK_SIG_STATUS) {
+		pr_err("Invalid status check mask\n");
+		ret = -EINVAL;
+		goto done;
+	}
+
+	mr_status->fail_status = 0;
+	if (check_mask & IB_MR_CHECK_SIG_STATUS) {
+		if (!mmr->sig) {
+			ret = -EINVAL;
+			pr_err("signature status check requested on a non-signature enabled MR\n");
+			goto done;
+		}
+
+		mmr->sig->sig_status_checked = true;
+		if (!mmr->sig->sig_err_exists)
+			goto done;
+
+		if (ibmr->lkey == mmr->sig->err_item.key)
+			memcpy(&mr_status->sig_err, &mmr->sig->err_item,
+			       sizeof(mr_status->sig_err));
+		else {
+			mr_status->sig_err.err_type = IB_SIG_BAD_GUARD;
+			mr_status->sig_err.sig_err_offset = 0;
+			mr_status->sig_err.key = mmr->sig->err_item.key;
+		}
+
+		mmr->sig->sig_err_exists = false;
+		mr_status->fail_status |= IB_MR_CHECK_SIG_STATUS;
+	}
+
+done:
+	return ret;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/odp.c b/kernel/drivers/infiniband/hw/mlx5/odp.c
new file mode 100644
index 000000000..5099db08a
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/odp.c
@@ -0,0 +1,798 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_umem.h>
+#include <rdma/ib_umem_odp.h>
+
+#include "mlx5_ib.h"
+
+#define MAX_PREFETCH_LEN (4*1024*1024U)
+
+/* Timeout in ms to wait for an active mmu notifier to complete when handling
+ * a pagefault. */
+#define MMU_NOTIFIER_TIMEOUT 1000
+
+struct workqueue_struct *mlx5_ib_page_fault_wq;
+
+void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start,
+			      unsigned long end)
+{
+	struct mlx5_ib_mr *mr;
+	const u64 umr_block_mask = (MLX5_UMR_MTT_ALIGNMENT / sizeof(u64)) - 1;
+	u64 idx = 0, blk_start_idx = 0;
+	int in_block = 0;
+	u64 addr;
+
+	if (!umem || !umem->odp_data) {
+		pr_err("invalidation called on NULL umem or non-ODP umem\n");
+		return;
+	}
+
+	mr = umem->odp_data->private;
+
+	if (!mr || !mr->ibmr.pd)
+		return;
+
+	start = max_t(u64, ib_umem_start(umem), start);
+	end = min_t(u64, ib_umem_end(umem), end);
+
+	/*
+	 * Iteration one - zap the HW's MTTs. The notifiers_count ensures that
+	 * while we are doing the invalidation, no page fault will attempt to
+	 * overwrite the same MTTs.  Concurent invalidations might race us,
+	 * but they will write 0s as well, so no difference in the end result.
+	 */
+
+	for (addr = start; addr < end; addr += (u64)umem->page_size) {
+		idx = (addr - ib_umem_start(umem)) / PAGE_SIZE;
+		/*
+		 * Strive to write the MTTs in chunks, but avoid overwriting
+		 * non-existing MTTs. The huristic here can be improved to
+		 * estimate the cost of another UMR vs. the cost of bigger
+		 * UMR.
+		 */
+		if (umem->odp_data->dma_list[idx] &
+		    (ODP_READ_ALLOWED_BIT | ODP_WRITE_ALLOWED_BIT)) {
+			if (!in_block) {
+				blk_start_idx = idx;
+				in_block = 1;
+			}
+		} else {
+			u64 umr_offset = idx & umr_block_mask;
+
+			if (in_block && umr_offset == 0) {
+				mlx5_ib_update_mtt(mr, blk_start_idx,
+						   idx - blk_start_idx, 1);
+				in_block = 0;
+			}
+		}
+	}
+	if (in_block)
+		mlx5_ib_update_mtt(mr, blk_start_idx, idx - blk_start_idx + 1,
+				   1);
+
+	/*
+	 * We are now sure that the device will not access the
+	 * memory. We can safely unmap it, and mark it as dirty if
+	 * needed.
+	 */
+
+	ib_umem_odp_unmap_dma_pages(umem, start, end);
+}
+
+#define COPY_ODP_BIT_MLX_TO_IB(reg, ib_caps, field_name, bit_name) do {	\
+	if (be32_to_cpu(reg.field_name) & MLX5_ODP_SUPPORT_##bit_name)	\
+		ib_caps->field_name |= IB_ODP_SUPPORT_##bit_name;	\
+} while (0)
+
+int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev *dev)
+{
+	int err;
+	struct mlx5_odp_caps hw_caps;
+	struct ib_odp_caps *caps = &dev->odp_caps;
+
+	memset(caps, 0, sizeof(*caps));
+
+	if (!(dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_ON_DMND_PG))
+		return 0;
+
+	err = mlx5_query_odp_caps(dev->mdev, &hw_caps);
+	if (err)
+		goto out;
+
+	caps->general_caps = IB_ODP_SUPPORT;
+	COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.ud_odp_caps,
+			       SEND);
+	COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps,
+			       SEND);
+	COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps,
+			       RECV);
+	COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps,
+			       WRITE);
+	COPY_ODP_BIT_MLX_TO_IB(hw_caps, caps, per_transport_caps.rc_odp_caps,
+			       READ);
+
+out:
+	return err;
+}
+
+static struct mlx5_ib_mr *mlx5_ib_odp_find_mr_lkey(struct mlx5_ib_dev *dev,
+						   u32 key)
+{
+	u32 base_key = mlx5_base_mkey(key);
+	struct mlx5_core_mr *mmr = __mlx5_mr_lookup(dev->mdev, base_key);
+	struct mlx5_ib_mr *mr = container_of(mmr, struct mlx5_ib_mr, mmr);
+
+	if (!mmr || mmr->key != key || !mr->live)
+		return NULL;
+
+	return container_of(mmr, struct mlx5_ib_mr, mmr);
+}
+
+static void mlx5_ib_page_fault_resume(struct mlx5_ib_qp *qp,
+				      struct mlx5_ib_pfault *pfault,
+				      int error) {
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device);
+	int ret = mlx5_core_page_fault_resume(dev->mdev, qp->mqp.qpn,
+					      pfault->mpfault.flags,
+					      error);
+	if (ret)
+		pr_err("Failed to resolve the page fault on QP 0x%x\n",
+		       qp->mqp.qpn);
+}
+
+/*
+ * Handle a single data segment in a page-fault WQE.
+ *
+ * Returns number of pages retrieved on success. The caller will continue to
+ * the next data segment.
+ * Can return the following error codes:
+ * -EAGAIN to designate a temporary error. The caller will abort handling the
+ *  page fault and resolve it.
+ * -EFAULT when there's an error mapping the requested pages. The caller will
+ *  abort the page fault handling and possibly move the QP to an error state.
+ * On other errors the QP should also be closed with an error.
+ */
+static int pagefault_single_data_segment(struct mlx5_ib_qp *qp,
+					 struct mlx5_ib_pfault *pfault,
+					 u32 key, u64 io_virt, size_t bcnt,
+					 u32 *bytes_mapped)
+{
+	struct mlx5_ib_dev *mib_dev = to_mdev(qp->ibqp.pd->device);
+	int srcu_key;
+	unsigned int current_seq;
+	u64 start_idx;
+	int npages = 0, ret = 0;
+	struct mlx5_ib_mr *mr;
+	u64 access_mask = ODP_READ_ALLOWED_BIT;
+
+	srcu_key = srcu_read_lock(&mib_dev->mr_srcu);
+	mr = mlx5_ib_odp_find_mr_lkey(mib_dev, key);
+	/*
+	 * If we didn't find the MR, it means the MR was closed while we were
+	 * handling the ODP event. In this case we return -EFAULT so that the
+	 * QP will be closed.
+	 */
+	if (!mr || !mr->ibmr.pd) {
+		pr_err("Failed to find relevant mr for lkey=0x%06x, probably the MR was destroyed\n",
+		       key);
+		ret = -EFAULT;
+		goto srcu_unlock;
+	}
+	if (!mr->umem->odp_data) {
+		pr_debug("skipping non ODP MR (lkey=0x%06x) in page fault handler.\n",
+			 key);
+		if (bytes_mapped)
+			*bytes_mapped +=
+				(bcnt - pfault->mpfault.bytes_committed);
+		goto srcu_unlock;
+	}
+	if (mr->ibmr.pd != qp->ibqp.pd) {
+		pr_err("Page-fault with different PDs for QP and MR.\n");
+		ret = -EFAULT;
+		goto srcu_unlock;
+	}
+
+	current_seq = ACCESS_ONCE(mr->umem->odp_data->notifiers_seq);
+	/*
+	 * Ensure the sequence number is valid for some time before we call
+	 * gup.
+	 */
+	smp_rmb();
+
+	/*
+	 * Avoid branches - this code will perform correctly
+	 * in all iterations (in iteration 2 and above,
+	 * bytes_committed == 0).
+	 */
+	io_virt += pfault->mpfault.bytes_committed;
+	bcnt -= pfault->mpfault.bytes_committed;
+
+	start_idx = (io_virt - (mr->mmr.iova & PAGE_MASK)) >> PAGE_SHIFT;
+
+	if (mr->umem->writable)
+		access_mask |= ODP_WRITE_ALLOWED_BIT;
+	npages = ib_umem_odp_map_dma_pages(mr->umem, io_virt, bcnt,
+					   access_mask, current_seq);
+	if (npages < 0) {
+		ret = npages;
+		goto srcu_unlock;
+	}
+
+	if (npages > 0) {
+		mutex_lock(&mr->umem->odp_data->umem_mutex);
+		if (!ib_umem_mmu_notifier_retry(mr->umem, current_seq)) {
+			/*
+			 * No need to check whether the MTTs really belong to
+			 * this MR, since ib_umem_odp_map_dma_pages already
+			 * checks this.
+			 */
+			ret = mlx5_ib_update_mtt(mr, start_idx, npages, 0);
+		} else {
+			ret = -EAGAIN;
+		}
+		mutex_unlock(&mr->umem->odp_data->umem_mutex);
+		if (ret < 0) {
+			if (ret != -EAGAIN)
+				pr_err("Failed to update mkey page tables\n");
+			goto srcu_unlock;
+		}
+
+		if (bytes_mapped) {
+			u32 new_mappings = npages * PAGE_SIZE -
+				(io_virt - round_down(io_virt, PAGE_SIZE));
+			*bytes_mapped += min_t(u32, new_mappings, bcnt);
+		}
+	}
+
+srcu_unlock:
+	if (ret == -EAGAIN) {
+		if (!mr->umem->odp_data->dying) {
+			struct ib_umem_odp *odp_data = mr->umem->odp_data;
+			unsigned long timeout =
+				msecs_to_jiffies(MMU_NOTIFIER_TIMEOUT);
+
+			if (!wait_for_completion_timeout(
+					&odp_data->notifier_completion,
+					timeout)) {
+				pr_warn("timeout waiting for mmu notifier completion\n");
+			}
+		} else {
+			/* The MR is being killed, kill the QP as well. */
+			ret = -EFAULT;
+		}
+	}
+	srcu_read_unlock(&mib_dev->mr_srcu, srcu_key);
+	pfault->mpfault.bytes_committed = 0;
+	return ret ? ret : npages;
+}
+
+/**
+ * Parse a series of data segments for page fault handling.
+ *
+ * @qp the QP on which the fault occurred.
+ * @pfault contains page fault information.
+ * @wqe points at the first data segment in the WQE.
+ * @wqe_end points after the end of the WQE.
+ * @bytes_mapped receives the number of bytes that the function was able to
+ *               map. This allows the caller to decide intelligently whether
+ *               enough memory was mapped to resolve the page fault
+ *               successfully (e.g. enough for the next MTU, or the entire
+ *               WQE).
+ * @total_wqe_bytes receives the total data size of this WQE in bytes (minus
+ *                  the committed bytes).
+ *
+ * Returns the number of pages loaded if positive, zero for an empty WQE, or a
+ * negative error code.
+ */
+static int pagefault_data_segments(struct mlx5_ib_qp *qp,
+				   struct mlx5_ib_pfault *pfault, void *wqe,
+				   void *wqe_end, u32 *bytes_mapped,
+				   u32 *total_wqe_bytes, int receive_queue)
+{
+	int ret = 0, npages = 0;
+	u64 io_virt;
+	u32 key;
+	u32 byte_count;
+	size_t bcnt;
+	int inline_segment;
+
+	/* Skip SRQ next-WQE segment. */
+	if (receive_queue && qp->ibqp.srq)
+		wqe += sizeof(struct mlx5_wqe_srq_next_seg);
+
+	if (bytes_mapped)
+		*bytes_mapped = 0;
+	if (total_wqe_bytes)
+		*total_wqe_bytes = 0;
+
+	while (wqe < wqe_end) {
+		struct mlx5_wqe_data_seg *dseg = wqe;
+
+		io_virt = be64_to_cpu(dseg->addr);
+		key = be32_to_cpu(dseg->lkey);
+		byte_count = be32_to_cpu(dseg->byte_count);
+		inline_segment = !!(byte_count &  MLX5_INLINE_SEG);
+		bcnt	       = byte_count & ~MLX5_INLINE_SEG;
+
+		if (inline_segment) {
+			bcnt = bcnt & MLX5_WQE_INLINE_SEG_BYTE_COUNT_MASK;
+			wqe += ALIGN(sizeof(struct mlx5_wqe_inline_seg) + bcnt,
+				     16);
+		} else {
+			wqe += sizeof(*dseg);
+		}
+
+		/* receive WQE end of sg list. */
+		if (receive_queue && bcnt == 0 && key == MLX5_INVALID_LKEY &&
+		    io_virt == 0)
+			break;
+
+		if (!inline_segment && total_wqe_bytes) {
+			*total_wqe_bytes += bcnt - min_t(size_t, bcnt,
+					pfault->mpfault.bytes_committed);
+		}
+
+		/* A zero length data segment designates a length of 2GB. */
+		if (bcnt == 0)
+			bcnt = 1U << 31;
+
+		if (inline_segment || bcnt <= pfault->mpfault.bytes_committed) {
+			pfault->mpfault.bytes_committed -=
+				min_t(size_t, bcnt,
+				      pfault->mpfault.bytes_committed);
+			continue;
+		}
+
+		ret = pagefault_single_data_segment(qp, pfault, key, io_virt,
+						    bcnt, bytes_mapped);
+		if (ret < 0)
+			break;
+		npages += ret;
+	}
+
+	return ret < 0 ? ret : npages;
+}
+
+/*
+ * Parse initiator WQE. Advances the wqe pointer to point at the
+ * scatter-gather list, and set wqe_end to the end of the WQE.
+ */
+static int mlx5_ib_mr_initiator_pfault_handler(
+	struct mlx5_ib_qp *qp, struct mlx5_ib_pfault *pfault,
+	void **wqe, void **wqe_end, int wqe_length)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device);
+	struct mlx5_wqe_ctrl_seg *ctrl = *wqe;
+	u16 wqe_index = pfault->mpfault.wqe.wqe_index;
+	unsigned ds, opcode;
+#if defined(DEBUG)
+	u32 ctrl_wqe_index, ctrl_qpn;
+#endif
+
+	ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK;
+	if (ds * MLX5_WQE_DS_UNITS > wqe_length) {
+		mlx5_ib_err(dev, "Unable to read the complete WQE. ds = 0x%x, ret = 0x%x\n",
+			    ds, wqe_length);
+		return -EFAULT;
+	}
+
+	if (ds == 0) {
+		mlx5_ib_err(dev, "Got WQE with zero DS. wqe_index=%x, qpn=%x\n",
+			    wqe_index, qp->mqp.qpn);
+		return -EFAULT;
+	}
+
+#if defined(DEBUG)
+	ctrl_wqe_index = (be32_to_cpu(ctrl->opmod_idx_opcode) &
+			MLX5_WQE_CTRL_WQE_INDEX_MASK) >>
+			MLX5_WQE_CTRL_WQE_INDEX_SHIFT;
+	if (wqe_index != ctrl_wqe_index) {
+		mlx5_ib_err(dev, "Got WQE with invalid wqe_index. wqe_index=0x%x, qpn=0x%x ctrl->wqe_index=0x%x\n",
+			    wqe_index, qp->mqp.qpn,
+			    ctrl_wqe_index);
+		return -EFAULT;
+	}
+
+	ctrl_qpn = (be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_QPN_MASK) >>
+		MLX5_WQE_CTRL_QPN_SHIFT;
+	if (qp->mqp.qpn != ctrl_qpn) {
+		mlx5_ib_err(dev, "Got WQE with incorrect QP number. wqe_index=0x%x, qpn=0x%x ctrl->qpn=0x%x\n",
+			    wqe_index, qp->mqp.qpn,
+			    ctrl_qpn);
+		return -EFAULT;
+	}
+#endif /* DEBUG */
+
+	*wqe_end = *wqe + ds * MLX5_WQE_DS_UNITS;
+	*wqe += sizeof(*ctrl);
+
+	opcode = be32_to_cpu(ctrl->opmod_idx_opcode) &
+		 MLX5_WQE_CTRL_OPCODE_MASK;
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_RC:
+		switch (opcode) {
+		case MLX5_OPCODE_SEND:
+		case MLX5_OPCODE_SEND_IMM:
+		case MLX5_OPCODE_SEND_INVAL:
+			if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
+			      IB_ODP_SUPPORT_SEND))
+				goto invalid_transport_or_opcode;
+			break;
+		case MLX5_OPCODE_RDMA_WRITE:
+		case MLX5_OPCODE_RDMA_WRITE_IMM:
+			if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
+			      IB_ODP_SUPPORT_WRITE))
+				goto invalid_transport_or_opcode;
+			*wqe += sizeof(struct mlx5_wqe_raddr_seg);
+			break;
+		case MLX5_OPCODE_RDMA_READ:
+			if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
+			      IB_ODP_SUPPORT_READ))
+				goto invalid_transport_or_opcode;
+			*wqe += sizeof(struct mlx5_wqe_raddr_seg);
+			break;
+		default:
+			goto invalid_transport_or_opcode;
+		}
+		break;
+	case IB_QPT_UD:
+		switch (opcode) {
+		case MLX5_OPCODE_SEND:
+		case MLX5_OPCODE_SEND_IMM:
+			if (!(dev->odp_caps.per_transport_caps.ud_odp_caps &
+			      IB_ODP_SUPPORT_SEND))
+				goto invalid_transport_or_opcode;
+			*wqe += sizeof(struct mlx5_wqe_datagram_seg);
+			break;
+		default:
+			goto invalid_transport_or_opcode;
+		}
+		break;
+	default:
+invalid_transport_or_opcode:
+		mlx5_ib_err(dev, "ODP fault on QP of an unsupported opcode or transport. transport: 0x%x opcode: 0x%x.\n",
+			    qp->ibqp.qp_type, opcode);
+		return -EFAULT;
+	}
+
+	return 0;
+}
+
+/*
+ * Parse responder WQE. Advances the wqe pointer to point at the
+ * scatter-gather list, and set wqe_end to the end of the WQE.
+ */
+static int mlx5_ib_mr_responder_pfault_handler(
+	struct mlx5_ib_qp *qp, struct mlx5_ib_pfault *pfault,
+	void **wqe, void **wqe_end, int wqe_length)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device);
+	struct mlx5_ib_wq *wq = &qp->rq;
+	int wqe_size = 1 << wq->wqe_shift;
+
+	if (qp->ibqp.srq) {
+		mlx5_ib_err(dev, "ODP fault on SRQ is not supported\n");
+		return -EFAULT;
+	}
+
+	if (qp->wq_sig) {
+		mlx5_ib_err(dev, "ODP fault with WQE signatures is not supported\n");
+		return -EFAULT;
+	}
+
+	if (wqe_size > wqe_length) {
+		mlx5_ib_err(dev, "Couldn't read all of the receive WQE's content\n");
+		return -EFAULT;
+	}
+
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_RC:
+		if (!(dev->odp_caps.per_transport_caps.rc_odp_caps &
+		      IB_ODP_SUPPORT_RECV))
+			goto invalid_transport_or_opcode;
+		break;
+	default:
+invalid_transport_or_opcode:
+		mlx5_ib_err(dev, "ODP fault on QP of an unsupported transport. transport: 0x%x\n",
+			    qp->ibqp.qp_type);
+		return -EFAULT;
+	}
+
+	*wqe_end = *wqe + wqe_size;
+
+	return 0;
+}
+
+static void mlx5_ib_mr_wqe_pfault_handler(struct mlx5_ib_qp *qp,
+					  struct mlx5_ib_pfault *pfault)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.pd->device);
+	int ret;
+	void *wqe, *wqe_end;
+	u32 bytes_mapped, total_wqe_bytes;
+	char *buffer = NULL;
+	int resume_with_error = 0;
+	u16 wqe_index = pfault->mpfault.wqe.wqe_index;
+	int requestor = pfault->mpfault.flags & MLX5_PFAULT_REQUESTOR;
+
+	buffer = (char *)__get_free_page(GFP_KERNEL);
+	if (!buffer) {
+		mlx5_ib_err(dev, "Error allocating memory for IO page fault handling.\n");
+		resume_with_error = 1;
+		goto resolve_page_fault;
+	}
+
+	ret = mlx5_ib_read_user_wqe(qp, requestor, wqe_index, buffer,
+				    PAGE_SIZE);
+	if (ret < 0) {
+		mlx5_ib_err(dev, "Failed reading a WQE following page fault, error=%x, wqe_index=%x, qpn=%x\n",
+			    -ret, wqe_index, qp->mqp.qpn);
+		resume_with_error = 1;
+		goto resolve_page_fault;
+	}
+
+	wqe = buffer;
+	if (requestor)
+		ret = mlx5_ib_mr_initiator_pfault_handler(qp, pfault, &wqe,
+							  &wqe_end, ret);
+	else
+		ret = mlx5_ib_mr_responder_pfault_handler(qp, pfault, &wqe,
+							  &wqe_end, ret);
+	if (ret < 0) {
+		resume_with_error = 1;
+		goto resolve_page_fault;
+	}
+
+	if (wqe >= wqe_end) {
+		mlx5_ib_err(dev, "ODP fault on invalid WQE.\n");
+		resume_with_error = 1;
+		goto resolve_page_fault;
+	}
+
+	ret = pagefault_data_segments(qp, pfault, wqe, wqe_end, &bytes_mapped,
+				      &total_wqe_bytes, !requestor);
+	if (ret == -EAGAIN) {
+		goto resolve_page_fault;
+	} else if (ret < 0 || total_wqe_bytes > bytes_mapped) {
+		mlx5_ib_err(dev, "Error getting user pages for page fault. Error: 0x%x\n",
+			    -ret);
+		resume_with_error = 1;
+		goto resolve_page_fault;
+	}
+
+resolve_page_fault:
+	mlx5_ib_page_fault_resume(qp, pfault, resume_with_error);
+	mlx5_ib_dbg(dev, "PAGE FAULT completed. QP 0x%x resume_with_error=%d, flags: 0x%x\n",
+		    qp->mqp.qpn, resume_with_error, pfault->mpfault.flags);
+
+	free_page((unsigned long)buffer);
+}
+
+static int pages_in_range(u64 address, u32 length)
+{
+	return (ALIGN(address + length, PAGE_SIZE) -
+		(address & PAGE_MASK)) >> PAGE_SHIFT;
+}
+
+static void mlx5_ib_mr_rdma_pfault_handler(struct mlx5_ib_qp *qp,
+					   struct mlx5_ib_pfault *pfault)
+{
+	struct mlx5_pagefault *mpfault = &pfault->mpfault;
+	u64 address;
+	u32 length;
+	u32 prefetch_len = mpfault->bytes_committed;
+	int prefetch_activated = 0;
+	u32 rkey = mpfault->rdma.r_key;
+	int ret;
+
+	/* The RDMA responder handler handles the page fault in two parts.
+	 * First it brings the necessary pages for the current packet
+	 * (and uses the pfault context), and then (after resuming the QP)
+	 * prefetches more pages. The second operation cannot use the pfault
+	 * context and therefore uses the dummy_pfault context allocated on
+	 * the stack */
+	struct mlx5_ib_pfault dummy_pfault = {};
+
+	dummy_pfault.mpfault.bytes_committed = 0;
+
+	mpfault->rdma.rdma_va += mpfault->bytes_committed;
+	mpfault->rdma.rdma_op_len -= min(mpfault->bytes_committed,
+					 mpfault->rdma.rdma_op_len);
+	mpfault->bytes_committed = 0;
+
+	address = mpfault->rdma.rdma_va;
+	length  = mpfault->rdma.rdma_op_len;
+
+	/* For some operations, the hardware cannot tell the exact message
+	 * length, and in those cases it reports zero. Use prefetch
+	 * logic. */
+	if (length == 0) {
+		prefetch_activated = 1;
+		length = mpfault->rdma.packet_size;
+		prefetch_len = min(MAX_PREFETCH_LEN, prefetch_len);
+	}
+
+	ret = pagefault_single_data_segment(qp, pfault, rkey, address, length,
+					    NULL);
+	if (ret == -EAGAIN) {
+		/* We're racing with an invalidation, don't prefetch */
+		prefetch_activated = 0;
+	} else if (ret < 0 || pages_in_range(address, length) > ret) {
+		mlx5_ib_page_fault_resume(qp, pfault, 1);
+		return;
+	}
+
+	mlx5_ib_page_fault_resume(qp, pfault, 0);
+
+	/* At this point, there might be a new pagefault already arriving in
+	 * the eq, switch to the dummy pagefault for the rest of the
+	 * processing. We're still OK with the objects being alive as the
+	 * work-queue is being fenced. */
+
+	if (prefetch_activated) {
+		ret = pagefault_single_data_segment(qp, &dummy_pfault, rkey,
+						    address,
+						    prefetch_len,
+						    NULL);
+		if (ret < 0) {
+			pr_warn("Prefetch failed (ret = %d, prefetch_activated = %d) for QPN %d, address: 0x%.16llx, length = 0x%.16x\n",
+				ret, prefetch_activated,
+				qp->ibqp.qp_num, address, prefetch_len);
+		}
+	}
+}
+
+void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
+			       struct mlx5_ib_pfault *pfault)
+{
+	u8 event_subtype = pfault->mpfault.event_subtype;
+
+	switch (event_subtype) {
+	case MLX5_PFAULT_SUBTYPE_WQE:
+		mlx5_ib_mr_wqe_pfault_handler(qp, pfault);
+		break;
+	case MLX5_PFAULT_SUBTYPE_RDMA:
+		mlx5_ib_mr_rdma_pfault_handler(qp, pfault);
+		break;
+	default:
+		pr_warn("Invalid page fault event subtype: 0x%x\n",
+			event_subtype);
+		mlx5_ib_page_fault_resume(qp, pfault, 1);
+		break;
+	}
+}
+
+static void mlx5_ib_qp_pfault_action(struct work_struct *work)
+{
+	struct mlx5_ib_pfault *pfault = container_of(work,
+						     struct mlx5_ib_pfault,
+						     work);
+	enum mlx5_ib_pagefault_context context =
+		mlx5_ib_get_pagefault_context(&pfault->mpfault);
+	struct mlx5_ib_qp *qp = container_of(pfault, struct mlx5_ib_qp,
+					     pagefaults[context]);
+	mlx5_ib_mr_pfault_handler(qp, pfault);
+}
+
+void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->disable_page_faults_lock, flags);
+	qp->disable_page_faults = 1;
+	spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags);
+
+	/*
+	 * Note that at this point, we are guarenteed that no more
+	 * work queue elements will be posted to the work queue with
+	 * the QP we are closing.
+	 */
+	flush_workqueue(mlx5_ib_page_fault_wq);
+}
+
+void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->disable_page_faults_lock, flags);
+	qp->disable_page_faults = 0;
+	spin_unlock_irqrestore(&qp->disable_page_faults_lock, flags);
+}
+
+static void mlx5_ib_pfault_handler(struct mlx5_core_qp *qp,
+				   struct mlx5_pagefault *pfault)
+{
+	/*
+	 * Note that we will only get one fault event per QP per context
+	 * (responder/initiator, read/write), until we resolve the page fault
+	 * with the mlx5_ib_page_fault_resume command. Since this function is
+	 * called from within the work element, there is no risk of missing
+	 * events.
+	 */
+	struct mlx5_ib_qp *mibqp = to_mibqp(qp);
+	enum mlx5_ib_pagefault_context context =
+		mlx5_ib_get_pagefault_context(pfault);
+	struct mlx5_ib_pfault *qp_pfault = &mibqp->pagefaults[context];
+
+	qp_pfault->mpfault = *pfault;
+
+	/* No need to stop interrupts here since we are in an interrupt */
+	spin_lock(&mibqp->disable_page_faults_lock);
+	if (!mibqp->disable_page_faults)
+		queue_work(mlx5_ib_page_fault_wq, &qp_pfault->work);
+	spin_unlock(&mibqp->disable_page_faults_lock);
+}
+
+void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)
+{
+	int i;
+
+	qp->disable_page_faults = 1;
+	spin_lock_init(&qp->disable_page_faults_lock);
+
+	qp->mqp.pfault_handler	= mlx5_ib_pfault_handler;
+
+	for (i = 0; i < MLX5_IB_PAGEFAULT_CONTEXTS; ++i)
+		INIT_WORK(&qp->pagefaults[i].work, mlx5_ib_qp_pfault_action);
+}
+
+int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev)
+{
+	int ret;
+
+	ret = init_srcu_struct(&ibdev->mr_srcu);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)
+{
+	cleanup_srcu_struct(&ibdev->mr_srcu);
+}
+
+int __init mlx5_ib_odp_init(void)
+{
+	mlx5_ib_page_fault_wq =
+		create_singlethread_workqueue("mlx5_ib_page_faults");
+	if (!mlx5_ib_page_fault_wq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void mlx5_ib_odp_cleanup(void)
+{
+	destroy_workqueue(mlx5_ib_page_fault_wq);
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/qp.c b/kernel/drivers/infiniband/hw/mlx5/qp.c
new file mode 100644
index 000000000..d35f62d4f
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/qp.c
@@ -0,0 +1,3174 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <rdma/ib_umem.h>
+#include "mlx5_ib.h"
+#include "user.h"
+
+/* not supported currently */
+static int wq_signature;
+
+enum {
+	MLX5_IB_ACK_REQ_FREQ	= 8,
+};
+
+enum {
+	MLX5_IB_DEFAULT_SCHED_QUEUE	= 0x83,
+	MLX5_IB_DEFAULT_QP0_SCHED_QUEUE	= 0x3f,
+	MLX5_IB_LINK_TYPE_IB		= 0,
+	MLX5_IB_LINK_TYPE_ETH		= 1
+};
+
+enum {
+	MLX5_IB_SQ_STRIDE	= 6,
+	MLX5_IB_CACHE_LINE_SIZE	= 64,
+};
+
+static const u32 mlx5_ib_opcode[] = {
+	[IB_WR_SEND]				= MLX5_OPCODE_SEND,
+	[IB_WR_SEND_WITH_IMM]			= MLX5_OPCODE_SEND_IMM,
+	[IB_WR_RDMA_WRITE]			= MLX5_OPCODE_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM]		= MLX5_OPCODE_RDMA_WRITE_IMM,
+	[IB_WR_RDMA_READ]			= MLX5_OPCODE_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP]		= MLX5_OPCODE_ATOMIC_CS,
+	[IB_WR_ATOMIC_FETCH_AND_ADD]		= MLX5_OPCODE_ATOMIC_FA,
+	[IB_WR_SEND_WITH_INV]			= MLX5_OPCODE_SEND_INVAL,
+	[IB_WR_LOCAL_INV]			= MLX5_OPCODE_UMR,
+	[IB_WR_FAST_REG_MR]			= MLX5_OPCODE_UMR,
+	[IB_WR_MASKED_ATOMIC_CMP_AND_SWP]	= MLX5_OPCODE_ATOMIC_MASKED_CS,
+	[IB_WR_MASKED_ATOMIC_FETCH_AND_ADD]	= MLX5_OPCODE_ATOMIC_MASKED_FA,
+	[MLX5_IB_WR_UMR]			= MLX5_OPCODE_UMR,
+};
+
+
+static int is_qp0(enum ib_qp_type qp_type)
+{
+	return qp_type == IB_QPT_SMI;
+}
+
+static int is_qp1(enum ib_qp_type qp_type)
+{
+	return qp_type == IB_QPT_GSI;
+}
+
+static int is_sqp(enum ib_qp_type qp_type)
+{
+	return is_qp0(qp_type) || is_qp1(qp_type);
+}
+
+static void *get_wqe(struct mlx5_ib_qp *qp, int offset)
+{
+	return mlx5_buf_offset(&qp->buf, offset);
+}
+
+static void *get_recv_wqe(struct mlx5_ib_qp *qp, int n)
+{
+	return get_wqe(qp, qp->rq.offset + (n << qp->rq.wqe_shift));
+}
+
+void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n)
+{
+	return get_wqe(qp, qp->sq.offset + (n << MLX5_IB_SQ_STRIDE));
+}
+
+/**
+ * mlx5_ib_read_user_wqe() - Copy a user-space WQE to kernel space.
+ *
+ * @qp: QP to copy from.
+ * @send: copy from the send queue when non-zero, use the receive queue
+ *	  otherwise.
+ * @wqe_index:  index to start copying from. For send work queues, the
+ *		wqe_index is in units of MLX5_SEND_WQE_BB.
+ *		For receive work queue, it is the number of work queue
+ *		element in the queue.
+ * @buffer: destination buffer.
+ * @length: maximum number of bytes to copy.
+ *
+ * Copies at least a single WQE, but may copy more data.
+ *
+ * Return: the number of bytes copied, or an error code.
+ */
+int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index,
+			  void *buffer, u32 length)
+{
+	struct ib_device *ibdev = qp->ibqp.device;
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_wq *wq = send ? &qp->sq : &qp->rq;
+	size_t offset;
+	size_t wq_end;
+	struct ib_umem *umem = qp->umem;
+	u32 first_copy_length;
+	int wqe_length;
+	int ret;
+
+	if (wq->wqe_cnt == 0) {
+		mlx5_ib_dbg(dev, "mlx5_ib_read_user_wqe for a QP with wqe_cnt == 0. qp_type: 0x%x\n",
+			    qp->ibqp.qp_type);
+		return -EINVAL;
+	}
+
+	offset = wq->offset + ((wqe_index % wq->wqe_cnt) << wq->wqe_shift);
+	wq_end = wq->offset + (wq->wqe_cnt << wq->wqe_shift);
+
+	if (send && length < sizeof(struct mlx5_wqe_ctrl_seg))
+		return -EINVAL;
+
+	if (offset > umem->length ||
+	    (send && offset + sizeof(struct mlx5_wqe_ctrl_seg) > umem->length))
+		return -EINVAL;
+
+	first_copy_length = min_t(u32, offset + length, wq_end) - offset;
+	ret = ib_umem_copy_from(buffer, umem, offset, first_copy_length);
+	if (ret)
+		return ret;
+
+	if (send) {
+		struct mlx5_wqe_ctrl_seg *ctrl = buffer;
+		int ds = be32_to_cpu(ctrl->qpn_ds) & MLX5_WQE_CTRL_DS_MASK;
+
+		wqe_length = ds * MLX5_WQE_DS_UNITS;
+	} else {
+		wqe_length = 1 << wq->wqe_shift;
+	}
+
+	if (wqe_length <= first_copy_length)
+		return first_copy_length;
+
+	ret = ib_umem_copy_from(buffer + first_copy_length, umem, wq->offset,
+				wqe_length - first_copy_length);
+	if (ret)
+		return ret;
+
+	return wqe_length;
+}
+
+static void mlx5_ib_qp_event(struct mlx5_core_qp *qp, int type)
+{
+	struct ib_qp *ibqp = &to_mibqp(qp)->ibqp;
+	struct ib_event event;
+
+	if (type == MLX5_EVENT_TYPE_PATH_MIG)
+		to_mibqp(qp)->port = to_mibqp(qp)->alt_port;
+
+	if (ibqp->event_handler) {
+		event.device     = ibqp->device;
+		event.element.qp = ibqp;
+		switch (type) {
+		case MLX5_EVENT_TYPE_PATH_MIG:
+			event.event = IB_EVENT_PATH_MIG;
+			break;
+		case MLX5_EVENT_TYPE_COMM_EST:
+			event.event = IB_EVENT_COMM_EST;
+			break;
+		case MLX5_EVENT_TYPE_SQ_DRAINED:
+			event.event = IB_EVENT_SQ_DRAINED;
+			break;
+		case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
+			event.event = IB_EVENT_QP_LAST_WQE_REACHED;
+			break;
+		case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
+			event.event = IB_EVENT_QP_FATAL;
+			break;
+		case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
+			event.event = IB_EVENT_PATH_MIG_ERR;
+			break;
+		case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+			event.event = IB_EVENT_QP_REQ_ERR;
+			break;
+		case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
+			event.event = IB_EVENT_QP_ACCESS_ERR;
+			break;
+		default:
+			pr_warn("mlx5_ib: Unexpected event type %d on QP %06x\n", type, qp->qpn);
+			return;
+		}
+
+		ibqp->event_handler(&event, ibqp->qp_context);
+	}
+}
+
+static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap,
+		       int has_rq, struct mlx5_ib_qp *qp, struct mlx5_ib_create_qp *ucmd)
+{
+	struct mlx5_general_caps *gen;
+	int wqe_size;
+	int wq_size;
+
+	gen = &dev->mdev->caps.gen;
+	/* Sanity check RQ size before proceeding */
+	if (cap->max_recv_wr  > gen->max_wqes)
+		return -EINVAL;
+
+	if (!has_rq) {
+		qp->rq.max_gs = 0;
+		qp->rq.wqe_cnt = 0;
+		qp->rq.wqe_shift = 0;
+	} else {
+		if (ucmd) {
+			qp->rq.wqe_cnt = ucmd->rq_wqe_count;
+			qp->rq.wqe_shift = ucmd->rq_wqe_shift;
+			qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig;
+			qp->rq.max_post = qp->rq.wqe_cnt;
+		} else {
+			wqe_size = qp->wq_sig ? sizeof(struct mlx5_wqe_signature_seg) : 0;
+			wqe_size += cap->max_recv_sge * sizeof(struct mlx5_wqe_data_seg);
+			wqe_size = roundup_pow_of_two(wqe_size);
+			wq_size = roundup_pow_of_two(cap->max_recv_wr) * wqe_size;
+			wq_size = max_t(int, wq_size, MLX5_SEND_WQE_BB);
+			qp->rq.wqe_cnt = wq_size / wqe_size;
+			if (wqe_size > gen->max_rq_desc_sz) {
+				mlx5_ib_dbg(dev, "wqe_size %d, max %d\n",
+					    wqe_size,
+					    gen->max_rq_desc_sz);
+				return -EINVAL;
+			}
+			qp->rq.wqe_shift = ilog2(wqe_size);
+			qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig;
+			qp->rq.max_post = qp->rq.wqe_cnt;
+		}
+	}
+
+	return 0;
+}
+
+static int sq_overhead(enum ib_qp_type qp_type)
+{
+	int size = 0;
+
+	switch (qp_type) {
+	case IB_QPT_XRC_INI:
+		size += sizeof(struct mlx5_wqe_xrc_seg);
+		/* fall through */
+	case IB_QPT_RC:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_atomic_seg) +
+			sizeof(struct mlx5_wqe_raddr_seg);
+		break;
+
+	case IB_QPT_XRC_TGT:
+		return 0;
+
+	case IB_QPT_UC:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_raddr_seg) +
+			sizeof(struct mlx5_wqe_umr_ctrl_seg) +
+			sizeof(struct mlx5_mkey_seg);
+		break;
+
+	case IB_QPT_UD:
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_datagram_seg);
+		break;
+
+	case MLX5_IB_QPT_REG_UMR:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_umr_ctrl_seg) +
+			sizeof(struct mlx5_mkey_seg);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	return size;
+}
+
+static int calc_send_wqe(struct ib_qp_init_attr *attr)
+{
+	int inl_size = 0;
+	int size;
+
+	size = sq_overhead(attr->qp_type);
+	if (size < 0)
+		return size;
+
+	if (attr->cap.max_inline_data) {
+		inl_size = size + sizeof(struct mlx5_wqe_inline_seg) +
+			attr->cap.max_inline_data;
+	}
+
+	size += attr->cap.max_send_sge * sizeof(struct mlx5_wqe_data_seg);
+	if (attr->create_flags & IB_QP_CREATE_SIGNATURE_EN &&
+	    ALIGN(max_t(int, inl_size, size), MLX5_SEND_WQE_BB) < MLX5_SIG_WQE_SIZE)
+			return MLX5_SIG_WQE_SIZE;
+	else
+		return ALIGN(max_t(int, inl_size, size), MLX5_SEND_WQE_BB);
+}
+
+static int calc_sq_size(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr,
+			struct mlx5_ib_qp *qp)
+{
+	struct mlx5_general_caps *gen;
+	int wqe_size;
+	int wq_size;
+
+	gen = &dev->mdev->caps.gen;
+	if (!attr->cap.max_send_wr)
+		return 0;
+
+	wqe_size = calc_send_wqe(attr);
+	mlx5_ib_dbg(dev, "wqe_size %d\n", wqe_size);
+	if (wqe_size < 0)
+		return wqe_size;
+
+	if (wqe_size > gen->max_sq_desc_sz) {
+		mlx5_ib_dbg(dev, "wqe_size(%d) > max_sq_desc_sz(%d)\n",
+			    wqe_size, gen->max_sq_desc_sz);
+		return -EINVAL;
+	}
+
+	qp->max_inline_data = wqe_size - sq_overhead(attr->qp_type) -
+		sizeof(struct mlx5_wqe_inline_seg);
+	attr->cap.max_inline_data = qp->max_inline_data;
+
+	if (attr->create_flags & IB_QP_CREATE_SIGNATURE_EN)
+		qp->signature_en = true;
+
+	wq_size = roundup_pow_of_two(attr->cap.max_send_wr * wqe_size);
+	qp->sq.wqe_cnt = wq_size / MLX5_SEND_WQE_BB;
+	if (qp->sq.wqe_cnt > gen->max_wqes) {
+		mlx5_ib_dbg(dev, "wqe count(%d) exceeds limits(%d)\n",
+			    qp->sq.wqe_cnt, gen->max_wqes);
+		return -ENOMEM;
+	}
+	qp->sq.wqe_shift = ilog2(MLX5_SEND_WQE_BB);
+	qp->sq.max_gs = attr->cap.max_send_sge;
+	qp->sq.max_post = wq_size / wqe_size;
+	attr->cap.max_send_wr = qp->sq.max_post;
+
+	return wq_size;
+}
+
+static int set_user_buf_size(struct mlx5_ib_dev *dev,
+			    struct mlx5_ib_qp *qp,
+			    struct mlx5_ib_create_qp *ucmd)
+{
+	struct mlx5_general_caps *gen;
+	int desc_sz = 1 << qp->sq.wqe_shift;
+
+	gen = &dev->mdev->caps.gen;
+	if (desc_sz > gen->max_sq_desc_sz) {
+		mlx5_ib_warn(dev, "desc_sz %d, max_sq_desc_sz %d\n",
+			     desc_sz, gen->max_sq_desc_sz);
+		return -EINVAL;
+	}
+
+	if (ucmd->sq_wqe_count && ((1 << ilog2(ucmd->sq_wqe_count)) != ucmd->sq_wqe_count)) {
+		mlx5_ib_warn(dev, "sq_wqe_count %d, sq_wqe_count %d\n",
+			     ucmd->sq_wqe_count, ucmd->sq_wqe_count);
+		return -EINVAL;
+	}
+
+	qp->sq.wqe_cnt = ucmd->sq_wqe_count;
+
+	if (qp->sq.wqe_cnt > gen->max_wqes) {
+		mlx5_ib_warn(dev, "wqe_cnt %d, max_wqes %d\n",
+			     qp->sq.wqe_cnt, gen->max_wqes);
+		return -EINVAL;
+	}
+
+	qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
+		(qp->sq.wqe_cnt << 6);
+
+	return 0;
+}
+
+static int qp_has_rq(struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_XRC_INI ||
+	    attr->qp_type == IB_QPT_XRC_TGT || attr->srq ||
+	    attr->qp_type == MLX5_IB_QPT_REG_UMR ||
+	    !attr->cap.max_recv_wr)
+		return 0;
+
+	return 1;
+}
+
+static int first_med_uuar(void)
+{
+	return 1;
+}
+
+static int next_uuar(int n)
+{
+	n++;
+
+	while (((n % 4) & 2))
+		n++;
+
+	return n;
+}
+
+static int num_med_uuar(struct mlx5_uuar_info *uuari)
+{
+	int n;
+
+	n = uuari->num_uars * MLX5_NON_FP_BF_REGS_PER_PAGE -
+		uuari->num_low_latency_uuars - 1;
+
+	return n >= 0 ? n : 0;
+}
+
+static int max_uuari(struct mlx5_uuar_info *uuari)
+{
+	return uuari->num_uars * 4;
+}
+
+static int first_hi_uuar(struct mlx5_uuar_info *uuari)
+{
+	int med;
+	int i;
+	int t;
+
+	med = num_med_uuar(uuari);
+	for (t = 0, i = first_med_uuar();; i = next_uuar(i)) {
+		t++;
+		if (t == med)
+			return next_uuar(i);
+	}
+
+	return 0;
+}
+
+static int alloc_high_class_uuar(struct mlx5_uuar_info *uuari)
+{
+	int i;
+
+	for (i = first_hi_uuar(uuari); i < max_uuari(uuari); i = next_uuar(i)) {
+		if (!test_bit(i, uuari->bitmap)) {
+			set_bit(i, uuari->bitmap);
+			uuari->count[i]++;
+			return i;
+		}
+	}
+
+	return -ENOMEM;
+}
+
+static int alloc_med_class_uuar(struct mlx5_uuar_info *uuari)
+{
+	int minidx = first_med_uuar();
+	int i;
+
+	for (i = first_med_uuar(); i < first_hi_uuar(uuari); i = next_uuar(i)) {
+		if (uuari->count[i] < uuari->count[minidx])
+			minidx = i;
+	}
+
+	uuari->count[minidx]++;
+	return minidx;
+}
+
+static int alloc_uuar(struct mlx5_uuar_info *uuari,
+		      enum mlx5_ib_latency_class lat)
+{
+	int uuarn = -EINVAL;
+
+	mutex_lock(&uuari->lock);
+	switch (lat) {
+	case MLX5_IB_LATENCY_CLASS_LOW:
+		uuarn = 0;
+		uuari->count[uuarn]++;
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_MEDIUM:
+		if (uuari->ver < 2)
+			uuarn = -ENOMEM;
+		else
+			uuarn = alloc_med_class_uuar(uuari);
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_HIGH:
+		if (uuari->ver < 2)
+			uuarn = -ENOMEM;
+		else
+			uuarn = alloc_high_class_uuar(uuari);
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_FAST_PATH:
+		uuarn = 2;
+		break;
+	}
+	mutex_unlock(&uuari->lock);
+
+	return uuarn;
+}
+
+static void free_med_class_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	clear_bit(uuarn, uuari->bitmap);
+	--uuari->count[uuarn];
+}
+
+static void free_high_class_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	clear_bit(uuarn, uuari->bitmap);
+	--uuari->count[uuarn];
+}
+
+static void free_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	int nuuars = uuari->num_uars * MLX5_BF_REGS_PER_PAGE;
+	int high_uuar = nuuars - uuari->num_low_latency_uuars;
+
+	mutex_lock(&uuari->lock);
+	if (uuarn == 0) {
+		--uuari->count[uuarn];
+		goto out;
+	}
+
+	if (uuarn < high_uuar) {
+		free_med_class_uuar(uuari, uuarn);
+		goto out;
+	}
+
+	free_high_class_uuar(uuari, uuarn);
+
+out:
+	mutex_unlock(&uuari->lock);
+}
+
+static enum mlx5_qp_state to_mlx5_state(enum ib_qp_state state)
+{
+	switch (state) {
+	case IB_QPS_RESET:	return MLX5_QP_STATE_RST;
+	case IB_QPS_INIT:	return MLX5_QP_STATE_INIT;
+	case IB_QPS_RTR:	return MLX5_QP_STATE_RTR;
+	case IB_QPS_RTS:	return MLX5_QP_STATE_RTS;
+	case IB_QPS_SQD:	return MLX5_QP_STATE_SQD;
+	case IB_QPS_SQE:	return MLX5_QP_STATE_SQER;
+	case IB_QPS_ERR:	return MLX5_QP_STATE_ERR;
+	default:		return -1;
+	}
+}
+
+static int to_mlx5_st(enum ib_qp_type type)
+{
+	switch (type) {
+	case IB_QPT_RC:			return MLX5_QP_ST_RC;
+	case IB_QPT_UC:			return MLX5_QP_ST_UC;
+	case IB_QPT_UD:			return MLX5_QP_ST_UD;
+	case MLX5_IB_QPT_REG_UMR:	return MLX5_QP_ST_REG_UMR;
+	case IB_QPT_XRC_INI:
+	case IB_QPT_XRC_TGT:		return MLX5_QP_ST_XRC;
+	case IB_QPT_SMI:		return MLX5_QP_ST_QP0;
+	case IB_QPT_GSI:		return MLX5_QP_ST_QP1;
+	case IB_QPT_RAW_IPV6:		return MLX5_QP_ST_RAW_IPV6;
+	case IB_QPT_RAW_ETHERTYPE:	return MLX5_QP_ST_RAW_ETHERTYPE;
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:		return -EINVAL;
+	}
+}
+
+static int uuarn_to_uar_index(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	return uuari->uars[uuarn / MLX5_BF_REGS_PER_PAGE].index;
+}
+
+static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			  struct mlx5_ib_qp *qp, struct ib_udata *udata,
+			  struct mlx5_create_qp_mbox_in **in,
+			  struct mlx5_ib_create_qp_resp *resp, int *inlen)
+{
+	struct mlx5_ib_ucontext *context;
+	struct mlx5_ib_create_qp ucmd;
+	int page_shift = 0;
+	int uar_index;
+	int npages;
+	u32 offset = 0;
+	int uuarn;
+	int ncont = 0;
+	int err;
+
+	err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
+	if (err) {
+		mlx5_ib_dbg(dev, "copy failed\n");
+		return err;
+	}
+
+	context = to_mucontext(pd->uobject->context);
+	/*
+	 * TBD: should come from the verbs when we have the API
+	 */
+	uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_HIGH);
+	if (uuarn < 0) {
+		mlx5_ib_dbg(dev, "failed to allocate low latency UUAR\n");
+		mlx5_ib_dbg(dev, "reverting to medium latency\n");
+		uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_MEDIUM);
+		if (uuarn < 0) {
+			mlx5_ib_dbg(dev, "failed to allocate medium latency UUAR\n");
+			mlx5_ib_dbg(dev, "reverting to high latency\n");
+			uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_LOW);
+			if (uuarn < 0) {
+				mlx5_ib_warn(dev, "uuar allocation failed\n");
+				return uuarn;
+			}
+		}
+	}
+
+	uar_index = uuarn_to_uar_index(&context->uuari, uuarn);
+	mlx5_ib_dbg(dev, "uuarn 0x%x, uar_index 0x%x\n", uuarn, uar_index);
+
+	qp->rq.offset = 0;
+	qp->sq.wqe_shift = ilog2(MLX5_SEND_WQE_BB);
+	qp->sq.offset = qp->rq.wqe_cnt << qp->rq.wqe_shift;
+
+	err = set_user_buf_size(dev, qp, &ucmd);
+	if (err)
+		goto err_uuar;
+
+	if (ucmd.buf_addr && qp->buf_size) {
+		qp->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
+				       qp->buf_size, 0, 0);
+		if (IS_ERR(qp->umem)) {
+			mlx5_ib_dbg(dev, "umem_get failed\n");
+			err = PTR_ERR(qp->umem);
+			goto err_uuar;
+		}
+	} else {
+		qp->umem = NULL;
+	}
+
+	if (qp->umem) {
+		mlx5_ib_cont_pages(qp->umem, ucmd.buf_addr, &npages, &page_shift,
+				   &ncont, NULL);
+		err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift, &offset);
+		if (err) {
+			mlx5_ib_warn(dev, "bad offset\n");
+			goto err_umem;
+		}
+		mlx5_ib_dbg(dev, "addr 0x%llx, size %d, npages %d, page_shift %d, ncont %d, offset %d\n",
+			    ucmd.buf_addr, qp->buf_size, npages, page_shift, ncont, offset);
+	}
+
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * ncont;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_umem;
+	}
+	if (qp->umem)
+		mlx5_ib_populate_pas(dev, qp->umem, page_shift, (*in)->pas, 0);
+	(*in)->ctx.log_pg_sz_remote_qpn =
+		cpu_to_be32((page_shift - MLX5_ADAPTER_PAGE_SHIFT) << 24);
+	(*in)->ctx.params2 = cpu_to_be32(offset << 6);
+
+	(*in)->ctx.qp_counter_set_usr_page = cpu_to_be32(uar_index);
+	resp->uuar_index = uuarn;
+	qp->uuarn = uuarn;
+
+	err = mlx5_ib_db_map_user(context, ucmd.db_addr, &qp->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "map failed\n");
+		goto err_free;
+	}
+
+	err = ib_copy_to_udata(udata, resp, sizeof(*resp));
+	if (err) {
+		mlx5_ib_dbg(dev, "copy failed\n");
+		goto err_unmap;
+	}
+	qp->create_type = MLX5_QP_USER;
+
+	return 0;
+
+err_unmap:
+	mlx5_ib_db_unmap_user(context, &qp->db);
+
+err_free:
+	kvfree(*in);
+
+err_umem:
+	if (qp->umem)
+		ib_umem_release(qp->umem);
+
+err_uuar:
+	free_uuar(&context->uuari, uuarn);
+	return err;
+}
+
+static void destroy_qp_user(struct ib_pd *pd, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_ucontext *context;
+
+	context = to_mucontext(pd->uobject->context);
+	mlx5_ib_db_unmap_user(context, &qp->db);
+	if (qp->umem)
+		ib_umem_release(qp->umem);
+	free_uuar(&context->uuari, qp->uuarn);
+}
+
+static int create_kernel_qp(struct mlx5_ib_dev *dev,
+			    struct ib_qp_init_attr *init_attr,
+			    struct mlx5_ib_qp *qp,
+			    struct mlx5_create_qp_mbox_in **in, int *inlen)
+{
+	enum mlx5_ib_latency_class lc = MLX5_IB_LATENCY_CLASS_LOW;
+	struct mlx5_uuar_info *uuari;
+	int uar_index;
+	int uuarn;
+	int err;
+
+	uuari = &dev->mdev->priv.uuari;
+	if (init_attr->create_flags & ~(IB_QP_CREATE_SIGNATURE_EN | IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK))
+		return -EINVAL;
+
+	if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR)
+		lc = MLX5_IB_LATENCY_CLASS_FAST_PATH;
+
+	uuarn = alloc_uuar(uuari, lc);
+	if (uuarn < 0) {
+		mlx5_ib_dbg(dev, "\n");
+		return -ENOMEM;
+	}
+
+	qp->bf = &uuari->bfs[uuarn];
+	uar_index = qp->bf->uar->index;
+
+	err = calc_sq_size(dev, init_attr, qp);
+	if (err < 0) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_uuar;
+	}
+
+	qp->rq.offset = 0;
+	qp->sq.offset = qp->rq.wqe_cnt << qp->rq.wqe_shift;
+	qp->buf_size = err + (qp->rq.wqe_cnt << qp->rq.wqe_shift);
+
+	err = mlx5_buf_alloc(dev->mdev, qp->buf_size, PAGE_SIZE * 2, &qp->buf);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_uuar;
+	}
+
+	qp->sq.qend = mlx5_get_send_wqe(qp, qp->sq.wqe_cnt);
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * qp->buf.npages;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	(*in)->ctx.qp_counter_set_usr_page = cpu_to_be32(uar_index);
+	(*in)->ctx.log_pg_sz_remote_qpn =
+		cpu_to_be32((qp->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT) << 24);
+	/* Set "fast registration enabled" for all kernel QPs */
+	(*in)->ctx.params1 |= cpu_to_be32(1 << 11);
+	(*in)->ctx.sq_crq_size |= cpu_to_be16(1 << 4);
+
+	mlx5_fill_page_array(&qp->buf, (*in)->pas);
+
+	err = mlx5_db_alloc(dev->mdev, &qp->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_free;
+	}
+
+	qp->sq.wrid = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wrid), GFP_KERNEL);
+	qp->sq.wr_data = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wr_data), GFP_KERNEL);
+	qp->rq.wrid = kmalloc(qp->rq.wqe_cnt * sizeof(*qp->rq.wrid), GFP_KERNEL);
+	qp->sq.w_list = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.w_list), GFP_KERNEL);
+	qp->sq.wqe_head = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wqe_head), GFP_KERNEL);
+
+	if (!qp->sq.wrid || !qp->sq.wr_data || !qp->rq.wrid ||
+	    !qp->sq.w_list || !qp->sq.wqe_head) {
+		err = -ENOMEM;
+		goto err_wrid;
+	}
+	qp->create_type = MLX5_QP_KERNEL;
+
+	return 0;
+
+err_wrid:
+	mlx5_db_free(dev->mdev, &qp->db);
+	kfree(qp->sq.wqe_head);
+	kfree(qp->sq.w_list);
+	kfree(qp->sq.wrid);
+	kfree(qp->sq.wr_data);
+	kfree(qp->rq.wrid);
+
+err_free:
+	kvfree(*in);
+
+err_buf:
+	mlx5_buf_free(dev->mdev, &qp->buf);
+
+err_uuar:
+	free_uuar(&dev->mdev->priv.uuari, uuarn);
+	return err;
+}
+
+static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
+{
+	mlx5_db_free(dev->mdev, &qp->db);
+	kfree(qp->sq.wqe_head);
+	kfree(qp->sq.w_list);
+	kfree(qp->sq.wrid);
+	kfree(qp->sq.wr_data);
+	kfree(qp->rq.wrid);
+	mlx5_buf_free(dev->mdev, &qp->buf);
+	free_uuar(&dev->mdev->priv.uuari, qp->bf->uuarn);
+}
+
+static __be32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
+{
+	if (attr->srq || (attr->qp_type == IB_QPT_XRC_TGT) ||
+	    (attr->qp_type == IB_QPT_XRC_INI))
+		return cpu_to_be32(MLX5_SRQ_RQ);
+	else if (!qp->has_rq)
+		return cpu_to_be32(MLX5_ZERO_LEN_RQ);
+	else
+		return cpu_to_be32(MLX5_NON_ZERO_RQ);
+}
+
+static int is_connected(enum ib_qp_type qp_type)
+{
+	if (qp_type == IB_QPT_RC || qp_type == IB_QPT_UC)
+		return 1;
+
+	return 0;
+}
+
+static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	struct mlx5_ib_create_qp_resp resp;
+	struct mlx5_create_qp_mbox_in *in;
+	struct mlx5_general_caps *gen;
+	struct mlx5_ib_create_qp ucmd;
+	int inlen = sizeof(*in);
+	int err;
+
+	mlx5_ib_odp_create_qp(qp);
+
+	gen = &dev->mdev->caps.gen;
+	mutex_init(&qp->mutex);
+	spin_lock_init(&qp->sq.lock);
+	spin_lock_init(&qp->rq.lock);
+
+	if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
+		if (!(gen->flags & MLX5_DEV_CAP_FLAG_BLOCK_MCAST)) {
+			mlx5_ib_dbg(dev, "block multicast loopback isn't supported\n");
+			return -EINVAL;
+		} else {
+			qp->flags |= MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK;
+		}
+	}
+
+	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
+		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
+
+	if (pd && pd->uobject) {
+		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
+			mlx5_ib_dbg(dev, "copy failed\n");
+			return -EFAULT;
+		}
+
+		qp->wq_sig = !!(ucmd.flags & MLX5_QP_FLAG_SIGNATURE);
+		qp->scat_cqe = !!(ucmd.flags & MLX5_QP_FLAG_SCATTER_CQE);
+	} else {
+		qp->wq_sig = !!wq_signature;
+	}
+
+	qp->has_rq = qp_has_rq(init_attr);
+	err = set_rq_size(dev, &init_attr->cap, qp->has_rq,
+			  qp, (pd && pd->uobject) ? &ucmd : NULL);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		return err;
+	}
+
+	if (pd) {
+		if (pd->uobject) {
+			mlx5_ib_dbg(dev, "requested sq_wqe_count (%d)\n", ucmd.sq_wqe_count);
+			if (ucmd.rq_wqe_shift != qp->rq.wqe_shift ||
+			    ucmd.rq_wqe_count != qp->rq.wqe_cnt) {
+				mlx5_ib_dbg(dev, "invalid rq params\n");
+				return -EINVAL;
+			}
+			if (ucmd.sq_wqe_count > gen->max_wqes) {
+				mlx5_ib_dbg(dev, "requested sq_wqe_count (%d) > max allowed (%d)\n",
+					    ucmd.sq_wqe_count, gen->max_wqes);
+				return -EINVAL;
+			}
+			err = create_user_qp(dev, pd, qp, udata, &in, &resp, &inlen);
+			if (err)
+				mlx5_ib_dbg(dev, "err %d\n", err);
+		} else {
+			err = create_kernel_qp(dev, init_attr, qp, &in, &inlen);
+			if (err)
+				mlx5_ib_dbg(dev, "err %d\n", err);
+			else
+				qp->pa_lkey = to_mpd(pd)->pa_lkey;
+		}
+
+		if (err)
+			return err;
+	} else {
+		in = mlx5_vzalloc(sizeof(*in));
+		if (!in)
+			return -ENOMEM;
+
+		qp->create_type = MLX5_QP_EMPTY;
+	}
+
+	if (is_sqp(init_attr->qp_type))
+		qp->port = init_attr->port_num;
+
+	in->ctx.flags = cpu_to_be32(to_mlx5_st(init_attr->qp_type) << 16 |
+				    MLX5_QP_PM_MIGRATED << 11);
+
+	if (init_attr->qp_type != MLX5_IB_QPT_REG_UMR)
+		in->ctx.flags_pd = cpu_to_be32(to_mpd(pd ? pd : devr->p0)->pdn);
+	else
+		in->ctx.flags_pd = cpu_to_be32(MLX5_QP_LAT_SENSITIVE);
+
+	if (qp->wq_sig)
+		in->ctx.flags_pd |= cpu_to_be32(MLX5_QP_ENABLE_SIG);
+
+	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+		in->ctx.flags_pd |= cpu_to_be32(MLX5_QP_BLOCK_MCAST);
+
+	if (qp->scat_cqe && is_connected(init_attr->qp_type)) {
+		int rcqe_sz;
+		int scqe_sz;
+
+		rcqe_sz = mlx5_ib_get_cqe_size(dev, init_attr->recv_cq);
+		scqe_sz = mlx5_ib_get_cqe_size(dev, init_attr->send_cq);
+
+		if (rcqe_sz == 128)
+			in->ctx.cs_res = MLX5_RES_SCAT_DATA64_CQE;
+		else
+			in->ctx.cs_res = MLX5_RES_SCAT_DATA32_CQE;
+
+		if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) {
+			if (scqe_sz == 128)
+				in->ctx.cs_req = MLX5_REQ_SCAT_DATA64_CQE;
+			else
+				in->ctx.cs_req = MLX5_REQ_SCAT_DATA32_CQE;
+		}
+	}
+
+	if (qp->rq.wqe_cnt) {
+		in->ctx.rq_size_stride = (qp->rq.wqe_shift - 4);
+		in->ctx.rq_size_stride |= ilog2(qp->rq.wqe_cnt) << 3;
+	}
+
+	in->ctx.rq_type_srqn = get_rx_type(qp, init_attr);
+
+	if (qp->sq.wqe_cnt)
+		in->ctx.sq_crq_size |= cpu_to_be16(ilog2(qp->sq.wqe_cnt) << 11);
+	else
+		in->ctx.sq_crq_size |= cpu_to_be16(0x8000);
+
+	/* Set default resources */
+	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.cqn_send = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		in->ctx.xrcd = cpu_to_be32(to_mxrcd(init_attr->xrcd)->xrcdn);
+		break;
+	case IB_QPT_XRC_INI:
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x1)->xrcdn);
+		in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		break;
+	default:
+		if (init_attr->srq) {
+			in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x0)->xrcdn);
+			in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(init_attr->srq)->msrq.srqn);
+		} else {
+			in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x1)->xrcdn);
+			in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		}
+	}
+
+	if (init_attr->send_cq)
+		in->ctx.cqn_send = cpu_to_be32(to_mcq(init_attr->send_cq)->mcq.cqn);
+
+	if (init_attr->recv_cq)
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(init_attr->recv_cq)->mcq.cqn);
+
+	in->ctx.db_rec_addr = cpu_to_be64(qp->db.dma);
+
+	err = mlx5_core_create_qp(dev->mdev, &qp->mqp, in, inlen);
+	if (err) {
+		mlx5_ib_dbg(dev, "create qp failed\n");
+		goto err_create;
+	}
+
+	kvfree(in);
+	/* Hardware wants QPN written in big-endian order (after
+	 * shifting) for send doorbell.  Precompute this value to save
+	 * a little bit when posting sends.
+	 */
+	qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);
+
+	qp->mqp.event = mlx5_ib_qp_event;
+
+	return 0;
+
+err_create:
+	if (qp->create_type == MLX5_QP_USER)
+		destroy_qp_user(pd, qp);
+	else if (qp->create_type == MLX5_QP_KERNEL)
+		destroy_qp_kernel(dev, qp);
+
+	kvfree(in);
+	return err;
+}
+
+static void mlx5_ib_lock_cqs(struct mlx5_ib_cq *send_cq, struct mlx5_ib_cq *recv_cq)
+	__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
+{
+	if (send_cq) {
+		if (recv_cq) {
+			if (send_cq->mcq.cqn < recv_cq->mcq.cqn)  {
+				spin_lock_irq(&send_cq->lock);
+				spin_lock_nested(&recv_cq->lock,
+						 SINGLE_DEPTH_NESTING);
+			} else if (send_cq->mcq.cqn == recv_cq->mcq.cqn) {
+				spin_lock_irq(&send_cq->lock);
+				__acquire(&recv_cq->lock);
+			} else {
+				spin_lock_irq(&recv_cq->lock);
+				spin_lock_nested(&send_cq->lock,
+						 SINGLE_DEPTH_NESTING);
+			}
+		} else {
+			spin_lock_irq(&send_cq->lock);
+			__acquire(&recv_cq->lock);
+		}
+	} else if (recv_cq) {
+		spin_lock_irq(&recv_cq->lock);
+		__acquire(&send_cq->lock);
+	} else {
+		__acquire(&send_cq->lock);
+		__acquire(&recv_cq->lock);
+	}
+}
+
+static void mlx5_ib_unlock_cqs(struct mlx5_ib_cq *send_cq, struct mlx5_ib_cq *recv_cq)
+	__releases(&send_cq->lock) __releases(&recv_cq->lock)
+{
+	if (send_cq) {
+		if (recv_cq) {
+			if (send_cq->mcq.cqn < recv_cq->mcq.cqn)  {
+				spin_unlock(&recv_cq->lock);
+				spin_unlock_irq(&send_cq->lock);
+			} else if (send_cq->mcq.cqn == recv_cq->mcq.cqn) {
+				__release(&recv_cq->lock);
+				spin_unlock_irq(&send_cq->lock);
+			} else {
+				spin_unlock(&send_cq->lock);
+				spin_unlock_irq(&recv_cq->lock);
+			}
+		} else {
+			__release(&recv_cq->lock);
+			spin_unlock_irq(&send_cq->lock);
+		}
+	} else if (recv_cq) {
+		__release(&send_cq->lock);
+		spin_unlock_irq(&recv_cq->lock);
+	} else {
+		__release(&recv_cq->lock);
+		__release(&send_cq->lock);
+	}
+}
+
+static struct mlx5_ib_pd *get_pd(struct mlx5_ib_qp *qp)
+{
+	return to_mpd(qp->ibqp.pd);
+}
+
+static void get_cqs(struct mlx5_ib_qp *qp,
+		    struct mlx5_ib_cq **send_cq, struct mlx5_ib_cq **recv_cq)
+{
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_XRC_TGT:
+		*send_cq = NULL;
+		*recv_cq = NULL;
+		break;
+	case MLX5_IB_QPT_REG_UMR:
+	case IB_QPT_XRC_INI:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = NULL;
+		break;
+
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_RAW_IPV6:
+	case IB_QPT_RAW_ETHERTYPE:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = to_mcq(qp->ibqp.recv_cq);
+		break;
+
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:
+		*send_cq = NULL;
+		*recv_cq = NULL;
+		break;
+	}
+}
+
+static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_cq *send_cq, *recv_cq;
+	struct mlx5_modify_qp_mbox_in *in;
+	int err;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return;
+
+	if (qp->state != IB_QPS_RESET) {
+		mlx5_ib_qp_disable_pagefaults(qp);
+		if (mlx5_core_qp_modify(dev->mdev, to_mlx5_state(qp->state),
+					MLX5_QP_STATE_RST, in, 0, &qp->mqp))
+			mlx5_ib_warn(dev, "mlx5_ib: modify QP %06x to RESET failed\n",
+				     qp->mqp.qpn);
+	}
+
+	get_cqs(qp, &send_cq, &recv_cq);
+
+	if (qp->create_type == MLX5_QP_KERNEL) {
+		mlx5_ib_lock_cqs(send_cq, recv_cq);
+		__mlx5_ib_cq_clean(recv_cq, qp->mqp.qpn,
+				   qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
+		if (send_cq != recv_cq)
+			__mlx5_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
+		mlx5_ib_unlock_cqs(send_cq, recv_cq);
+	}
+
+	err = mlx5_core_destroy_qp(dev->mdev, &qp->mqp);
+	if (err)
+		mlx5_ib_warn(dev, "failed to destroy QP 0x%x\n", qp->mqp.qpn);
+	kfree(in);
+
+
+	if (qp->create_type == MLX5_QP_KERNEL)
+		destroy_qp_kernel(dev, qp);
+	else if (qp->create_type == MLX5_QP_USER)
+		destroy_qp_user(&get_pd(qp)->ibpd, qp);
+}
+
+static const char *ib_qp_type_str(enum ib_qp_type type)
+{
+	switch (type) {
+	case IB_QPT_SMI:
+		return "IB_QPT_SMI";
+	case IB_QPT_GSI:
+		return "IB_QPT_GSI";
+	case IB_QPT_RC:
+		return "IB_QPT_RC";
+	case IB_QPT_UC:
+		return "IB_QPT_UC";
+	case IB_QPT_UD:
+		return "IB_QPT_UD";
+	case IB_QPT_RAW_IPV6:
+		return "IB_QPT_RAW_IPV6";
+	case IB_QPT_RAW_ETHERTYPE:
+		return "IB_QPT_RAW_ETHERTYPE";
+	case IB_QPT_XRC_INI:
+		return "IB_QPT_XRC_INI";
+	case IB_QPT_XRC_TGT:
+		return "IB_QPT_XRC_TGT";
+	case IB_QPT_RAW_PACKET:
+		return "IB_QPT_RAW_PACKET";
+	case MLX5_IB_QPT_REG_UMR:
+		return "MLX5_IB_QPT_REG_UMR";
+	case IB_QPT_MAX:
+	default:
+		return "Invalid QP type";
+	}
+}
+
+struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
+				struct ib_qp_init_attr *init_attr,
+				struct ib_udata *udata)
+{
+	struct mlx5_general_caps *gen;
+	struct mlx5_ib_dev *dev;
+	struct mlx5_ib_qp *qp;
+	u16 xrcdn = 0;
+	int err;
+
+	if (pd) {
+		dev = to_mdev(pd->device);
+	} else {
+		/* being cautious here */
+		if (init_attr->qp_type != IB_QPT_XRC_TGT &&
+		    init_attr->qp_type != MLX5_IB_QPT_REG_UMR) {
+			pr_warn("%s: no PD for transport %s\n", __func__,
+				ib_qp_type_str(init_attr->qp_type));
+			return ERR_PTR(-EINVAL);
+		}
+		dev = to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
+	}
+	gen = &dev->mdev->caps.gen;
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+	case IB_QPT_XRC_INI:
+		if (!(gen->flags & MLX5_DEV_CAP_FLAG_XRC)) {
+			mlx5_ib_dbg(dev, "XRC not supported\n");
+			return ERR_PTR(-ENOSYS);
+		}
+		init_attr->recv_cq = NULL;
+		if (init_attr->qp_type == IB_QPT_XRC_TGT) {
+			xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+			init_attr->send_cq = NULL;
+		}
+
+		/* fall through */
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	case MLX5_IB_QPT_REG_UMR:
+		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		err = create_qp_common(dev, pd, init_attr, udata, qp);
+		if (err) {
+			mlx5_ib_dbg(dev, "create_qp_common failed\n");
+			kfree(qp);
+			return ERR_PTR(err);
+		}
+
+		if (is_qp0(init_attr->qp_type))
+			qp->ibqp.qp_num = 0;
+		else if (is_qp1(init_attr->qp_type))
+			qp->ibqp.qp_num = 1;
+		else
+			qp->ibqp.qp_num = qp->mqp.qpn;
+
+		mlx5_ib_dbg(dev, "ib qpnum 0x%x, mlx qpn 0x%x, rcqn 0x%x, scqn 0x%x\n",
+			    qp->ibqp.qp_num, qp->mqp.qpn, to_mcq(init_attr->recv_cq)->mcq.cqn,
+			    to_mcq(init_attr->send_cq)->mcq.cqn);
+
+		qp->xrcdn = xrcdn;
+
+		break;
+
+	case IB_QPT_RAW_IPV6:
+	case IB_QPT_RAW_ETHERTYPE:
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:
+		mlx5_ib_dbg(dev, "unsupported qp type %d\n",
+			    init_attr->qp_type);
+		/* Don't support raw QPs */
+		return ERR_PTR(-EINVAL);
+	}
+
+	return &qp->ibqp;
+}
+
+int mlx5_ib_destroy_qp(struct ib_qp *qp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->device);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+
+	destroy_qp_common(dev, mqp);
+
+	kfree(mqp);
+
+	return 0;
+}
+
+static __be32 to_mlx5_access_flags(struct mlx5_ib_qp *qp, const struct ib_qp_attr *attr,
+				   int attr_mask)
+{
+	u32 hw_access_flags = 0;
+	u8 dest_rd_atomic;
+	u32 access_flags;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		dest_rd_atomic = attr->max_dest_rd_atomic;
+	else
+		dest_rd_atomic = qp->resp_depth;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		access_flags = attr->qp_access_flags;
+	else
+		access_flags = qp->atomic_rd_en;
+
+	if (!dest_rd_atomic)
+		access_flags &= IB_ACCESS_REMOTE_WRITE;
+
+	if (access_flags & IB_ACCESS_REMOTE_READ)
+		hw_access_flags |= MLX5_QP_BIT_RRE;
+	if (access_flags & IB_ACCESS_REMOTE_ATOMIC)
+		hw_access_flags |= (MLX5_QP_BIT_RAE | MLX5_ATOMIC_MODE_CX);
+	if (access_flags & IB_ACCESS_REMOTE_WRITE)
+		hw_access_flags |= MLX5_QP_BIT_RWE;
+
+	return cpu_to_be32(hw_access_flags);
+}
+
+enum {
+	MLX5_PATH_FLAG_FL	= 1 << 0,
+	MLX5_PATH_FLAG_FREE_AR	= 1 << 1,
+	MLX5_PATH_FLAG_COUNTER	= 1 << 2,
+};
+
+static int ib_rate_to_mlx5(struct mlx5_ib_dev *dev, u8 rate)
+{
+	struct mlx5_general_caps *gen;
+
+	gen = &dev->mdev->caps.gen;
+	if (rate == IB_RATE_PORT_CURRENT) {
+		return 0;
+	} else if (rate < IB_RATE_2_5_GBPS || rate > IB_RATE_300_GBPS) {
+		return -EINVAL;
+	} else {
+		while (rate != IB_RATE_2_5_GBPS &&
+		       !(1 << (rate + MLX5_STAT_RATE_OFFSET) &
+			 gen->stat_rate_support))
+			--rate;
+	}
+
+	return rate + MLX5_STAT_RATE_OFFSET;
+}
+
+static int mlx5_set_path(struct mlx5_ib_dev *dev, const struct ib_ah_attr *ah,
+			 struct mlx5_qp_path *path, u8 port, int attr_mask,
+			 u32 path_flags, const struct ib_qp_attr *attr)
+{
+	struct mlx5_general_caps *gen;
+	int err;
+
+	gen = &dev->mdev->caps.gen;
+	path->fl = (path_flags & MLX5_PATH_FLAG_FL) ? 0x80 : 0;
+	path->free_ar = (path_flags & MLX5_PATH_FLAG_FREE_AR) ? 0x80 : 0;
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		path->pkey_index = attr->pkey_index;
+
+	path->grh_mlid	= ah->src_path_bits & 0x7f;
+	path->rlid	= cpu_to_be16(ah->dlid);
+
+	if (ah->ah_flags & IB_AH_GRH) {
+		if (ah->grh.sgid_index >= gen->port[port - 1].gid_table_len) {
+			pr_err("sgid_index (%u) too large. max is %d\n",
+			       ah->grh.sgid_index, gen->port[port - 1].gid_table_len);
+			return -EINVAL;
+		}
+		path->grh_mlid |= 1 << 7;
+		path->mgid_index = ah->grh.sgid_index;
+		path->hop_limit  = ah->grh.hop_limit;
+		path->tclass_flowlabel =
+			cpu_to_be32((ah->grh.traffic_class << 20) |
+				    (ah->grh.flow_label));
+		memcpy(path->rgid, ah->grh.dgid.raw, 16);
+	}
+
+	err = ib_rate_to_mlx5(dev, ah->static_rate);
+	if (err < 0)
+		return err;
+	path->static_rate = err;
+	path->port = port;
+
+	if (attr_mask & IB_QP_TIMEOUT)
+		path->ackto_lt = attr->timeout << 3;
+
+	path->sl = ah->sl & 0xf;
+
+	return 0;
+}
+
+static enum mlx5_qp_optpar opt_mask[MLX5_QP_NUM_STATE][MLX5_QP_NUM_STATE][MLX5_QP_ST_MAX] = {
+	[MLX5_QP_STATE_INIT] = {
+		[MLX5_QP_STATE_INIT] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_Q_KEY		|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+		},
+		[MLX5_QP_STATE_RTR] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH  |
+					  MLX5_QP_OPTPAR_RRE            |
+					  MLX5_QP_OPTPAR_RAE            |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH  |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_PKEY_INDEX     |
+					  MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_MLX] = MLX5_QP_OPTPAR_PKEY_INDEX	|
+					   MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_XRC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH |
+					  MLX5_QP_OPTPAR_RRE            |
+					  MLX5_QP_OPTPAR_RAE            |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+		},
+	},
+	[MLX5_QP_STATE_RTR] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH	|
+					  MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_RNR_TIMEOUT,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH	|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_Q_KEY,
+		},
+	},
+	[MLX5_QP_STATE_RTS] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_RNR_TIMEOUT	|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_ALT_ADDR_PATH,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_ALT_ADDR_PATH,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_Q_KEY		|
+					  MLX5_QP_OPTPAR_SRQN		|
+					  MLX5_QP_OPTPAR_CQN_RCV,
+		},
+	},
+	[MLX5_QP_STATE_SQER] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_UD]	 = MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_MLX] = MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_UC]	 = MLX5_QP_OPTPAR_RWE,
+			[MLX5_QP_ST_RC]	 = MLX5_QP_OPTPAR_RNR_TIMEOUT	|
+					   MLX5_QP_OPTPAR_RWE		|
+					   MLX5_QP_OPTPAR_RAE		|
+					   MLX5_QP_OPTPAR_RRE,
+		},
+	},
+};
+
+static int ib_nr_to_mlx5_nr(int ib_mask)
+{
+	switch (ib_mask) {
+	case IB_QP_STATE:
+		return 0;
+	case IB_QP_CUR_STATE:
+		return 0;
+	case IB_QP_EN_SQD_ASYNC_NOTIFY:
+		return 0;
+	case IB_QP_ACCESS_FLAGS:
+		return MLX5_QP_OPTPAR_RWE | MLX5_QP_OPTPAR_RRE |
+			MLX5_QP_OPTPAR_RAE;
+	case IB_QP_PKEY_INDEX:
+		return MLX5_QP_OPTPAR_PKEY_INDEX;
+	case IB_QP_PORT:
+		return MLX5_QP_OPTPAR_PRI_PORT;
+	case IB_QP_QKEY:
+		return MLX5_QP_OPTPAR_Q_KEY;
+	case IB_QP_AV:
+		return MLX5_QP_OPTPAR_PRIMARY_ADDR_PATH |
+			MLX5_QP_OPTPAR_PRI_PORT;
+	case IB_QP_PATH_MTU:
+		return 0;
+	case IB_QP_TIMEOUT:
+		return MLX5_QP_OPTPAR_ACK_TIMEOUT;
+	case IB_QP_RETRY_CNT:
+		return MLX5_QP_OPTPAR_RETRY_COUNT;
+	case IB_QP_RNR_RETRY:
+		return MLX5_QP_OPTPAR_RNR_RETRY;
+	case IB_QP_RQ_PSN:
+		return 0;
+	case IB_QP_MAX_QP_RD_ATOMIC:
+		return MLX5_QP_OPTPAR_SRA_MAX;
+	case IB_QP_ALT_PATH:
+		return MLX5_QP_OPTPAR_ALT_ADDR_PATH;
+	case IB_QP_MIN_RNR_TIMER:
+		return MLX5_QP_OPTPAR_RNR_TIMEOUT;
+	case IB_QP_SQ_PSN:
+		return 0;
+	case IB_QP_MAX_DEST_RD_ATOMIC:
+		return MLX5_QP_OPTPAR_RRA_MAX | MLX5_QP_OPTPAR_RWE |
+			MLX5_QP_OPTPAR_RRE | MLX5_QP_OPTPAR_RAE;
+	case IB_QP_PATH_MIG_STATE:
+		return MLX5_QP_OPTPAR_PM_STATE;
+	case IB_QP_CAP:
+		return 0;
+	case IB_QP_DEST_QPN:
+		return 0;
+	}
+	return 0;
+}
+
+static int ib_mask_to_mlx5_opt(int ib_mask)
+{
+	int result = 0;
+	int i;
+
+	for (i = 0; i < 8 * sizeof(int); i++) {
+		if ((1 << i) & ib_mask)
+			result |= ib_nr_to_mlx5_nr(1 << i);
+	}
+
+	return result;
+}
+
+static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
+			       const struct ib_qp_attr *attr, int attr_mask,
+			       enum ib_qp_state cur_state, enum ib_qp_state new_state)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_ib_cq *send_cq, *recv_cq;
+	struct mlx5_qp_context *context;
+	struct mlx5_general_caps *gen;
+	struct mlx5_modify_qp_mbox_in *in;
+	struct mlx5_ib_pd *pd;
+	enum mlx5_qp_state mlx5_cur, mlx5_new;
+	enum mlx5_qp_optpar optpar;
+	int sqd_event;
+	int mlx5_st;
+	int err;
+
+	gen = &dev->mdev->caps.gen;
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	context = &in->ctx;
+	err = to_mlx5_st(ibqp->qp_type);
+	if (err < 0)
+		goto out;
+
+	context->flags = cpu_to_be32(err << 16);
+
+	if (!(attr_mask & IB_QP_PATH_MIG_STATE)) {
+		context->flags |= cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
+	} else {
+		switch (attr->path_mig_state) {
+		case IB_MIG_MIGRATED:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
+			break;
+		case IB_MIG_REARM:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_REARM << 11);
+			break;
+		case IB_MIG_ARMED:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_ARMED << 11);
+			break;
+		}
+	}
+
+	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI) {
+		context->mtu_msgmax = (IB_MTU_256 << 5) | 8;
+	} else if (ibqp->qp_type == IB_QPT_UD ||
+		   ibqp->qp_type == MLX5_IB_QPT_REG_UMR) {
+		context->mtu_msgmax = (IB_MTU_4096 << 5) | 12;
+	} else if (attr_mask & IB_QP_PATH_MTU) {
+		if (attr->path_mtu < IB_MTU_256 ||
+		    attr->path_mtu > IB_MTU_4096) {
+			mlx5_ib_warn(dev, "invalid mtu %d\n", attr->path_mtu);
+			err = -EINVAL;
+			goto out;
+		}
+		context->mtu_msgmax = (attr->path_mtu << 5) | gen->log_max_msg;
+	}
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		context->log_pg_sz_remote_qpn = cpu_to_be32(attr->dest_qp_num);
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		context->pri_path.pkey_index = attr->pkey_index;
+
+	/* todo implement counter_index functionality */
+
+	if (is_sqp(ibqp->qp_type))
+		context->pri_path.port = qp->port;
+
+	if (attr_mask & IB_QP_PORT)
+		context->pri_path.port = attr->port_num;
+
+	if (attr_mask & IB_QP_AV) {
+		err = mlx5_set_path(dev, &attr->ah_attr, &context->pri_path,
+				    attr_mask & IB_QP_PORT ? attr->port_num : qp->port,
+				    attr_mask, 0, attr);
+		if (err)
+			goto out;
+	}
+
+	if (attr_mask & IB_QP_TIMEOUT)
+		context->pri_path.ackto_lt |= attr->timeout << 3;
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		err = mlx5_set_path(dev, &attr->alt_ah_attr, &context->alt_path,
+				    attr->alt_port_num, attr_mask, 0, attr);
+		if (err)
+			goto out;
+	}
+
+	pd = get_pd(qp);
+	get_cqs(qp, &send_cq, &recv_cq);
+
+	context->flags_pd = cpu_to_be32(pd ? pd->pdn : to_mpd(dev->devr.p0)->pdn);
+	context->cqn_send = send_cq ? cpu_to_be32(send_cq->mcq.cqn) : 0;
+	context->cqn_recv = recv_cq ? cpu_to_be32(recv_cq->mcq.cqn) : 0;
+	context->params1  = cpu_to_be32(MLX5_IB_ACK_REQ_FREQ << 28);
+
+	if (attr_mask & IB_QP_RNR_RETRY)
+		context->params1 |= cpu_to_be32(attr->rnr_retry << 13);
+
+	if (attr_mask & IB_QP_RETRY_CNT)
+		context->params1 |= cpu_to_be32(attr->retry_cnt << 16);
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+		if (attr->max_rd_atomic)
+			context->params1 |=
+				cpu_to_be32(fls(attr->max_rd_atomic - 1) << 21);
+	}
+
+	if (attr_mask & IB_QP_SQ_PSN)
+		context->next_send_psn = cpu_to_be32(attr->sq_psn);
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
+		if (attr->max_dest_rd_atomic)
+			context->params2 |=
+				cpu_to_be32(fls(attr->max_dest_rd_atomic - 1) << 21);
+	}
+
+	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
+		context->params2 |= to_mlx5_access_flags(qp, attr, attr_mask);
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		context->rnr_nextrecvpsn |= cpu_to_be32(attr->min_rnr_timer << 24);
+
+	if (attr_mask & IB_QP_RQ_PSN)
+		context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
+
+	if (attr_mask & IB_QP_QKEY)
+		context->qkey = cpu_to_be32(attr->qkey);
+
+	if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->db_rec_addr = cpu_to_be64(qp->db.dma);
+
+	if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD	&&
+	    attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY && attr->en_sqd_async_notify)
+		sqd_event = 1;
+	else
+		sqd_event = 0;
+
+	if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->sq_crq_size |= cpu_to_be16(1 << 4);
+
+
+	mlx5_cur = to_mlx5_state(cur_state);
+	mlx5_new = to_mlx5_state(new_state);
+	mlx5_st = to_mlx5_st(ibqp->qp_type);
+	if (mlx5_st < 0)
+		goto out;
+
+	/* If moving to a reset or error state, we must disable page faults on
+	 * this QP and flush all current page faults. Otherwise a stale page
+	 * fault may attempt to work on this QP after it is reset and moved
+	 * again to RTS, and may cause the driver and the device to get out of
+	 * sync. */
+	if (cur_state != IB_QPS_RESET && cur_state != IB_QPS_ERR &&
+	    (new_state == IB_QPS_RESET || new_state == IB_QPS_ERR))
+		mlx5_ib_qp_disable_pagefaults(qp);
+
+	optpar = ib_mask_to_mlx5_opt(attr_mask);
+	optpar &= opt_mask[mlx5_cur][mlx5_new][mlx5_st];
+	in->optparam = cpu_to_be32(optpar);
+	err = mlx5_core_qp_modify(dev->mdev, to_mlx5_state(cur_state),
+				  to_mlx5_state(new_state), in, sqd_event,
+				  &qp->mqp);
+	if (err)
+		goto out;
+
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		mlx5_ib_qp_enable_pagefaults(qp);
+
+	qp->state = new_state;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		qp->atomic_rd_en = attr->qp_access_flags;
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		qp->resp_depth = attr->max_dest_rd_atomic;
+	if (attr_mask & IB_QP_PORT)
+		qp->port = attr->port_num;
+	if (attr_mask & IB_QP_ALT_PATH)
+		qp->alt_port = attr->alt_port_num;
+
+	/*
+	 * If we moved a kernel QP to RESET, clean up all old CQ
+	 * entries and reinitialize the QP.
+	 */
+	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
+		mlx5_ib_cq_clean(recv_cq, qp->mqp.qpn,
+				 ibqp->srq ? to_msrq(ibqp->srq) : NULL);
+		if (send_cq != recv_cq)
+			mlx5_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
+
+		qp->rq.head = 0;
+		qp->rq.tail = 0;
+		qp->sq.head = 0;
+		qp->sq.tail = 0;
+		qp->sq.cur_post = 0;
+		qp->sq.last_poll = 0;
+		qp->db.db[MLX5_RCV_DBR] = 0;
+		qp->db.db[MLX5_SND_DBR] = 0;
+	}
+
+out:
+	kfree(in);
+	return err;
+}
+
+int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	enum ib_qp_state cur_state, new_state;
+	struct mlx5_general_caps *gen;
+	int err = -EINVAL;
+	int port;
+
+	gen = &dev->mdev->caps.gen;
+	mutex_lock(&qp->mutex);
+
+	cur_state = attr_mask & IB_QP_CUR_STATE ? attr->cur_qp_state : qp->state;
+	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
+
+	if (ibqp->qp_type != MLX5_IB_QPT_REG_UMR &&
+	    !ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask,
+				IB_LINK_LAYER_UNSPECIFIED))
+		goto out;
+
+	if ((attr_mask & IB_QP_PORT) &&
+	    (attr->port_num == 0 || attr->port_num > gen->num_ports))
+		goto out;
+
+	if (attr_mask & IB_QP_PKEY_INDEX) {
+		port = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
+		if (attr->pkey_index >= gen->port[port - 1].pkey_table_len)
+			goto out;
+	}
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
+	    attr->max_rd_atomic > (1 << gen->log_max_ra_res_qp))
+		goto out;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
+	    attr->max_dest_rd_atomic > (1 << gen->log_max_ra_req_qp))
+		goto out;
+
+	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
+		err = 0;
+		goto out;
+	}
+
+	err = __mlx5_ib_modify_qp(ibqp, attr, attr_mask, cur_state, new_state);
+
+out:
+	mutex_unlock(&qp->mutex);
+	return err;
+}
+
+static int mlx5_wq_overflow(struct mlx5_ib_wq *wq, int nreq, struct ib_cq *ib_cq)
+{
+	struct mlx5_ib_cq *cq;
+	unsigned cur;
+
+	cur = wq->head - wq->tail;
+	if (likely(cur + nreq < wq->max_post))
+		return 0;
+
+	cq = to_mcq(ib_cq);
+	spin_lock(&cq->lock);
+	cur = wq->head - wq->tail;
+	spin_unlock(&cq->lock);
+
+	return cur + nreq >= wq->max_post;
+}
+
+static __always_inline void set_raddr_seg(struct mlx5_wqe_raddr_seg *rseg,
+					  u64 remote_addr, u32 rkey)
+{
+	rseg->raddr    = cpu_to_be64(remote_addr);
+	rseg->rkey     = cpu_to_be32(rkey);
+	rseg->reserved = 0;
+}
+
+static void set_datagram_seg(struct mlx5_wqe_datagram_seg *dseg,
+			     struct ib_send_wr *wr)
+{
+	memcpy(&dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof(struct mlx5_av));
+	dseg->av.dqp_dct = cpu_to_be32(wr->wr.ud.remote_qpn | MLX5_EXTENDED_UD_AV);
+	dseg->av.key.qkey.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+}
+
+static void set_data_ptr_seg(struct mlx5_wqe_data_seg *dseg, struct ib_sge *sg)
+{
+	dseg->byte_count = cpu_to_be32(sg->length);
+	dseg->lkey       = cpu_to_be32(sg->lkey);
+	dseg->addr       = cpu_to_be64(sg->addr);
+}
+
+static __be16 get_klm_octo(int npages)
+{
+	return cpu_to_be16(ALIGN(npages, 8) / 2);
+}
+
+static __be64 frwr_mkey_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_LEN		|
+		MLX5_MKEY_MASK_PAGE_SIZE	|
+		MLX5_MKEY_MASK_START_ADDR	|
+		MLX5_MKEY_MASK_EN_RINVAL	|
+		MLX5_MKEY_MASK_KEY		|
+		MLX5_MKEY_MASK_LR		|
+		MLX5_MKEY_MASK_LW		|
+		MLX5_MKEY_MASK_RR		|
+		MLX5_MKEY_MASK_RW		|
+		MLX5_MKEY_MASK_A		|
+		MLX5_MKEY_MASK_SMALL_FENCE	|
+		MLX5_MKEY_MASK_FREE;
+
+	return cpu_to_be64(result);
+}
+
+static __be64 sig_mkey_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_LEN		|
+		MLX5_MKEY_MASK_PAGE_SIZE	|
+		MLX5_MKEY_MASK_START_ADDR	|
+		MLX5_MKEY_MASK_EN_SIGERR	|
+		MLX5_MKEY_MASK_EN_RINVAL	|
+		MLX5_MKEY_MASK_KEY		|
+		MLX5_MKEY_MASK_LR		|
+		MLX5_MKEY_MASK_LW		|
+		MLX5_MKEY_MASK_RR		|
+		MLX5_MKEY_MASK_RW		|
+		MLX5_MKEY_MASK_SMALL_FENCE	|
+		MLX5_MKEY_MASK_FREE		|
+		MLX5_MKEY_MASK_BSF_EN;
+
+	return cpu_to_be64(result);
+}
+
+static void set_frwr_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				 struct ib_send_wr *wr, int li)
+{
+	memset(umr, 0, sizeof(*umr));
+
+	if (li) {
+		umr->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
+		umr->flags = 1 << 7;
+		return;
+	}
+
+	umr->flags = (1 << 5); /* fail if not free */
+	umr->klm_octowords = get_klm_octo(wr->wr.fast_reg.page_list_len);
+	umr->mkey_mask = frwr_mkey_mask();
+}
+
+static __be64 get_umr_reg_mr_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_LEN		|
+		 MLX5_MKEY_MASK_PAGE_SIZE	|
+		 MLX5_MKEY_MASK_START_ADDR	|
+		 MLX5_MKEY_MASK_PD		|
+		 MLX5_MKEY_MASK_LR		|
+		 MLX5_MKEY_MASK_LW		|
+		 MLX5_MKEY_MASK_KEY		|
+		 MLX5_MKEY_MASK_RR		|
+		 MLX5_MKEY_MASK_RW		|
+		 MLX5_MKEY_MASK_A		|
+		 MLX5_MKEY_MASK_FREE;
+
+	return cpu_to_be64(result);
+}
+
+static __be64 get_umr_unreg_mr_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_FREE;
+
+	return cpu_to_be64(result);
+}
+
+static __be64 get_umr_update_mtt_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_FREE;
+
+	return cpu_to_be64(result);
+}
+
+static void set_reg_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				struct ib_send_wr *wr)
+{
+	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+
+	memset(umr, 0, sizeof(*umr));
+
+	if (wr->send_flags & MLX5_IB_SEND_UMR_FAIL_IF_FREE)
+		umr->flags = MLX5_UMR_CHECK_FREE; /* fail if free */
+	else
+		umr->flags = MLX5_UMR_CHECK_NOT_FREE; /* fail if not free */
+
+	if (!(wr->send_flags & MLX5_IB_SEND_UMR_UNREG)) {
+		umr->klm_octowords = get_klm_octo(umrwr->npages);
+		if (wr->send_flags & MLX5_IB_SEND_UMR_UPDATE_MTT) {
+			umr->mkey_mask = get_umr_update_mtt_mask();
+			umr->bsf_octowords = get_klm_octo(umrwr->target.offset);
+			umr->flags |= MLX5_UMR_TRANSLATION_OFFSET_EN;
+		} else {
+			umr->mkey_mask = get_umr_reg_mr_mask();
+		}
+	} else {
+		umr->mkey_mask = get_umr_unreg_mr_mask();
+	}
+
+	if (!wr->num_sge)
+		umr->flags |= MLX5_UMR_INLINE;
+}
+
+static u8 get_umr_flags(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
+		MLX5_PERM_LOCAL_READ | MLX5_PERM_UMR_EN;
+}
+
+static void set_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr,
+			     int li, int *writ)
+{
+	memset(seg, 0, sizeof(*seg));
+	if (li) {
+		seg->status = MLX5_MKEY_STATUS_FREE;
+		return;
+	}
+
+	seg->flags = get_umr_flags(wr->wr.fast_reg.access_flags) |
+		     MLX5_ACCESS_MODE_MTT;
+	*writ = seg->flags & (MLX5_PERM_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE);
+	seg->qpn_mkey7_0 = cpu_to_be32((wr->wr.fast_reg.rkey & 0xff) | 0xffffff00);
+	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL);
+	seg->start_addr = cpu_to_be64(wr->wr.fast_reg.iova_start);
+	seg->len = cpu_to_be64(wr->wr.fast_reg.length);
+	seg->xlt_oct_size = cpu_to_be32((wr->wr.fast_reg.page_list_len + 1) / 2);
+	seg->log2_page_size = wr->wr.fast_reg.page_shift;
+}
+
+static void set_reg_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr)
+{
+	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+
+	memset(seg, 0, sizeof(*seg));
+	if (wr->send_flags & MLX5_IB_SEND_UMR_UNREG) {
+		seg->status = MLX5_MKEY_STATUS_FREE;
+		return;
+	}
+
+	seg->flags = convert_access(umrwr->access_flags);
+	if (!(wr->send_flags & MLX5_IB_SEND_UMR_UPDATE_MTT)) {
+		seg->flags_pd = cpu_to_be32(to_mpd(umrwr->pd)->pdn);
+		seg->start_addr = cpu_to_be64(umrwr->target.virt_addr);
+	}
+	seg->len = cpu_to_be64(umrwr->length);
+	seg->log2_page_size = umrwr->page_shift;
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff00 |
+				       mlx5_mkey_variant(umrwr->mkey));
+}
+
+static void set_frwr_pages(struct mlx5_wqe_data_seg *dseg,
+			   struct ib_send_wr *wr,
+			   struct mlx5_core_dev *mdev,
+			   struct mlx5_ib_pd *pd,
+			   int writ)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->wr.fast_reg.page_list);
+	u64 *page_list = wr->wr.fast_reg.page_list->page_list;
+	u64 perm = MLX5_EN_RD | (writ ? MLX5_EN_WR : 0);
+	int i;
+
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++)
+		mfrpl->mapped_page_list[i] = cpu_to_be64(page_list[i] | perm);
+	dseg->addr = cpu_to_be64(mfrpl->map);
+	dseg->byte_count = cpu_to_be32(ALIGN(sizeof(u64) * wr->wr.fast_reg.page_list_len, 64));
+	dseg->lkey = cpu_to_be32(pd->pa_lkey);
+}
+
+static __be32 send_ieth(struct ib_send_wr *wr)
+{
+	switch (wr->opcode) {
+	case IB_WR_SEND_WITH_IMM:
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		return wr->ex.imm_data;
+
+	case IB_WR_SEND_WITH_INV:
+		return cpu_to_be32(wr->ex.invalidate_rkey);
+
+	default:
+		return 0;
+	}
+}
+
+static u8 calc_sig(void *wqe, int size)
+{
+	u8 *p = wqe;
+	u8 res = 0;
+	int i;
+
+	for (i = 0; i < size; i++)
+		res ^= p[i];
+
+	return ~res;
+}
+
+static u8 wq_sig(void *wqe)
+{
+	return calc_sig(wqe, (*((u8 *)wqe + 8) & 0x3f) << 4);
+}
+
+static int set_data_inl_seg(struct mlx5_ib_qp *qp, struct ib_send_wr *wr,
+			    void *wqe, int *sz)
+{
+	struct mlx5_wqe_inline_seg *seg;
+	void *qend = qp->sq.qend;
+	void *addr;
+	int inl = 0;
+	int copy;
+	int len;
+	int i;
+
+	seg = wqe;
+	wqe += sizeof(*seg);
+	for (i = 0; i < wr->num_sge; i++) {
+		addr = (void *)(unsigned long)(wr->sg_list[i].addr);
+		len  = wr->sg_list[i].length;
+		inl += len;
+
+		if (unlikely(inl > qp->max_inline_data))
+			return -ENOMEM;
+
+		if (unlikely(wqe + len > qend)) {
+			copy = qend - wqe;
+			memcpy(wqe, addr, copy);
+			addr += copy;
+			len -= copy;
+			wqe = mlx5_get_send_wqe(qp, 0);
+		}
+		memcpy(wqe, addr, len);
+		wqe += len;
+	}
+
+	seg->byte_count = cpu_to_be32(inl | MLX5_INLINE_SEG);
+
+	*sz = ALIGN(inl + sizeof(seg->byte_count), 16) / 16;
+
+	return 0;
+}
+
+static u16 prot_field_size(enum ib_signature_type type)
+{
+	switch (type) {
+	case IB_SIG_TYPE_T10_DIF:
+		return MLX5_DIF_SIZE;
+	default:
+		return 0;
+	}
+}
+
+static u8 bs_selector(int block_size)
+{
+	switch (block_size) {
+	case 512:	    return 0x1;
+	case 520:	    return 0x2;
+	case 4096:	    return 0x3;
+	case 4160:	    return 0x4;
+	case 1073741824:    return 0x5;
+	default:	    return 0;
+	}
+}
+
+static void mlx5_fill_inl_bsf(struct ib_sig_domain *domain,
+			      struct mlx5_bsf_inl *inl)
+{
+	/* Valid inline section and allow BSF refresh */
+	inl->vld_refresh = cpu_to_be16(MLX5_BSF_INL_VALID |
+				       MLX5_BSF_REFRESH_DIF);
+	inl->dif_apptag = cpu_to_be16(domain->sig.dif.app_tag);
+	inl->dif_reftag = cpu_to_be32(domain->sig.dif.ref_tag);
+	/* repeating block */
+	inl->rp_inv_seed = MLX5_BSF_REPEAT_BLOCK;
+	inl->sig_type = domain->sig.dif.bg_type == IB_T10DIF_CRC ?
+			MLX5_DIF_CRC : MLX5_DIF_IPCS;
+
+	if (domain->sig.dif.ref_remap)
+		inl->dif_inc_ref_guard_check |= MLX5_BSF_INC_REFTAG;
+
+	if (domain->sig.dif.app_escape) {
+		if (domain->sig.dif.ref_escape)
+			inl->dif_inc_ref_guard_check |= MLX5_BSF_APPREF_ESCAPE;
+		else
+			inl->dif_inc_ref_guard_check |= MLX5_BSF_APPTAG_ESCAPE;
+	}
+
+	inl->dif_app_bitmask_check =
+		cpu_to_be16(domain->sig.dif.apptag_check_mask);
+}
+
+static int mlx5_set_bsf(struct ib_mr *sig_mr,
+			struct ib_sig_attrs *sig_attrs,
+			struct mlx5_bsf *bsf, u32 data_size)
+{
+	struct mlx5_core_sig_ctx *msig = to_mmr(sig_mr)->sig;
+	struct mlx5_bsf_basic *basic = &bsf->basic;
+	struct ib_sig_domain *mem = &sig_attrs->mem;
+	struct ib_sig_domain *wire = &sig_attrs->wire;
+
+	memset(bsf, 0, sizeof(*bsf));
+
+	/* Basic + Extended + Inline */
+	basic->bsf_size_sbs = 1 << 7;
+	/* Input domain check byte mask */
+	basic->check_byte_mask = sig_attrs->check_mask;
+	basic->raw_data_size = cpu_to_be32(data_size);
+
+	/* Memory domain */
+	switch (sig_attrs->mem.sig_type) {
+	case IB_SIG_TYPE_NONE:
+		break;
+	case IB_SIG_TYPE_T10_DIF:
+		basic->mem.bs_selector = bs_selector(mem->sig.dif.pi_interval);
+		basic->m_bfs_psv = cpu_to_be32(msig->psv_memory.psv_idx);
+		mlx5_fill_inl_bsf(mem, &bsf->m_inl);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	/* Wire domain */
+	switch (sig_attrs->wire.sig_type) {
+	case IB_SIG_TYPE_NONE:
+		break;
+	case IB_SIG_TYPE_T10_DIF:
+		if (mem->sig.dif.pi_interval == wire->sig.dif.pi_interval &&
+		    mem->sig_type == wire->sig_type) {
+			/* Same block structure */
+			basic->bsf_size_sbs |= 1 << 4;
+			if (mem->sig.dif.bg_type == wire->sig.dif.bg_type)
+				basic->wire.copy_byte_mask |= MLX5_CPY_GRD_MASK;
+			if (mem->sig.dif.app_tag == wire->sig.dif.app_tag)
+				basic->wire.copy_byte_mask |= MLX5_CPY_APP_MASK;
+			if (mem->sig.dif.ref_tag == wire->sig.dif.ref_tag)
+				basic->wire.copy_byte_mask |= MLX5_CPY_REF_MASK;
+		} else
+			basic->wire.bs_selector = bs_selector(wire->sig.dif.pi_interval);
+
+		basic->w_bfs_psv = cpu_to_be32(msig->psv_wire.psv_idx);
+		mlx5_fill_inl_bsf(wire, &bsf->w_inl);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int set_sig_data_segment(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
+				void **seg, int *size)
+{
+	struct ib_sig_attrs *sig_attrs = wr->wr.sig_handover.sig_attrs;
+	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	struct mlx5_bsf *bsf;
+	u32 data_len = wr->sg_list->length;
+	u32 data_key = wr->sg_list->lkey;
+	u64 data_va = wr->sg_list->addr;
+	int ret;
+	int wqe_size;
+
+	if (!wr->wr.sig_handover.prot ||
+	    (data_key == wr->wr.sig_handover.prot->lkey &&
+	     data_va == wr->wr.sig_handover.prot->addr &&
+	     data_len == wr->wr.sig_handover.prot->length)) {
+		/**
+		 * Source domain doesn't contain signature information
+		 * or data and protection are interleaved in memory.
+		 * So need construct:
+		 *                  ------------------
+		 *                 |     data_klm     |
+		 *                  ------------------
+		 *                 |       BSF        |
+		 *                  ------------------
+		 **/
+		struct mlx5_klm *data_klm = *seg;
+
+		data_klm->bcount = cpu_to_be32(data_len);
+		data_klm->key = cpu_to_be32(data_key);
+		data_klm->va = cpu_to_be64(data_va);
+		wqe_size = ALIGN(sizeof(*data_klm), 64);
+	} else {
+		/**
+		 * Source domain contains signature information
+		 * So need construct a strided block format:
+		 *               ---------------------------
+		 *              |     stride_block_ctrl     |
+		 *               ---------------------------
+		 *              |          data_klm         |
+		 *               ---------------------------
+		 *              |          prot_klm         |
+		 *               ---------------------------
+		 *              |             BSF           |
+		 *               ---------------------------
+		 **/
+		struct mlx5_stride_block_ctrl_seg *sblock_ctrl;
+		struct mlx5_stride_block_entry *data_sentry;
+		struct mlx5_stride_block_entry *prot_sentry;
+		u32 prot_key = wr->wr.sig_handover.prot->lkey;
+		u64 prot_va = wr->wr.sig_handover.prot->addr;
+		u16 block_size = sig_attrs->mem.sig.dif.pi_interval;
+		int prot_size;
+
+		sblock_ctrl = *seg;
+		data_sentry = (void *)sblock_ctrl + sizeof(*sblock_ctrl);
+		prot_sentry = (void *)data_sentry + sizeof(*data_sentry);
+
+		prot_size = prot_field_size(sig_attrs->mem.sig_type);
+		if (!prot_size) {
+			pr_err("Bad block size given: %u\n", block_size);
+			return -EINVAL;
+		}
+		sblock_ctrl->bcount_per_cycle = cpu_to_be32(block_size +
+							    prot_size);
+		sblock_ctrl->op = cpu_to_be32(MLX5_STRIDE_BLOCK_OP);
+		sblock_ctrl->repeat_count = cpu_to_be32(data_len / block_size);
+		sblock_ctrl->num_entries = cpu_to_be16(2);
+
+		data_sentry->bcount = cpu_to_be16(block_size);
+		data_sentry->key = cpu_to_be32(data_key);
+		data_sentry->va = cpu_to_be64(data_va);
+		data_sentry->stride = cpu_to_be16(block_size);
+
+		prot_sentry->bcount = cpu_to_be16(prot_size);
+		prot_sentry->key = cpu_to_be32(prot_key);
+		prot_sentry->va = cpu_to_be64(prot_va);
+		prot_sentry->stride = cpu_to_be16(prot_size);
+
+		wqe_size = ALIGN(sizeof(*sblock_ctrl) + sizeof(*data_sentry) +
+				 sizeof(*prot_sentry), 64);
+	}
+
+	*seg += wqe_size;
+	*size += wqe_size / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	bsf = *seg;
+	ret = mlx5_set_bsf(sig_mr, sig_attrs, bsf, data_len);
+	if (ret)
+		return -EINVAL;
+
+	*seg += sizeof(*bsf);
+	*size += sizeof(*bsf) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	return 0;
+}
+
+static void set_sig_mkey_segment(struct mlx5_mkey_seg *seg,
+				 struct ib_send_wr *wr, u32 nelements,
+				 u32 length, u32 pdn)
+{
+	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	u32 sig_key = sig_mr->rkey;
+	u8 sigerr = to_mmr(sig_mr)->sig->sigerr_count & 1;
+
+	memset(seg, 0, sizeof(*seg));
+
+	seg->flags = get_umr_flags(wr->wr.sig_handover.access_flags) |
+				   MLX5_ACCESS_MODE_KLM;
+	seg->qpn_mkey7_0 = cpu_to_be32((sig_key & 0xff) | 0xffffff00);
+	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL | sigerr << 26 |
+				    MLX5_MKEY_BSF_EN | pdn);
+	seg->len = cpu_to_be64(length);
+	seg->xlt_oct_size = cpu_to_be32(be16_to_cpu(get_klm_octo(nelements)));
+	seg->bsfs_octo_size = cpu_to_be32(MLX5_MKEY_BSF_OCTO_SIZE);
+}
+
+static void set_sig_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				struct ib_send_wr *wr, u32 nelements)
+{
+	memset(umr, 0, sizeof(*umr));
+
+	umr->flags = MLX5_FLAGS_INLINE | MLX5_FLAGS_CHECK_FREE;
+	umr->klm_octowords = get_klm_octo(nelements);
+	umr->bsf_octowords = cpu_to_be16(MLX5_MKEY_BSF_OCTO_SIZE);
+	umr->mkey_mask = sig_mkey_mask();
+}
+
+
+static int set_sig_umr_wr(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
+			  void **seg, int *size)
+{
+	struct mlx5_ib_mr *sig_mr = to_mmr(wr->wr.sig_handover.sig_mr);
+	u32 pdn = get_pd(qp)->pdn;
+	u32 klm_oct_size;
+	int region_len, ret;
+
+	if (unlikely(wr->num_sge != 1) ||
+	    unlikely(wr->wr.sig_handover.access_flags &
+		     IB_ACCESS_REMOTE_ATOMIC) ||
+	    unlikely(!sig_mr->sig) || unlikely(!qp->signature_en) ||
+	    unlikely(!sig_mr->sig->sig_status_checked))
+		return -EINVAL;
+
+	/* length of the protected region, data + protection */
+	region_len = wr->sg_list->length;
+	if (wr->wr.sig_handover.prot &&
+	    (wr->wr.sig_handover.prot->lkey != wr->sg_list->lkey  ||
+	     wr->wr.sig_handover.prot->addr != wr->sg_list->addr  ||
+	     wr->wr.sig_handover.prot->length != wr->sg_list->length))
+		region_len += wr->wr.sig_handover.prot->length;
+
+	/**
+	 * KLM octoword size - if protection was provided
+	 * then we use strided block format (3 octowords),
+	 * else we use single KLM (1 octoword)
+	 **/
+	klm_oct_size = wr->wr.sig_handover.prot ? 3 : 1;
+
+	set_sig_umr_segment(*seg, wr, klm_oct_size);
+	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	set_sig_mkey_segment(*seg, wr, klm_oct_size, region_len, pdn);
+	*seg += sizeof(struct mlx5_mkey_seg);
+	*size += sizeof(struct mlx5_mkey_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	ret = set_sig_data_segment(wr, qp, seg, size);
+	if (ret)
+		return ret;
+
+	sig_mr->sig->sig_status_checked = false;
+	return 0;
+}
+
+static int set_psv_wr(struct ib_sig_domain *domain,
+		      u32 psv_idx, void **seg, int *size)
+{
+	struct mlx5_seg_set_psv *psv_seg = *seg;
+
+	memset(psv_seg, 0, sizeof(*psv_seg));
+	psv_seg->psv_num = cpu_to_be32(psv_idx);
+	switch (domain->sig_type) {
+	case IB_SIG_TYPE_NONE:
+		break;
+	case IB_SIG_TYPE_T10_DIF:
+		psv_seg->transient_sig = cpu_to_be32(domain->sig.dif.bg << 16 |
+						     domain->sig.dif.app_tag);
+		psv_seg->ref_tag = cpu_to_be32(domain->sig.dif.ref_tag);
+		break;
+	default:
+		pr_err("Bad signature type given.\n");
+		return 1;
+	}
+
+	*seg += sizeof(*psv_seg);
+	*size += sizeof(*psv_seg) / 16;
+
+	return 0;
+}
+
+static int set_frwr_li_wr(void **seg, struct ib_send_wr *wr, int *size,
+			  struct mlx5_core_dev *mdev, struct mlx5_ib_pd *pd, struct mlx5_ib_qp *qp)
+{
+	int writ = 0;
+	int li;
+
+	li = wr->opcode == IB_WR_LOCAL_INV ? 1 : 0;
+	if (unlikely(wr->send_flags & IB_SEND_INLINE))
+		return -EINVAL;
+
+	set_frwr_umr_segment(*seg, wr, li);
+	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+	set_mkey_segment(*seg, wr, li, &writ);
+	*seg += sizeof(struct mlx5_mkey_seg);
+	*size += sizeof(struct mlx5_mkey_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+	if (!li) {
+		if (unlikely(wr->wr.fast_reg.page_list_len >
+			     wr->wr.fast_reg.page_list->max_page_list_len))
+			return	-ENOMEM;
+
+		set_frwr_pages(*seg, wr, mdev, pd, writ);
+		*seg += sizeof(struct mlx5_wqe_data_seg);
+		*size += (sizeof(struct mlx5_wqe_data_seg) / 16);
+	}
+	return 0;
+}
+
+static void dump_wqe(struct mlx5_ib_qp *qp, int idx, int size_16)
+{
+	__be32 *p = NULL;
+	int tidx = idx;
+	int i, j;
+
+	pr_debug("dump wqe at %p\n", mlx5_get_send_wqe(qp, tidx));
+	for (i = 0, j = 0; i < size_16 * 4; i += 4, j += 4) {
+		if ((i & 0xf) == 0) {
+			void *buf = mlx5_get_send_wqe(qp, tidx);
+			tidx = (tidx + 1) & (qp->sq.wqe_cnt - 1);
+			p = buf;
+			j = 0;
+		}
+		pr_debug("%08x %08x %08x %08x\n", be32_to_cpu(p[j]),
+			 be32_to_cpu(p[j + 1]), be32_to_cpu(p[j + 2]),
+			 be32_to_cpu(p[j + 3]));
+	}
+}
+
+static void mlx5_bf_copy(u64 __iomem *dst, u64 *src,
+			 unsigned bytecnt, struct mlx5_ib_qp *qp)
+{
+	while (bytecnt > 0) {
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		bytecnt -= 64;
+		if (unlikely(src == qp->sq.qend))
+			src = mlx5_get_send_wqe(qp, 0);
+	}
+}
+
+static u8 get_fence(u8 fence, struct ib_send_wr *wr)
+{
+	if (unlikely(wr->opcode == IB_WR_LOCAL_INV &&
+		     wr->send_flags & IB_SEND_FENCE))
+		return MLX5_FENCE_MODE_STRONG_ORDERING;
+
+	if (unlikely(fence)) {
+		if (wr->send_flags & IB_SEND_FENCE)
+			return MLX5_FENCE_MODE_SMALL_AND_FENCE;
+		else
+			return fence;
+
+	} else {
+		return 0;
+	}
+}
+
+static int begin_wqe(struct mlx5_ib_qp *qp, void **seg,
+		     struct mlx5_wqe_ctrl_seg **ctrl,
+		     struct ib_send_wr *wr, unsigned *idx,
+		     int *size, int nreq)
+{
+	int err = 0;
+
+	if (unlikely(mlx5_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq))) {
+		err = -ENOMEM;
+		return err;
+	}
+
+	*idx = qp->sq.cur_post & (qp->sq.wqe_cnt - 1);
+	*seg = mlx5_get_send_wqe(qp, *idx);
+	*ctrl = *seg;
+	*(uint32_t *)(*seg + 8) = 0;
+	(*ctrl)->imm = send_ieth(wr);
+	(*ctrl)->fm_ce_se = qp->sq_signal_bits |
+		(wr->send_flags & IB_SEND_SIGNALED ?
+		 MLX5_WQE_CTRL_CQ_UPDATE : 0) |
+		(wr->send_flags & IB_SEND_SOLICITED ?
+		 MLX5_WQE_CTRL_SOLICITED : 0);
+
+	*seg += sizeof(**ctrl);
+	*size = sizeof(**ctrl) / 16;
+
+	return err;
+}
+
+static void finish_wqe(struct mlx5_ib_qp *qp,
+		       struct mlx5_wqe_ctrl_seg *ctrl,
+		       u8 size, unsigned idx, u64 wr_id,
+		       int nreq, u8 fence, u8 next_fence,
+		       u32 mlx5_opcode)
+{
+	u8 opmod = 0;
+
+	ctrl->opmod_idx_opcode = cpu_to_be32(((u32)(qp->sq.cur_post) << 8) |
+					     mlx5_opcode | ((u32)opmod << 24));
+	ctrl->qpn_ds = cpu_to_be32(size | (qp->mqp.qpn << 8));
+	ctrl->fm_ce_se |= fence;
+	qp->fm_cache = next_fence;
+	if (unlikely(qp->wq_sig))
+		ctrl->signature = wq_sig(ctrl);
+
+	qp->sq.wrid[idx] = wr_id;
+	qp->sq.w_list[idx].opcode = mlx5_opcode;
+	qp->sq.wqe_head[idx] = qp->sq.head + nreq;
+	qp->sq.cur_post += DIV_ROUND_UP(size * 16, MLX5_SEND_WQE_BB);
+	qp->sq.w_list[idx].next = qp->sq.cur_post;
+}
+
+
+int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr)
+{
+	struct mlx5_wqe_ctrl_seg *ctrl = NULL;  /* compiler warning */
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_core_dev *mdev = dev->mdev;
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_ib_mr *mr;
+	struct mlx5_wqe_data_seg *dpseg;
+	struct mlx5_wqe_xrc_seg *xrc;
+	struct mlx5_bf *bf = qp->bf;
+	int uninitialized_var(size);
+	void *qend = qp->sq.qend;
+	unsigned long flags;
+	unsigned idx;
+	int err = 0;
+	int inl = 0;
+	int num_sge;
+	void *seg;
+	int nreq;
+	int i;
+	u8 next_fence = 0;
+	u8 fence;
+
+	spin_lock_irqsave(&qp->sq.lock, flags);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (unlikely(wr->opcode >= ARRAY_SIZE(mlx5_ib_opcode))) {
+			mlx5_ib_warn(dev, "\n");
+			err = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		fence = qp->fm_cache;
+		num_sge = wr->num_sge;
+		if (unlikely(num_sge > qp->sq.max_gs)) {
+			mlx5_ib_warn(dev, "\n");
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		err = begin_wqe(qp, &seg, &ctrl, wr, &idx, &size, nreq);
+		if (err) {
+			mlx5_ib_warn(dev, "\n");
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		switch (ibqp->qp_type) {
+		case IB_QPT_XRC_INI:
+			xrc = seg;
+			xrc->xrc_srqn = htonl(wr->xrc_remote_srq_num);
+			seg += sizeof(*xrc);
+			size += sizeof(*xrc) / 16;
+			/* fall through */
+		case IB_QPT_RC:
+			switch (wr->opcode) {
+			case IB_WR_RDMA_READ:
+			case IB_WR_RDMA_WRITE:
+			case IB_WR_RDMA_WRITE_WITH_IMM:
+				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
+					      wr->wr.rdma.rkey);
+				seg += sizeof(struct mlx5_wqe_raddr_seg);
+				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
+				break;
+
+			case IB_WR_ATOMIC_CMP_AND_SWP:
+			case IB_WR_ATOMIC_FETCH_AND_ADD:
+			case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
+				mlx5_ib_warn(dev, "Atomic operations are not supported yet\n");
+				err = -ENOSYS;
+				*bad_wr = wr;
+				goto out;
+
+			case IB_WR_LOCAL_INV:
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				qp->sq.wr_data[idx] = IB_WR_LOCAL_INV;
+				ctrl->imm = cpu_to_be32(wr->ex.invalidate_rkey);
+				err = set_frwr_li_wr(&seg, wr, &size, mdev, to_mpd(ibqp->pd), qp);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+				num_sge = 0;
+				break;
+
+			case IB_WR_FAST_REG_MR:
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				qp->sq.wr_data[idx] = IB_WR_FAST_REG_MR;
+				ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+				err = set_frwr_li_wr(&seg, wr, &size, mdev, to_mpd(ibqp->pd), qp);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+				num_sge = 0;
+				break;
+
+			case IB_WR_REG_SIG_MR:
+				qp->sq.wr_data[idx] = IB_WR_REG_SIG_MR;
+				mr = to_mmr(wr->wr.sig_handover.sig_mr);
+
+				ctrl->imm = cpu_to_be32(mr->ibmr.rkey);
+				err = set_sig_umr_wr(wr, qp, &seg, &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_UMR);
+				/*
+				 * SET_PSV WQEs are not signaled and solicited
+				 * on error
+				 */
+				wr->send_flags &= ~IB_SEND_SIGNALED;
+				wr->send_flags |= IB_SEND_SOLICITED;
+				err = begin_wqe(qp, &seg, &ctrl, wr,
+						&idx, &size, nreq);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					err = -ENOMEM;
+					*bad_wr = wr;
+					goto out;
+				}
+
+				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->mem,
+						 mr->sig->psv_memory.psv_idx, &seg,
+						 &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_SET_PSV);
+				err = begin_wqe(qp, &seg, &ctrl, wr,
+						&idx, &size, nreq);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					err = -ENOMEM;
+					*bad_wr = wr;
+					goto out;
+				}
+
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->wire,
+						 mr->sig->psv_wire.psv_idx, &seg,
+						 &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_SET_PSV);
+				num_sge = 0;
+				goto skip_psv;
+
+			default:
+				break;
+			}
+			break;
+
+		case IB_QPT_UC:
+			switch (wr->opcode) {
+			case IB_WR_RDMA_WRITE:
+			case IB_WR_RDMA_WRITE_WITH_IMM:
+				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
+					      wr->wr.rdma.rkey);
+				seg  += sizeof(struct mlx5_wqe_raddr_seg);
+				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
+				break;
+
+			default:
+				break;
+			}
+			break;
+
+		case IB_QPT_UD:
+		case IB_QPT_SMI:
+		case IB_QPT_GSI:
+			set_datagram_seg(seg, wr);
+			seg += sizeof(struct mlx5_wqe_datagram_seg);
+			size += sizeof(struct mlx5_wqe_datagram_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			break;
+
+		case MLX5_IB_QPT_REG_UMR:
+			if (wr->opcode != MLX5_IB_WR_UMR) {
+				err = -EINVAL;
+				mlx5_ib_warn(dev, "bad opcode\n");
+				goto out;
+			}
+			qp->sq.wr_data[idx] = MLX5_IB_WR_UMR;
+			ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+			set_reg_umr_segment(seg, wr);
+			seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+			size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			set_reg_mkey_segment(seg, wr);
+			seg += sizeof(struct mlx5_mkey_seg);
+			size += sizeof(struct mlx5_mkey_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			break;
+
+		default:
+			break;
+		}
+
+		if (wr->send_flags & IB_SEND_INLINE && num_sge) {
+			int uninitialized_var(sz);
+
+			err = set_data_inl_seg(qp, wr, seg, &sz);
+			if (unlikely(err)) {
+				mlx5_ib_warn(dev, "\n");
+				*bad_wr = wr;
+				goto out;
+			}
+			inl = 1;
+			size += sz;
+		} else {
+			dpseg = seg;
+			for (i = 0; i < num_sge; i++) {
+				if (unlikely(dpseg == qend)) {
+					seg = mlx5_get_send_wqe(qp, 0);
+					dpseg = seg;
+				}
+				if (likely(wr->sg_list[i].length)) {
+					set_data_ptr_seg(dpseg, wr->sg_list + i);
+					size += sizeof(struct mlx5_wqe_data_seg) / 16;
+					dpseg++;
+				}
+			}
+		}
+
+		finish_wqe(qp, ctrl, size, idx, wr->wr_id, nreq,
+			   get_fence(fence, wr), next_fence,
+			   mlx5_ib_opcode[wr->opcode]);
+skip_psv:
+		if (0)
+			dump_wqe(qp, idx, size);
+	}
+
+out:
+	if (likely(nreq)) {
+		qp->sq.head += nreq;
+
+		/* Make sure that descriptors are written before
+		 * updating doorbell record and ringing the doorbell
+		 */
+		wmb();
+
+		qp->db.db[MLX5_SND_DBR] = cpu_to_be32(qp->sq.cur_post);
+
+		/* Make sure doorbell record is visible to the HCA before
+		 * we hit doorbell */
+		wmb();
+
+		if (bf->need_lock)
+			spin_lock(&bf->lock);
+		else
+			__acquire(&bf->lock);
+
+		/* TBD enable WC */
+		if (0 && nreq == 1 && bf->uuarn && inl && size > 1 && size <= bf->buf_size / 16) {
+			mlx5_bf_copy(bf->reg + bf->offset, (u64 *)ctrl, ALIGN(size * 16, 64), qp);
+			/* wc_wmb(); */
+		} else {
+			mlx5_write64((__be32 *)ctrl, bf->regreg + bf->offset,
+				     MLX5_GET_DOORBELL_LOCK(&bf->lock32));
+			/* Make sure doorbells don't leak out of SQ spinlock
+			 * and reach the HCA out of order.
+			 */
+			mmiowb();
+		}
+		bf->offset ^= bf->buf_size;
+		if (bf->need_lock)
+			spin_unlock(&bf->lock);
+		else
+			__release(&bf->lock);
+	}
+
+	spin_unlock_irqrestore(&qp->sq.lock, flags);
+
+	return err;
+}
+
+static void set_sig_seg(struct mlx5_rwqe_sig *sig, int size)
+{
+	sig->signature = calc_sig(sig, size);
+}
+
+int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr)
+{
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_wqe_data_seg *scat;
+	struct mlx5_rwqe_sig *sig;
+	unsigned long flags;
+	int err = 0;
+	int nreq;
+	int ind;
+	int i;
+
+	spin_lock_irqsave(&qp->rq.lock, flags);
+
+	ind = qp->rq.head & (qp->rq.wqe_cnt - 1);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (mlx5_wq_overflow(&qp->rq, nreq, qp->ibqp.recv_cq)) {
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		if (unlikely(wr->num_sge > qp->rq.max_gs)) {
+			err = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		scat = get_recv_wqe(qp, ind);
+		if (qp->wq_sig)
+			scat++;
+
+		for (i = 0; i < wr->num_sge; i++)
+			set_data_ptr_seg(scat + i, wr->sg_list + i);
+
+		if (i < qp->rq.max_gs) {
+			scat[i].byte_count = 0;
+			scat[i].lkey       = cpu_to_be32(MLX5_INVALID_LKEY);
+			scat[i].addr       = 0;
+		}
+
+		if (qp->wq_sig) {
+			sig = (struct mlx5_rwqe_sig *)scat;
+			set_sig_seg(sig, (qp->rq.max_gs + 1) << 2);
+		}
+
+		qp->rq.wrid[ind] = wr->wr_id;
+
+		ind = (ind + 1) & (qp->rq.wqe_cnt - 1);
+	}
+
+out:
+	if (likely(nreq)) {
+		qp->rq.head += nreq;
+
+		/* Make sure that descriptors are written before
+		 * doorbell record.
+		 */
+		wmb();
+
+		*qp->db.db = cpu_to_be32(qp->rq.head & 0xffff);
+	}
+
+	spin_unlock_irqrestore(&qp->rq.lock, flags);
+
+	return err;
+}
+
+static inline enum ib_qp_state to_ib_qp_state(enum mlx5_qp_state mlx5_state)
+{
+	switch (mlx5_state) {
+	case MLX5_QP_STATE_RST:      return IB_QPS_RESET;
+	case MLX5_QP_STATE_INIT:     return IB_QPS_INIT;
+	case MLX5_QP_STATE_RTR:      return IB_QPS_RTR;
+	case MLX5_QP_STATE_RTS:      return IB_QPS_RTS;
+	case MLX5_QP_STATE_SQ_DRAINING:
+	case MLX5_QP_STATE_SQD:      return IB_QPS_SQD;
+	case MLX5_QP_STATE_SQER:     return IB_QPS_SQE;
+	case MLX5_QP_STATE_ERR:      return IB_QPS_ERR;
+	default:		     return -1;
+	}
+}
+
+static inline enum ib_mig_state to_ib_mig_state(int mlx5_mig_state)
+{
+	switch (mlx5_mig_state) {
+	case MLX5_QP_PM_ARMED:		return IB_MIG_ARMED;
+	case MLX5_QP_PM_REARM:		return IB_MIG_REARM;
+	case MLX5_QP_PM_MIGRATED:	return IB_MIG_MIGRATED;
+	default: return -1;
+	}
+}
+
+static int to_ib_qp_access_flags(int mlx5_flags)
+{
+	int ib_flags = 0;
+
+	if (mlx5_flags & MLX5_QP_BIT_RRE)
+		ib_flags |= IB_ACCESS_REMOTE_READ;
+	if (mlx5_flags & MLX5_QP_BIT_RWE)
+		ib_flags |= IB_ACCESS_REMOTE_WRITE;
+	if (mlx5_flags & MLX5_QP_BIT_RAE)
+		ib_flags |= IB_ACCESS_REMOTE_ATOMIC;
+
+	return ib_flags;
+}
+
+static void to_ib_ah_attr(struct mlx5_ib_dev *ibdev, struct ib_ah_attr *ib_ah_attr,
+				struct mlx5_qp_path *path)
+{
+	struct mlx5_core_dev *dev = ibdev->mdev;
+
+	memset(ib_ah_attr, 0, sizeof(*ib_ah_attr));
+	ib_ah_attr->port_num	  = path->port;
+
+	if (ib_ah_attr->port_num == 0 ||
+	    ib_ah_attr->port_num > dev->caps.gen.num_ports)
+		return;
+
+	ib_ah_attr->sl = path->sl & 0xf;
+
+	ib_ah_attr->dlid	  = be16_to_cpu(path->rlid);
+	ib_ah_attr->src_path_bits = path->grh_mlid & 0x7f;
+	ib_ah_attr->static_rate   = path->static_rate ? path->static_rate - 5 : 0;
+	ib_ah_attr->ah_flags      = (path->grh_mlid & (1 << 7)) ? IB_AH_GRH : 0;
+	if (ib_ah_attr->ah_flags) {
+		ib_ah_attr->grh.sgid_index = path->mgid_index;
+		ib_ah_attr->grh.hop_limit  = path->hop_limit;
+		ib_ah_attr->grh.traffic_class =
+			(be32_to_cpu(path->tclass_flowlabel) >> 20) & 0xff;
+		ib_ah_attr->grh.flow_label =
+			be32_to_cpu(path->tclass_flowlabel) & 0xfffff;
+		memcpy(ib_ah_attr->grh.dgid.raw,
+		       path->rgid, sizeof(ib_ah_attr->grh.dgid.raw));
+	}
+}
+
+int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_query_qp_mbox_out *outb;
+	struct mlx5_qp_context *context;
+	int mlx5_state;
+	int err = 0;
+
+#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
+	/*
+	 * Wait for any outstanding page faults, in case the user frees memory
+	 * based upon this query's result.
+	 */
+	flush_workqueue(mlx5_ib_page_fault_wq);
+#endif
+
+	mutex_lock(&qp->mutex);
+	outb = kzalloc(sizeof(*outb), GFP_KERNEL);
+	if (!outb) {
+		err = -ENOMEM;
+		goto out;
+	}
+	context = &outb->ctx;
+	err = mlx5_core_qp_query(dev->mdev, &qp->mqp, outb, sizeof(*outb));
+	if (err)
+		goto out_free;
+
+	mlx5_state = be32_to_cpu(context->flags) >> 28;
+
+	qp->state		     = to_ib_qp_state(mlx5_state);
+	qp_attr->qp_state	     = qp->state;
+	qp_attr->path_mtu	     = context->mtu_msgmax >> 5;
+	qp_attr->path_mig_state	     =
+		to_ib_mig_state((be32_to_cpu(context->flags) >> 11) & 0x3);
+	qp_attr->qkey		     = be32_to_cpu(context->qkey);
+	qp_attr->rq_psn		     = be32_to_cpu(context->rnr_nextrecvpsn) & 0xffffff;
+	qp_attr->sq_psn		     = be32_to_cpu(context->next_send_psn) & 0xffffff;
+	qp_attr->dest_qp_num	     = be32_to_cpu(context->log_pg_sz_remote_qpn) & 0xffffff;
+	qp_attr->qp_access_flags     =
+		to_ib_qp_access_flags(be32_to_cpu(context->params2));
+
+	if (qp->ibqp.qp_type == IB_QPT_RC || qp->ibqp.qp_type == IB_QPT_UC) {
+		to_ib_ah_attr(dev, &qp_attr->ah_attr, &context->pri_path);
+		to_ib_ah_attr(dev, &qp_attr->alt_ah_attr, &context->alt_path);
+		qp_attr->alt_pkey_index = context->alt_path.pkey_index & 0x7f;
+		qp_attr->alt_port_num	= qp_attr->alt_ah_attr.port_num;
+	}
+
+	qp_attr->pkey_index = context->pri_path.pkey_index & 0x7f;
+	qp_attr->port_num = context->pri_path.port;
+
+	/* qp_attr->en_sqd_async_notify is only applicable in modify qp */
+	qp_attr->sq_draining = mlx5_state == MLX5_QP_STATE_SQ_DRAINING;
+
+	qp_attr->max_rd_atomic = 1 << ((be32_to_cpu(context->params1) >> 21) & 0x7);
+
+	qp_attr->max_dest_rd_atomic =
+		1 << ((be32_to_cpu(context->params2) >> 21) & 0x7);
+	qp_attr->min_rnr_timer	    =
+		(be32_to_cpu(context->rnr_nextrecvpsn) >> 24) & 0x1f;
+	qp_attr->timeout	    = context->pri_path.ackto_lt >> 3;
+	qp_attr->retry_cnt	    = (be32_to_cpu(context->params1) >> 16) & 0x7;
+	qp_attr->rnr_retry	    = (be32_to_cpu(context->params1) >> 13) & 0x7;
+	qp_attr->alt_timeout	    = context->alt_path.ackto_lt >> 3;
+	qp_attr->cur_qp_state	     = qp_attr->qp_state;
+	qp_attr->cap.max_recv_wr     = qp->rq.wqe_cnt;
+	qp_attr->cap.max_recv_sge    = qp->rq.max_gs;
+
+	if (!ibqp->uobject) {
+		qp_attr->cap.max_send_wr  = qp->sq.wqe_cnt;
+		qp_attr->cap.max_send_sge = qp->sq.max_gs;
+	} else {
+		qp_attr->cap.max_send_wr  = 0;
+		qp_attr->cap.max_send_sge = 0;
+	}
+
+	/* We don't support inline sends for kernel QPs (yet), and we
+	 * don't know what userspace's value should be.
+	 */
+	qp_attr->cap.max_inline_data = 0;
+
+	qp_init_attr->cap	     = qp_attr->cap;
+
+	qp_init_attr->create_flags = 0;
+	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+		qp_init_attr->create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
+
+	qp_init_attr->sq_sig_type = qp->sq_signal_bits & MLX5_WQE_CTRL_CQ_UPDATE ?
+		IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
+
+out_free:
+	kfree(outb);
+
+out:
+	mutex_unlock(&qp->mutex);
+	return err;
+}
+
+struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_general_caps *gen;
+	struct mlx5_ib_xrcd *xrcd;
+	int err;
+
+	gen = &dev->mdev->caps.gen;
+	if (!(gen->flags & MLX5_DEV_CAP_FLAG_XRC))
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = kmalloc(sizeof(*xrcd), GFP_KERNEL);
+	if (!xrcd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx5_core_xrcd_alloc(dev->mdev, &xrcd->xrcdn);
+	if (err) {
+		kfree(xrcd);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	return &xrcd->ibxrcd;
+}
+
+int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	struct mlx5_ib_dev *dev = to_mdev(xrcd->device);
+	u32 xrcdn = to_mxrcd(xrcd)->xrcdn;
+	int err;
+
+	err = mlx5_core_xrcd_dealloc(dev->mdev, xrcdn);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to dealloc xrcdn 0x%x\n", xrcdn);
+		return err;
+	}
+
+	kfree(xrcd);
+
+	return 0;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/srq.c b/kernel/drivers/infiniband/hw/mlx5/srq.c
new file mode 100644
index 000000000..02d77a297
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/srq.c
@@ -0,0 +1,485 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/mlx5/qp.h>
+#include <linux/mlx5/srq.h>
+#include <linux/slab.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "mlx5_ib.h"
+#include "user.h"
+
+/* not supported currently */
+static int srq_signature;
+
+static void *get_wqe(struct mlx5_ib_srq *srq, int n)
+{
+	return mlx5_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
+}
+
+static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
+{
+	struct ib_event event;
+	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
+
+	if (ibsrq->event_handler) {
+		event.device      = ibsrq->device;
+		event.element.srq = ibsrq;
+		switch (type) {
+		case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
+			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
+			break;
+		case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
+			event.event = IB_EVENT_SRQ_ERR;
+			break;
+		default:
+			pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
+				type, srq->srqn);
+			return;
+		}
+
+		ibsrq->event_handler(&event, ibsrq->srq_context);
+	}
+}
+
+static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq,
+			   struct mlx5_create_srq_mbox_in **in,
+			   struct ib_udata *udata, int buf_size, int *inlen)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_create_srq ucmd;
+	size_t ucmdlen;
+	int err;
+	int npages;
+	int page_shift;
+	int ncont;
+	u32 offset;
+
+	ucmdlen =
+		(udata->inlen - sizeof(struct ib_uverbs_cmd_hdr) <
+		 sizeof(ucmd)) ? (sizeof(ucmd) -
+				  sizeof(ucmd.reserved)) : sizeof(ucmd);
+
+	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
+		mlx5_ib_dbg(dev, "failed copy udata\n");
+		return -EFAULT;
+	}
+
+	if (ucmdlen == sizeof(ucmd) &&
+	    ucmd.reserved != 0)
+		return -EINVAL;
+
+	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
+
+	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size,
+				0, 0);
+	if (IS_ERR(srq->umem)) {
+		mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
+		err = PTR_ERR(srq->umem);
+		return err;
+	}
+
+	mlx5_ib_cont_pages(srq->umem, ucmd.buf_addr, &npages,
+			   &page_shift, &ncont, NULL);
+	err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift,
+				     &offset);
+	if (err) {
+		mlx5_ib_warn(dev, "bad offset\n");
+		goto err_umem;
+	}
+
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * ncont;
+	*in = mlx5_vzalloc(*inlen);
+	if (!(*in)) {
+		err = -ENOMEM;
+		goto err_umem;
+	}
+
+	mlx5_ib_populate_pas(dev, srq->umem, page_shift, (*in)->pas, 0);
+
+	err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context),
+				  ucmd.db_addr, &srq->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "map doorbell failed\n");
+		goto err_in;
+	}
+
+	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	(*in)->ctx.pgoff_cqn = cpu_to_be32(offset << 26);
+
+	return 0;
+
+err_in:
+	kvfree(*in);
+
+err_umem:
+	ib_umem_release(srq->umem);
+
+	return err;
+}
+
+static int create_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq,
+			     struct mlx5_create_srq_mbox_in **in, int buf_size,
+			     int *inlen)
+{
+	int err;
+	int i;
+	struct mlx5_wqe_srq_next_seg *next;
+	int page_shift;
+	int npages;
+
+	err = mlx5_db_alloc(dev->mdev, &srq->db);
+	if (err) {
+		mlx5_ib_warn(dev, "alloc dbell rec failed\n");
+		return err;
+	}
+
+	if (mlx5_buf_alloc(dev->mdev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
+		mlx5_ib_dbg(dev, "buf alloc failed\n");
+		err = -ENOMEM;
+		goto err_db;
+	}
+	page_shift = srq->buf.page_shift;
+
+	srq->head    = 0;
+	srq->tail    = srq->msrq.max - 1;
+	srq->wqe_ctr = 0;
+
+	for (i = 0; i < srq->msrq.max; i++) {
+		next = get_wqe(srq, i);
+		next->next_wqe_index =
+			cpu_to_be16((i + 1) & (srq->msrq.max - 1));
+	}
+
+	npages = DIV_ROUND_UP(srq->buf.npages, 1 << (page_shift - PAGE_SHIFT));
+	mlx5_ib_dbg(dev, "buf_size %d, page_shift %d, npages %d, calc npages %d\n",
+		    buf_size, page_shift, srq->buf.npages, npages);
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * npages;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	mlx5_fill_page_array(&srq->buf, (*in)->pas);
+
+	srq->wrid = kmalloc(srq->msrq.max * sizeof(u64), GFP_KERNEL);
+	if (!srq->wrid) {
+		mlx5_ib_dbg(dev, "kmalloc failed %lu\n",
+			    (unsigned long)(srq->msrq.max * sizeof(u64)));
+		err = -ENOMEM;
+		goto err_in;
+	}
+	srq->wq_sig = !!srq_signature;
+
+	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+
+	return 0;
+
+err_in:
+	kvfree(*in);
+
+err_buf:
+	mlx5_buf_free(dev->mdev, &srq->buf);
+
+err_db:
+	mlx5_db_free(dev->mdev, &srq->db);
+	return err;
+}
+
+static void destroy_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq)
+{
+	mlx5_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
+	ib_umem_release(srq->umem);
+}
+
+
+static void destroy_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq)
+{
+	kfree(srq->wrid);
+	mlx5_buf_free(dev->mdev, &srq->buf);
+	mlx5_db_free(dev->mdev, &srq->db);
+}
+
+struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
+				  struct ib_srq_init_attr *init_attr,
+				  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_general_caps *gen;
+	struct mlx5_ib_srq *srq;
+	int desc_size;
+	int buf_size;
+	int err;
+	struct mlx5_create_srq_mbox_in *uninitialized_var(in);
+	int uninitialized_var(inlen);
+	int is_xrc;
+	u32 flgs, xrcdn;
+
+	gen = &dev->mdev->caps.gen;
+	/* Sanity check SRQ size before proceeding */
+	if (init_attr->attr.max_wr >= gen->max_srq_wqes) {
+		mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
+			    init_attr->attr.max_wr,
+			    gen->max_srq_wqes);
+		return ERR_PTR(-EINVAL);
+	}
+
+	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&srq->mutex);
+	spin_lock_init(&srq->lock);
+	srq->msrq.max    = roundup_pow_of_two(init_attr->attr.max_wr + 1);
+	srq->msrq.max_gs = init_attr->attr.max_sge;
+
+	desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
+		    srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
+	desc_size = roundup_pow_of_two(desc_size);
+	desc_size = max_t(int, 32, desc_size);
+	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
+		sizeof(struct mlx5_wqe_data_seg);
+	srq->msrq.wqe_shift = ilog2(desc_size);
+	buf_size = srq->msrq.max * desc_size;
+	mlx5_ib_dbg(dev, "desc_size 0x%x, req wr 0x%x, srq size 0x%x, max_gs 0x%x, max_avail_gather 0x%x\n",
+		    desc_size, init_attr->attr.max_wr, srq->msrq.max, srq->msrq.max_gs,
+		    srq->msrq.max_avail_gather);
+
+	if (pd->uobject)
+		err = create_srq_user(pd, srq, &in, udata, buf_size, &inlen);
+	else
+		err = create_srq_kernel(dev, srq, &in, buf_size, &inlen);
+
+	if (err) {
+		mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
+			     pd->uobject ? "user" : "kernel", err);
+		goto err_srq;
+	}
+
+	is_xrc = (init_attr->srq_type == IB_SRQT_XRC);
+	in->ctx.state_log_sz = ilog2(srq->msrq.max);
+	flgs = ((srq->msrq.wqe_shift - 4) | (is_xrc << 5) | (srq->wq_sig << 7)) << 24;
+	xrcdn = 0;
+	if (is_xrc) {
+		xrcdn = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
+		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(init_attr->ext.xrc.cq)->mcq.cqn);
+	} else if (init_attr->srq_type == IB_SRQT_BASIC) {
+		xrcdn = to_mxrcd(dev->devr.x0)->xrcdn;
+		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(dev->devr.c0)->mcq.cqn);
+	}
+
+	in->ctx.flags_xrcd = cpu_to_be32((flgs & 0xFF000000) | (xrcdn & 0xFFFFFF));
+
+	in->ctx.pd = cpu_to_be32(to_mpd(pd)->pdn);
+	in->ctx.db_record = cpu_to_be64(srq->db.dma);
+	err = mlx5_core_create_srq(dev->mdev, &srq->msrq, in, inlen);
+	kvfree(in);
+	if (err) {
+		mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
+		goto err_usr_kern_srq;
+	}
+
+	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
+
+	srq->msrq.event = mlx5_ib_srq_event;
+	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
+
+	if (pd->uobject)
+		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof(__u32))) {
+			mlx5_ib_dbg(dev, "copy to user failed\n");
+			err = -EFAULT;
+			goto err_core;
+		}
+
+	init_attr->attr.max_wr = srq->msrq.max - 1;
+
+	return &srq->ibsrq;
+
+err_core:
+	mlx5_core_destroy_srq(dev->mdev, &srq->msrq);
+
+err_usr_kern_srq:
+	if (pd->uobject)
+		destroy_srq_user(pd, srq);
+	else
+		destroy_srq_kernel(dev, srq);
+
+err_srq:
+	kfree(srq);
+
+	return ERR_PTR(err);
+}
+
+int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	int ret;
+
+	/* We don't support resizing SRQs yet */
+	if (attr_mask & IB_SRQ_MAX_WR)
+		return -EINVAL;
+
+	if (attr_mask & IB_SRQ_LIMIT) {
+		if (attr->srq_limit >= srq->msrq.max)
+			return -EINVAL;
+
+		mutex_lock(&srq->mutex);
+		ret = mlx5_core_arm_srq(dev->mdev, &srq->msrq, attr->srq_limit, 1);
+		mutex_unlock(&srq->mutex);
+
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	int ret;
+	struct mlx5_query_srq_mbox_out *out;
+
+	out = kzalloc(sizeof(*out), GFP_KERNEL);
+	if (!out)
+		return -ENOMEM;
+
+	ret = mlx5_core_query_srq(dev->mdev, &srq->msrq, out);
+	if (ret)
+		goto out_box;
+
+	srq_attr->srq_limit = be16_to_cpu(out->ctx.lwm);
+	srq_attr->max_wr    = srq->msrq.max - 1;
+	srq_attr->max_sge   = srq->msrq.max_gs;
+
+out_box:
+	kfree(out);
+	return ret;
+}
+
+int mlx5_ib_destroy_srq(struct ib_srq *srq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(srq->device);
+	struct mlx5_ib_srq *msrq = to_msrq(srq);
+
+	mlx5_core_destroy_srq(dev->mdev, &msrq->msrq);
+
+	if (srq->uobject) {
+		mlx5_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
+		ib_umem_release(msrq->umem);
+	} else {
+		destroy_srq_kernel(dev, msrq);
+	}
+
+	kfree(srq);
+	return 0;
+}
+
+void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
+{
+	struct mlx5_wqe_srq_next_seg *next;
+
+	/* always called with interrupts disabled. */
+	spin_lock(&srq->lock);
+
+	next = get_wqe(srq, srq->tail);
+	next->next_wqe_index = cpu_to_be16(wqe_index);
+	srq->tail = wqe_index;
+
+	spin_unlock(&srq->lock);
+}
+
+int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			  struct ib_recv_wr **bad_wr)
+{
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	struct mlx5_wqe_srq_next_seg *next;
+	struct mlx5_wqe_data_seg *scat;
+	unsigned long flags;
+	int err = 0;
+	int nreq;
+	int i;
+
+	spin_lock_irqsave(&srq->lock, flags);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
+			err = -EINVAL;
+			*bad_wr = wr;
+			break;
+		}
+
+		if (unlikely(srq->head == srq->tail)) {
+			err = -ENOMEM;
+			*bad_wr = wr;
+			break;
+		}
+
+		srq->wrid[srq->head] = wr->wr_id;
+
+		next      = get_wqe(srq, srq->head);
+		srq->head = be16_to_cpu(next->next_wqe_index);
+		scat      = (struct mlx5_wqe_data_seg *)(next + 1);
+
+		for (i = 0; i < wr->num_sge; i++) {
+			scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
+			scat[i].lkey       = cpu_to_be32(wr->sg_list[i].lkey);
+			scat[i].addr       = cpu_to_be64(wr->sg_list[i].addr);
+		}
+
+		if (i < srq->msrq.max_avail_gather) {
+			scat[i].byte_count = 0;
+			scat[i].lkey       = cpu_to_be32(MLX5_INVALID_LKEY);
+			scat[i].addr       = 0;
+		}
+	}
+
+	if (likely(nreq)) {
+		srq->wqe_ctr += nreq;
+
+		/* Make sure that descriptors are written before
+		 * doorbell record.
+		 */
+		wmb();
+
+		*srq->db.db = cpu_to_be32(srq->wqe_ctr);
+	}
+
+	spin_unlock_irqrestore(&srq->lock, flags);
+
+	return err;
+}
diff --git a/kernel/drivers/infiniband/hw/mlx5/user.h b/kernel/drivers/infiniband/hw/mlx5/user.h
new file mode 100644
index 000000000..76fb7b927
--- /dev/null
+++ b/kernel/drivers/infiniband/hw/mlx5/user.h
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef MLX5_IB_USER_H
+#define MLX5_IB_USER_H
+
+#include <linux/types.h>
+
+enum {
+	MLX5_QP_FLAG_SIGNATURE		= 1 << 0,
+	MLX5_QP_FLAG_SCATTER_CQE	= 1 << 1,
+};
+
+enum {
+	MLX5_SRQ_FLAG_SIGNATURE		= 1 << 0,
+};
+
+
+/* Increment this value if any changes that break userspace ABI
+ * compatibility are made.
+ */
+#define MLX5_IB_UVERBS_ABI_VERSION	1
+
+/* Make sure that all structs defined in this file remain laid out so
+ * that they pack the same way on 32-bit and 64-bit architectures (to
+ * avoid incompatibility between 32-bit userspace and 64-bit kernels).
+ * In particular do not use pointer types -- pass pointers in __u64
+ * instead.
+ */
+
+struct mlx5_ib_alloc_ucontext_req {
+	__u32	total_num_uuars;
+	__u32	num_low_latency_uuars;
+};
+
+struct mlx5_ib_alloc_ucontext_req_v2 {
+	__u32	total_num_uuars;
+	__u32	num_low_latency_uuars;
+	__u32	flags;
+	__u32	reserved;
+};
+
+struct mlx5_ib_alloc_ucontext_resp {
+	__u32	qp_tab_size;
+	__u32	bf_reg_size;
+	__u32	tot_uuars;
+	__u32	cache_line_size;
+	__u16	max_sq_desc_sz;
+	__u16	max_rq_desc_sz;
+	__u32	max_send_wqebb;
+	__u32	max_recv_wr;
+	__u32	max_srq_recv_wr;
+	__u16	num_ports;
+	__u16	reserved;
+};
+
+struct mlx5_ib_alloc_pd_resp {
+	__u32	pdn;
+};
+
+struct mlx5_ib_create_cq {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	cqe_size;
+	__u32	reserved; /* explicit padding (optional on i386) */
+};
+
+struct mlx5_ib_create_cq_resp {
+	__u32	cqn;
+	__u32	reserved;
+};
+
+struct mlx5_ib_resize_cq {
+	__u64	buf_addr;
+	__u16	cqe_size;
+	__u16	reserved0;
+	__u32	reserved1;
+};
+
+struct mlx5_ib_create_srq {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	flags;
+	__u32	reserved; /* explicit padding (optional on i386) */
+};
+
+struct mlx5_ib_create_srq_resp {
+	__u32	srqn;
+	__u32	reserved;
+};
+
+struct mlx5_ib_create_qp {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	sq_wqe_count;
+	__u32	rq_wqe_count;
+	__u32	rq_wqe_shift;
+	__u32	flags;
+};
+
+struct mlx5_ib_create_qp_resp {
+	__u32	uuar_index;
+};
+#endif /* MLX5_IB_USER_H */