mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations

Previously, the structure of a guest's proxy QPs followed the
structure of the PPF special qps (qp0 port 1, qp0 port 2, qp1 port 1,
qp1 port 2, ...).  The guest then did offset calculations on the
sqp_base qp number that the PPF passed to it in QUERY_FUNC_CAP().

This is now changed so that the guest does no offset calculations
regarding proxy or tunnel QPs to use.  This change frees the PPF from
needing to adhere to a specific order in allocating proxy and tunnel
QPs.

Now QUERY_FUNC_CAP provides each port individually with its proxy
qp0, proxy qp1, tunnel qp0, and tunnel qp1 QP numbers, and these are
used directly where required (with no offset calculations).

To accomplish this change, several fields were added to the phys_caps
structure for use by the PPF and by non-SR-IOV mode:

    base_sqpn -- in non-sriov mode, this was formerly sqp_start.
    base_proxy_sqpn -- the first physical proxy qp number -- used by PPF
    base_tunnel_sqpn -- the first physical tunnel qp number -- used by PPF.

The current code in the PPF still adheres to the previous layout of
sqps, proxy-sqps and tunnel-sqps.  However, the PPF can change this
layout without affecting VF or (paravirtualized) PF code.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <roland@purestorage.com>

6 files changed, 280 insertions, 155 deletions

diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 603a114b3dfe..658a622791fb 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -505,7 +505,7 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 	} else
 		tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
 
-	dqpn = dev->dev->caps.sqp_start + 8 * slave + port + (dest_qpt * 2) - 1;
+	dqpn = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave + port + (dest_qpt * 2) - 1;
 
 	/* get tunnel tx data buf for slave */
 	src_qp = tun_qp->qp;
@@ -1074,9 +1074,9 @@ static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
 
 static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
 {
-	int slave_start = dev->dev->caps.sqp_start + 8 * slave;
+	int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave;
 
-	return (qpn >= slave_start && qpn <= slave_start + 1);
+	return (qpn >= proxy_start && qpn <= proxy_start + 1);
 }
 
 
@@ -1191,14 +1191,14 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc
 	int slave;
 
 	/* Get slave that sent this packet */
-	if (wc->src_qp < dev->dev->caps.sqp_start ||
-	    wc->src_qp >= dev->dev->caps.base_tunnel_sqpn ||
+	if (wc->src_qp < dev->dev->phys_caps.base_proxy_sqpn ||
+	    wc->src_qp >= dev->dev->phys_caps.base_proxy_sqpn + 8 * MLX4_MFUNC_MAX ||
 	    (wc->src_qp & 0x1) != ctx->port - 1 ||
 	    wc->src_qp & 0x4) {
 		mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp);
 		return;
 	}
-	slave = ((wc->src_qp & ~0x7) - dev->dev->caps.sqp_start) / 8;
+	slave = ((wc->src_qp & ~0x7) - dev->dev->phys_caps.base_proxy_sqpn) / 8;
 	if (slave != ctx->slave) {
 		mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: "
 			     "belongs to another slave\n", wc->src_qp);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index a8622510de42..96fe103f5973 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -116,33 +116,57 @@ static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 	if (!mlx4_is_master(dev->dev))
 		return 0;
 
-	return qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
-	       qp->mqp.qpn < dev->dev->caps.base_sqpn +
-	       8 + 16 * MLX4_MFUNC_MAX;
+	return qp->mqp.qpn >= dev->dev->phys_caps.base_tunnel_sqpn &&
+	       qp->mqp.qpn < dev->dev->phys_caps.base_tunnel_sqpn +
+		8 * MLX4_MFUNC_MAX;
 }
 
 static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 {
-	return ((mlx4_is_master(dev->dev) &&
-		 qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
-		 qp->mqp.qpn <= dev->dev->caps.base_sqpn + 3) ||
-		(qp->mqp.qpn >= dev->dev->caps.sqp_start &&
-		 qp->mqp.qpn <= dev->dev->caps.sqp_start + 3));
+	int proxy_sqp = 0;
+	int real_sqp = 0;
+	int i;
+	/* PPF or Native -- real SQP */
+	real_sqp = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+		    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+		    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 3);
+	if (real_sqp)
+		return 1;
+	/* VF or PF -- proxy SQP */
+	if (mlx4_is_mfunc(dev->dev)) {
+		for (i = 0; i < dev->dev->caps.num_ports; i++) {
+			if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i] ||
+			    qp->mqp.qpn == dev->dev->caps.qp1_proxy[i]) {
+				proxy_sqp = 1;
+				break;
+			}
+		}
+	}
+	return proxy_sqp;
 }
 
 /* used for INIT/CLOSE port logic */
 static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 {
-	int qp0;
-
-	/* qp0 is either the proxy qp0, or the real qp0 */
-	qp0 = (qp->mqp.qpn >= dev->dev->caps.sqp_start &&
-		qp->mqp.qpn <= dev->dev->caps.sqp_start + 1) ||
-		(mlx4_is_master(dev->dev) &&
-		 qp->mqp.qpn >= dev->dev->caps.base_sqpn &&
-		 qp->mqp.qpn <= dev->dev->caps.base_sqpn + 1);
-
-	return qp0;
+	int proxy_qp0 = 0;
+	int real_qp0 = 0;
+	int i;
+	/* PPF or Native -- real QP0 */
+	real_qp0 = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+		    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+		    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 1);
+	if (real_qp0)
+		return 1;
+	/* VF or PF -- proxy QP0 */
+	if (mlx4_is_mfunc(dev->dev)) {
+		for (i = 0; i < dev->dev->caps.num_ports; i++) {
+			if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i]) {
+				proxy_qp0 = 1;
+				break;
+			}
+		}
+	}
+	return proxy_qp0;
 }
 
 static void *get_wqe(struct mlx4_ib_qp *qp, int offset)
@@ -607,8 +631,10 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 			qp_type = MLX4_IB_QPT_TUN_SMI_OWNER;
 		else
 			qp_type = MLX4_IB_QPT_TUN_SMI;
-		qpn = dev->dev->caps.base_tunnel_sqpn + 8 * tnl_init->slave +
-		      tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
+		/* we are definitely in the PPF here, since we are creating
+		 * tunnel QPs. base_tunnel_sqpn is therefore valid. */
+		qpn = dev->dev->phys_caps.base_tunnel_sqpn + 8 * tnl_init->slave
+			+ tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
 		sqpn = qpn;
 	}
 
@@ -630,12 +656,6 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 
 	qp->mlx4_ib_qp_type = qp_type;
 
-	if (mlx4_is_mfunc(dev->dev) &&
-	    (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI)) {
-		qpn -= 8;
-		sqpn -= 8;
-	}
-
 	mutex_init(&qp->mutex);
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
@@ -935,6 +955,23 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 	del_gid_entries(qp);
 }
 
+static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr)
+{
+	/* Native or PPF */
+	if (!mlx4_is_mfunc(dev->dev) ||
+	    (mlx4_is_master(dev->dev) &&
+	     attr->create_flags & MLX4_IB_SRIOV_SQP)) {
+		return  dev->dev->phys_caps.base_sqpn +
+			(attr->qp_type == IB_QPT_SMI ? 0 : 2) +
+			attr->port_num - 1;
+	}
+	/* PF or VF -- creating proxies */
+	if (attr->qp_type == IB_QPT_SMI)
+		return dev->dev->caps.qp0_proxy[attr->port_num - 1];
+	else
+		return dev->dev->caps.qp1_proxy[attr->port_num - 1];
+}
+
 struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
@@ -998,9 +1035,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 			return ERR_PTR(-EINVAL);
 
 		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
-				       to_mdev(pd->device)->dev->caps.sqp_start +
-				       (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
-				       init_attr->port_num - 1,
+				       get_sqp_num(to_mdev(pd->device), init_attr),
 				       &qp);
 		if (err)
 			return ERR_PTR(err);
@@ -1643,8 +1678,7 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 		sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
 	else
 		sqp->ud_header.bth.destination_qpn =
-			cpu_to_be32(mdev->dev->caps.base_tunnel_sqpn +
-				    sqp->qp.port - 1);
+			cpu_to_be32(mdev->dev->caps.qp0_tunnel[sqp->qp.port - 1]);
 
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
 	if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
@@ -2012,10 +2046,10 @@ static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
 			cpu_to_be32(0xf0000000);
 
 	memcpy(dseg->av, &sqp_av, sizeof (struct mlx4_av));
-	dseg->dqpn = cpu_to_be32(dev->dev->caps.base_tunnel_sqpn +
-				 qpt * 2 + port - 1);
-	/* use well-known qkey from the QPC */
-	dseg->qkey = cpu_to_be32(0x80000000);
+	/* This function used only for sending on QP1 proxies */
+	dseg->dqpn = cpu_to_be32(dev->dev->caps.qp1_tunnel[port - 1]);
+	/* Use QKEY from the QP context, which is set by master */
+	dseg->qkey = cpu_to_be32(IB_QP_SET_QKEY);
 }
 
 static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len)
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 1d74e8541146..e2ed36e2fb29 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -184,8 +184,6 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET		0x28
 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET		0x2c
 #define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET	0X30
-#define QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET	0x44
-#define QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET	0x48
 
 #define QUERY_FUNC_CAP_FMR_FLAG			0x80
 #define QUERY_FUNC_CAP_FLAG_RDMA		0x40
@@ -196,21 +194,39 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 #define QUERY_FUNC_CAP_RDMA_PROPS_OFFSET	0x8
 #define QUERY_FUNC_CAP_ETH_PROPS_OFFSET		0xc
 
+#define QUERY_FUNC_CAP_QP0_TUNNEL		0x10
+#define QUERY_FUNC_CAP_QP0_PROXY		0x14
+#define QUERY_FUNC_CAP_QP1_TUNNEL		0x18
+#define QUERY_FUNC_CAP_QP1_PROXY		0x1c
+
 #define QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC	0x40
 #define QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN	0x80
 
 #define QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID 0x80
 
 	if (vhcr->op_modifier == 1) {
-		field = vhcr->in_modifier;
-		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
-
 		field = 0;
 		/* ensure force vlan and force mac bits are not set */
 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
 		/* ensure that phy_wqe_gid bit is not set */
 		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
 
+		field = vhcr->in_modifier; /* phys-port = logical-port */
+		MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
+
+		/* size is now the QP number */
+		size = dev->phys_caps.base_tunnel_sqpn + 8 * slave + field - 1;
+		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_TUNNEL);
+
+		size += 2;
+		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_TUNNEL);
+
+		size = dev->phys_caps.base_proxy_sqpn + 8 * slave + field - 1;
+		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_PROXY);
+
+		size += 2;
+		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_PROXY);
+
 	} else if (vhcr->op_modifier == 0) {
 		/* enable rdma and ethernet interfaces */
 		field = (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA);
@@ -249,117 +265,124 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 		size = dev->caps.num_mgms + dev->caps.num_amgms;
 		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
 
-		size = dev->caps.base_tunnel_sqpn + 8 * slave;
-		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
-
-		size = dev->caps.sqp_start + 8 * slave;
-		MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
-
 	} else
 		err = -EINVAL;
 
 	return err;
 }
 
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap)
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+			struct mlx4_func_cap *func_cap)
 {
 	struct mlx4_cmd_mailbox *mailbox;
 	u32			*outbox;
-	u8			field;
+	u8			field, op_modifier;
 	u32			size;
-	int			i;
 	int			err = 0;
 
+	op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */
 
 	mailbox = mlx4_alloc_cmd_mailbox(dev);
 	if (IS_ERR(mailbox))
 		return PTR_ERR(mailbox);
 
-	err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_FUNC_CAP,
+	err = mlx4_cmd_box(dev, 0, mailbox->dma, gen_or_port, op_modifier,
+			   MLX4_CMD_QUERY_FUNC_CAP,
 			   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
 	if (err)
 		goto out;
 
 	outbox = mailbox->buf;
 
-	MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
-	if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
-		mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
-		err = -EPROTONOSUPPORT;
-		goto out;
-	}
-	func_cap->flags = field;
-
-	MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
-	func_cap->num_ports = field;
+	if (!op_modifier) {
+		MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET);
+		if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) {
+			mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n");
+			err = -EPROTONOSUPPORT;
+			goto out;
+		}
+		func_cap->flags = field;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
-	func_cap->pf_context_behaviour = size;
+		MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET);
+		func_cap->num_ports = field;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
-	func_cap->qp_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET);
+		func_cap->pf_context_behaviour = size;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
-	func_cap->srq_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET);
+		func_cap->qp_quota = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
-	func_cap->cq_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET);
+		func_cap->srq_quota = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
-	func_cap->max_eq = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET);
+		func_cap->cq_quota = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
-	func_cap->reserved_eq = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET);
+		func_cap->max_eq = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
-	func_cap->mpt_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET);
+		func_cap->reserved_eq = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
-	func_cap->mtt_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET);
+		func_cap->mpt_quota = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
-	func_cap->mcg_quota = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET);
+		func_cap->mtt_quota = size & 0xFFFFFF;
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET);
-	func_cap->base_tunnel_qpn = size & 0xFFFFFF;
+		MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET);
+		func_cap->mcg_quota = size & 0xFFFFFF;
+		goto out;
+	}
 
-	MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET);
-	func_cap->base_proxy_qpn = size & 0xFFFFFF;
+	/* logical port query */
+	if (gen_or_port > dev->caps.num_ports) {
+		err = -EINVAL;
+		goto out;
+	}
 
-	for (i = 1; i <= func_cap->num_ports; ++i) {
-		err = mlx4_cmd_box(dev, 0, mailbox->dma, i, 1,
-				   MLX4_CMD_QUERY_FUNC_CAP,
-				   MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
-		if (err)
+	if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_ETH) {
+		MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
+		if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) {
+			mlx4_err(dev, "VLAN is enforced on this port\n");
+			err = -EPROTONOSUPPORT;
 			goto out;
+		}
 
-		if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH) {
-			MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET);
-			if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) {
-				mlx4_err(dev, "VLAN is enforced on this port\n");
-				err = -EPROTONOSUPPORT;
-				goto out;
-			}
-
-			if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) {
-				mlx4_err(dev, "Force mac is enabled on this port\n");
-				err = -EPROTONOSUPPORT;
-				goto out;
-			}
-		} else if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB) {
-			MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
-			if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) {
-				mlx4_err(dev, "phy_wqe_gid is "
-					 "enforced on this ib port\n");
-				err = -EPROTONOSUPPORT;
-				goto out;
-			}
+		if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) {
+			mlx4_err(dev, "Force mac is enabled on this port\n");
+			err = -EPROTONOSUPPORT;
+			goto out;
+		}
+	} else if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_IB) {
+		MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET);
+		if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) {
+			mlx4_err(dev, "phy_wqe_gid is "
+				 "enforced on this ib port\n");
+			err = -EPROTONOSUPPORT;
+			goto out;
 		}
+	}
 
-		MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
-		func_cap->physical_port[i] = field;
+	MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET);
+	func_cap->physical_port = field;
+	if (func_cap->physical_port != gen_or_port) {
+		err = -ENOSYS;
+		goto out;
 	}
 
+	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_TUNNEL);
+	func_cap->qp0_tunnel_qpn = size & 0xFFFFFF;
+
+	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_PROXY);
+	func_cap->qp0_proxy_qpn = size & 0xFFFFFF;
+
+	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_TUNNEL);
+	func_cap->qp1_tunnel_qpn = size & 0xFFFFFF;
+
+	MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_PROXY);
+	func_cap->qp1_proxy_qpn = size & 0xFFFFFF;
+
 	/* All other resources are allocated by the master, but we still report
 	 * 'num' and 'reserved' capabilities as follows:
 	 * - num remains the maximum resource index
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h
index ced1de57c818..85abe9c11a22 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.h
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.h
@@ -134,11 +134,12 @@ struct mlx4_func_cap {
 	int	max_eq;
 	int	reserved_eq;
 	int	mcg_quota;
-	u32	base_qpn;
-	u32	base_tunnel_qpn;
-	u32	base_proxy_qpn;
-	u8	physical_port[MLX4_MAX_PORTS + 1];
-	u8	port_flags[MLX4_MAX_PORTS + 1];
+	u32	qp0_tunnel_qpn;
+	u32	qp0_proxy_qpn;
+	u32	qp1_tunnel_qpn;
+	u32	qp1_proxy_qpn;
+	u8	physical_port;
+	u8	port_flags;
 };
 
 struct mlx4_adapter {
@@ -195,7 +196,8 @@ struct mlx4_set_ib_param {
 };
 
 int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap);
-int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap);
+int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port,
+			struct mlx4_func_cap *func_cap);
 int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave,
 				struct mlx4_vhcr *vhcr,
 				struct mlx4_cmd_mailbox *inbox,
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index 76f69fdd01d5..1e2ab9d00f1c 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -410,15 +410,16 @@ static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
 {
 	u32 qk = MLX4_RESERVED_QKEY_BASE;
-	if (qpn >= dev->caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
-	    qpn < dev->caps.sqp_start)
+
+	if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
+	    qpn < dev->phys_caps.base_proxy_sqpn)
 		return -EINVAL;
 
-	if (qpn >= dev->caps.base_tunnel_sqpn)
+	if (qpn >= dev->phys_caps.base_tunnel_sqpn)
 		/* tunnel qp */
-		qk += qpn - dev->caps.base_tunnel_sqpn;
+		qk += qpn - dev->phys_caps.base_tunnel_sqpn;
 	else
-		qk += qpn - dev->caps.sqp_start;
+		qk += qpn - dev->phys_caps.base_proxy_sqpn;
 	*qkey = qk;
 	return 0;
 }
@@ -527,9 +528,10 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
 	}
 
 	memset(&func_cap, 0, sizeof(func_cap));
-	err = mlx4_QUERY_FUNC_CAP(dev, &func_cap);
+	err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
 	if (err) {
-		mlx4_err(dev, "QUERY_FUNC_CAP command failed, aborting.\n");
+		mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
+			  err);
 		return err;
 	}
 
@@ -557,12 +559,33 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
 		return -ENODEV;
 	}
 
+	dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+	dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+	dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+	dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+
+	if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
+	    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
+		err = -ENOMEM;
+		goto err_mem;
+	}
+
 	for (i = 1; i <= dev->caps.num_ports; ++i) {
+		err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
+		if (err) {
+			mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
+				 " port %d, aborting (%d).\n", i, err);
+			goto err_mem;
+		}
+		dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
+		dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
+		dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
+		dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
 		dev->caps.port_mask[i] = dev->caps.port_type[i];
 		if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
 						    &dev->caps.gid_table_len[i],
 						    &dev->caps.pkey_table_len[i]))
-			return -ENODEV;
+			goto err_mem;
 	}
 
 	if (dev->caps.uar_page_size * (dev->caps.num_uars -
@@ -572,14 +595,20 @@ static int mlx4_slave_cap(struct mlx4_dev *dev)
 			 "PCI resource 2 size of 0x%llx, aborting.\n",
 			 dev->caps.uar_page_size * dev->caps.num_uars,
 			 (unsigned long long) pci_resource_len(dev->pdev, 2));
-		return -ENODEV;
+		goto err_mem;
 	}
 
-	/* Calculate our sqp_start */
-	dev->caps.sqp_start = func_cap.base_proxy_qpn;
-	dev->caps.base_tunnel_sqpn = func_cap.base_tunnel_qpn;
-
 	return 0;
+
+err_mem:
+	kfree(dev->caps.qp0_tunnel);
+	kfree(dev->caps.qp0_proxy);
+	kfree(dev->caps.qp1_tunnel);
+	kfree(dev->caps.qp1_proxy);
+	dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
+		dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
+
+	return err;
 }
 
 /*
@@ -2261,6 +2290,12 @@ static void mlx4_remove_one(struct pci_dev *pdev)
 
 		if (!mlx4_is_slave(dev))
 			mlx4_free_ownership(dev);
+
+		kfree(dev->caps.qp0_tunnel);
+		kfree(dev->caps.qp0_proxy);
+		kfree(dev->caps.qp1_tunnel);
+		kfree(dev->caps.qp1_proxy);
+
 		kfree(priv);
 		pci_release_regions(pdev);
 		pci_disable_device(pdev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c
index 436ef6c69add..81e2abe07bbb 100644
--- a/drivers/net/ethernet/mellanox/mlx4/qp.c
+++ b/drivers/net/ethernet/mellanox/mlx4/qp.c
@@ -68,15 +68,15 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type)
 }
 
 /* used for INIT/CLOSE port logic */
-static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0)
+static int is_master_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0)
 {
+	/* this procedure is called after we already know we are on the master */
 	/* qp0 is either the proxy qp0, or the real qp0 */
-	*proxy_qp0 = qp->qpn >= dev->caps.sqp_start &&
-		qp->qpn <= dev->caps.sqp_start + 1;
+	u32 pf_proxy_offset = dev->phys_caps.base_proxy_sqpn + 8 * mlx4_master_func_num(dev);
+	*proxy_qp0 = qp->qpn >= pf_proxy_offset && qp->qpn <= pf_proxy_offset + 1;
 
-	*real_qp0 = mlx4_is_master(dev) &&
-		qp->qpn >= dev->caps.base_sqpn &&
-		qp->qpn <= dev->caps.base_sqpn + 1;
+	*real_qp0 = qp->qpn >= dev->phys_caps.base_sqpn &&
+		qp->qpn <= dev->phys_caps.base_sqpn + 1;
 
 	return *real_qp0 || *proxy_qp0;
 }
@@ -143,7 +143,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
 			MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native);
 		if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR &&
 		    cur_state != MLX4_QP_STATE_RST &&
-		    is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
+		    is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
 			port = (qp->qpn & 1) + 1;
 			if (proxy_qp0)
 				priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0;
@@ -175,7 +175,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
 		       new_state == MLX4_QP_STATE_RST ? 2 : 0,
 		       op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native);
 
-	if (mlx4_is_master(dev) && is_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
+	if (mlx4_is_master(dev) && is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) {
 		port = (qp->qpn & 1) + 1;
 		if (cur_state != MLX4_QP_STATE_ERR &&
 		    cur_state != MLX4_QP_STATE_RST &&
@@ -422,6 +422,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table;
 	int err;
 	int reserved_from_top = 0;
+	int k;
 
 	spin_lock_init(&qp_table->lock);
 	INIT_RADIX_TREE(&dev->qp_table_tree, GFP_ATOMIC);
@@ -436,7 +437,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	 * We also reserve the MSB of the 24-bit QP number to indicate
 	 * that a QP is an XRC QP.
 	 */
-	dev->caps.base_sqpn =
+	dev->phys_caps.base_sqpn =
 		ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);
 
 	{
@@ -479,24 +480,54 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	*/
 
 	err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps,
-			       (1 << 23) - 1, dev->caps.base_sqpn + 8 +
+			       (1 << 23) - 1, dev->phys_caps.base_sqpn + 8 +
 			       16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev),
 			       reserved_from_top);
+	if (err)
+		return err;
 
-	/* In mfunc, sqp_start is the base of the proxy SQPs, since the PF also
-	 * uses paravirtualized SQPs.
-	 * In native mode, sqp_start is the base of the real SQPs. */
 	if (mlx4_is_mfunc(dev)) {
-		dev->caps.sqp_start = dev->caps.base_sqpn +
-			8 * (mlx4_master_func_num(dev) + 1);
-		dev->caps.base_tunnel_sqpn = dev->caps.sqp_start + 8 * MLX4_MFUNC_MAX;
-	} else
-		dev->caps.sqp_start = dev->caps.base_sqpn;
+		/* for PPF use */
+		dev->phys_caps.base_proxy_sqpn = dev->phys_caps.base_sqpn + 8;
+		dev->phys_caps.base_tunnel_sqpn = dev->phys_caps.base_sqpn + 8 + 8 * MLX4_MFUNC_MAX;
+
+		/* In mfunc, calculate proxy and tunnel qp offsets for the PF here,
+		 * since the PF does not call mlx4_slave_caps */
+		dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+		dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+		dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+		dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
+
+		if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
+		    !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
+			err = -ENOMEM;
+			goto err_mem;
+		}
 
+		for (k = 0; k < dev->caps.num_ports; k++) {
+			dev->caps.qp0_proxy[k] = dev->phys_caps.base_proxy_sqpn +
+				8 * mlx4_master_func_num(dev) + k;
+			dev->caps.qp0_tunnel[k] = dev->caps.qp0_proxy[k] + 8 * MLX4_MFUNC_MAX;
+			dev->caps.qp1_proxy[k] = dev->phys_caps.base_proxy_sqpn +
+				8 * mlx4_master_func_num(dev) + MLX4_MAX_PORTS + k;
+			dev->caps.qp1_tunnel[k] = dev->caps.qp1_proxy[k] + 8 * MLX4_MFUNC_MAX;
+		}
+	}
+
+
+	err = mlx4_CONF_SPECIAL_QP(dev, dev->phys_caps.base_sqpn);
 	if (err)
-		return err;
+		goto err_mem;
+	return 0;
 
-	return mlx4_CONF_SPECIAL_QP(dev, dev->caps.base_sqpn);
+err_mem:
+	kfree(dev->caps.qp0_tunnel);
+	kfree(dev->caps.qp0_proxy);
+	kfree(dev->caps.qp1_tunnel);
+	kfree(dev->caps.qp1_proxy);
+	dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
+		dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
+	return err;
 }
 
 void mlx4_cleanup_qp_table(struct mlx4_dev *dev)