[PATCH rdma-next v1 00/36] Refactor mlx5_ib_create_qp

[PATCH rdma-next v1 00/36] Refactor mlx5_ib_create_qp

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Changelog
v1: * Combined both series to easy apply
    * Rebsed on top of rdma-next + mlx5-next
    * Fixed create_qp flags check
v0: https://lore.kernel.org/lkml/20200420151105.282848-1-leon@kernel.org
    https://lore.kernel.org/lkml/20200423190303.12856-1-leon@kernel.org

-----------------------------------------------------------------------------

Hi,

This is first part of series which tries to return some sanity
to mlx5_ib_create_qp() function. Such refactoring is required
to make extension of that function with less worries of breaking
driver.

Extra goal of such refactoring is to ensure that QP is allocated
at the beginning of function and released at the end. It will allow
us to move QP allocation to be under IB/core responsibility.

It is based on previously sent [1] "[PATCH mlx5-next 00/24] Mass
conversion to light mlx5 command interface"

Thanks

[1]
https://lore.kernel.org/linux-rdma/20200420114136.264924-1-leon@kernel.org

Aharon Landau (1):
  RDMA/mlx5: Verify that QP is created with RQ or SQ

Leon Romanovsky (35):
  RDMA/mlx5: Organize QP types checks in one place
  RDMA/mlx5: Delete impossible GSI port check
  RDMA/mlx5: Perform check if QP creation flow is valid
  RDMA/mlx5: Prepare QP allocation for future removal
  RDMA/mlx5: Avoid setting redundant NULL for XRC QPs
  RDMA/mlx5: Set QP subtype immediately when it is known
  RDMA/mlx5: Separate create QP flows to be based on type
  RDMA/mlx5: Split scatter CQE configuration for DCT QP
  RDMA/mlx5: Update all DRIVER QP places to use QP subtype
  RDMA/mlx5: Move DRIVER QP flags check into separate function
  RDMA/mlx5: Remove second copy from user for non RSS RAW QPs
  RDMA/mlx5: Initial separation of RAW_PACKET QP from common flow
  RDMA/mlx5: Delete create QP flags obfuscation
  RDMA/mlx5: Process create QP flags in one place
  RDMA/mlx5: Use flags_en mechanism to mark QP created with WQE
    signature
  RDMA/mlx5: Change scatter CQE flag to be set like other vendor flags
  RDMA/mlx5: Return all configured create flags through query QP
  RDMA/mlx5: Process all vendor flags in one place
  RDMA/mlx5: Delete unsupported QP types
  RDMA/mlx5: Store QP type in the vendor QP structure
  RDMA/mlx5: Promote RSS RAW QP attribute check in higher level
  RDMA/mlx5: Combine copy of create QP command in RSS RAW QP
  RDMA/mlx5: Remove second user copy in create_user_qp
  RDMA/mlx5: Rely on existence of udata to separate kernel/user flows
  RDMA/mlx5: Delete impossible inlen check
  RDMA/mlx5: Globally parse DEVX UID
  RDMA/mlx5: Separate XRC_TGT QP creation from common flow
  RDMA/mlx5: Separate to user/kernel create QP flows
  RDMA/mlx5: Reduce amount of duplication in QP destroy
  RDMA/mlx5: Group all create QP parameters to simplify in-kernel
    interfaces
  RDMA/mlx5: Promote RSS RAW QP flags check to higher level
  RDMA/mlx5: Handle udate outlen checks in one place
  RDMA/mlx5: Copy response to the user in one place
  RDMA/mlx5: Remove redundant destroy QP call
  RDMA/mlx5: Consolidate into special function all create QP calls

 drivers/infiniband/hw/mlx5/devx.c    |    2 +-
 drivers/infiniband/hw/mlx5/flow.c    |    2 +-
 drivers/infiniband/hw/mlx5/gsi.c     |    9 -
 drivers/infiniband/hw/mlx5/main.c    |    6 +-
 drivers/infiniband/hw/mlx5/mlx5_ib.h |   46 +-
 drivers/infiniband/hw/mlx5/odp.c     |    5 +-
 drivers/infiniband/hw/mlx5/qp.c      | 1625 ++++++++++++++------------
 7 files changed, 905 insertions(+), 790 deletions(-)

--
2.25.3

[PATCH rdma-next v1 01/36] RDMA/mlx5: Organize QP types checks in one place

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Perform check if QP type is supported in one place at the beginning of
the create_qp function instead of current implementation with checks
buried inside of the code.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 129 +++++++++++++++++---------------
 1 file changed, 68 insertions(+), 61 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index af599c8b88aa..fdab5b6db1e5 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2677,12 +2677,42 @@ static int set_mlx_qp_type(struct mlx5_ib_dev *dev,
 		}
 	}
 
-	if (!MLX5_CAP_GEN(dev->mdev, dct)) {
-		mlx5_ib_dbg(dev, "DC transport is not supported\n");
-		return -EOPNOTSUPP;
+	return 0;
+}
+
+static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_DRIVER && !MLX5_CAP_GEN(dev->mdev, dct))
+		goto out;
+
+	switch (attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+	case IB_QPT_XRC_INI:
+		if (!MLX5_CAP_GEN(dev->mdev, xrc))
+			goto out;
+		fallthrough;
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_SMI:
+	case MLX5_IB_QPT_HW_GSI:
+	case MLX5_IB_QPT_REG_UMR:
+	case IB_QPT_DRIVER:
+	case IB_QPT_GSI:
+		return 0;
+	case IB_QPT_RAW_IPV6:
+	case IB_QPT_RAW_ETHERTYPE:
+	case IB_QPT_MAX:
+	default:
+		goto out;
 	}
 
 	return 0;
+
+out:
+	mlx5_ib_dbg(dev, "Unsupported QP type %d\n", attr->qp_type);
+	return -EOPNOTSUPP;
 }
 
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
@@ -2698,9 +2728,17 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
 
-	if (pd) {
-		dev = to_mdev(pd->device);
+	dev = pd ? to_mdev(pd->device) :
+		   to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
 
+	err = check_qp_type(dev, init_attr);
+	if (err) {
+		mlx5_ib_dbg(dev, "Unsupported QP type %d\n",
+			    init_attr->qp_type);
+		return ERR_PTR(err);
+	}
+
+	if (pd) {
 		if (init_attr->qp_type == IB_QPT_RAW_PACKET) {
 			if (!ucontext) {
 				mlx5_ib_dbg(dev, "Raw Packet QP is not supported for kernel consumers\n");
@@ -2718,7 +2756,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				ib_qp_type_str(init_attr->qp_type));
 			return ERR_PTR(-EINVAL);
 		}
-		dev = to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
 	}
 
 	if (init_attr->qp_type == IB_QPT_DRIVER) {
@@ -2741,67 +2778,37 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		}
 	}
 
-	switch (init_attr->qp_type) {
-	case IB_QPT_XRC_TGT:
-	case IB_QPT_XRC_INI:
-		if (!MLX5_CAP_GEN(dev->mdev, 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_RAW_PACKET:
-	case IB_QPT_RC:
-	case IB_QPT_UC:
-	case IB_QPT_UD:
-	case IB_QPT_SMI:
-	case MLX5_IB_QPT_HW_GSI:
-	case MLX5_IB_QPT_REG_UMR:
-	case MLX5_IB_QPT_DCI:
-		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 (init_attr->qp_type == IB_QPT_GSI)
+		return mlx5_ib_gsi_create_qp(pd, init_attr);
 
-		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->trans_qp.base.mqp.qpn;
+	if (init_attr->qp_type == IB_QPT_XRC_TGT) {
+		init_attr->recv_cq = NULL;
+		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+		init_attr->send_cq = NULL;
+	}
 
-		mlx5_ib_dbg(dev, "ib qpnum 0x%x, mlx qpn 0x%x, rcqn 0x%x, scqn 0x%x\n",
-			    qp->ibqp.qp_num, qp->trans_qp.base.mqp.qpn,
-			    init_attr->recv_cq ? to_mcq(init_attr->recv_cq)->mcq.cqn : -1,
-			    init_attr->send_cq ? to_mcq(init_attr->send_cq)->mcq.cqn : -1);
+	if (init_attr->qp_type == IB_QPT_XRC_INI)
+		init_attr->recv_cq = NULL;
 
-		qp->trans_qp.xrcdn = xrcdn;
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
 
-		break;
+	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);
+	}
 
-	case IB_QPT_GSI:
-		return mlx5_ib_gsi_create_qp(pd, init_attr);
+	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->trans_qp.base.mqp.qpn;
 
-	case IB_QPT_RAW_IPV6:
-	case IB_QPT_RAW_ETHERTYPE:
-	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(-EOPNOTSUPP);
-	}
+	qp->trans_qp.xrcdn = xrcdn;
 
 	if (verbs_init_attr->qp_type == IB_QPT_DRIVER)
 		qp->qp_sub_type = init_attr->qp_type;
-- 
2.25.3

[PATCH rdma-next v1 02/36] RDMA/mlx5: Delete impossible GSI port check

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

GSI QP is created in the kernel with very strict parameters,
there is no possible way that port number will be wrong in
such flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/gsi.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index 1ae6fd95acaa..1afbf03d1a98 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -123,15 +123,6 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 	const int num_qps = mlx5_ib_deth_sqpn_cap(dev) ? num_pkeys : 0;
 	int ret;
 
-	mlx5_ib_dbg(dev, "creating GSI QP\n");
-
-	if (port_num > ARRAY_SIZE(dev->devr.ports) || port_num < 1) {
-		mlx5_ib_warn(dev,
-			     "invalid port number %d during GSI QP creation\n",
-			     port_num);
-		return ERR_PTR(-EINVAL);
-	}
-
 	gsi = kzalloc(sizeof(*gsi), GFP_KERNEL);
 	if (!gsi)
 		return ERR_PTR(-ENOMEM);
-- 
2.25.3

[PATCH rdma-next v1 03/36] RDMA/mlx5: Perform check if QP creation flow is valid

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Fast check that kernel and user flows provides enough
data to create QP.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 128 +++++++++++++++-----------------
 1 file changed, 61 insertions(+), 67 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index fdab5b6db1e5..91d6151c349c 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1666,9 +1666,6 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	size_t required_cmd_sz;
 	u8 lb_flag = 0;
 
-	if (init_attr->qp_type != IB_QPT_RAW_PACKET)
-		return -EOPNOTSUPP;
-
 	if (init_attr->create_flags || init_attr->send_cq)
 		return -EINVAL;
 
@@ -2032,13 +2029,8 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (mlx5_st < 0)
 		return -EINVAL;
 
-	if (init_attr->rwq_ind_tbl) {
-		if (!udata)
-			return -ENOSYS;
-
-		err = create_rss_raw_qp_tir(dev, qp, pd, init_attr, udata);
-		return err;
-	}
+	if (init_attr->rwq_ind_tbl)
+		return create_rss_raw_qp_tir(dev, qp, pd, init_attr, udata);
 
 	if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
 		if (!MLX5_CAP_GEN(mdev, block_lb_mc)) {
@@ -2565,39 +2557,6 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 		destroy_qp_user(dev, &get_pd(qp)->ibpd, qp, base, udata);
 }
 
-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_DRIVER:
-		return "IB_QPT_DRIVER";
-	case IB_QPT_MAX:
-	default:
-		return "Invalid QP type";
-	}
-}
-
 static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd,
 					struct ib_qp_init_attr *attr,
 					struct mlx5_ib_create_qp *ucmd,
@@ -2655,9 +2614,6 @@ static int set_mlx_qp_type(struct mlx5_ib_dev *dev,
 	enum { MLX_QP_FLAGS = MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI };
 	int err;
 
-	if (!udata)
-		return -EINVAL;
-
 	if (udata->inlen < sizeof(*ucmd)) {
 		mlx5_ib_dbg(dev, "create_qp user command is smaller than expected\n");
 		return -EINVAL;
@@ -2715,6 +2671,62 @@ static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
 	return -EOPNOTSUPP;
 }
 
+static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			    struct ib_qp_init_attr *attr,
+			    struct ib_udata *udata)
+{
+	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
+		udata, struct mlx5_ib_ucontext, ibucontext);
+
+	if (!udata) {
+		/* Kernel create_qp callers */
+		if (attr->rwq_ind_tbl)
+			return -EOPNOTSUPP;
+
+		switch (attr->qp_type) {
+		case IB_QPT_RAW_PACKET:
+		case IB_QPT_DRIVER:
+			return -EOPNOTSUPP;
+		default:
+			return 0;
+		}
+	}
+
+	/* Userspace create_qp callers */
+	if (attr->qp_type == IB_QPT_RAW_PACKET && !ucontext->cqe_version) {
+		mlx5_ib_dbg(dev,
+			"Raw Packet QP is only supported for CQE version > 0\n");
+		return -EINVAL;
+	}
+
+	if (attr->qp_type != IB_QPT_RAW_PACKET && attr->rwq_ind_tbl) {
+		mlx5_ib_dbg(dev,
+			    "Wrong QP type %d for the RWQ indirect table\n",
+			    attr->qp_type);
+		return -EINVAL;
+	}
+
+	switch (attr->qp_type) {
+	case IB_QPT_SMI:
+	case MLX5_IB_QPT_HW_GSI:
+	case MLX5_IB_QPT_REG_UMR:
+	case IB_QPT_GSI:
+		mlx5_ib_dbg(dev, "Kernel doesn't support QP type %d\n",
+			    attr->qp_type);
+		return -EINVAL;
+	default:
+		break;
+	}
+
+	/*
+	 * We don't need to see this warning, it means that kernel code
+	 * missing ib_pd. Placed here to catch developer's mistakes.
+	 */
+	WARN_ONCE(!pd && attr->qp_type != IB_QPT_XRC_TGT,
+		  "There is a missing PD pointer assignment\n");
+	return 0;
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *verbs_init_attr,
 				struct ib_udata *udata)
@@ -2725,8 +2737,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	int err;
 	struct ib_qp_init_attr mlx_init_attr;
 	struct ib_qp_init_attr *init_attr = verbs_init_attr;
-	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
-		udata, struct mlx5_ib_ucontext, ibucontext);
 
 	dev = pd ? to_mdev(pd->device) :
 		   to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
@@ -2738,25 +2748,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		return ERR_PTR(err);
 	}
 
-	if (pd) {
-		if (init_attr->qp_type == IB_QPT_RAW_PACKET) {
-			if (!ucontext) {
-				mlx5_ib_dbg(dev, "Raw Packet QP is not supported for kernel consumers\n");
-				return ERR_PTR(-EINVAL);
-			} else if (!ucontext->cqe_version) {
-				mlx5_ib_dbg(dev, "Raw Packet QP is only supported for CQE version > 0\n");
-				return ERR_PTR(-EINVAL);
-			}
-		}
-	} 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);
-		}
-	}
+	err = check_valid_flow(dev, pd, init_attr, udata);
+	if (err)
+		return ERR_PTR(err);
 
 	if (init_attr->qp_type == IB_QPT_DRIVER) {
 		struct mlx5_ib_create_qp ucmd;
-- 
2.25.3

[PATCH rdma-next v1 04/36] RDMA/mlx5: Prepare QP allocation for future removal

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Unify the QP memory allocation across different paths,
so it will be in one place.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 45 +++++++++++++++------------------
 1 file changed, 20 insertions(+), 25 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 91d6151c349c..07df470e0d58 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2557,14 +2557,13 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 		destroy_qp_user(dev, &get_pd(qp)->ibpd, qp, base, udata);
 }
 
-static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd,
+static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 					struct ib_qp_init_attr *attr,
 					struct mlx5_ib_create_qp *ucmd,
 					struct ib_udata *udata)
 {
 	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
-	struct mlx5_ib_qp *qp;
 	int err = 0;
 	u32 uidx = MLX5_IB_DEFAULT_UIDX;
 	void *dctc;
@@ -2576,15 +2575,9 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd,
 	if (err)
 		return ERR_PTR(err);
 
-	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
-	if (!qp)
-		return ERR_PTR(-ENOMEM);
-
 	qp->dct.in = kzalloc(MLX5_ST_SZ_BYTES(create_dct_in), GFP_KERNEL);
-	if (!qp->dct.in) {
-		err = -ENOMEM;
-		goto err_free;
-	}
+	if (!qp->dct.in)
+		return ERR_PTR(-ENOMEM);
 
 	MLX5_SET(create_dct_in, qp->dct.in, uid, to_mpd(pd)->uid);
 	dctc = MLX5_ADDR_OF(create_dct_in, qp->dct.in, dct_context_entry);
@@ -2601,9 +2594,6 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd,
 	qp->state = IB_QPS_RESET;
 
 	return &qp->ibqp;
-err_free:
-	kfree(qp);
-	return ERR_PTR(err);
 }
 
 static int set_mlx_qp_type(struct mlx5_ib_dev *dev,
@@ -2752,6 +2742,13 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (err)
 		return ERR_PTR(err);
 
+	if (init_attr->qp_type == IB_QPT_GSI)
+		return mlx5_ib_gsi_create_qp(pd, init_attr);
+
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
+
 	if (init_attr->qp_type == IB_QPT_DRIVER) {
 		struct mlx5_ib_create_qp ucmd;
 
@@ -2759,22 +2756,21 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		memcpy(init_attr, verbs_init_attr, sizeof(*verbs_init_attr));
 		err = set_mlx_qp_type(dev, init_attr, &ucmd, udata);
 		if (err)
-			return ERR_PTR(err);
+			goto free_qp;
 
 		if (init_attr->qp_type == MLX5_IB_QPT_DCI) {
 			if (init_attr->cap.max_recv_wr ||
 			    init_attr->cap.max_recv_sge) {
 				mlx5_ib_dbg(dev, "DCI QP requires zero size receive queue\n");
-				return ERR_PTR(-EINVAL);
+				err = -EINVAL;
+				goto free_qp;
 			}
 		} else {
-			return mlx5_ib_create_dct(pd, init_attr, &ucmd, udata);
+			return mlx5_ib_create_dct(pd, qp, init_attr, &ucmd,
+						  udata);
 		}
 	}
 
-	if (init_attr->qp_type == IB_QPT_GSI)
-		return mlx5_ib_gsi_create_qp(pd, init_attr);
-
 	if (init_attr->qp_type == IB_QPT_XRC_TGT) {
 		init_attr->recv_cq = NULL;
 		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
@@ -2784,15 +2780,10 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (init_attr->qp_type == IB_QPT_XRC_INI)
 		init_attr->recv_cq = NULL;
 
-	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);
+		goto free_qp;
 	}
 
 	if (is_qp0(init_attr->qp_type))
@@ -2808,6 +2799,10 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		qp->qp_sub_type = init_attr->qp_type;
 
 	return &qp->ibqp;
+
+free_qp:
+	kfree(qp);
+	return ERR_PTR(err);
 }
 
 static int mlx5_ib_destroy_dct(struct mlx5_ib_qp *mqp)
-- 
2.25.3

[PATCH rdma-next v1 05/36] RDMA/mlx5: Avoid setting redundant NULL for XRC QPs

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

There is no need to set NULL in recv_cq and send_cq, they are already
set to NULL by the IB/core logic.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 07df470e0d58..86933a2023dc 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2771,14 +2771,8 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		}
 	}
 
-	if (init_attr->qp_type == IB_QPT_XRC_TGT) {
-		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;
-	}
-
-	if (init_attr->qp_type == IB_QPT_XRC_INI)
-		init_attr->recv_cq = NULL;
 
 	err = create_qp_common(dev, pd, init_attr, udata, qp);
 	if (err) {
-- 
2.25.3

[PATCH rdma-next v1 06/36] RDMA/mlx5: Set QP subtype immediately when it is known

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

There is no need to delay QP subtype assignment to the end of the
create_qp() function and it is better to move it to be immediately
after it is checked so we would be able to rewrite later checks
to be based on it and not on over-written struct ib_qp_init_attr.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 86933a2023dc..d991c33c4d9b 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2581,7 +2581,6 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 
 	MLX5_SET(create_dct_in, qp->dct.in, uid, to_mpd(pd)->uid);
 	dctc = MLX5_ADDR_OF(create_dct_in, qp->dct.in, dct_context_entry);
-	qp->qp_sub_type = MLX5_IB_QPT_DCT;
 	MLX5_SET(dctc, dctc, pd, to_mpd(pd)->pdn);
 	MLX5_SET(dctc, dctc, srqn_xrqn, to_msrq(attr->srq)->msrq.srqn);
 	MLX5_SET(dctc, dctc, cqn, to_mcq(attr->recv_cq)->mcq.cqn);
@@ -2765,7 +2764,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				err = -EINVAL;
 				goto free_qp;
 			}
+			qp->qp_sub_type = MLX5_IB_QPT_DCI;
 		} else {
+			qp->qp_sub_type = MLX5_IB_QPT_DCT;
 			return mlx5_ib_create_dct(pd, qp, init_attr, &ucmd,
 						  udata);
 		}
@@ -2789,9 +2790,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 
 	qp->trans_qp.xrcdn = xrcdn;
 
-	if (verbs_init_attr->qp_type == IB_QPT_DRIVER)
-		qp->qp_sub_type = init_attr->qp_type;
-
 	return &qp->ibqp;
 
 free_qp:
-- 
2.25.3

[PATCH rdma-next v1 07/36] RDMA/mlx5: Separate create QP flows to be based on type

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Move driver QP creation flow to separate functions to simplify
the create_qp() and allow future separation of create_qp_common()
to subtypes.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 54 ++++++++++++++++++++++++---------
 1 file changed, 39 insertions(+), 15 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index d991c33c4d9b..ae336c1eed74 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2557,10 +2557,9 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 		destroy_qp_user(dev, &get_pd(qp)->ibpd, qp, base, udata);
 }
 
-static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-					struct ib_qp_init_attr *attr,
-					struct mlx5_ib_create_qp *ucmd,
-					struct ib_udata *udata)
+static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
+		      struct ib_qp_init_attr *attr,
+		      struct mlx5_ib_create_qp *ucmd, struct ib_udata *udata)
 {
 	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
@@ -2568,16 +2567,13 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	u32 uidx = MLX5_IB_DEFAULT_UIDX;
 	void *dctc;
 
-	if (!attr->srq || !attr->recv_cq)
-		return ERR_PTR(-EINVAL);
-
 	err = get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), &uidx);
 	if (err)
-		return ERR_PTR(err);
+		return err;
 
 	qp->dct.in = kzalloc(MLX5_ST_SZ_BYTES(create_dct_in), GFP_KERNEL);
 	if (!qp->dct.in)
-		return ERR_PTR(-ENOMEM);
+		return -ENOMEM;
 
 	MLX5_SET(create_dct_in, qp->dct.in, uid, to_mpd(pd)->uid);
 	dctc = MLX5_ADDR_OF(create_dct_in, qp->dct.in, dct_context_entry);
@@ -2592,7 +2588,7 @@ static struct ib_qp *mlx5_ib_create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 
 	qp->state = IB_QPS_RESET;
 
-	return &qp->ibqp;
+	return 0;
 }
 
 static int set_mlx_qp_type(struct mlx5_ib_dev *dev,
@@ -2716,10 +2712,36 @@ static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 }
 
+static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
+			    struct ib_qp_init_attr *attr,
+			    struct mlx5_ib_create_qp *ucmd,
+			    struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
+	int ret = -EINVAL;
+
+	switch (qp->qp_sub_type) {
+	case MLX5_IB_QPT_DCT:
+		if (!attr->srq || !attr->recv_cq)
+			goto out;
+
+		ret = create_dct(pd, qp, attr, ucmd, udata);
+		break;
+	case MLX5_IB_QPT_DCI:
+		ret = create_qp_common(mdev, pd, attr, udata, qp);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+out:	return ret;
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *verbs_init_attr,
 				struct ib_udata *udata)
 {
+	struct mlx5_ib_create_qp ucmd = {};
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
 	u16 xrcdn = 0;
@@ -2749,8 +2771,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		return ERR_PTR(-ENOMEM);
 
 	if (init_attr->qp_type == IB_QPT_DRIVER) {
-		struct mlx5_ib_create_qp ucmd;
-
 		init_attr = &mlx_init_attr;
 		memcpy(init_attr, verbs_init_attr, sizeof(*verbs_init_attr));
 		err = set_mlx_qp_type(dev, init_attr, &ucmd, udata);
@@ -2767,15 +2787,19 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 			qp->qp_sub_type = MLX5_IB_QPT_DCI;
 		} else {
 			qp->qp_sub_type = MLX5_IB_QPT_DCT;
-			return mlx5_ib_create_dct(pd, qp, init_attr, &ucmd,
-						  udata);
 		}
 	}
 
 	if (init_attr->qp_type == IB_QPT_XRC_TGT)
 		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
 
-	err = create_qp_common(dev, pd, init_attr, udata, qp);
+	switch (init_attr->qp_type) {
+	case IB_QPT_DRIVER:
+		err = create_driver_qp(pd, qp, init_attr, &ucmd, udata);
+		break;
+	default:
+		err = create_qp_common(dev, pd, init_attr, udata, qp);
+	}
 	if (err) {
 		mlx5_ib_dbg(dev, "create_qp_common failed\n");
 		goto free_qp;
-- 
2.25.3

[PATCH rdma-next v1 08/36] RDMA/mlx5: Split scatter CQE configuration for DCT QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

DCT QPs have separate creation flow and can be easily extracted
from configure_responder_scat_cqe(), this makes both updated
functions more clear.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index ae336c1eed74..d0e8d27305e9 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1907,13 +1907,6 @@ static void configure_responder_scat_cqe(struct ib_qp_init_attr *init_attr,
 
 	rcqe_sz = mlx5_ib_get_cqe_size(init_attr->recv_cq);
 
-	if (init_attr->qp_type == MLX5_IB_QPT_DCT) {
-		if (rcqe_sz == 128)
-			MLX5_SET(dctc, qpc, cs_res, MLX5_RES_SCAT_DATA64_CQE);
-
-		return;
-	}
-
 	MLX5_SET(qpc, qpc, cs_res,
 		 rcqe_sz == 128 ? MLX5_RES_SCAT_DATA64_CQE :
 				  MLX5_RES_SCAT_DATA32_CQE);
@@ -2583,8 +2576,12 @@ static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	MLX5_SET64(dctc, dctc, dc_access_key, ucmd->access_key);
 	MLX5_SET(dctc, dctc, user_index, uidx);
 
-	if (ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE)
-		configure_responder_scat_cqe(attr, dctc);
+	if (ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE) {
+		int rcqe_sz = mlx5_ib_get_cqe_size(attr->recv_cq);
+
+		if (rcqe_sz == 128)
+			MLX5_SET(dctc, dctc, cs_res, MLX5_RES_SCAT_DATA64_CQE);
+	}
 
 	qp->state = IB_QPS_RESET;
 
-- 
2.25.3

[PATCH rdma-next v1 09/36] RDMA/mlx5: Update all DRIVER QP places to use QP subtype

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Instead of overwriting QP init attributes with driver QP subtype,
use that subtype directly. This change will allow us to remove
logic which cached QP init attributes.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 48 +++++++++++----------------------
 1 file changed, 15 insertions(+), 33 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index d0e8d27305e9..0b2090bcb8e8 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1232,7 +1232,7 @@ static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
 static u32 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 == MLX5_IB_QPT_DCI) ||
+	    (qp->qp_sub_type == MLX5_IB_QPT_DCI) ||
 	    (attr->qp_type == IB_QPT_XRC_INI))
 		return MLX5_SRQ_RQ;
 	else if (!qp->has_rq)
@@ -1241,15 +1241,6 @@ static u32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
 		return 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 ||
-	    qp_type == MLX5_IB_QPT_DCI)
-		return 1;
-
-	return 0;
-}
-
 static int create_raw_packet_qp_tis(struct mlx5_ib_dev *dev,
 				    struct mlx5_ib_qp *qp,
 				    struct mlx5_ib_sq *sq, u32 tdn,
@@ -1897,33 +1888,14 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	return err;
 }
 
-static void configure_responder_scat_cqe(struct ib_qp_init_attr *init_attr,
-					 void *qpc)
-{
-	int rcqe_sz;
-
-	if (init_attr->qp_type == MLX5_IB_QPT_DCI)
-		return;
-
-	rcqe_sz = mlx5_ib_get_cqe_size(init_attr->recv_cq);
-
-	MLX5_SET(qpc, qpc, cs_res,
-		 rcqe_sz == 128 ? MLX5_RES_SCAT_DATA64_CQE :
-				  MLX5_RES_SCAT_DATA32_CQE);
-}
-
 static void configure_requester_scat_cqe(struct mlx5_ib_dev *dev,
 					 struct ib_qp_init_attr *init_attr,
 					 struct mlx5_ib_create_qp *ucmd,
 					 void *qpc)
 {
-	enum ib_qp_type qpt = init_attr->qp_type;
 	int scqe_sz;
 	bool allow_scat_cqe = false;
 
-	if (qpt == IB_QPT_UC || qpt == IB_QPT_UD)
-		return;
-
 	if (ucmd)
 		allow_scat_cqe = ucmd->flags & MLX5_QP_FLAG_ALLOW_SCATTER_CQE;
 
@@ -2018,7 +1990,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
 
-	mlx5_st = to_mlx5_st(init_attr->qp_type);
+	mlx5_st = to_mlx5_st((init_attr->qp_type != IB_QPT_DRIVER) ?
+				     init_attr->qp_type :
+				     qp->qp_sub_type);
 	if (mlx5_st < 0)
 		return -EINVAL;
 
@@ -2240,12 +2214,20 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		MLX5_SET(qpc, qpc, cd_slave_receive, 1);
 	if (qp->flags & MLX5_IB_QP_PACKET_BASED_CREDIT)
 		MLX5_SET(qpc, qpc, req_e2e_credit_mode, 1);
-	if (qp->scat_cqe && is_connected(init_attr->qp_type)) {
-		configure_responder_scat_cqe(init_attr, qpc);
+	if (qp->scat_cqe && (init_attr->qp_type == IB_QPT_RC ||
+			     init_attr->qp_type == IB_QPT_UC)) {
+		int rcqe_sz = rcqe_sz =
+			mlx5_ib_get_cqe_size(init_attr->recv_cq);
+
+		MLX5_SET(qpc, qpc, cs_res,
+			 rcqe_sz == 128 ? MLX5_RES_SCAT_DATA64_CQE :
+					  MLX5_RES_SCAT_DATA32_CQE);
+	}
+	if (qp->scat_cqe && (qp->qp_sub_type == MLX5_IB_QPT_DCI ||
+			     init_attr->qp_type == IB_QPT_RC))
 		configure_requester_scat_cqe(dev, init_attr,
 					     udata ? &ucmd : NULL,
 					     qpc);
-	}
 
 	if (qp->rq.wqe_cnt) {
 		MLX5_SET(qpc, qpc, log_rq_stride, qp->rq.wqe_shift - 4);
-- 
2.25.3

[PATCH rdma-next v1 10/36] RDMA/mlx5: Move DRIVER QP flags check into separate function

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Perform validation of DRIVER QP in relevant function.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 91 ++++++++++++++++-----------------
 1 file changed, 43 insertions(+), 48 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 0b2090bcb8e8..5e4c73c4a7b4 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2570,36 +2570,6 @@ static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	return 0;
 }
 
-static int set_mlx_qp_type(struct mlx5_ib_dev *dev,
-			   struct ib_qp_init_attr *init_attr,
-			   struct mlx5_ib_create_qp *ucmd,
-			   struct ib_udata *udata)
-{
-	enum { MLX_QP_FLAGS = MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI };
-	int err;
-
-	if (udata->inlen < sizeof(*ucmd)) {
-		mlx5_ib_dbg(dev, "create_qp user command is smaller than expected\n");
-		return -EINVAL;
-	}
-	err = ib_copy_from_udata(ucmd, udata, sizeof(*ucmd));
-	if (err)
-		return err;
-
-	if ((ucmd->flags & MLX_QP_FLAGS) == MLX5_QP_FLAG_TYPE_DCI) {
-		init_attr->qp_type = MLX5_IB_QPT_DCI;
-	} else {
-		if ((ucmd->flags & MLX_QP_FLAGS) == MLX5_QP_FLAG_TYPE_DCT) {
-			init_attr->qp_type = MLX5_IB_QPT_DCT;
-		} else {
-			mlx5_ib_dbg(dev, "Invalid QP flags\n");
-			return -EINVAL;
-		}
-	}
-
-	return 0;
-}
-
 static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
 {
 	if (attr->qp_type == IB_QPT_DRIVER && !MLX5_CAP_GEN(dev->mdev, dct))
@@ -2691,6 +2661,24 @@ static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 }
 
+static int process_vendor_flags(struct mlx5_ib_qp *qp,
+				struct ib_qp_init_attr *attr,
+				struct mlx5_ib_create_qp *ucmd)
+{
+	switch (ucmd->flags & (MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI)) {
+	case MLX5_QP_FLAG_TYPE_DCI:
+		qp->qp_sub_type = MLX5_IB_QPT_DCI;
+		break;
+	case MLX5_QP_FLAG_TYPE_DCT:
+		qp->qp_sub_type = MLX5_IB_QPT_DCT;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 			    struct ib_qp_init_attr *attr,
 			    struct mlx5_ib_create_qp *ucmd,
@@ -2707,6 +2695,9 @@ static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		ret = create_dct(pd, qp, attr, ucmd, udata);
 		break;
 	case MLX5_IB_QPT_DCI:
+		if (attr->cap.max_recv_wr || attr->cap.max_recv_sge)
+			goto out;
+
 		ret = create_qp_common(mdev, pd, attr, udata, qp);
 		break;
 	default:
@@ -2716,8 +2707,16 @@ static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 out:	return ret;
 }
 
+static size_t process_udata_size(struct ib_qp_init_attr *attr,
+				 struct ib_udata *udata)
+{
+	size_t ucmd = sizeof(struct mlx5_ib_create_qp);
+
+	return (udata->inlen < ucmd) ? 0 : ucmd;
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
-				struct ib_qp_init_attr *verbs_init_attr,
+				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
 	struct mlx5_ib_create_qp ucmd = {};
@@ -2725,8 +2724,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	struct mlx5_ib_qp *qp;
 	u16 xrcdn = 0;
 	int err;
-	struct ib_qp_init_attr mlx_init_attr;
-	struct ib_qp_init_attr *init_attr = verbs_init_attr;
 
 	dev = pd ? to_mdev(pd->device) :
 		   to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
@@ -2745,28 +2742,26 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (init_attr->qp_type == IB_QPT_GSI)
 		return mlx5_ib_gsi_create_qp(pd, init_attr);
 
+	if (udata && init_attr->qp_type == IB_QPT_DRIVER) {
+		size_t inlen =
+			process_udata_size(init_attr, udata);
+
+		if (!inlen)
+			return ERR_PTR(-EINVAL);
+
+		err = ib_copy_from_udata(&ucmd, udata, inlen);
+		if (err)
+			return ERR_PTR(err);
+	}
+
 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
 	if (!qp)
 		return ERR_PTR(-ENOMEM);
 
 	if (init_attr->qp_type == IB_QPT_DRIVER) {
-		init_attr = &mlx_init_attr;
-		memcpy(init_attr, verbs_init_attr, sizeof(*verbs_init_attr));
-		err = set_mlx_qp_type(dev, init_attr, &ucmd, udata);
+		err = process_vendor_flags(qp, init_attr, &ucmd);
 		if (err)
 			goto free_qp;
-
-		if (init_attr->qp_type == MLX5_IB_QPT_DCI) {
-			if (init_attr->cap.max_recv_wr ||
-			    init_attr->cap.max_recv_sge) {
-				mlx5_ib_dbg(dev, "DCI QP requires zero size receive queue\n");
-				err = -EINVAL;
-				goto free_qp;
-			}
-			qp->qp_sub_type = MLX5_IB_QPT_DCI;
-		} else {
-			qp->qp_sub_type = MLX5_IB_QPT_DCT;
-		}
 	}
 
 	if (init_attr->qp_type == IB_QPT_XRC_TGT)
-- 
2.25.3

[PATCH rdma-next v1 11/36] RDMA/mlx5: Remove second copy from user for non RSS RAW QPs

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Change the common code to use already copied user command buffer.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 56 ++++++++++++++++-----------------
 1 file changed, 27 insertions(+), 29 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 5e4c73c4a7b4..91a2c9994b59 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1967,6 +1967,7 @@ static inline bool check_flags_mask(uint64_t input, uint64_t supported)
 
 static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
+			    struct mlx5_ib_create_qp *ucmd,
 			    struct ib_udata *udata, struct mlx5_ib_qp *qp)
 {
 	struct mlx5_ib_resources *devr = &dev->devr;
@@ -1979,7 +1980,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	struct mlx5_ib_cq *recv_cq;
 	unsigned long flags;
 	u32 uidx = MLX5_IB_DEFAULT_UIDX;
-	struct mlx5_ib_create_qp ucmd;
 	struct mlx5_ib_qp_base *base;
 	int mlx5_st;
 	void *qpc;
@@ -2056,12 +2056,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	}
 
 	if (udata) {
-		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
-			mlx5_ib_dbg(dev, "copy failed\n");
-			return -EFAULT;
-		}
-
-		if (!check_flags_mask(ucmd.flags,
+		if (!check_flags_mask(ucmd->flags,
 				      MLX5_QP_FLAG_ALLOW_SCATTER_CQE |
 				      MLX5_QP_FLAG_BFREG_INDEX |
 				      MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE |
@@ -2075,14 +2070,15 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 				      MLX5_QP_FLAG_TYPE_DCT))
 			return -EINVAL;
 
-		err = get_qp_user_index(ucontext, &ucmd, udata->inlen, &uidx);
+		err = get_qp_user_index(ucontext, ucmd, udata->inlen, &uidx);
 		if (err)
 			return err;
 
-		qp->wq_sig = !!(ucmd.flags & MLX5_QP_FLAG_SIGNATURE);
+		qp->wq_sig = !!(ucmd->flags & MLX5_QP_FLAG_SIGNATURE);
 		if (MLX5_CAP_GEN(dev->mdev, sctr_data_cqe))
-			qp->scat_cqe = !!(ucmd.flags & MLX5_QP_FLAG_SCATTER_CQE);
-		if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
+			qp->scat_cqe =
+				!!(ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE);
+		if (ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
 			if (init_attr->qp_type != IB_QPT_RAW_PACKET ||
 			    !tunnel_offload_supported(mdev)) {
 				mlx5_ib_dbg(dev, "Tunnel offload isn't supported\n");
@@ -2091,7 +2087,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			qp->flags_en |= MLX5_QP_FLAG_TUNNEL_OFFLOADS;
 		}
 
-		if (ucmd.flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC) {
+		if (ucmd->flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC) {
 			if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
 				mlx5_ib_dbg(dev, "Self-LB UC isn't supported\n");
 				return -EOPNOTSUPP;
@@ -2099,7 +2095,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC;
 		}
 
-		if (ucmd.flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC) {
+		if (ucmd->flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC) {
 			if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
 				mlx5_ib_dbg(dev, "Self-LB UM isn't supported\n");
 				return -EOPNOTSUPP;
@@ -2107,7 +2103,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC;
 		}
 
-		if (ucmd.flags & MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE) {
+		if (ucmd->flags & MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE) {
 			if (init_attr->qp_type != IB_QPT_RC ||
 				!MLX5_CAP_GEN(dev->mdev, qp_packet_based)) {
 				mlx5_ib_dbg(dev, "packet based credit mode isn't supported\n");
@@ -2138,8 +2134,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	       &qp->trans_qp.base;
 
 	qp->has_rq = qp_has_rq(init_attr);
-	err = set_rq_size(dev, &init_attr->cap, qp->has_rq,
-			  qp, udata ? &ucmd : NULL);
+	err = set_rq_size(dev, &init_attr->cap, qp->has_rq, qp, ucmd);
 	if (err) {
 		mlx5_ib_dbg(dev, "err %d\n", err);
 		return err;
@@ -2149,15 +2144,16 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		if (udata) {
 			__u32 max_wqes =
 				1 << MLX5_CAP_GEN(mdev, log_max_qp_sz);
-			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, "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 > max_wqes) {
+			if (ucmd->sq_wqe_count > max_wqes) {
 				mlx5_ib_dbg(dev, "requested sq_wqe_count (%d) > max allowed (%d)\n",
-					    ucmd.sq_wqe_count, max_wqes);
+					    ucmd->sq_wqe_count, max_wqes);
 				return -EINVAL;
 			}
 			if (init_attr->create_flags &
@@ -2225,9 +2221,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	}
 	if (qp->scat_cqe && (qp->qp_sub_type == MLX5_IB_QPT_DCI ||
 			     init_attr->qp_type == IB_QPT_RC))
-		configure_requester_scat_cqe(dev, init_attr,
-					     udata ? &ucmd : NULL,
-					     qpc);
+		configure_requester_scat_cqe(dev, init_attr, ucmd, qpc);
 
 	if (qp->rq.wqe_cnt) {
 		MLX5_SET(qpc, qpc, log_rq_stride, qp->rq.wqe_shift - 4);
@@ -2308,7 +2302,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 
 	if (init_attr->qp_type == IB_QPT_RAW_PACKET ||
 	    qp->flags & MLX5_IB_QP_UNDERLAY) {
-		qp->raw_packet_qp.sq.ubuffer.buf_addr = ucmd.sq_buf_addr;
+		qp->raw_packet_qp.sq.ubuffer.buf_addr = ucmd->sq_buf_addr;
 		raw_packet_qp_copy_info(qp, &qp->raw_packet_qp);
 		err = create_raw_packet_qp(dev, qp, in, inlen, pd, udata,
 					   &resp);
@@ -2698,7 +2692,7 @@ static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		if (attr->cap.max_recv_wr || attr->cap.max_recv_sge)
 			goto out;
 
-		ret = create_qp_common(mdev, pd, attr, udata, qp);
+		ret = create_qp_common(mdev, pd, attr, ucmd, udata, qp);
 		break;
 	default:
 		return -EINVAL;
@@ -2712,7 +2706,10 @@ static size_t process_udata_size(struct ib_qp_init_attr *attr,
 {
 	size_t ucmd = sizeof(struct mlx5_ib_create_qp);
 
-	return (udata->inlen < ucmd) ? 0 : ucmd;
+	if (attr->qp_type == IB_QPT_DRIVER)
+		return (udata->inlen < ucmd) ? 0 : ucmd;
+
+	return ucmd;
 }
 
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
@@ -2742,7 +2739,7 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (init_attr->qp_type == IB_QPT_GSI)
 		return mlx5_ib_gsi_create_qp(pd, init_attr);
 
-	if (udata && init_attr->qp_type == IB_QPT_DRIVER) {
+	if (udata && !init_attr->rwq_ind_tbl) {
 		size_t inlen =
 			process_udata_size(init_attr, udata);
 
@@ -2772,7 +2769,8 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		err = create_driver_qp(pd, qp, init_attr, &ucmd, udata);
 		break;
 	default:
-		err = create_qp_common(dev, pd, init_attr, udata, qp);
+		err = create_qp_common(dev, pd, init_attr,
+				       (udata) ? &ucmd : NULL, udata, qp);
 	}
 	if (err) {
 		mlx5_ib_dbg(dev, "create_qp_common failed\n");
-- 
2.25.3

[PATCH rdma-next v1 12/36] RDMA/mlx5: Initial separation of RAW_PACKET QP from common flow

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Create initial function for IB_QPT_RAW_PACKET flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 91a2c9994b59..a514b4eca06e 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1634,13 +1634,13 @@ static void destroy_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *q
 			     to_mpd(qp->ibqp.pd)->uid);
 }
 
-static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
-				 struct ib_pd *pd,
+static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 				 struct ib_qp_init_attr *init_attr,
 				 struct ib_udata *udata)
 {
 	struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	struct mlx5_ib_create_qp_resp resp = {};
 	int inlen;
 	int outlen;
@@ -1996,9 +1996,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (mlx5_st < 0)
 		return -EINVAL;
 
-	if (init_attr->rwq_ind_tbl)
-		return create_rss_raw_qp_tir(dev, qp, pd, init_attr, udata);
-
 	if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
 		if (!MLX5_CAP_GEN(mdev, block_lb_mc)) {
 			mlx5_ib_dbg(dev, "block multicast loopback isn't supported\n");
@@ -2712,6 +2709,18 @@ static size_t process_udata_size(struct ib_qp_init_attr *attr,
 	return ucmd;
 }
 
+static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
+			 struct ib_qp_init_attr *attr,
+			 struct mlx5_ib_create_qp *ucmd, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+
+	if (attr->rwq_ind_tbl)
+		return create_rss_raw_qp_tir(pd, qp, attr, udata);
+
+	return create_qp_common(dev, pd, attr, ucmd, udata, qp);
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
@@ -2768,6 +2777,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	case IB_QPT_DRIVER:
 		err = create_driver_qp(pd, qp, init_attr, &ucmd, udata);
 		break;
+	case IB_QPT_RAW_PACKET:
+		err = create_raw_qp(pd, qp, init_attr, &ucmd, udata);
+		break;
 	default:
 		err = create_qp_common(dev, pd, init_attr,
 				       (udata) ? &ucmd : NULL, udata, qp);
-- 
2.25.3

[PATCH rdma-next v1 13/36] RDMA/mlx5: Delete create QP flags obfuscation

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

There is no point in redefinition of stable and exposed to users create
flags. Their values won't be changed and it is equal to used by the
mlx5. Delete the mlx5 definitions and use IB/core fields.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/devx.c    |  2 +-
 drivers/infiniband/hw/mlx5/flow.c    |  2 +-
 drivers/infiniband/hw/mlx5/main.c    |  6 +--
 drivers/infiniband/hw/mlx5/mlx5_ib.h | 21 +-------
 drivers/infiniband/hw/mlx5/qp.c      | 80 ++++++++++++++--------------
 5 files changed, 47 insertions(+), 64 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index 35b98c2d64d5..1d7feed6d3cb 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -615,7 +615,7 @@ static bool devx_is_valid_obj_id(struct uverbs_attr_bundle *attrs,
 		enum ib_qp_type	qp_type = qp->ibqp.qp_type;
 
 		if (qp_type == IB_QPT_RAW_PACKET ||
-		    (qp->flags & MLX5_IB_QP_UNDERLAY)) {
+		    (qp->flags & IB_QP_CREATE_SOURCE_QPN)) {
 			struct mlx5_ib_raw_packet_qp *raw_packet_qp =
 							 &qp->raw_packet_qp;
 			struct mlx5_ib_rq *rq = &raw_packet_qp->rq;
diff --git a/drivers/infiniband/hw/mlx5/flow.c b/drivers/infiniband/hw/mlx5/flow.c
index 862b7bf3e646..1b44f691b571 100644
--- a/drivers/infiniband/hw/mlx5/flow.c
+++ b/drivers/infiniband/hw/mlx5/flow.c
@@ -142,7 +142,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_CREATE_FLOW)(
 			return -EINVAL;
 
 		mqp = to_mqp(qp);
-		if (mqp->flags & MLX5_IB_QP_RSS)
+		if (mqp->is_rss)
 			dest_id = mqp->rss_qp.tirn;
 		else
 			dest_id = mqp->raw_packet_qp.rq.tirn;
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index f10675213115..f8d752ca92f7 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -3967,7 +3967,7 @@ static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
 		dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
 	} else {
 		dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
-		if (mqp->flags & MLX5_IB_QP_RSS)
+		if (mqp->is_rss)
 			dst->tir_num = mqp->rss_qp.tirn;
 		else
 			dst->tir_num = mqp->raw_packet_qp.rq.tirn;
@@ -3978,7 +3978,7 @@ static struct ib_flow *mlx5_ib_create_flow(struct ib_qp *qp,
 			handler = create_dont_trap_rule(dev, ft_prio,
 							flow_attr, dst);
 		} else {
-			underlay_qpn = (mqp->flags & MLX5_IB_QP_UNDERLAY) ?
+			underlay_qpn = (mqp->flags & IB_QP_CREATE_SOURCE_QPN) ?
 					mqp->underlay_qpn : 0;
 			handler = _create_flow_rule(dev, ft_prio, flow_attr,
 						    dst, underlay_qpn, ucmd);
@@ -4447,7 +4447,7 @@ static int mlx5_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	uid = ibqp->pd ?
 		to_mpd(ibqp->pd)->uid : 0;
 
-	if (mqp->flags & MLX5_IB_QP_UNDERLAY) {
+	if (mqp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		mlx5_ib_dbg(dev, "Attaching a multi cast group to underlay QP is not supported\n");
 		return -EOPNOTSUPP;
 	}
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index aaabb8a98eed..b144660a47e1 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -450,7 +450,8 @@ struct mlx5_ib_qp {
 	int			scat_cqe;
 	int			max_inline_data;
 	struct mlx5_bf	        bf;
-	int			has_rq;
+	u8			has_rq:1;
+	u8			is_rss:1;
 
 	/* only for user space QPs. For kernel
 	 * we have it from the bf object
@@ -481,24 +482,6 @@ struct mlx5_ib_cq_buf {
 	int			nent;
 };
 
-enum mlx5_ib_qp_flags {
-	MLX5_IB_QP_LSO                          = IB_QP_CREATE_IPOIB_UD_LSO,
-	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
-	MLX5_IB_QP_CROSS_CHANNEL            = IB_QP_CREATE_CROSS_CHANNEL,
-	MLX5_IB_QP_MANAGED_SEND             = IB_QP_CREATE_MANAGED_SEND,
-	MLX5_IB_QP_MANAGED_RECV             = IB_QP_CREATE_MANAGED_RECV,
-	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 5,
-	/* QP uses 1 as its source QP number */
-	MLX5_IB_QP_SQPN_QP1			= 1 << 6,
-	MLX5_IB_QP_CAP_SCATTER_FCS		= 1 << 7,
-	MLX5_IB_QP_RSS				= 1 << 8,
-	MLX5_IB_QP_CVLAN_STRIPPING		= 1 << 9,
-	MLX5_IB_QP_UNDERLAY			= 1 << 10,
-	MLX5_IB_QP_PCI_WRITE_END_PADDING	= 1 << 11,
-	MLX5_IB_QP_TUNNEL_OFFLOAD		= 1 << 12,
-	MLX5_IB_QP_PACKET_BASED_CREDIT		= 1 << 13,
-};
-
 struct mlx5_umr_wr {
 	struct ib_send_wr		wr;
 	u64				virt_addr;
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index a514b4eca06e..cdbb837138c9 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -596,7 +596,7 @@ static int set_user_buf_size(struct mlx5_ib_dev *dev,
 	}
 
 	if (attr->qp_type == IB_QPT_RAW_PACKET ||
-	    qp->flags & MLX5_IB_QP_UNDERLAY) {
+	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		base->ubuffer.buf_size = qp->rq.wqe_cnt << qp->rq.wqe_shift;
 		qp->raw_packet_qp.sq.ubuffer.buf_size = qp->sq.wqe_cnt << 6;
 	} else {
@@ -951,7 +951,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		bfregn = MLX5_IB_INVALID_BFREG;
 		break;
 	case 0:
-		if (qp->flags & MLX5_IB_QP_CROSS_CHANNEL)
+		if (qp->flags & IB_QP_CREATE_CROSS_CHANNEL)
 			return -EINVAL;
 		bfregn = alloc_bfreg(dev, &context->bfregi);
 		if (bfregn < 0)
@@ -1169,7 +1169,7 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev,
 
 	if (init_attr->create_flags & MLX5_IB_QP_CREATE_SQPN_QP1) {
 		MLX5_SET(qpc, qpc, deth_sqpn, 1);
-		qp->flags |= MLX5_IB_QP_SQPN_QP1;
+		qp->flags |= MLX5_IB_QP_CREATE_SQPN_QP1;
 	}
 
 	mlx5_fill_page_frag_array(&qp->buf,
@@ -1251,7 +1251,7 @@ static int create_raw_packet_qp_tis(struct mlx5_ib_dev *dev,
 
 	MLX5_SET(create_tis_in, in, uid, to_mpd(pd)->uid);
 	MLX5_SET(tisc, tisc, transport_domain, tdn);
-	if (qp->flags & MLX5_IB_QP_UNDERLAY)
+	if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
 		MLX5_SET(tisc, tisc, underlay_qpn, qp->underlay_qpn);
 
 	return mlx5_core_create_tis(dev->mdev, in, &sq->tisn);
@@ -1400,7 +1400,7 @@ static int create_raw_packet_qp_rq(struct mlx5_ib_dev *dev,
 	MLX5_SET(rqc, rqc, user_index, MLX5_GET(qpc, qpc, user_index));
 	MLX5_SET(rqc, rqc, cqn, MLX5_GET(qpc, qpc, cqn_rcv));
 
-	if (mqp->flags & MLX5_IB_QP_CAP_SCATTER_FCS)
+	if (mqp->flags & IB_QP_CREATE_SCATTER_FCS)
 		MLX5_SET(rqc, rqc, scatter_fcs, 1);
 
 	wq = MLX5_ADDR_OF(rqc, rqc, wq);
@@ -1538,9 +1538,9 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	if (qp->rq.wqe_cnt) {
 		rq->base.container_mibqp = qp;
 
-		if (qp->flags & MLX5_IB_QP_CVLAN_STRIPPING)
+		if (qp->flags & IB_QP_CREATE_CVLAN_STRIPPING)
 			rq->flags |= MLX5_IB_RQ_CVLAN_STRIPPING;
-		if (qp->flags & MLX5_IB_QP_PCI_WRITE_END_PADDING)
+		if (qp->flags & IB_QP_CREATE_PCI_WRITE_END_PADDING)
 			rq->flags |= MLX5_IB_RQ_PCI_WRITE_END_PADDING;
 		err = create_raw_packet_qp_rq(dev, rq, in, inlen, pd);
 		if (err)
@@ -1878,7 +1878,7 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	kvfree(in);
 	/* qpn is reserved for that QP */
 	qp->trans_qp.base.mqp.qpn = 0;
-	qp->flags |= MLX5_IB_QP_RSS;
+	qp->is_rss = true;
 	return 0;
 
 err_copy:
@@ -2001,7 +2001,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			mlx5_ib_dbg(dev, "block multicast loopback isn't supported\n");
 			return -EINVAL;
 		} else {
-			qp->flags |= MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK;
+			qp->flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
 		}
 	}
 
@@ -2014,11 +2014,11 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			return -EINVAL;
 		}
 		if (init_attr->create_flags & IB_QP_CREATE_CROSS_CHANNEL)
-			qp->flags |= MLX5_IB_QP_CROSS_CHANNEL;
+			qp->flags |= IB_QP_CREATE_CROSS_CHANNEL;
 		if (init_attr->create_flags & IB_QP_CREATE_MANAGED_SEND)
-			qp->flags |= MLX5_IB_QP_MANAGED_SEND;
+			qp->flags |= IB_QP_CREATE_MANAGED_SEND;
 		if (init_attr->create_flags & IB_QP_CREATE_MANAGED_RECV)
-			qp->flags |= MLX5_IB_QP_MANAGED_RECV;
+			qp->flags |= IB_QP_CREATE_MANAGED_RECV;
 	}
 
 	if (init_attr->qp_type == IB_QPT_UD &&
@@ -2038,7 +2038,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			mlx5_ib_dbg(dev, "Scatter FCS isn't supported\n");
 			return -EOPNOTSUPP;
 		}
-		qp->flags |= MLX5_IB_QP_CAP_SCATTER_FCS;
+		qp->flags |= IB_QP_CREATE_SCATTER_FCS;
 	}
 
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
@@ -2049,7 +2049,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		      MLX5_CAP_ETH(dev->mdev, vlan_cap)) ||
 		    (init_attr->qp_type != IB_QPT_RAW_PACKET))
 			return -EOPNOTSUPP;
-		qp->flags |= MLX5_IB_QP_CVLAN_STRIPPING;
+		qp->flags |= IB_QP_CREATE_CVLAN_STRIPPING;
 	}
 
 	if (udata) {
@@ -2106,7 +2106,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 				mlx5_ib_dbg(dev, "packet based credit mode isn't supported\n");
 				return -EOPNOTSUPP;
 			}
-			qp->flags |= MLX5_IB_QP_PACKET_BASED_CREDIT;
+			qp->flags_en |= MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE;
 		}
 
 		if (init_attr->create_flags & IB_QP_CREATE_SOURCE_QPN) {
@@ -2118,7 +2118,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 				return -EOPNOTSUPP;
 			}
 
-			qp->flags |= MLX5_IB_QP_UNDERLAY;
+			qp->flags |= IB_QP_CREATE_SOURCE_QPN;
 			qp->underlay_qpn = init_attr->source_qpn;
 		}
 	} else {
@@ -2126,7 +2126,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	}
 
 	base = (init_attr->qp_type == IB_QPT_RAW_PACKET ||
-		qp->flags & MLX5_IB_QP_UNDERLAY) ?
+		qp->flags & IB_QP_CREATE_SOURCE_QPN) ?
 	       &qp->raw_packet_qp.rq.base :
 	       &qp->trans_qp.base;
 
@@ -2196,16 +2196,16 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (qp->wq_sig)
 		MLX5_SET(qpc, qpc, wq_signature, 1);
 
-	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
 		MLX5_SET(qpc, qpc, block_lb_mc, 1);
 
-	if (qp->flags & MLX5_IB_QP_CROSS_CHANNEL)
+	if (qp->flags & IB_QP_CREATE_CROSS_CHANNEL)
 		MLX5_SET(qpc, qpc, cd_master, 1);
-	if (qp->flags & MLX5_IB_QP_MANAGED_SEND)
+	if (qp->flags & IB_QP_CREATE_MANAGED_SEND)
 		MLX5_SET(qpc, qpc, cd_slave_send, 1);
-	if (qp->flags & MLX5_IB_QP_MANAGED_RECV)
+	if (qp->flags & IB_QP_CREATE_MANAGED_RECV)
 		MLX5_SET(qpc, qpc, cd_slave_receive, 1);
-	if (qp->flags & MLX5_IB_QP_PACKET_BASED_CREDIT)
+	if (qp->flags_en & MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE)
 		MLX5_SET(qpc, qpc, req_e2e_credit_mode, 1);
 	if (qp->scat_cqe && (init_attr->qp_type == IB_QPT_RC ||
 			     init_attr->qp_type == IB_QPT_UC)) {
@@ -2276,7 +2276,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (init_attr->qp_type == IB_QPT_UD &&
 	    (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)) {
 		MLX5_SET(qpc, qpc, ulp_stateless_offload_mode, 1);
-		qp->flags |= MLX5_IB_QP_LSO;
+		qp->flags |= IB_QP_CREATE_IPOIB_UD_LSO;
 	}
 
 	if (init_attr->create_flags & IB_QP_CREATE_PCI_WRITE_END_PADDING) {
@@ -2288,7 +2288,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			MLX5_SET(qpc, qpc, end_padding_mode,
 				 MLX5_WQ_END_PAD_MODE_ALIGN);
 		} else {
-			qp->flags |= MLX5_IB_QP_PCI_WRITE_END_PADDING;
+			qp->flags |= IB_QP_CREATE_PCI_WRITE_END_PADDING;
 		}
 	}
 
@@ -2298,7 +2298,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	}
 
 	if (init_attr->qp_type == IB_QPT_RAW_PACKET ||
-	    qp->flags & MLX5_IB_QP_UNDERLAY) {
+	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		qp->raw_packet_qp.sq.ubuffer.buf_addr = ucmd->sq_buf_addr;
 		raw_packet_qp_copy_info(qp, &qp->raw_packet_qp);
 		err = create_raw_packet_qp(dev, qp, in, inlen, pd, udata,
@@ -2463,13 +2463,13 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	}
 
 	base = (qp->ibqp.qp_type == IB_QPT_RAW_PACKET ||
-		qp->flags & MLX5_IB_QP_UNDERLAY) ?
+		qp->flags & IB_QP_CREATE_SOURCE_QPN) ?
 	       &qp->raw_packet_qp.rq.base :
 	       &qp->trans_qp.base;
 
 	if (qp->state != IB_QPS_RESET) {
 		if (qp->ibqp.qp_type != IB_QPT_RAW_PACKET &&
-		    !(qp->flags & MLX5_IB_QP_UNDERLAY)) {
+		    !(qp->flags & IB_QP_CREATE_SOURCE_QPN)) {
 			err = mlx5_core_qp_modify(dev, MLX5_CMD_OP_2RST_QP, 0,
 						  NULL, &base->mqp);
 		} else {
@@ -2508,7 +2508,7 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
 
 	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET ||
-	    qp->flags & MLX5_IB_QP_UNDERLAY) {
+	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		destroy_raw_packet_qp(dev, qp);
 	} else {
 		err = mlx5_core_destroy_qp(dev, &base->mqp);
@@ -3550,7 +3550,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 	if ((cur_state == IB_QPS_RESET) && (new_state == IB_QPS_INIT)) {
 		if ((ibqp->qp_type == IB_QPT_RC) ||
 		    (ibqp->qp_type == IB_QPT_UD &&
-		     !(qp->flags & MLX5_IB_QP_SQPN_QP1)) ||
+		     !(qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1)) ||
 		    (ibqp->qp_type == IB_QPT_UC) ||
 		    (ibqp->qp_type == IB_QPT_RAW_PACKET) ||
 		    (ibqp->qp_type == IB_QPT_XRC_INI) ||
@@ -3567,7 +3567,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 	if (is_sqp(ibqp->qp_type)) {
 		context->mtu_msgmax = (IB_MTU_256 << 5) | 8;
 	} else if ((ibqp->qp_type == IB_QPT_UD &&
-		    !(qp->flags & MLX5_IB_QP_UNDERLAY)) ||
+		    !(qp->flags & IB_QP_CREATE_SOURCE_QPN)) ||
 		   ibqp->qp_type == MLX5_IB_QPT_REG_UMR) {
 		context->mtu_msgmax = (IB_MTU_4096 << 5) | 12;
 	} else if (attr_mask & IB_QP_PATH_MTU) {
@@ -3672,7 +3672,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 			       qp->port) - 1;
 
 		/* Underlay port should be used - index 0 function per port */
-		if (qp->flags & MLX5_IB_QP_UNDERLAY)
+		if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
 			port_num = 0;
 
 		if (ibqp->counter)
@@ -3686,7 +3686,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 	if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
 		context->sq_crq_size |= cpu_to_be16(1 << 4);
 
-	if (qp->flags & MLX5_IB_QP_SQPN_QP1)
+	if (qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1)
 		context->deth_sqpn = cpu_to_be32(1);
 
 	mlx5_cur = to_mlx5_state(cur_state);
@@ -3703,7 +3703,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 	optpar &= opt_mask[mlx5_cur][mlx5_new][mlx5_st];
 
 	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET ||
-	    qp->flags & MLX5_IB_QP_UNDERLAY) {
+	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		struct mlx5_modify_raw_qp_param raw_qp_param = {};
 
 		raw_qp_param.operation = op;
@@ -3999,7 +3999,7 @@ int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		port = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
 	}
 
-	if (qp->flags & MLX5_IB_QP_UNDERLAY) {
+	if (qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		if (attr_mask & ~(IB_QP_STATE | IB_QP_CUR_STATE)) {
 			mlx5_ib_dbg(dev, "invalid attr_mask 0x%x when underlay QP is used\n",
 				    attr_mask);
@@ -5831,7 +5831,7 @@ int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 	mutex_lock(&qp->mutex);
 
 	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET ||
-	    qp->flags & MLX5_IB_QP_UNDERLAY) {
+	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		err = query_raw_packet_qp_state(dev, qp, &raw_packet_qp_state);
 		if (err)
 			goto out;
@@ -5866,16 +5866,16 @@ int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 	qp_init_attr->cap	     = qp_attr->cap;
 
 	qp_init_attr->create_flags = 0;
-	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
 		qp_init_attr->create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
 
-	if (qp->flags & MLX5_IB_QP_CROSS_CHANNEL)
+	if (qp->flags & IB_QP_CREATE_CROSS_CHANNEL)
 		qp_init_attr->create_flags |= IB_QP_CREATE_CROSS_CHANNEL;
-	if (qp->flags & MLX5_IB_QP_MANAGED_SEND)
+	if (qp->flags & IB_QP_CREATE_MANAGED_SEND)
 		qp_init_attr->create_flags |= IB_QP_CREATE_MANAGED_SEND;
-	if (qp->flags & MLX5_IB_QP_MANAGED_RECV)
+	if (qp->flags & IB_QP_CREATE_MANAGED_RECV)
 		qp_init_attr->create_flags |= IB_QP_CREATE_MANAGED_RECV;
-	if (qp->flags & MLX5_IB_QP_SQPN_QP1)
+	if (qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1)
 		qp_init_attr->create_flags |= MLX5_IB_QP_CREATE_SQPN_QP1;
 
 	qp_init_attr->sq_sig_type = qp->sq_signal_bits & MLX5_WQE_CTRL_CQ_UPDATE ?
-- 
2.25.3

[PATCH rdma-next v1 14/36] RDMA/mlx5: Process create QP flags in one place

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

create_flags is checked in too many places and scattered across all
the code, consolidate all the checks inside one function, so we will
be easily see the flow. As part of such change, delete unreachable code,
because IB/core is responsible sanitize the input.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 200 ++++++++++++++++----------------
 1 file changed, 101 insertions(+), 99 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index cdbb837138c9..02eb03484b91 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1097,17 +1097,9 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev,
 	void *qpc;
 	int err;
 
-	if (init_attr->create_flags & ~(IB_QP_CREATE_INTEGRITY_EN |
-					IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |
-					IB_QP_CREATE_IPOIB_UD_LSO |
-					IB_QP_CREATE_NETIF_QP |
-					MLX5_IB_QP_CREATE_SQPN_QP1 |
-					MLX5_IB_QP_CREATE_WC_TEST))
-		return -EINVAL;
-
 	if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR)
 		qp->bf.bfreg = &dev->fp_bfreg;
-	else if (init_attr->create_flags & MLX5_IB_QP_CREATE_WC_TEST)
+	else if (qp->flags & MLX5_IB_QP_CREATE_WC_TEST)
 		qp->bf.bfreg = &dev->wc_bfreg;
 	else
 		qp->bf.bfreg = &dev->bfreg;
@@ -1167,10 +1159,8 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev,
 	MLX5_SET(qpc, qpc, fre, 1);
 	MLX5_SET(qpc, qpc, rlky, 1);
 
-	if (init_attr->create_flags & MLX5_IB_QP_CREATE_SQPN_QP1) {
+	if (qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1)
 		MLX5_SET(qpc, qpc, deth_sqpn, 1);
-		qp->flags |= MLX5_IB_QP_CREATE_SQPN_QP1;
-	}
 
 	mlx5_fill_page_frag_array(&qp->buf,
 				  (__be64 *)MLX5_ADDR_OF(create_qp_in,
@@ -1657,7 +1647,7 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	size_t required_cmd_sz;
 	u8 lb_flag = 0;
 
-	if (init_attr->create_flags || init_attr->send_cq)
+	if (init_attr->send_cq)
 		return -EINVAL;
 
 	min_resp_len = offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
@@ -1996,62 +1986,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (mlx5_st < 0)
 		return -EINVAL;
 
-	if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
-		if (!MLX5_CAP_GEN(mdev, block_lb_mc)) {
-			mlx5_ib_dbg(dev, "block multicast loopback isn't supported\n");
-			return -EINVAL;
-		} else {
-			qp->flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
-		}
-	}
-
-	if (init_attr->create_flags &
-			(IB_QP_CREATE_CROSS_CHANNEL |
-			 IB_QP_CREATE_MANAGED_SEND |
-			 IB_QP_CREATE_MANAGED_RECV)) {
-		if (!MLX5_CAP_GEN(mdev, cd)) {
-			mlx5_ib_dbg(dev, "cross-channel isn't supported\n");
-			return -EINVAL;
-		}
-		if (init_attr->create_flags & IB_QP_CREATE_CROSS_CHANNEL)
-			qp->flags |= IB_QP_CREATE_CROSS_CHANNEL;
-		if (init_attr->create_flags & IB_QP_CREATE_MANAGED_SEND)
-			qp->flags |= IB_QP_CREATE_MANAGED_SEND;
-		if (init_attr->create_flags & IB_QP_CREATE_MANAGED_RECV)
-			qp->flags |= IB_QP_CREATE_MANAGED_RECV;
-	}
-
-	if (init_attr->qp_type == IB_QPT_UD &&
-	    (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO))
-		if (!MLX5_CAP_GEN(mdev, ipoib_basic_offloads)) {
-			mlx5_ib_dbg(dev, "ipoib UD lso qp isn't supported\n");
-			return -EOPNOTSUPP;
-		}
-
-	if (init_attr->create_flags & IB_QP_CREATE_SCATTER_FCS) {
-		if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
-			mlx5_ib_dbg(dev, "Scatter FCS is supported only for Raw Packet QPs");
-			return -EOPNOTSUPP;
-		}
-		if (!MLX5_CAP_GEN(dev->mdev, eth_net_offloads) ||
-		    !MLX5_CAP_ETH(dev->mdev, scatter_fcs)) {
-			mlx5_ib_dbg(dev, "Scatter FCS isn't supported\n");
-			return -EOPNOTSUPP;
-		}
-		qp->flags |= IB_QP_CREATE_SCATTER_FCS;
-	}
-
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
 		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
 
-	if (init_attr->create_flags & IB_QP_CREATE_CVLAN_STRIPPING) {
-		if (!(MLX5_CAP_GEN(dev->mdev, eth_net_offloads) &&
-		      MLX5_CAP_ETH(dev->mdev, vlan_cap)) ||
-		    (init_attr->qp_type != IB_QPT_RAW_PACKET))
-			return -EOPNOTSUPP;
-		qp->flags |= IB_QP_CREATE_CVLAN_STRIPPING;
-	}
-
 	if (udata) {
 		if (!check_flags_mask(ucmd->flags,
 				      MLX5_QP_FLAG_ALLOW_SCATTER_CQE |
@@ -2108,23 +2045,13 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			}
 			qp->flags_en |= MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE;
 		}
-
-		if (init_attr->create_flags & IB_QP_CREATE_SOURCE_QPN) {
-			if (init_attr->qp_type != IB_QPT_UD ||
-			    (MLX5_CAP_GEN(dev->mdev, port_type) !=
-			     MLX5_CAP_PORT_TYPE_IB) ||
-			    !mlx5_get_flow_namespace(dev->mdev, MLX5_FLOW_NAMESPACE_BYPASS)) {
-				mlx5_ib_dbg(dev, "Source QP option isn't supported\n");
-				return -EOPNOTSUPP;
-			}
-
-			qp->flags |= IB_QP_CREATE_SOURCE_QPN;
-			qp->underlay_qpn = init_attr->source_qpn;
-		}
 	} else {
 		qp->wq_sig = !!wq_signature;
 	}
 
+	if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
+		qp->underlay_qpn = init_attr->source_qpn;
+
 	base = (init_attr->qp_type == IB_QPT_RAW_PACKET ||
 		qp->flags & IB_QP_CREATE_SOURCE_QPN) ?
 	       &qp->raw_packet_qp.rq.base :
@@ -2153,11 +2080,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 					    ucmd->sq_wqe_count, max_wqes);
 				return -EINVAL;
 			}
-			if (init_attr->create_flags &
-			    MLX5_IB_QP_CREATE_SQPN_QP1) {
-				mlx5_ib_dbg(dev, "user-space is not allowed to create UD QPs spoofing as QP1\n");
-				return -EINVAL;
-			}
 			err = create_user_qp(dev, pd, qp, udata, init_attr, &in,
 					     &resp, &inlen, base);
 			if (err)
@@ -2273,23 +2195,15 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		MLX5_SET(qpc, qpc, user_index, uidx);
 
 	/* we use IB_QP_CREATE_IPOIB_UD_LSO to indicates ipoib qp */
-	if (init_attr->qp_type == IB_QPT_UD &&
-	    (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)) {
+	if (qp->flags & IB_QP_CREATE_IPOIB_UD_LSO)
 		MLX5_SET(qpc, qpc, ulp_stateless_offload_mode, 1);
-		qp->flags |= IB_QP_CREATE_IPOIB_UD_LSO;
-	}
 
-	if (init_attr->create_flags & IB_QP_CREATE_PCI_WRITE_END_PADDING) {
-		if (!MLX5_CAP_GEN(dev->mdev, end_pad)) {
-			mlx5_ib_dbg(dev, "scatter end padding is not supported\n");
-			err = -EOPNOTSUPP;
-			goto err;
-		} else if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
-			MLX5_SET(qpc, qpc, end_padding_mode,
-				 MLX5_WQ_END_PAD_MODE_ALIGN);
-		} else {
-			qp->flags |= IB_QP_CREATE_PCI_WRITE_END_PADDING;
-		}
+	if (qp->flags & IB_QP_CREATE_PCI_WRITE_END_PADDING &&
+	    init_attr->qp_type != IB_QPT_RAW_PACKET) {
+		MLX5_SET(qpc, qpc, end_padding_mode,
+			 MLX5_WQ_END_PAD_MODE_ALIGN);
+		/* Special case to clean flag */
+		qp->flags &= ~IB_QP_CREATE_PCI_WRITE_END_PADDING;
 	}
 
 	if (inlen < 0) {
@@ -2670,6 +2584,91 @@ static int process_vendor_flags(struct mlx5_ib_qp *qp,
 	return 0;
 }
 
+static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
+				bool cond, struct mlx5_ib_qp *qp)
+{
+	if (!(*flags & flag))
+		return;
+
+	if (cond) {
+		qp->flags |= flag;
+		*flags &= ~flag;
+		return;
+	}
+
+	if (flag == MLX5_IB_QP_CREATE_WC_TEST) {
+		/*
+		 * Special case, if condition didn't meet, it won't be error,
+		 * just different in-kernel flow.
+		 */
+		*flags &= ~MLX5_IB_QP_CREATE_WC_TEST;
+		return;
+	}
+	mlx5_ib_dbg(dev, "Verbs create QP flag 0x%X is not supported\n", flag);
+}
+
+static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
+				struct ib_qp_init_attr *attr)
+{
+	enum ib_qp_type qp_type = attr->qp_type;
+	struct mlx5_core_dev *mdev = dev->mdev;
+	int create_flags = attr->create_flags;
+	bool cond;
+
+	if (qp->qp_sub_type == MLX5_IB_QPT_DCT)
+		return (create_flags) ? -EINVAL : 0;
+
+	if (qp_type == IB_QPT_RAW_PACKET && attr->rwq_ind_tbl)
+		return (create_flags) ? -EINVAL : 0;
+
+	process_create_flag(dev, &create_flags,
+			    IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
+			    MLX5_CAP_GEN(mdev, block_lb_mc), qp);
+	process_create_flag(dev, &create_flags, IB_QP_CREATE_CROSS_CHANNEL,
+			    MLX5_CAP_GEN(mdev, cd), qp);
+	process_create_flag(dev, &create_flags, IB_QP_CREATE_MANAGED_SEND,
+			    MLX5_CAP_GEN(mdev, cd), qp);
+	process_create_flag(dev, &create_flags, IB_QP_CREATE_MANAGED_RECV,
+			    MLX5_CAP_GEN(mdev, cd), qp);
+
+	if (qp_type == IB_QPT_UD) {
+		process_create_flag(dev, &create_flags,
+				    IB_QP_CREATE_IPOIB_UD_LSO,
+				    MLX5_CAP_GEN(mdev, ipoib_basic_offloads),
+				    qp);
+		cond = MLX5_CAP_GEN(mdev, port_type) == MLX5_CAP_PORT_TYPE_IB;
+		process_create_flag(dev, &create_flags, IB_QP_CREATE_SOURCE_QPN,
+				    cond, qp);
+	}
+
+	if (qp_type == IB_QPT_RAW_PACKET) {
+		cond = MLX5_CAP_GEN(mdev, eth_net_offloads) &&
+		       MLX5_CAP_ETH(mdev, scatter_fcs);
+		process_create_flag(dev, &create_flags,
+				    IB_QP_CREATE_SCATTER_FCS, cond, qp);
+
+		cond = MLX5_CAP_GEN(mdev, eth_net_offloads) &&
+		       MLX5_CAP_ETH(mdev, vlan_cap);
+		process_create_flag(dev, &create_flags,
+				    IB_QP_CREATE_CVLAN_STRIPPING, cond, qp);
+	}
+
+	process_create_flag(dev, &create_flags,
+			    IB_QP_CREATE_PCI_WRITE_END_PADDING,
+			    MLX5_CAP_GEN(mdev, end_pad), qp);
+
+	process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_WC_TEST,
+			    qp_type != MLX5_IB_QPT_REG_UMR, qp);
+	process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_SQPN_QP1,
+			    true, qp);
+
+	if (create_flags)
+		mlx5_ib_dbg(dev, "Create QP has unsupported flags 0x%X\n",
+			    create_flags);
+
+	return (create_flags) ? -EINVAL : 0;
+}
+
 static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 			    struct ib_qp_init_attr *attr,
 			    struct mlx5_ib_create_qp *ucmd,
@@ -2769,6 +2768,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		if (err)
 			goto free_qp;
 	}
+	err = process_create_flags(dev, qp, init_attr);
+	if (err)
+		goto free_qp;
 
 	if (init_attr->qp_type == IB_QPT_XRC_TGT)
 		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
-- 
2.25.3

[PATCH rdma-next v1 15/36] RDMA/mlx5: Use flags_en mechanism to mark QP created with WQE signature

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

MLX5_QP_FLAG_SIGNATURE is exposed to the users but in the kernel
the create_qp flow treated it differently from other MLX5_QP_FLAG_*s.
Fix it by ditching wq_sig boolean variable and use general flag_en
mechanism.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  1 -
 drivers/infiniband/hw/mlx5/odp.c     |  2 +-
 drivers/infiniband/hw/mlx5/qp.c      | 36 +++++++++++++++++-----------
 3 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index b144660a47e1..61a96c1dd125 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -446,7 +446,6 @@ struct mlx5_ib_qp {
 	u32			flags;
 	u8			port;
 	u8			state;
-	int			wq_sig;
 	int			scat_cqe;
 	int			max_inline_data;
 	struct mlx5_bf	        bf;
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 16af1105cfcf..e4759310c0e2 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -1190,7 +1190,7 @@ static int mlx5_ib_mr_responder_pfault_handler_rq(struct mlx5_ib_dev *dev,
 	struct mlx5_ib_wq *wq = &qp->rq;
 	int wqe_size = 1 << wq->wqe_shift;
 
-	if (qp->wq_sig) {
+	if (qp->flags_en & MLX5_QP_FLAG_SIGNATURE) {
 		mlx5_ib_err(dev, "ODP fault with WQE signatures is not supported\n");
 		return -EFAULT;
 	}
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 02eb03484b91..9d29b84242f9 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -41,9 +41,6 @@
 #include "cmd.h"
 #include "qp.h"
 
-/* not supported currently */
-static int wq_signature;
-
 enum {
 	MLX5_IB_ACK_REQ_FREQ	= 8,
 };
@@ -392,17 +389,26 @@ static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap,
 		cap->max_recv_wr = 0;
 		cap->max_recv_sge = 0;
 	} else {
+		int wq_sig = !!(qp->flags_en & MLX5_QP_FLAG_SIGNATURE);
+
 		if (ucmd) {
 			qp->rq.wqe_cnt = ucmd->rq_wqe_count;
 			if (ucmd->rq_wqe_shift > BITS_PER_BYTE * sizeof(ucmd->rq_wqe_shift))
 				return -EINVAL;
 			qp->rq.wqe_shift = ucmd->rq_wqe_shift;
-			if ((1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) < qp->wq_sig)
+			if ((1 << qp->rq.wqe_shift) /
+				    sizeof(struct mlx5_wqe_data_seg) <
+			    wq_sig)
 				return -EINVAL;
-			qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig;
+			qp->rq.max_gs =
+				(1 << qp->rq.wqe_shift) /
+					sizeof(struct mlx5_wqe_data_seg) -
+				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 =
+				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;
@@ -416,7 +422,10 @@ static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap,
 				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_gs =
+				(1 << qp->rq.wqe_shift) /
+					sizeof(struct mlx5_wqe_data_seg) -
+				wq_sig;
 			qp->rq.max_post = qp->rq.wqe_cnt;
 		}
 	}
@@ -2008,7 +2017,8 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			return err;
 
-		qp->wq_sig = !!(ucmd->flags & MLX5_QP_FLAG_SIGNATURE);
+		if (ucmd->flags & MLX5_QP_FLAG_SIGNATURE)
+			qp->flags_en |= MLX5_QP_FLAG_SIGNATURE;
 		if (MLX5_CAP_GEN(dev->mdev, sctr_data_cqe))
 			qp->scat_cqe =
 				!!(ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE);
@@ -2045,8 +2055,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			}
 			qp->flags_en |= MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE;
 		}
-	} else {
-		qp->wq_sig = !!wq_signature;
 	}
 
 	if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
@@ -2115,7 +2123,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		MLX5_SET(qpc, qpc, latency_sensitive, 1);
 
 
-	if (qp->wq_sig)
+	if (qp->flags_en & MLX5_QP_FLAG_SIGNATURE)
 		MLX5_SET(qpc, qpc, wq_signature, 1);
 
 	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
@@ -4997,7 +5005,7 @@ static void finish_wqe(struct mlx5_ib_qp *qp,
 					     mlx5_opcode | ((u32)opmod << 24));
 	ctrl->qpn_ds = cpu_to_be32(size | (qp->trans_qp.base.mqp.qpn << 8));
 	ctrl->fm_ce_se |= fence;
-	if (unlikely(qp->wq_sig))
+	if (unlikely(qp->flags_en & MLX5_QP_FLAG_SIGNATURE))
 		ctrl->signature = wq_sig(ctrl);
 
 	qp->sq.wrid[idx] = wr_id;
@@ -5449,7 +5457,7 @@ static int _mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
 		}
 
 		scat = mlx5_frag_buf_get_wqe(&qp->rq.fbc, ind);
-		if (qp->wq_sig)
+		if (qp->flags_en & MLX5_QP_FLAG_SIGNATURE)
 			scat++;
 
 		for (i = 0; i < wr->num_sge; i++)
@@ -5461,7 +5469,7 @@ static int _mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
 			scat[i].addr       = 0;
 		}
 
-		if (qp->wq_sig) {
+		if (qp->flags_en & MLX5_QP_FLAG_SIGNATURE) {
 			sig = (struct mlx5_rwqe_sig *)scat;
 			set_sig_seg(sig, (qp->rq.max_gs + 1) << 2);
 		}
-- 
2.25.3

[PATCH rdma-next v1 16/36] RDMA/mlx5: Change scatter CQE flag to be set like other vendor flags

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

In similar way to wqe_sig, the scat_cqe was treated differently from
other create QP vendor flags. Change it to be similar to other flags
and use flags_en mechanism.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  1 -
 drivers/infiniband/hw/mlx5/qp.c      | 17 ++++++++++-------
 2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 61a96c1dd125..b6467cadc384 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -446,7 +446,6 @@ struct mlx5_ib_qp {
 	u32			flags;
 	u8			port;
 	u8			state;
-	int			scat_cqe;
 	int			max_inline_data;
 	struct mlx5_bf	        bf;
 	u8			has_rq:1;
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 9d29b84242f9..6a4b20c71b40 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2019,9 +2019,10 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 
 		if (ucmd->flags & MLX5_QP_FLAG_SIGNATURE)
 			qp->flags_en |= MLX5_QP_FLAG_SIGNATURE;
-		if (MLX5_CAP_GEN(dev->mdev, sctr_data_cqe))
-			qp->scat_cqe =
-				!!(ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE);
+		if (ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE &&
+		    MLX5_CAP_GEN(dev->mdev, sctr_data_cqe))
+			qp->flags_en |= MLX5_QP_FLAG_SCATTER_CQE;
+
 		if (ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
 			if (init_attr->qp_type != IB_QPT_RAW_PACKET ||
 			    !tunnel_offload_supported(mdev)) {
@@ -2137,8 +2138,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		MLX5_SET(qpc, qpc, cd_slave_receive, 1);
 	if (qp->flags_en & MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE)
 		MLX5_SET(qpc, qpc, req_e2e_credit_mode, 1);
-	if (qp->scat_cqe && (init_attr->qp_type == IB_QPT_RC ||
-			     init_attr->qp_type == IB_QPT_UC)) {
+	if ((qp->flags_en & MLX5_QP_FLAG_SCATTER_CQE) &&
+	    (init_attr->qp_type == IB_QPT_RC ||
+	     init_attr->qp_type == IB_QPT_UC)) {
 		int rcqe_sz = rcqe_sz =
 			mlx5_ib_get_cqe_size(init_attr->recv_cq);
 
@@ -2146,8 +2148,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			 rcqe_sz == 128 ? MLX5_RES_SCAT_DATA64_CQE :
 					  MLX5_RES_SCAT_DATA32_CQE);
 	}
-	if (qp->scat_cqe && (qp->qp_sub_type == MLX5_IB_QPT_DCI ||
-			     init_attr->qp_type == IB_QPT_RC))
+	if ((qp->flags_en & MLX5_QP_FLAG_SCATTER_CQE) &&
+	    (qp->qp_sub_type == MLX5_IB_QPT_DCI ||
+	     init_attr->qp_type == IB_QPT_RC))
 		configure_requester_scat_cqe(dev, init_attr, ucmd, qpc);
 
 	if (qp->rq.wqe_cnt) {
-- 
2.25.3

[PATCH rdma-next v1 17/36] RDMA/mlx5: Return all configured create flags through query QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

The "flags" field in struct mlx5_ib_qp contains all UAPI flags
configured at the create QP stage. Return all the data as is
without masking.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  1 +
 drivers/infiniband/hw/mlx5/qp.c      | 13 +------------
 2 files changed, 2 insertions(+), 12 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index b6467cadc384..9b2baf119823 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -443,6 +443,7 @@ struct mlx5_ib_qp {
 	/* serialize qp state modifications
 	 */
 	struct mutex		mutex;
+	/* cached variant of create_flags from struct ib_qp_init_attr */
 	u32			flags;
 	u8			port;
 	u8			state;
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 6a4b20c71b40..f0385965a694 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -5878,18 +5878,7 @@ int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 
 	qp_init_attr->cap	     = qp_attr->cap;
 
-	qp_init_attr->create_flags = 0;
-	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
-		qp_init_attr->create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
-
-	if (qp->flags & IB_QP_CREATE_CROSS_CHANNEL)
-		qp_init_attr->create_flags |= IB_QP_CREATE_CROSS_CHANNEL;
-	if (qp->flags & IB_QP_CREATE_MANAGED_SEND)
-		qp_init_attr->create_flags |= IB_QP_CREATE_MANAGED_SEND;
-	if (qp->flags & IB_QP_CREATE_MANAGED_RECV)
-		qp_init_attr->create_flags |= IB_QP_CREATE_MANAGED_RECV;
-	if (qp->flags & MLX5_IB_QP_CREATE_SQPN_QP1)
-		qp_init_attr->create_flags |= MLX5_IB_QP_CREATE_SQPN_QP1;
+	qp_init_attr->create_flags = qp->flags;
 
 	qp_init_attr->sq_sig_type = qp->sq_signal_bits & MLX5_WQE_CTRL_CQ_UPDATE ?
 		IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
-- 
2.25.3

[PATCH rdma-next v1 18/36] RDMA/mlx5: Process all vendor flags in one place

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Check that vendor flags provided through ucmd are valid.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 156 +++++++++++++++-----------------
 1 file changed, 71 insertions(+), 85 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index f0385965a694..2673678f1899 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1430,13 +1430,6 @@ static void destroy_raw_packet_qp_rq(struct mlx5_ib_dev *dev,
 	mlx5_core_destroy_rq_tracked(dev, &rq->base.mqp);
 }
 
-static bool tunnel_offload_supported(struct mlx5_core_dev *dev)
-{
-	return  (MLX5_CAP_ETH(dev, tunnel_stateless_vxlan) ||
-		 MLX5_CAP_ETH(dev, tunnel_stateless_gre) ||
-		 MLX5_CAP_ETH(dev, tunnel_stateless_geneve_rx));
-}
-
 static void destroy_raw_packet_qp_tir(struct mlx5_ib_dev *dev,
 				      struct mlx5_ib_rq *rq,
 				      u32 qp_flags_en,
@@ -1693,27 +1686,20 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		return -EOPNOTSUPP;
 	}
 
-	if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS &&
-	    !tunnel_offload_supported(dev->mdev)) {
-		mlx5_ib_dbg(dev, "tunnel offloads isn't supported\n");
-		return -EOPNOTSUPP;
-	}
-
 	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_INNER &&
 	    !(ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)) {
 		mlx5_ib_dbg(dev, "Tunnel offloads must be set for inner RSS\n");
 		return -EOPNOTSUPP;
 	}
 
-	if (ucmd.flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC || dev->is_rep) {
-		lb_flag |= MLX5_TIRC_SELF_LB_BLOCK_BLOCK_UNICAST;
+	if (dev->is_rep)
 		qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC;
-	}
 
-	if (ucmd.flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC) {
+	if (qp->flags_en & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC)
+		lb_flag |= MLX5_TIRC_SELF_LB_BLOCK_BLOCK_UNICAST;
+
+	if (qp->flags_en & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC)
 		lb_flag |= MLX5_TIRC_SELF_LB_BLOCK_BLOCK_MULTICAST;
-		qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC;
-	}
 
 	err = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));
 	if (err) {
@@ -1959,11 +1945,6 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev,
 	return atomic_mode;
 }
 
-static inline bool check_flags_mask(uint64_t input, uint64_t supported)
-{
-	return (input & ~supported) == 0;
-}
-
 static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
 			    struct mlx5_ib_create_qp *ucmd,
@@ -1999,63 +1980,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
 
 	if (udata) {
-		if (!check_flags_mask(ucmd->flags,
-				      MLX5_QP_FLAG_ALLOW_SCATTER_CQE |
-				      MLX5_QP_FLAG_BFREG_INDEX |
-				      MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE |
-				      MLX5_QP_FLAG_SCATTER_CQE |
-				      MLX5_QP_FLAG_SIGNATURE |
-				      MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC |
-				      MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC |
-				      MLX5_QP_FLAG_TUNNEL_OFFLOADS |
-				      MLX5_QP_FLAG_UAR_PAGE_INDEX |
-				      MLX5_QP_FLAG_TYPE_DCI |
-				      MLX5_QP_FLAG_TYPE_DCT))
-			return -EINVAL;
-
 		err = get_qp_user_index(ucontext, ucmd, udata->inlen, &uidx);
 		if (err)
 			return err;
-
-		if (ucmd->flags & MLX5_QP_FLAG_SIGNATURE)
-			qp->flags_en |= MLX5_QP_FLAG_SIGNATURE;
-		if (ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE &&
-		    MLX5_CAP_GEN(dev->mdev, sctr_data_cqe))
-			qp->flags_en |= MLX5_QP_FLAG_SCATTER_CQE;
-
-		if (ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS) {
-			if (init_attr->qp_type != IB_QPT_RAW_PACKET ||
-			    !tunnel_offload_supported(mdev)) {
-				mlx5_ib_dbg(dev, "Tunnel offload isn't supported\n");
-				return -EOPNOTSUPP;
-			}
-			qp->flags_en |= MLX5_QP_FLAG_TUNNEL_OFFLOADS;
-		}
-
-		if (ucmd->flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC) {
-			if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
-				mlx5_ib_dbg(dev, "Self-LB UC isn't supported\n");
-				return -EOPNOTSUPP;
-			}
-			qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC;
-		}
-
-		if (ucmd->flags & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC) {
-			if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
-				mlx5_ib_dbg(dev, "Self-LB UM isn't supported\n");
-				return -EOPNOTSUPP;
-			}
-			qp->flags_en |= MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC;
-		}
-
-		if (ucmd->flags & MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE) {
-			if (init_attr->qp_type != IB_QPT_RC ||
-				!MLX5_CAP_GEN(dev->mdev, qp_packet_based)) {
-				mlx5_ib_dbg(dev, "packet based credit mode isn't supported\n");
-				return -EOPNOTSUPP;
-			}
-			qp->flags_en |= MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE;
-		}
 	}
 
 	if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
@@ -2474,7 +2401,7 @@ static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	MLX5_SET64(dctc, dctc, dc_access_key, ucmd->access_key);
 	MLX5_SET(dctc, dctc, user_index, uidx);
 
-	if (ucmd->flags & MLX5_QP_FLAG_SCATTER_CQE) {
+	if (qp->flags_en & MLX5_QP_FLAG_SCATTER_CQE) {
 		int rcqe_sz = mlx5_ib_get_cqe_size(attr->recv_cq);
 
 		if (rcqe_sz == 128)
@@ -2577,22 +2504,81 @@ static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 }
 
-static int process_vendor_flags(struct mlx5_ib_qp *qp,
+static void process_vendor_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
+				bool cond, struct mlx5_ib_qp *qp)
+{
+	if (!(*flags & flag))
+		return;
+
+	if (cond) {
+		qp->flags_en |= flag;
+		*flags &= ~flag;
+		return;
+	}
+
+	if (flag == MLX5_QP_FLAG_SCATTER_CQE) {
+		/*
+		 * We don't return error if this flag was provided,
+		 * and mlx5 doesn't have right capability.
+		 */
+		*flags &= ~MLX5_QP_FLAG_SCATTER_CQE;
+		return;
+	}
+	mlx5_ib_dbg(dev, "Vendor create QP flag 0x%X is not supported\n", flag);
+}
+
+static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				struct ib_qp_init_attr *attr,
 				struct mlx5_ib_create_qp *ucmd)
 {
-	switch (ucmd->flags & (MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI)) {
+	struct mlx5_core_dev *mdev = dev->mdev;
+	int flags = ucmd->flags;
+	bool cond;
+
+	switch (flags & (MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI)) {
 	case MLX5_QP_FLAG_TYPE_DCI:
 		qp->qp_sub_type = MLX5_IB_QPT_DCI;
 		break;
 	case MLX5_QP_FLAG_TYPE_DCT:
 		qp->qp_sub_type = MLX5_IB_QPT_DCT;
-		break;
+		fallthrough;
 	default:
+		break;
+	}
+
+	if (attr->qp_type == IB_QPT_DRIVER && !qp->qp_sub_type)
 		return -EINVAL;
+
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_TYPE_DCI, true, qp);
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_TYPE_DCT, true, qp);
+
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_SIGNATURE, true, qp);
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_SCATTER_CQE,
+			    MLX5_CAP_GEN(mdev, sctr_data_cqe), qp);
+
+	if (attr->qp_type == IB_QPT_RAW_PACKET) {
+		cond = MLX5_CAP_ETH(mdev, tunnel_stateless_vxlan) ||
+		       MLX5_CAP_ETH(mdev, tunnel_stateless_gre) ||
+		       MLX5_CAP_ETH(mdev, tunnel_stateless_geneve_rx);
+		process_vendor_flag(dev, &flags, MLX5_QP_FLAG_TUNNEL_OFFLOADS,
+				    cond, qp);
+		process_vendor_flag(dev, &flags,
+				    MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC, true,
+				    qp);
+		process_vendor_flag(dev, &flags,
+				    MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC, true,
+				    qp);
 	}
 
-	return 0;
+	if (attr->qp_type == IB_QPT_RC)
+		process_vendor_flag(dev, &flags,
+				    MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE,
+				    MLX5_CAP_GEN(mdev, qp_packet_based), qp);
+
+	if (flags)
+		mlx5_ib_dbg(dev, "udata has unsupported flags 0x%X\n", flags);
+
+	return (flags) ? -EINVAL : 0;
 }
 
 static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
@@ -2774,8 +2760,8 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (!qp)
 		return ERR_PTR(-ENOMEM);
 
-	if (init_attr->qp_type == IB_QPT_DRIVER) {
-		err = process_vendor_flags(qp, init_attr, &ucmd);
+	if (udata) {
+		err = process_vendor_flags(dev, qp, init_attr, &ucmd);
 		if (err)
 			goto free_qp;
 	}
-- 
2.25.3

[PATCH rdma-next v1 19/36] RDMA/mlx5: Delete unsupported QP types

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

There is no need to explicitly check unsupported QP types,
rely on  "default" keyword in switch-case to catch them.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 12 +-----------
 1 file changed, 1 insertion(+), 11 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 2673678f1899..5e156b02816a 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -760,10 +760,7 @@ static int to_mlx5_st(enum ib_qp_type type)
 	case IB_QPT_SMI:		return MLX5_QP_ST_QP0;
 	case MLX5_IB_QPT_HW_GSI:	return MLX5_QP_ST_QP1;
 	case MLX5_IB_QPT_DCI:		return MLX5_QP_ST_DCI;
-	case IB_QPT_RAW_IPV6:		return MLX5_QP_ST_RAW_IPV6;
-	case IB_QPT_RAW_PACKET:
-	case IB_QPT_RAW_ETHERTYPE:	return MLX5_QP_ST_RAW_ETHERTYPE;
-	case IB_QPT_MAX:
+	case IB_QPT_RAW_PACKET:		return MLX5_QP_ST_RAW_ETHERTYPE;
 	default:		return -EINVAL;
 	}
 }
@@ -2282,14 +2279,10 @@ static void get_cqs(enum ib_qp_type qp_type,
 	case IB_QPT_RC:
 	case IB_QPT_UC:
 	case IB_QPT_UD:
-	case IB_QPT_RAW_IPV6:
-	case IB_QPT_RAW_ETHERTYPE:
 	case IB_QPT_RAW_PACKET:
 		*send_cq = ib_send_cq ? to_mcq(ib_send_cq) : NULL;
 		*recv_cq = ib_recv_cq ? to_mcq(ib_recv_cq) : NULL;
 		break;
-
-	case IB_QPT_MAX:
 	default:
 		*send_cq = NULL;
 		*recv_cq = NULL;
@@ -2434,9 +2427,6 @@ static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
 	case IB_QPT_DRIVER:
 	case IB_QPT_GSI:
 		return 0;
-	case IB_QPT_RAW_IPV6:
-	case IB_QPT_RAW_ETHERTYPE:
-	case IB_QPT_MAX:
 	default:
 		goto out;
 	}
-- 
2.25.3

[PATCH rdma-next v1 20/36] RDMA/mlx5: Store QP type in the vendor QP structure

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

QP type is stored in the IB/core QP struct, but it doesn't have all the
needed information, like internal QP type used in the driver itself.
Update mlx5_ib to have cached QP type which includes both IBTA and
Mellanox specific one.

Such change allows us to make even further cleanup of QP creation flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/mlx5_ib.h |   8 +-
 drivers/infiniband/hw/mlx5/odp.c     |   3 +-
 drivers/infiniband/hw/mlx5/qp.c      | 136 +++++++++++++--------------
 3 files changed, 74 insertions(+), 73 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 9b2baf119823..82ea01a211dd 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -465,8 +465,12 @@ struct mlx5_ib_qp {
 	struct mlx5_rate_limit	rl;
 	u32                     underlay_qpn;
 	u32			flags_en;
-	/* storage for qp sub type when core qp type is IB_QPT_DRIVER */
-	enum ib_qp_type		qp_sub_type;
+	/*
+	 * IB/core doesn't store low-level QP types, so
+	 * store both MLX and IBTA types in the field below.
+	 * IB_QPT_DRIVER will be break to DCI/DCT subtypes.
+	 */
+	enum ib_qp_type		type;
 	/* A flag to indicate if there's a new counter is configured
 	 * but not take effective
 	 */
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index e4759310c0e2..70577d546567 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -1136,8 +1136,7 @@ static int mlx5_ib_mr_initiator_pfault_handler(
 	if (qp->ibqp.qp_type == IB_QPT_XRC_INI)
 		*wqe += sizeof(struct mlx5_wqe_xrc_seg);
 
-	if (qp->ibqp.qp_type == IB_QPT_UD ||
-	    qp->qp_sub_type == MLX5_IB_QPT_DCI) {
+	if (qp->type == IB_QPT_UD || qp->type == MLX5_IB_QPT_DCI) {
 		av = *wqe;
 		if (av->dqp_dct & cpu_to_be32(MLX5_EXTENDED_UD_AV))
 			*wqe += sizeof(struct mlx5_av);
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 5e156b02816a..0d3f4bafe448 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1227,14 +1227,13 @@ static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
 
 static u32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
 {
-	if (attr->srq || (attr->qp_type == IB_QPT_XRC_TGT) ||
-	    (qp->qp_sub_type == MLX5_IB_QPT_DCI) ||
-	    (attr->qp_type == IB_QPT_XRC_INI))
+	if (attr->srq || (qp->type == IB_QPT_XRC_TGT) ||
+	    (qp->type == MLX5_IB_QPT_DCI) || (qp->type == IB_QPT_XRC_INI))
 		return MLX5_SRQ_RQ;
 	else if (!qp->has_rq)
 		return MLX5_ZERO_LEN_RQ;
-	else
-		return MLX5_NON_ZERO_RQ;
+
+	return MLX5_NON_ZERO_RQ;
 }
 
 static int create_raw_packet_qp_tis(struct mlx5_ib_dev *dev,
@@ -1967,9 +1966,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
 
-	mlx5_st = to_mlx5_st((init_attr->qp_type != IB_QPT_DRIVER) ?
-				     init_attr->qp_type :
-				     qp->qp_sub_type);
+	mlx5_st = to_mlx5_st(qp->type);
 	if (mlx5_st < 0)
 		return -EINVAL;
 
@@ -2073,8 +2070,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 					  MLX5_RES_SCAT_DATA32_CQE);
 	}
 	if ((qp->flags_en & MLX5_QP_FLAG_SCATTER_CQE) &&
-	    (qp->qp_sub_type == MLX5_IB_QPT_DCI ||
-	     init_attr->qp_type == IB_QPT_RC))
+	    (qp->type == MLX5_IB_QPT_DCI || qp->type == IB_QPT_RC))
 		configure_requester_scat_cqe(dev, init_attr, ucmd, qpc);
 
 	if (qp->rq.wqe_cnt) {
@@ -2166,7 +2162,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	base->container_mibqp = qp;
 	base->mqp.event = mlx5_ib_qp_event;
 
-	get_cqs(init_attr->qp_type, init_attr->send_cq, init_attr->recv_cq,
+	get_cqs(qp->type, init_attr->send_cq, init_attr->recv_cq,
 		&send_cq, &recv_cq);
 	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
 	mlx5_ib_lock_cqs(send_cq, recv_cq);
@@ -2406,7 +2402,8 @@ static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	return 0;
 }
 
-static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
+static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr,
+			 enum ib_qp_type *type)
 {
 	if (attr->qp_type == IB_QPT_DRIVER && !MLX5_CAP_GEN(dev->mdev, dct))
 		goto out;
@@ -2426,11 +2423,12 @@ static int check_qp_type(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr)
 	case MLX5_IB_QPT_REG_UMR:
 	case IB_QPT_DRIVER:
 	case IB_QPT_GSI:
-		return 0;
+		break;
 	default:
 		goto out;
 	}
 
+	*type = attr->qp_type;
 	return 0;
 
 out:
@@ -2518,7 +2516,6 @@ static void process_vendor_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
 }
 
 static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
-				struct ib_qp_init_attr *attr,
 				struct mlx5_ib_create_qp *ucmd)
 {
 	struct mlx5_core_dev *mdev = dev->mdev;
@@ -2527,17 +2524,20 @@ static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 
 	switch (flags & (MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI)) {
 	case MLX5_QP_FLAG_TYPE_DCI:
-		qp->qp_sub_type = MLX5_IB_QPT_DCI;
+		qp->type = MLX5_IB_QPT_DCI;
 		break;
 	case MLX5_QP_FLAG_TYPE_DCT:
-		qp->qp_sub_type = MLX5_IB_QPT_DCT;
-		fallthrough;
-	default:
+		qp->type = MLX5_IB_QPT_DCT;
 		break;
-	}
-
-	if (attr->qp_type == IB_QPT_DRIVER && !qp->qp_sub_type)
+	default:
+		if (qp->type != IB_QPT_DRIVER)
+			break;
+		/*
+		 * It is IB_QPT_DRIVER and or no subtype or
+		 * wrong subtype were provided.
+		 */
 		return -EINVAL;
+	}
 
 	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_TYPE_DCI, true, qp);
 	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_TYPE_DCT, true, qp);
@@ -2546,7 +2546,7 @@ static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_SCATTER_CQE,
 			    MLX5_CAP_GEN(mdev, sctr_data_cqe), qp);
 
-	if (attr->qp_type == IB_QPT_RAW_PACKET) {
+	if (qp->type == IB_QPT_RAW_PACKET) {
 		cond = MLX5_CAP_ETH(mdev, tunnel_stateless_vxlan) ||
 		       MLX5_CAP_ETH(mdev, tunnel_stateless_gre) ||
 		       MLX5_CAP_ETH(mdev, tunnel_stateless_geneve_rx);
@@ -2560,7 +2560,7 @@ static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				    qp);
 	}
 
-	if (attr->qp_type == IB_QPT_RC)
+	if (qp->type == IB_QPT_RC)
 		process_vendor_flag(dev, &flags,
 				    MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE,
 				    MLX5_CAP_GEN(mdev, qp_packet_based), qp);
@@ -2597,12 +2597,12 @@ static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
 static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				struct ib_qp_init_attr *attr)
 {
-	enum ib_qp_type qp_type = attr->qp_type;
+	enum ib_qp_type qp_type = qp->type;
 	struct mlx5_core_dev *mdev = dev->mdev;
 	int create_flags = attr->create_flags;
 	bool cond;
 
-	if (qp->qp_sub_type == MLX5_IB_QPT_DCT)
+	if (qp_type == MLX5_IB_QPT_DCT)
 		return (create_flags) ? -EINVAL : 0;
 
 	if (qp_type == IB_QPT_RAW_PACKET && attr->rwq_ind_tbl)
@@ -2656,34 +2656,6 @@ static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	return (create_flags) ? -EINVAL : 0;
 }
 
-static int create_driver_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-			    struct ib_qp_init_attr *attr,
-			    struct mlx5_ib_create_qp *ucmd,
-			    struct ib_udata *udata)
-{
-	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
-	int ret = -EINVAL;
-
-	switch (qp->qp_sub_type) {
-	case MLX5_IB_QPT_DCT:
-		if (!attr->srq || !attr->recv_cq)
-			goto out;
-
-		ret = create_dct(pd, qp, attr, ucmd, udata);
-		break;
-	case MLX5_IB_QPT_DCI:
-		if (attr->cap.max_recv_wr || attr->cap.max_recv_sge)
-			goto out;
-
-		ret = create_qp_common(mdev, pd, attr, ucmd, udata, qp);
-		break;
-	default:
-		return -EINVAL;
-	}
-
-out:	return ret;
-}
-
 static size_t process_udata_size(struct ib_qp_init_attr *attr,
 				 struct ib_udata *udata)
 {
@@ -2707,6 +2679,30 @@ static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	return create_qp_common(dev, pd, attr, ucmd, udata, qp);
 }
 
+static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
+			 struct ib_qp_init_attr *attr)
+{
+	int ret = 0;
+
+	switch (qp->type) {
+	case MLX5_IB_QPT_DCT:
+		ret = (!attr->srq || !attr->recv_cq) ? -EINVAL : 0;
+		break;
+	case MLX5_IB_QPT_DCI:
+		ret = (attr->cap.max_recv_wr || attr->cap.max_recv_sge) ?
+			      -EINVAL :
+			      0;
+		break;
+	default:
+		break;
+	}
+
+	if (ret)
+		mlx5_ib_dbg(dev, "QP type %d has wrong attributes\n", qp->type);
+
+	return ret;
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
@@ -2714,13 +2710,14 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	struct mlx5_ib_create_qp ucmd = {};
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
+	enum ib_qp_type type;
 	u16 xrcdn = 0;
 	int err;
 
 	dev = pd ? to_mdev(pd->device) :
 		   to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
 
-	err = check_qp_type(dev, init_attr);
+	err = check_qp_type(dev, init_attr, &type);
 	if (err) {
 		mlx5_ib_dbg(dev, "Unsupported QP type %d\n",
 			    init_attr->qp_type);
@@ -2750,8 +2747,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (!qp)
 		return ERR_PTR(-ENOMEM);
 
+	qp->type = type;
 	if (udata) {
-		err = process_vendor_flags(dev, qp, init_attr, &ucmd);
+		err = process_vendor_flags(dev, qp, &ucmd);
 		if (err)
 			goto free_qp;
 	}
@@ -2759,16 +2757,20 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (err)
 		goto free_qp;
 
-	if (init_attr->qp_type == IB_QPT_XRC_TGT)
+	if (qp->type == IB_QPT_XRC_TGT)
 		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
 
-	switch (init_attr->qp_type) {
-	case IB_QPT_DRIVER:
-		err = create_driver_qp(pd, qp, init_attr, &ucmd, udata);
-		break;
+	err = check_qp_attr(dev, qp, init_attr);
+	if (err)
+		goto free_qp;
+
+	switch (qp->type) {
 	case IB_QPT_RAW_PACKET:
 		err = create_raw_qp(pd, qp, init_attr, &ucmd, udata);
 		break;
+	case MLX5_IB_QPT_DCT:
+		err = create_dct(pd, qp, init_attr, &ucmd, udata);
+		break;
 	default:
 		err = create_qp_common(dev, pd, init_attr,
 				       (udata) ? &ucmd : NULL, udata, qp);
@@ -2821,7 +2823,7 @@ int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 	if (unlikely(qp->qp_type == IB_QPT_GSI))
 		return mlx5_ib_gsi_destroy_qp(qp);
 
-	if (mqp->qp_sub_type == MLX5_IB_QPT_DCT)
+	if (mqp->type == MLX5_IB_QPT_DCT)
 		return mlx5_ib_destroy_dct(mqp);
 
 	destroy_qp_common(dev, mqp, udata);
@@ -3508,8 +3510,7 @@ static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
 	u16 op;
 	u8 tx_affinity = 0;
 
-	mlx5_st = to_mlx5_st(ibqp->qp_type == IB_QPT_DRIVER ?
-			     qp->qp_sub_type : ibqp->qp_type);
+	mlx5_st = to_mlx5_st(qp->type);
 	if (mlx5_st < 0)
 		return -EINVAL;
 
@@ -3970,11 +3971,8 @@ int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	if (unlikely(ibqp->qp_type == IB_QPT_GSI))
 		return mlx5_ib_gsi_modify_qp(ibqp, attr, attr_mask);
 
-	if (ibqp->qp_type == IB_QPT_DRIVER)
-		qp_type = qp->qp_sub_type;
-	else
-		qp_type = (unlikely(ibqp->qp_type == MLX5_IB_QPT_HW_GSI)) ?
-			IB_QPT_GSI : ibqp->qp_type;
+	qp_type = (unlikely(ibqp->qp_type == MLX5_IB_QPT_HW_GSI)) ? IB_QPT_GSI :
+								    qp->type;
 
 	if (qp_type == MLX5_IB_QPT_DCT)
 		return mlx5_ib_modify_dct(ibqp, attr, attr_mask, udata);
@@ -5813,7 +5811,7 @@ int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
 	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
 	memset(qp_attr, 0, sizeof(*qp_attr));
 
-	if (unlikely(qp->qp_sub_type == MLX5_IB_QPT_DCT))
+	if (unlikely(qp->type == MLX5_IB_QPT_DCT))
 		return mlx5_ib_dct_query_qp(dev, qp, qp_attr,
 					    qp_attr_mask, qp_init_attr);
 
-- 
2.25.3

[PATCH rdma-next v1 21/36] RDMA/mlx5: Promote RSS RAW QP attribute check in higher level

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Perform check of attributes of RAW PACKET QP in separate function.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 0d3f4bafe448..454433a18b97 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1645,9 +1645,6 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	size_t required_cmd_sz;
 	u8 lb_flag = 0;
 
-	if (init_attr->send_cq)
-		return -EINVAL;
-
 	min_resp_len = offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
 	if (udata->outlen < min_resp_len)
 		return -EINVAL;
@@ -2693,6 +2690,9 @@ static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			      -EINVAL :
 			      0;
 		break;
+	case IB_QPT_RAW_PACKET:
+		ret = (attr->rwq_ind_tbl && attr->send_cq) ? -EINVAL : 0;
+		break;
 	default:
 		break;
 	}
-- 
2.25.3

[PATCH rdma-next v1 22/36] RDMA/mlx5: Combine copy of create QP command in RSS RAW QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Change the create QP flow to handle all copy_from_user() operations in
one place.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 156 +++++++++++++++++---------------
 1 file changed, 82 insertions(+), 74 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 454433a18b97..4f69105f082b 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1624,6 +1624,7 @@ static void destroy_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *q
 
 static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 				 struct ib_qp_init_attr *init_attr,
+				 struct mlx5_ib_create_qp_rss *ucmd,
 				 struct ib_udata *udata)
 {
 	struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context(
@@ -1641,46 +1642,26 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	u32 outer_l4;
 	size_t min_resp_len;
 	u32 tdn = mucontext->tdn;
-	struct mlx5_ib_create_qp_rss ucmd = {};
-	size_t required_cmd_sz;
 	u8 lb_flag = 0;
 
 	min_resp_len = offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
 	if (udata->outlen < min_resp_len)
 		return -EINVAL;
 
-	required_cmd_sz = offsetof(typeof(ucmd), flags) + sizeof(ucmd.flags);
-	if (udata->inlen < required_cmd_sz) {
-		mlx5_ib_dbg(dev, "invalid inlen\n");
-		return -EINVAL;
-	}
-
-	if (udata->inlen > sizeof(ucmd) &&
-	    !ib_is_udata_cleared(udata, sizeof(ucmd),
-				 udata->inlen - sizeof(ucmd))) {
-		mlx5_ib_dbg(dev, "inlen is not supported\n");
-		return -EOPNOTSUPP;
-	}
-
-	if (ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata->inlen))) {
-		mlx5_ib_dbg(dev, "copy failed\n");
-		return -EFAULT;
-	}
-
-	if (ucmd.comp_mask) {
+	if (ucmd->comp_mask) {
 		mlx5_ib_dbg(dev, "invalid comp mask\n");
 		return -EOPNOTSUPP;
 	}
 
-	if (ucmd.flags & ~(MLX5_QP_FLAG_TUNNEL_OFFLOADS |
+	if (ucmd->flags & ~(MLX5_QP_FLAG_TUNNEL_OFFLOADS |
 			   MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC |
 			   MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC)) {
 		mlx5_ib_dbg(dev, "invalid flags\n");
 		return -EOPNOTSUPP;
 	}
 
-	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_INNER &&
-	    !(ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)) {
+	if (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_INNER &&
+	    !(ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)) {
 		mlx5_ib_dbg(dev, "Tunnel offloads must be set for inner RSS\n");
 		return -EOPNOTSUPP;
 	}
@@ -1717,29 +1698,29 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 
 	hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
 
-	if (ucmd.flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)
+	if (ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)
 		MLX5_SET(tirc, tirc, tunneled_offload_en, 1);
 
 	MLX5_SET(tirc, tirc, self_lb_block, lb_flag);
 
-	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_INNER)
+	if (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_INNER)
 		hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_inner);
 	else
 		hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer);
 
-	switch (ucmd.rx_hash_function) {
+	switch (ucmd->rx_hash_function) {
 	case MLX5_RX_HASH_FUNC_TOEPLITZ:
 	{
 		void *rss_key = MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key);
 		size_t len = MLX5_FLD_SZ_BYTES(tirc, rx_hash_toeplitz_key);
 
-		if (len != ucmd.rx_key_len) {
+		if (len != ucmd->rx_key_len) {
 			err = -EINVAL;
 			goto err;
 		}
 
 		MLX5_SET(tirc, tirc, rx_hash_fn, MLX5_RX_HASH_FN_TOEPLITZ);
-		memcpy(rss_key, ucmd.rx_hash_key, len);
+		memcpy(rss_key, ucmd->rx_hash_key, len);
 		break;
 	}
 	default:
@@ -1747,7 +1728,7 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		goto err;
 	}
 
-	if (!ucmd.rx_hash_fields_mask) {
+	if (!ucmd->rx_hash_fields_mask) {
 		/* special case when this TIR serves as steering entry without hashing */
 		if (!init_attr->rwq_ind_tbl->log_ind_tbl_size)
 			goto create_tir;
@@ -1755,29 +1736,31 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		goto err;
 	}
 
-	if (((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
-	     (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4)) &&
-	     ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6) ||
-	     (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))) {
+	if (((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
+	     (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4)) &&
+	     ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6) ||
+	     (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))) {
 		err = -EINVAL;
 		goto err;
 	}
 
 	/* If none of IPV4 & IPV6 SRC/DST was set - this bit field is ignored */
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4))
 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
 			 MLX5_L3_PROT_TYPE_IPV4);
-	else if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6) ||
-		 (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))
+	else if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6) ||
+		 (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))
 		MLX5_SET(rx_hash_field_select, hfso, l3_prot_type,
 			 MLX5_L3_PROT_TYPE_IPV6);
 
-	outer_l4 = ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
-		    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP)) << 0 |
-		   ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP) ||
-		    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP)) << 1 |
-		   (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_IPSEC_SPI) << 2;
+	outer_l4 = ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
+		    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP))
+			   << 0 |
+		   ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP) ||
+		    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP))
+			   << 1 |
+		   (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_IPSEC_SPI) << 2;
 
 	/* Check that only one l4 protocol is set */
 	if (outer_l4 & (outer_l4 - 1)) {
@@ -1786,32 +1769,32 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	}
 
 	/* If none of TCP & UDP SRC/DST was set - this bit field is ignored */
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP))
 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
 			 MLX5_L4_PROT_TYPE_TCP);
-	else if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP) ||
-		 (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP))
+	else if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP) ||
+		 (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP))
 		MLX5_SET(rx_hash_field_select, hfso, l4_prot_type,
 			 MLX5_L4_PROT_TYPE_UDP);
 
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV4) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_IPV6))
 		selected_fields |= MLX5_HASH_FIELD_SEL_SRC_IP;
 
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV4) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_IPV6))
 		selected_fields |= MLX5_HASH_FIELD_SEL_DST_IP;
 
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_TCP) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_SRC_PORT_UDP))
 		selected_fields |= MLX5_HASH_FIELD_SEL_L4_SPORT;
 
-	if ((ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP) ||
-	    (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP))
+	if ((ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_TCP) ||
+	    (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_DST_PORT_UDP))
 		selected_fields |= MLX5_HASH_FIELD_SEL_L4_DPORT;
 
-	if (ucmd.rx_hash_fields_mask & MLX5_RX_HASH_IPSEC_SPI)
+	if (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_IPSEC_SPI)
 		selected_fields |= MLX5_HASH_FIELD_SEL_IPSEC_SPI;
 
 	MLX5_SET(rx_hash_field_select, hfso, selected_fields, selected_fields);
@@ -2513,11 +2496,16 @@ static void process_vendor_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
 }
 
 static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
-				struct mlx5_ib_create_qp *ucmd)
+				void *ucmd, struct ib_qp_init_attr *attr)
 {
 	struct mlx5_core_dev *mdev = dev->mdev;
-	int flags = ucmd->flags;
 	bool cond;
+	int flags;
+
+	if (attr->rwq_ind_tbl)
+		flags = ((struct mlx5_ib_create_qp_rss *)ucmd)->flags;
+	else
+		flags = ((struct mlx5_ib_create_qp *)ucmd)->flags;
 
 	switch (flags & (MLX5_QP_FLAG_TYPE_DCT | MLX5_QP_FLAG_TYPE_DCI)) {
 	case MLX5_QP_FLAG_TYPE_DCI:
@@ -2657,21 +2645,32 @@ static size_t process_udata_size(struct ib_qp_init_attr *attr,
 				 struct ib_udata *udata)
 {
 	size_t ucmd = sizeof(struct mlx5_ib_create_qp);
+	size_t inlen = udata->inlen;
 
 	if (attr->qp_type == IB_QPT_DRIVER)
-		return (udata->inlen < ucmd) ? 0 : ucmd;
+		return (inlen < ucmd) ? 0 : ucmd;
+
+	if (!attr->rwq_ind_tbl)
+		return ucmd;
+
+	if (inlen < offsetofend(struct mlx5_ib_create_qp_rss, flags))
+		return 0;
+
+	ucmd = sizeof(struct mlx5_ib_create_qp_rss);
+	if (inlen > ucmd && !ib_is_udata_cleared(udata, ucmd, inlen - ucmd))
+		return 0;
 
-	return ucmd;
+	return min(ucmd, inlen);
 }
 
 static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-			 struct ib_qp_init_attr *attr,
-			 struct mlx5_ib_create_qp *ucmd, struct ib_udata *udata)
+			 struct ib_qp_init_attr *attr, void *ucmd,
+			 struct ib_udata *udata)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 
 	if (attr->rwq_ind_tbl)
-		return create_rss_raw_qp_tir(pd, qp, attr, udata);
+		return create_rss_raw_qp_tir(pd, qp, attr, ucmd, udata);
 
 	return create_qp_common(dev, pd, attr, ucmd, udata, qp);
 }
@@ -2707,10 +2706,10 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
-	struct mlx5_ib_create_qp ucmd = {};
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
 	enum ib_qp_type type;
+	void *ucmd = NULL;
 	u16 xrcdn = 0;
 	int err;
 
@@ -2731,25 +2730,31 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (init_attr->qp_type == IB_QPT_GSI)
 		return mlx5_ib_gsi_create_qp(pd, init_attr);
 
-	if (udata && !init_attr->rwq_ind_tbl) {
+	if (udata) {
 		size_t inlen =
 			process_udata_size(init_attr, udata);
 
 		if (!inlen)
 			return ERR_PTR(-EINVAL);
 
-		err = ib_copy_from_udata(&ucmd, udata, inlen);
+		ucmd = kzalloc(inlen, GFP_KERNEL);
+		if (!ucmd)
+			return ERR_PTR(-ENOMEM);
+
+		err = ib_copy_from_udata(ucmd, udata, inlen);
 		if (err)
-			return ERR_PTR(err);
+			goto free_ucmd;
 	}
 
 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
-	if (!qp)
-		return ERR_PTR(-ENOMEM);
+	if (!qp) {
+		err = -ENOMEM;
+		goto free_ucmd;
+	}
 
 	qp->type = type;
 	if (udata) {
-		err = process_vendor_flags(dev, qp, &ucmd);
+		err = process_vendor_flags(dev, qp, ucmd, init_attr);
 		if (err)
 			goto free_qp;
 	}
@@ -2766,20 +2771,21 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 
 	switch (qp->type) {
 	case IB_QPT_RAW_PACKET:
-		err = create_raw_qp(pd, qp, init_attr, &ucmd, udata);
+		err = create_raw_qp(pd, qp, init_attr, ucmd, udata);
 		break;
 	case MLX5_IB_QPT_DCT:
-		err = create_dct(pd, qp, init_attr, &ucmd, udata);
+		err = create_dct(pd, qp, init_attr, ucmd, udata);
 		break;
 	default:
-		err = create_qp_common(dev, pd, init_attr,
-				       (udata) ? &ucmd : NULL, udata, qp);
+		err = create_qp_common(dev, pd, init_attr, ucmd, udata, qp);
 	}
 	if (err) {
 		mlx5_ib_dbg(dev, "create_qp_common failed\n");
 		goto free_qp;
 	}
 
+	kfree(ucmd);
+
 	if (is_qp0(init_attr->qp_type))
 		qp->ibqp.qp_num = 0;
 	else if (is_qp1(init_attr->qp_type))
@@ -2793,6 +2799,8 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 
 free_qp:
 	kfree(qp);
+free_ucmd:
+	kfree(ucmd);
 	return ERR_PTR(err);
 }
 
-- 
2.25.3

[PATCH rdma-next v1 23/36] RDMA/mlx5: Remove second user copy in create_user_qp

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Combine copy_from_user() from create_user_qp() and general code.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 34 +++++++++++++++------------------
 1 file changed, 15 insertions(+), 19 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 4f69105f082b..495f03905fbf 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -914,13 +914,12 @@ static int adjust_bfregn(struct mlx5_ib_dev *dev,
 
 static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			  struct mlx5_ib_qp *qp, struct ib_udata *udata,
-			  struct ib_qp_init_attr *attr,
-			  u32 **in,
+			  struct ib_qp_init_attr *attr, u32 **in,
 			  struct mlx5_ib_create_qp_resp *resp, int *inlen,
-			  struct mlx5_ib_qp_base *base)
+			  struct mlx5_ib_qp_base *base,
+			  struct mlx5_ib_create_qp *ucmd)
 {
 	struct mlx5_ib_ucontext *context;
-	struct mlx5_ib_create_qp ucmd;
 	struct mlx5_ib_ubuffer *ubuffer = &base->ubuffer;
 	int page_shift = 0;
 	int uar_index = 0;
@@ -934,24 +933,18 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	u16 uid;
 	u32 uar_flags;
 
-	err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
-	if (err) {
-		mlx5_ib_dbg(dev, "copy failed\n");
-		return err;
-	}
-
 	context = rdma_udata_to_drv_context(udata, struct mlx5_ib_ucontext,
 					    ibucontext);
-	uar_flags = ucmd.flags & (MLX5_QP_FLAG_UAR_PAGE_INDEX |
-				  MLX5_QP_FLAG_BFREG_INDEX);
+	uar_flags = qp->flags_en &
+		    (MLX5_QP_FLAG_UAR_PAGE_INDEX | MLX5_QP_FLAG_BFREG_INDEX);
 	switch (uar_flags) {
 	case MLX5_QP_FLAG_UAR_PAGE_INDEX:
-		uar_index = ucmd.bfreg_index;
+		uar_index = ucmd->bfreg_index;
 		bfregn = MLX5_IB_INVALID_BFREG;
 		break;
 	case MLX5_QP_FLAG_BFREG_INDEX:
 		uar_index = bfregn_to_uar_index(dev, &context->bfregi,
-						ucmd.bfreg_index, true);
+						ucmd->bfreg_index, true);
 		if (uar_index < 0)
 			return uar_index;
 		bfregn = MLX5_IB_INVALID_BFREG;
@@ -976,12 +969,12 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	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, base, attr);
+	err = set_user_buf_size(dev, qp, ucmd, base, attr);
 	if (err)
 		goto err_bfreg;
 
-	if (ucmd.buf_addr && ubuffer->buf_size) {
-		ubuffer->buf_addr = ucmd.buf_addr;
+	if (ucmd->buf_addr && ubuffer->buf_size) {
+		ubuffer->buf_addr = ucmd->buf_addr;
 		err = mlx5_ib_umem_get(dev, udata, ubuffer->buf_addr,
 				       ubuffer->buf_size, &ubuffer->umem,
 				       &npages, &page_shift, &ncont, &offset);
@@ -1018,7 +1011,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		resp->bfreg_index = MLX5_IB_INVALID_BFREG;
 	qp->bfregn = bfregn;
 
-	err = mlx5_ib_db_map_user(context, udata, ucmd.db_addr, &qp->db);
+	err = mlx5_ib_db_map_user(context, udata, ucmd->db_addr, &qp->db);
 	if (err) {
 		mlx5_ib_dbg(dev, "map failed\n");
 		goto err_free;
@@ -1991,7 +1984,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 				return -EINVAL;
 			}
 			err = create_user_qp(dev, pd, qp, udata, init_attr, &in,
-					     &resp, &inlen, base);
+					     &resp, &inlen, base, ucmd);
 			if (err)
 				mlx5_ib_dbg(dev, "err %d\n", err);
 		} else {
@@ -2550,6 +2543,9 @@ static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				    MLX5_QP_FLAG_PACKET_BASED_CREDIT_MODE,
 				    MLX5_CAP_GEN(mdev, qp_packet_based), qp);
 
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_BFREG_INDEX, true, qp);
+	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_UAR_PAGE_INDEX, true, qp);
+
 	if (flags)
 		mlx5_ib_dbg(dev, "udata has unsupported flags 0x%X\n", flags);
 
-- 
2.25.3

[PATCH rdma-next v1 24/36] RDMA/mlx5: Rely on existence of udata to separate kernel/user flows

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Instead of keeping special field to separate kernel/user create/destroy
flows, rely on existence of udata pointer. All allocation flows are
using kzalloc() and leave uninitialized pointers as NULL which makes
MLX5_QP_EMPTY and MLX5_QP_KERNEL flows to be the same.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/mlx5_ib.h | 14 --------------
 drivers/infiniband/hw/mlx5/qp.c      | 23 ++++++++++-------------
 2 files changed, 10 insertions(+), 27 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 82ea01a211dd..df375cb4efbb 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -337,7 +337,6 @@ struct mlx5_ib_rwq {
 	struct ib_umem		*umem;
 	size_t			buf_size;
 	unsigned int		page_shift;
-	int			create_type;
 	struct mlx5_db		db;
 	u32			user_index;
 	u32			wqe_count;
@@ -346,17 +345,6 @@ struct mlx5_ib_rwq {
 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
 };
 
-enum {
-	MLX5_QP_USER,
-	MLX5_QP_KERNEL,
-	MLX5_QP_EMPTY
-};
-
-enum {
-	MLX5_WQ_USER,
-	MLX5_WQ_KERNEL
-};
-
 struct mlx5_ib_rwq_ind_table {
 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
 	u32			rqtn;
@@ -457,8 +445,6 @@ struct mlx5_ib_qp {
 	 */
 	int			bfregn;
 
-	int			create_type;
-
 	struct list_head	qps_list;
 	struct list_head	cq_recv_list;
 	struct list_head	cq_send_list;
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 495f03905fbf..74f09cdb4a33 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -897,7 +897,6 @@ static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		goto err_umem;
 	}
 
-	rwq->create_type = MLX5_WQ_USER;
 	return 0;
 
 err_umem:
@@ -1022,7 +1021,6 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		mlx5_ib_dbg(dev, "copy failed\n");
 		goto err_unmap;
 	}
-	qp->create_type = MLX5_QP_USER;
 
 	return 0;
 
@@ -1187,7 +1185,6 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev,
 		err = -ENOMEM;
 		goto err_wrid;
 	}
-	qp->create_type = MLX5_QP_KERNEL;
 
 	return 0;
 
@@ -1214,8 +1211,10 @@ static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
 	kvfree(qp->sq.wrid);
 	kvfree(qp->sq.wr_data);
 	kvfree(qp->rq.wrid);
-	mlx5_db_free(dev->mdev, &qp->db);
-	mlx5_frag_buf_free(dev->mdev, &qp->buf);
+	if (qp->db.db)
+		mlx5_db_free(dev->mdev, &qp->db);
+	if (qp->buf.frags)
+		mlx5_frag_buf_free(dev->mdev, &qp->buf);
 }
 
 static u32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
@@ -2000,8 +1999,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		in = kvzalloc(inlen, GFP_KERNEL);
 		if (!in)
 			return -ENOMEM;
-
-		qp->create_type = MLX5_QP_EMPTY;
 	}
 
 	if (is_sqp(init_attr->qp_type))
@@ -2155,9 +2152,9 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 
 err_create:
-	if (qp->create_type == MLX5_QP_USER)
+	if (udata)
 		destroy_qp_user(dev, pd, qp, base, udata);
-	else if (qp->create_type == MLX5_QP_KERNEL)
+	else
 		destroy_qp_kernel(dev, qp);
 
 err:
@@ -2311,7 +2308,7 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	if (recv_cq)
 		list_del(&qp->cq_recv_list);
 
-	if (qp->create_type == MLX5_QP_KERNEL) {
+	if (!udata) {
 		__mlx5_ib_cq_clean(recv_cq, base->mqp.qpn,
 				   qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
 		if (send_cq != recv_cq)
@@ -2331,10 +2328,10 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				     base->mqp.qpn);
 	}
 
-	if (qp->create_type == MLX5_QP_KERNEL)
-		destroy_qp_kernel(dev, qp);
-	else if (qp->create_type == MLX5_QP_USER)
+	if (udata)
 		destroy_qp_user(dev, &get_pd(qp)->ibpd, qp, base, udata);
+	else
+		destroy_qp_kernel(dev, qp);
 }
 
 static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-- 
2.25.3

[PATCH rdma-next v1 25/36] RDMA/mlx5: Delete impossible inlen check

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

The inlen is set to be above zero in all flows before
and can't be negative at this stage.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 74f09cdb4a33..5a43128d651b 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2107,11 +2107,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		qp->flags &= ~IB_QP_CREATE_PCI_WRITE_END_PADDING;
 	}
 
-	if (inlen < 0) {
-		err = -EINVAL;
-		goto err;
-	}
-
 	if (init_attr->qp_type == IB_QPT_RAW_PACKET ||
 	    qp->flags & IB_QP_CREATE_SOURCE_QPN) {
 		qp->raw_packet_qp.sq.ubuffer.buf_addr = ucmd->sq_buf_addr;
@@ -2156,8 +2151,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		destroy_qp_user(dev, pd, qp, base, udata);
 	else
 		destroy_qp_kernel(dev, qp);
-
-err:
 	kvfree(in);
 	return err;
 }
-- 
2.25.3

[PATCH rdma-next v1 26/36] RDMA/mlx5: Globally parse DEVX UID

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Remove duplication in parsing of DEVX UID.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 51 +++++++++++++++++----------------
 1 file changed, 27 insertions(+), 24 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 5a43128d651b..b2174e0817f5 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1916,18 +1916,16 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev,
 static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
 			    struct mlx5_ib_create_qp *ucmd,
-			    struct ib_udata *udata, struct mlx5_ib_qp *qp)
+			    struct ib_udata *udata, struct mlx5_ib_qp *qp,
+			    u32 uidx)
 {
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
 	struct mlx5_core_dev *mdev = dev->mdev;
 	struct mlx5_ib_create_qp_resp resp = {};
-	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
-		udata, struct mlx5_ib_ucontext, ibucontext);
 	struct mlx5_ib_cq *send_cq;
 	struct mlx5_ib_cq *recv_cq;
 	unsigned long flags;
-	u32 uidx = MLX5_IB_DEFAULT_UIDX;
 	struct mlx5_ib_qp_base *base;
 	int mlx5_st;
 	void *qpc;
@@ -1945,12 +1943,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
 		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
 
-	if (udata) {
-		err = get_qp_user_index(ucontext, ucmd, udata->inlen, &uidx);
-		if (err)
-			return err;
-	}
-
 	if (qp->flags & IB_QP_CREATE_SOURCE_QPN)
 		qp->underlay_qpn = init_attr->source_qpn;
 
@@ -2329,18 +2321,10 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 
 static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 		      struct ib_qp_init_attr *attr,
-		      struct mlx5_ib_create_qp *ucmd, struct ib_udata *udata)
+		      struct mlx5_ib_create_qp *ucmd, u32 uidx)
 {
-	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
-		udata, struct mlx5_ib_ucontext, ibucontext);
-	int err = 0;
-	u32 uidx = MLX5_IB_DEFAULT_UIDX;
 	void *dctc;
 
-	err = get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), &uidx);
-	if (err)
-		return err;
-
 	qp->dct.in = kzalloc(MLX5_ST_SZ_BYTES(create_dct_in), GFP_KERNEL);
 	if (!qp->dct.in)
 		return -ENOMEM;
@@ -2651,14 +2635,14 @@ static size_t process_udata_size(struct ib_qp_init_attr *attr,
 
 static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 			 struct ib_qp_init_attr *attr, void *ucmd,
-			 struct ib_udata *udata)
+			 struct ib_udata *udata, u32 uidx)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 
 	if (attr->rwq_ind_tbl)
 		return create_rss_raw_qp_tir(pd, qp, attr, ucmd, udata);
 
-	return create_qp_common(dev, pd, attr, ucmd, udata, qp);
+	return create_qp_common(dev, pd, attr, ucmd, udata, qp, uidx);
 }
 
 static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
@@ -2688,10 +2672,24 @@ static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	return ret;
 }
 
+static int get_qp_uidx(struct mlx5_ib_qp *qp, struct ib_udata *udata,
+		       struct mlx5_ib_create_qp *ucmd,
+		       struct ib_qp_init_attr *attr, u32 *uidx)
+{
+	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
+		udata, struct mlx5_ib_ucontext, ibucontext);
+
+	if (attr->rwq_ind_tbl)
+		return 0;
+
+	return get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), uidx);
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
+	u32 uidx = MLX5_IB_DEFAULT_UIDX;
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
 	enum ib_qp_type type;
@@ -2743,6 +2741,10 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		err = process_vendor_flags(dev, qp, ucmd, init_attr);
 		if (err)
 			goto free_qp;
+
+		err = get_qp_uidx(qp, udata, ucmd, init_attr, &uidx);
+		if (err)
+			goto free_qp;
 	}
 	err = process_create_flags(dev, qp, init_attr);
 	if (err)
@@ -2757,13 +2759,14 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 
 	switch (qp->type) {
 	case IB_QPT_RAW_PACKET:
-		err = create_raw_qp(pd, qp, init_attr, ucmd, udata);
+		err = create_raw_qp(pd, qp, init_attr, ucmd, udata, uidx);
 		break;
 	case MLX5_IB_QPT_DCT:
-		err = create_dct(pd, qp, init_attr, ucmd, udata);
+		err = create_dct(pd, qp, init_attr, ucmd, uidx);
 		break;
 	default:
-		err = create_qp_common(dev, pd, init_attr, ucmd, udata, qp);
+		err = create_qp_common(dev, pd, init_attr, ucmd, udata, qp,
+				       uidx);
 	}
 	if (err) {
 		mlx5_ib_dbg(dev, "create_qp_common failed\n");
-- 
2.25.3

[PATCH rdma-next v1 27/36] RDMA/mlx5: Separate XRC_TGT QP creation from common flow

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

XRC_TGT QP doesn't fail into kernel or user flow separation. It is
initiated by the user, but is created through in-kernel verbs flow
and doesn't have PD and udata in similar way to kernel QPs.

So let's separate creation of that QP type from the common flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 158 +++++++++++++++++++++-----------
 1 file changed, 106 insertions(+), 52 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index b2174e0817f5..8890c172f7e5 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -991,8 +991,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		goto err_umem;
 	}
 
-	uid = (attr->qp_type != IB_QPT_XRC_TGT &&
-	       attr->qp_type != IB_QPT_XRC_INI) ? to_mpd(pd)->uid : 0;
+	uid = (attr->qp_type != IB_QPT_XRC_INI) ? to_mpd(pd)->uid : 0;
 	MLX5_SET(create_qp_in, *in, uid, uid);
 	pas = (__be64 *)MLX5_ADDR_OF(create_qp_in, *in, pas);
 	if (ubuffer->umem)
@@ -1913,6 +1912,81 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev,
 	return atomic_mode;
 }
 
+static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev,
+			     struct ib_qp_init_attr *attr,
+			     struct mlx5_ib_qp *qp, struct ib_udata *udata,
+			     u32 uidx)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
+	struct mlx5_core_dev *mdev = dev->mdev;
+	struct mlx5_ib_qp_base *base;
+	unsigned long flags;
+	void *qpc;
+	u32 *in;
+	int err;
+
+	mutex_init(&qp->mutex);
+
+	if (attr->sq_sig_type == IB_SIGNAL_ALL_WR)
+		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
+
+	in = kvzalloc(inlen, GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
+
+	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_XRC);
+	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
+	MLX5_SET(qpc, qpc, pd, to_mpd(devr->p0)->pdn);
+
+	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
+		MLX5_SET(qpc, qpc, block_lb_mc, 1);
+	if (qp->flags & IB_QP_CREATE_CROSS_CHANNEL)
+		MLX5_SET(qpc, qpc, cd_master, 1);
+	if (qp->flags & IB_QP_CREATE_MANAGED_SEND)
+		MLX5_SET(qpc, qpc, cd_slave_send, 1);
+	if (qp->flags & IB_QP_CREATE_MANAGED_RECV)
+		MLX5_SET(qpc, qpc, cd_slave_receive, 1);
+
+	MLX5_SET(qpc, qpc, rq_type, MLX5_SRQ_RQ);
+	MLX5_SET(qpc, qpc, no_sq, 1);
+	MLX5_SET(qpc, qpc, cqn_rcv, to_mcq(devr->c0)->mcq.cqn);
+	MLX5_SET(qpc, qpc, cqn_snd, to_mcq(devr->c0)->mcq.cqn);
+	MLX5_SET(qpc, qpc, srqn_rmpn_xrqn, to_msrq(devr->s0)->msrq.srqn);
+	MLX5_SET(qpc, qpc, xrcd, to_mxrcd(attr->xrcd)->xrcdn);
+	MLX5_SET64(qpc, qpc, dbr_addr, qp->db.dma);
+
+	/* 0xffffff means we ask to work with cqe version 0 */
+	if (MLX5_CAP_GEN(mdev, cqe_version) == MLX5_CQE_VERSION_V1)
+		MLX5_SET(qpc, qpc, user_index, uidx);
+
+	if (qp->flags & IB_QP_CREATE_PCI_WRITE_END_PADDING) {
+		MLX5_SET(qpc, qpc, end_padding_mode,
+			 MLX5_WQ_END_PAD_MODE_ALIGN);
+		/* Special case to clean flag */
+		qp->flags &= ~IB_QP_CREATE_PCI_WRITE_END_PADDING;
+	}
+
+	base = &qp->trans_qp.base;
+	err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
+	kvfree(in);
+	if (err) {
+		destroy_qp_user(dev, NULL, qp, base, udata);
+		return err;
+	}
+
+	base->container_mibqp = qp;
+	base->mqp.event = mlx5_ib_qp_event;
+
+	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
+	list_add_tail(&qp->qps_list, &dev->qp_list);
+	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
+
+	return 0;
+}
+
 static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
 			    struct mlx5_ib_create_qp *ucmd,
@@ -1958,40 +2032,30 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		return err;
 	}
 
-	if (pd) {
-		if (udata) {
-			__u32 max_wqes =
-				1 << MLX5_CAP_GEN(mdev, log_max_qp_sz);
-			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 > max_wqes) {
-				mlx5_ib_dbg(dev, "requested sq_wqe_count (%d) > max allowed (%d)\n",
-					    ucmd->sq_wqe_count, max_wqes);
-				return -EINVAL;
-			}
-			err = create_user_qp(dev, pd, qp, udata, init_attr, &in,
-					     &resp, &inlen, base, ucmd);
-			if (err)
-				mlx5_ib_dbg(dev, "err %d\n", err);
-		} else {
-			err = create_kernel_qp(dev, init_attr, qp, &in, &inlen,
-					       base);
-			if (err)
-				mlx5_ib_dbg(dev, "err %d\n", err);
+	if (udata) {
+		__u32 max_wqes = 1 << MLX5_CAP_GEN(mdev, log_max_qp_sz);
+
+		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 > max_wqes) {
+			mlx5_ib_dbg(
+				dev,
+				"requested sq_wqe_count (%d) > max allowed (%d)\n",
+				ucmd->sq_wqe_count, max_wqes);
+			return -EINVAL;
 		}
+		err = create_user_qp(dev, pd, qp, udata, init_attr, &in, &resp,
+				     &inlen, base, ucmd);
+	} else
+		err = create_kernel_qp(dev, init_attr, qp, &in, &inlen, base);
 
-		if (err)
-			return err;
-	} else {
-		in = kvzalloc(inlen, GFP_KERNEL);
-		if (!in)
-			return -ENOMEM;
-	}
+	if (err)
+		return err;
 
 	if (is_sqp(init_attr->qp_type))
 		qp->port = init_attr->port_num;
@@ -2054,12 +2118,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 
 	/* Set default resources */
 	switch (init_attr->qp_type) {
-	case IB_QPT_XRC_TGT:
-		MLX5_SET(qpc, qpc, cqn_rcv, to_mcq(devr->c0)->mcq.cqn);
-		MLX5_SET(qpc, qpc, cqn_snd, to_mcq(devr->c0)->mcq.cqn);
-		MLX5_SET(qpc, qpc, srqn_rmpn_xrqn, to_msrq(devr->s0)->msrq.srqn);
-		MLX5_SET(qpc, qpc, xrcd, to_mxrcd(init_attr->xrcd)->xrcdn);
-		break;
 	case IB_QPT_XRC_INI:
 		MLX5_SET(qpc, qpc, cqn_rcv, to_mcq(devr->c0)->mcq.cqn);
 		MLX5_SET(qpc, qpc, xrcd, to_mxrcd(devr->x1)->xrcdn);
@@ -2105,16 +2163,12 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		raw_packet_qp_copy_info(qp, &qp->raw_packet_qp);
 		err = create_raw_packet_qp(dev, qp, in, inlen, pd, udata,
 					   &resp);
-	} else {
+	} else
 		err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
-	}
-
-	if (err) {
-		mlx5_ib_dbg(dev, "create qp failed\n");
-		goto err_create;
-	}
 
 	kvfree(in);
+	if (err)
+		goto err_create;
 
 	base->container_mibqp = qp;
 	base->mqp.event = mlx5_ib_qp_event;
@@ -2143,7 +2197,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		destroy_qp_user(dev, pd, qp, base, udata);
 	else
 		destroy_qp_kernel(dev, qp);
-	kvfree(in);
 	return err;
 }
 
@@ -2750,9 +2803,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	if (err)
 		goto free_qp;
 
-	if (qp->type == IB_QPT_XRC_TGT)
-		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
-
 	err = check_qp_attr(dev, qp, init_attr);
 	if (err)
 		goto free_qp;
@@ -2764,12 +2814,16 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 	case MLX5_IB_QPT_DCT:
 		err = create_dct(pd, qp, init_attr, ucmd, uidx);
 		break;
+	case IB_QPT_XRC_TGT:
+		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+		err = create_xrc_tgt_qp(dev, init_attr, qp, udata, uidx);
+		break;
 	default:
 		err = create_qp_common(dev, pd, init_attr, ucmd, udata, qp,
 				       uidx);
 	}
 	if (err) {
-		mlx5_ib_dbg(dev, "create_qp_common failed\n");
+		mlx5_ib_dbg(dev, "create_qp failed %d\n", err);
 		goto free_qp;
 	}
 
-- 
2.25.3

[PATCH rdma-next v1 28/36] RDMA/mlx5: Separate to user/kernel create QP flows

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

The kernel and user create QP flows have very little common code,
separate them to simplify the future work of creating per-type
create_*_qp() functions.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 205 ++++++++++++++++++++++++--------
 1 file changed, 156 insertions(+), 49 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 8890c172f7e5..d5061001217e 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -911,12 +911,12 @@ static int adjust_bfregn(struct mlx5_ib_dev *dev,
 				bfregn % MLX5_NON_FP_BFREGS_PER_UAR;
 }
 
-static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			  struct mlx5_ib_qp *qp, struct ib_udata *udata,
-			  struct ib_qp_init_attr *attr, u32 **in,
-			  struct mlx5_ib_create_qp_resp *resp, int *inlen,
-			  struct mlx5_ib_qp_base *base,
-			  struct mlx5_ib_create_qp *ucmd)
+static int _create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			   struct mlx5_ib_qp *qp, struct ib_udata *udata,
+			   struct ib_qp_init_attr *attr, u32 **in,
+			   struct mlx5_ib_create_qp_resp *resp, int *inlen,
+			   struct mlx5_ib_qp_base *base,
+			   struct mlx5_ib_create_qp *ucmd)
 {
 	struct mlx5_ib_ucontext *context;
 	struct mlx5_ib_ubuffer *ubuffer = &base->ubuffer;
@@ -1083,11 +1083,10 @@ static void *get_sq_edge(struct mlx5_ib_wq *sq, u32 idx)
 	return fragment_end + MLX5_SEND_WQE_BB;
 }
 
-static int create_kernel_qp(struct mlx5_ib_dev *dev,
-			    struct ib_qp_init_attr *init_attr,
-			    struct mlx5_ib_qp *qp,
-			    u32 **in, int *inlen,
-			    struct mlx5_ib_qp_base *base)
+static int _create_kernel_qp(struct mlx5_ib_dev *dev,
+			     struct ib_qp_init_attr *init_attr,
+			     struct mlx5_ib_qp *qp, u32 **in, int *inlen,
+			     struct mlx5_ib_qp_base *base)
 {
 	int uar_index;
 	void *qpc;
@@ -1987,11 +1986,11 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev,
 	return 0;
 }
 
-static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			    struct ib_qp_init_attr *init_attr,
-			    struct mlx5_ib_create_qp *ucmd,
-			    struct ib_udata *udata, struct mlx5_ib_qp *qp,
-			    u32 uidx)
+static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			  struct ib_qp_init_attr *init_attr,
+			  struct mlx5_ib_create_qp *ucmd,
+			  struct ib_udata *udata, struct mlx5_ib_qp *qp,
+			  u32 uidx)
 {
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
@@ -2032,28 +2031,15 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		return err;
 	}
 
-	if (udata) {
-		__u32 max_wqes = 1 << MLX5_CAP_GEN(mdev, log_max_qp_sz);
+	if (ucmd->rq_wqe_shift != qp->rq.wqe_shift ||
+	    ucmd->rq_wqe_count != qp->rq.wqe_cnt)
+		return -EINVAL;
 
-		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 > max_wqes) {
-			mlx5_ib_dbg(
-				dev,
-				"requested sq_wqe_count (%d) > max allowed (%d)\n",
-				ucmd->sq_wqe_count, max_wqes);
-			return -EINVAL;
-		}
-		err = create_user_qp(dev, pd, qp, udata, init_attr, &in, &resp,
-				     &inlen, base, ucmd);
-	} else
-		err = create_kernel_qp(dev, init_attr, qp, &in, &inlen, base);
+	if (ucmd->sq_wqe_count > (1 << MLX5_CAP_GEN(mdev, log_max_qp_sz)))
+		return -EINVAL;
 
+	err = _create_user_qp(dev, pd, qp, udata, init_attr, &in, &resp, &inlen,
+			      base, ucmd);
 	if (err)
 		return err;
 
@@ -2064,12 +2050,7 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 
 	MLX5_SET(qpc, qpc, st, mlx5_st);
 	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
-
-	if (init_attr->qp_type != MLX5_IB_QPT_REG_UMR)
-		MLX5_SET(qpc, qpc, pd, to_mpd(pd ? pd : devr->p0)->pdn);
-	else
-		MLX5_SET(qpc, qpc, latency_sensitive, 1);
-
+	MLX5_SET(qpc, qpc, pd, to_mpd(pd)->pdn);
 
 	if (qp->flags_en & MLX5_QP_FLAG_SIGNATURE)
 		MLX5_SET(qpc, qpc, wq_signature, 1);
@@ -2145,10 +2126,6 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (MLX5_CAP_GEN(mdev, cqe_version) == MLX5_CQE_VERSION_V1)
 		MLX5_SET(qpc, qpc, user_index, uidx);
 
-	/* we use IB_QP_CREATE_IPOIB_UD_LSO to indicates ipoib qp */
-	if (qp->flags & IB_QP_CREATE_IPOIB_UD_LSO)
-		MLX5_SET(qpc, qpc, ulp_stateless_offload_mode, 1);
-
 	if (qp->flags & IB_QP_CREATE_PCI_WRITE_END_PADDING &&
 	    init_attr->qp_type != IB_QPT_RAW_PACKET) {
 		MLX5_SET(qpc, qpc, end_padding_mode,
@@ -2200,6 +2177,133 @@ static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return err;
 }
 
+static int create_kernel_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			    struct ib_qp_init_attr *attr, struct mlx5_ib_qp *qp,
+			    u32 uidx)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
+	struct mlx5_core_dev *mdev = dev->mdev;
+	struct mlx5_ib_cq *send_cq;
+	struct mlx5_ib_cq *recv_cq;
+	unsigned long flags;
+	struct mlx5_ib_qp_base *base;
+	int mlx5_st;
+	void *qpc;
+	u32 *in;
+	int err;
+
+	mutex_init(&qp->mutex);
+	spin_lock_init(&qp->sq.lock);
+	spin_lock_init(&qp->rq.lock);
+
+	mlx5_st = to_mlx5_st(qp->type);
+	if (mlx5_st < 0)
+		return -EINVAL;
+
+	if (attr->sq_sig_type == IB_SIGNAL_ALL_WR)
+		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
+
+	base = &qp->trans_qp.base;
+
+	qp->has_rq = qp_has_rq(attr);
+	err = set_rq_size(dev, &attr->cap, qp->has_rq, qp, NULL);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		return err;
+	}
+
+	err = _create_kernel_qp(dev, attr, qp, &in, &inlen, base);
+	if (err)
+		return err;
+
+	if (is_sqp(attr->qp_type))
+		qp->port = attr->port_num;
+
+	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
+
+	MLX5_SET(qpc, qpc, st, mlx5_st);
+	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
+
+	if (attr->qp_type != MLX5_IB_QPT_REG_UMR)
+		MLX5_SET(qpc, qpc, pd, to_mpd(pd ? pd : devr->p0)->pdn);
+	else
+		MLX5_SET(qpc, qpc, latency_sensitive, 1);
+
+
+	if (qp->flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
+		MLX5_SET(qpc, qpc, block_lb_mc, 1);
+
+	if (qp->rq.wqe_cnt) {
+		MLX5_SET(qpc, qpc, log_rq_stride, qp->rq.wqe_shift - 4);
+		MLX5_SET(qpc, qpc, log_rq_size, ilog2(qp->rq.wqe_cnt));
+	}
+
+	MLX5_SET(qpc, qpc, rq_type, get_rx_type(qp, attr));
+
+	if (qp->sq.wqe_cnt)
+		MLX5_SET(qpc, qpc, log_sq_size, ilog2(qp->sq.wqe_cnt));
+	else
+		MLX5_SET(qpc, qpc, no_sq, 1);
+
+	if (attr->srq) {
+		MLX5_SET(qpc, qpc, xrcd, to_mxrcd(devr->x0)->xrcdn);
+		MLX5_SET(qpc, qpc, srqn_rmpn_xrqn,
+			 to_msrq(attr->srq)->msrq.srqn);
+	} else {
+		MLX5_SET(qpc, qpc, xrcd, to_mxrcd(devr->x1)->xrcdn);
+		MLX5_SET(qpc, qpc, srqn_rmpn_xrqn,
+			 to_msrq(devr->s1)->msrq.srqn);
+	}
+
+	if (attr->send_cq)
+		MLX5_SET(qpc, qpc, cqn_snd, to_mcq(attr->send_cq)->mcq.cqn);
+
+	if (attr->recv_cq)
+		MLX5_SET(qpc, qpc, cqn_rcv, to_mcq(attr->recv_cq)->mcq.cqn);
+
+	MLX5_SET64(qpc, qpc, dbr_addr, qp->db.dma);
+
+	/* 0xffffff means we ask to work with cqe version 0 */
+	if (MLX5_CAP_GEN(mdev, cqe_version) == MLX5_CQE_VERSION_V1)
+		MLX5_SET(qpc, qpc, user_index, uidx);
+
+	/* we use IB_QP_CREATE_IPOIB_UD_LSO to indicates ipoib qp */
+	if (qp->flags & IB_QP_CREATE_IPOIB_UD_LSO)
+		MLX5_SET(qpc, qpc, ulp_stateless_offload_mode, 1);
+
+	err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
+	kvfree(in);
+	if (err)
+		goto err_create;
+
+	base->container_mibqp = qp;
+	base->mqp.event = mlx5_ib_qp_event;
+
+	get_cqs(qp->type, attr->send_cq, attr->recv_cq,
+		&send_cq, &recv_cq);
+	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
+	mlx5_ib_lock_cqs(send_cq, recv_cq);
+	/* Maintain device to QPs access, needed for further handling via reset
+	 * flow
+	 */
+	list_add_tail(&qp->qps_list, &dev->qp_list);
+	/* Maintain CQ to QPs access, needed for further handling via reset flow
+	 */
+	if (send_cq)
+		list_add_tail(&qp->cq_send_list, &send_cq->list_send_qp);
+	if (recv_cq)
+		list_add_tail(&qp->cq_recv_list, &recv_cq->list_recv_qp);
+	mlx5_ib_unlock_cqs(send_cq, recv_cq);
+	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
+
+	return 0;
+
+err_create:
+	destroy_qp_kernel(dev, qp);
+	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)
 {
@@ -2695,7 +2799,7 @@ static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	if (attr->rwq_ind_tbl)
 		return create_rss_raw_qp_tir(pd, qp, attr, ucmd, udata);
 
-	return create_qp_common(dev, pd, attr, ucmd, udata, qp, uidx);
+	return create_user_qp(dev, pd, attr, ucmd, udata, qp, uidx);
 }
 
 static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
@@ -2819,8 +2923,11 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 		err = create_xrc_tgt_qp(dev, init_attr, qp, udata, uidx);
 		break;
 	default:
-		err = create_qp_common(dev, pd, init_attr, ucmd, udata, qp,
-				       uidx);
+		if (udata)
+			err = create_user_qp(dev, pd, init_attr, ucmd, udata,
+					     qp, uidx);
+		else
+			err = create_kernel_qp(dev, pd, init_attr, qp, uidx);
 	}
 	if (err) {
 		mlx5_ib_dbg(dev, "create_qp failed %d\n", err);
-- 
2.25.3

[PATCH rdma-next v1 29/36] RDMA/mlx5: Reduce amount of duplication in QP destroy

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Delete both PD argument and checks if udata was provided, in favour
of unified destroy QP functions.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 70 +++++++++++++++------------------
 1 file changed, 31 insertions(+), 39 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index d5061001217e..6b390b0e43af 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1038,25 +1038,36 @@ static int _create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return err;
 }
 
-static void destroy_qp_user(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			    struct mlx5_ib_qp *qp, struct mlx5_ib_qp_base *base,
-			    struct ib_udata *udata)
+static void destroy_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
+		       struct mlx5_ib_qp_base *base, struct ib_udata *udata)
 {
-	struct mlx5_ib_ucontext *context =
-		rdma_udata_to_drv_context(
-			udata,
-			struct mlx5_ib_ucontext,
-			ibucontext);
+	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
+		udata, struct mlx5_ib_ucontext, ibucontext);
 
-	mlx5_ib_db_unmap_user(context, &qp->db);
-	ib_umem_release(base->ubuffer.umem);
+	if (udata) {
+		/* User QP */
+		mlx5_ib_db_unmap_user(context, &qp->db);
+		ib_umem_release(base->ubuffer.umem);
 
-	/*
-	 * Free only the BFREGs which are handled by the kernel.
-	 * BFREGs of UARs allocated dynamically are handled by user.
-	 */
-	if (qp->bfregn != MLX5_IB_INVALID_BFREG)
-		mlx5_ib_free_bfreg(dev, &context->bfregi, qp->bfregn);
+		/*
+		 * Free only the BFREGs which are handled by the kernel.
+		 * BFREGs of UARs allocated dynamically are handled by user.
+		 */
+		if (qp->bfregn != MLX5_IB_INVALID_BFREG)
+			mlx5_ib_free_bfreg(dev, &context->bfregi, qp->bfregn);
+		return;
+	}
+
+	/* Kernel QP */
+	kvfree(qp->sq.wqe_head);
+	kvfree(qp->sq.w_list);
+	kvfree(qp->sq.wrid);
+	kvfree(qp->sq.wr_data);
+	kvfree(qp->rq.wrid);
+	if (qp->db.db)
+		mlx5_db_free(dev->mdev, &qp->db);
+	if (qp->buf.frags)
+		mlx5_frag_buf_free(dev->mdev, &qp->buf);
 }
 
 /* get_sq_edge - Get the next nearby edge.
@@ -1202,19 +1213,6 @@ static int _create_kernel_qp(struct mlx5_ib_dev *dev,
 	return err;
 }
 
-static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
-{
-	kvfree(qp->sq.wqe_head);
-	kvfree(qp->sq.w_list);
-	kvfree(qp->sq.wrid);
-	kvfree(qp->sq.wr_data);
-	kvfree(qp->rq.wrid);
-	if (qp->db.db)
-		mlx5_db_free(dev->mdev, &qp->db);
-	if (qp->buf.frags)
-		mlx5_frag_buf_free(dev->mdev, &qp->buf);
-}
-
 static u32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
 {
 	if (attr->srq || (qp->type == IB_QPT_XRC_TGT) ||
@@ -1972,7 +1970,7 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev,
 	err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
 	kvfree(in);
 	if (err) {
-		destroy_qp_user(dev, NULL, qp, base, udata);
+		destroy_qp(dev, qp, base, udata);
 		return err;
 	}
 
@@ -2170,10 +2168,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 
 err_create:
-	if (udata)
-		destroy_qp_user(dev, pd, qp, base, udata);
-	else
-		destroy_qp_kernel(dev, qp);
+	destroy_qp(dev, qp, base, udata);
 	return err;
 }
 
@@ -2300,7 +2295,7 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	return 0;
 
 err_create:
-	destroy_qp_kernel(dev, qp);
+	destroy_qp(dev, qp, base, NULL);
 	return err;
 }
 
@@ -2470,10 +2465,7 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 				     base->mqp.qpn);
 	}
 
-	if (udata)
-		destroy_qp_user(dev, &get_pd(qp)->ibpd, qp, base, udata);
-	else
-		destroy_qp_kernel(dev, qp);
+	destroy_qp(dev, qp, base, udata);
 }
 
 static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-- 
2.25.3

[PATCH rdma-next v1 30/36] RDMA/mlx5: Group all create QP parameters to simplify in-kernel interfaces

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

The amount of parameters passed in and out between internal mlx5
create QP functions is too large to easily follow the flow. Change
it by grouping all create QP parameter into one structure.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 148 +++++++++++++++++---------------
 1 file changed, 81 insertions(+), 67 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 6b390b0e43af..3807e1687cb2 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1610,14 +1610,24 @@ static void destroy_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *q
 			     to_mpd(qp->ibqp.pd)->uid);
 }
 
-static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-				 struct ib_qp_init_attr *init_attr,
-				 struct mlx5_ib_create_qp_rss *ucmd,
-				 struct ib_udata *udata)
+struct mlx5_create_qp_params {
+	struct ib_udata *udata;
+	size_t inlen;
+	void *ucmd;
+	u8 is_rss_raw : 1;
+	struct ib_qp_init_attr *attr;
+	u32 uidx;
+};
+
+static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+				 struct mlx5_ib_qp *qp,
+				 struct mlx5_create_qp_params *params)
 {
+	struct ib_qp_init_attr *init_attr = params->attr;
+	struct mlx5_ib_create_qp_rss *ucmd = params->ucmd;
+	struct ib_udata *udata = params->udata;
 	struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
-	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	struct mlx5_ib_create_qp_resp resp = {};
 	int inlen;
 	int outlen;
@@ -1632,7 +1642,8 @@ static int create_rss_raw_qp_tir(struct ib_pd *pd, struct mlx5_ib_qp *qp,
 	u32 tdn = mucontext->tdn;
 	u8 lb_flag = 0;
 
-	min_resp_len = offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
+	min_resp_len =
+		offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
 	if (udata->outlen < min_resp_len)
 		return -EINVAL;
 
@@ -1909,11 +1920,12 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev,
 	return atomic_mode;
 }
 
-static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev,
-			     struct ib_qp_init_attr *attr,
-			     struct mlx5_ib_qp *qp, struct ib_udata *udata,
-			     u32 uidx)
+static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
+			     struct mlx5_create_qp_params *params)
 {
+	struct ib_qp_init_attr *attr = params->attr;
+	struct ib_udata *udata = params->udata;
+	u32 uidx = params->uidx;
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
 	struct mlx5_core_dev *mdev = dev->mdev;
@@ -1985,11 +1997,13 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev,
 }
 
 static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			  struct ib_qp_init_attr *init_attr,
-			  struct mlx5_ib_create_qp *ucmd,
-			  struct ib_udata *udata, struct mlx5_ib_qp *qp,
-			  u32 uidx)
+			  struct mlx5_ib_qp *qp,
+			  struct mlx5_create_qp_params *params)
 {
+	struct ib_qp_init_attr *init_attr = params->attr;
+	struct mlx5_ib_create_qp *ucmd = params->ucmd;
+	struct ib_udata *udata = params->udata;
+	u32 uidx = params->uidx;
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
 	struct mlx5_core_dev *mdev = dev->mdev;
@@ -2173,9 +2187,11 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 }
 
 static int create_kernel_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			    struct ib_qp_init_attr *attr, struct mlx5_ib_qp *qp,
-			    u32 uidx)
+			    struct mlx5_ib_qp *qp,
+			    struct mlx5_create_qp_params *params)
 {
+	struct ib_qp_init_attr *attr = params->attr;
+	u32 uidx = params->uidx;
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
 	struct mlx5_core_dev *mdev = dev->mdev;
@@ -2469,9 +2485,11 @@ static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 }
 
 static int create_dct(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-		      struct ib_qp_init_attr *attr,
-		      struct mlx5_ib_create_qp *ucmd, u32 uidx)
+		      struct mlx5_create_qp_params *params)
 {
+	struct ib_qp_init_attr *attr = params->attr;
+	struct mlx5_ib_create_qp *ucmd = params->ucmd;
+	u32 uidx = params->uidx;
 	void *dctc;
 
 	qp->dct.in = kzalloc(MLX5_ST_SZ_BYTES(create_dct_in), GFP_KERNEL);
@@ -2782,16 +2800,14 @@ static size_t process_udata_size(struct ib_qp_init_attr *attr,
 	return min(ucmd, inlen);
 }
 
-static int create_raw_qp(struct ib_pd *pd, struct mlx5_ib_qp *qp,
-			 struct ib_qp_init_attr *attr, void *ucmd,
-			 struct ib_udata *udata, u32 uidx)
+static int create_raw_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			 struct mlx5_ib_qp *qp,
+			 struct mlx5_create_qp_params *params)
 {
-	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	if (params->is_rss_raw)
+		return create_rss_raw_qp_tir(dev, pd, qp, params);
 
-	if (attr->rwq_ind_tbl)
-		return create_rss_raw_qp_tir(pd, qp, attr, ucmd, udata);
-
-	return create_user_qp(dev, pd, attr, ucmd, udata, qp, uidx);
+	return create_user_qp(dev, pd, qp, params);
 }
 
 static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
@@ -2821,60 +2837,59 @@ static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	return ret;
 }
 
-static int get_qp_uidx(struct mlx5_ib_qp *qp, struct ib_udata *udata,
-		       struct mlx5_ib_create_qp *ucmd,
-		       struct ib_qp_init_attr *attr, u32 *uidx)
+static int get_qp_uidx(struct mlx5_ib_qp *qp,
+		       struct mlx5_create_qp_params *params)
 {
+	struct mlx5_ib_create_qp *ucmd = params->ucmd;
+	struct ib_udata *udata = params->udata;
 	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
 
-	if (attr->rwq_ind_tbl)
+	if (params->is_rss_raw)
 		return 0;
 
-	return get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), uidx);
+	return get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), &params->uidx);
 }
 
-struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
-				struct ib_qp_init_attr *init_attr,
+struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 				struct ib_udata *udata)
 {
-	u32 uidx = MLX5_IB_DEFAULT_UIDX;
+	struct mlx5_create_qp_params params = {};
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
 	enum ib_qp_type type;
-	void *ucmd = NULL;
 	u16 xrcdn = 0;
 	int err;
 
 	dev = pd ? to_mdev(pd->device) :
-		   to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
+		   to_mdev(to_mxrcd(attr->xrcd)->ibxrcd.device);
 
-	err = check_qp_type(dev, init_attr, &type);
-	if (err) {
-		mlx5_ib_dbg(dev, "Unsupported QP type %d\n",
-			    init_attr->qp_type);
+	err = check_qp_type(dev, attr, &type);
+	if (err)
 		return ERR_PTR(err);
-	}
 
-	err = check_valid_flow(dev, pd, init_attr, udata);
+	err = check_valid_flow(dev, pd, attr, udata);
 	if (err)
 		return ERR_PTR(err);
 
-	if (init_attr->qp_type == IB_QPT_GSI)
-		return mlx5_ib_gsi_create_qp(pd, init_attr);
+	if (attr->qp_type == IB_QPT_GSI)
+		return mlx5_ib_gsi_create_qp(pd, attr);
 
-	if (udata) {
-		size_t inlen =
-			process_udata_size(init_attr, udata);
+	params.udata = udata;
+	params.uidx = MLX5_IB_DEFAULT_UIDX;
+	params.attr = attr;
+	params.is_rss_raw = !!attr->rwq_ind_tbl;
 
-		if (!inlen)
+	if (udata) {
+		params.inlen = process_udata_size(attr, udata);
+		if (!params.inlen)
 			return ERR_PTR(-EINVAL);
 
-		ucmd = kzalloc(inlen, GFP_KERNEL);
-		if (!ucmd)
+		params.ucmd = kzalloc(params.inlen, GFP_KERNEL);
+		if (!params.ucmd)
 			return ERR_PTR(-ENOMEM);
 
-		err = ib_copy_from_udata(ucmd, udata, inlen);
+		err = ib_copy_from_udata(params.ucmd, udata, params.inlen);
 		if (err)
 			goto free_ucmd;
 	}
@@ -2887,50 +2902,49 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 
 	qp->type = type;
 	if (udata) {
-		err = process_vendor_flags(dev, qp, ucmd, init_attr);
+		err = process_vendor_flags(dev, qp, params.ucmd, attr);
 		if (err)
 			goto free_qp;
 
-		err = get_qp_uidx(qp, udata, ucmd, init_attr, &uidx);
+		err = get_qp_uidx(qp, &params);
 		if (err)
 			goto free_qp;
 	}
-	err = process_create_flags(dev, qp, init_attr);
+	err = process_create_flags(dev, qp, attr);
 	if (err)
 		goto free_qp;
 
-	err = check_qp_attr(dev, qp, init_attr);
+	err = check_qp_attr(dev, qp, attr);
 	if (err)
 		goto free_qp;
 
 	switch (qp->type) {
 	case IB_QPT_RAW_PACKET:
-		err = create_raw_qp(pd, qp, init_attr, ucmd, udata, uidx);
+		err = create_raw_qp(dev, pd, qp, &params);
 		break;
 	case MLX5_IB_QPT_DCT:
-		err = create_dct(pd, qp, init_attr, ucmd, uidx);
+		err = create_dct(pd, qp, &params);
 		break;
 	case IB_QPT_XRC_TGT:
-		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
-		err = create_xrc_tgt_qp(dev, init_attr, qp, udata, uidx);
+		xrcdn = to_mxrcd(attr->xrcd)->xrcdn;
+		err = create_xrc_tgt_qp(dev, qp, &params);
 		break;
 	default:
 		if (udata)
-			err = create_user_qp(dev, pd, init_attr, ucmd, udata,
-					     qp, uidx);
+			err = create_user_qp(dev, pd, qp, &params);
 		else
-			err = create_kernel_qp(dev, pd, init_attr, qp, uidx);
+			err = create_kernel_qp(dev, pd, qp, &params);
 	}
 	if (err) {
-		mlx5_ib_dbg(dev, "create_qp failed %d\n", err);
+		mlx5_ib_err(dev, "create_qp failed %d\n", err);
 		goto free_qp;
 	}
 
-	kfree(ucmd);
+	kfree(params.ucmd);
 
-	if (is_qp0(init_attr->qp_type))
+	if (is_qp0(attr->qp_type))
 		qp->ibqp.qp_num = 0;
-	else if (is_qp1(init_attr->qp_type))
+	else if (is_qp1(attr->qp_type))
 		qp->ibqp.qp_num = 1;
 	else
 		qp->ibqp.qp_num = qp->trans_qp.base.mqp.qpn;
@@ -2942,7 +2956,7 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
 free_qp:
 	kfree(qp);
 free_ucmd:
-	kfree(ucmd);
+	kfree(params.ucmd);
 	return ERR_PTR(err);
 }
 
-- 
2.25.3

[PATCH rdma-next v1 31/36] RDMA/mlx5: Promote RSS RAW QP flags check to higher level

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Move check that user didn't supplied RSS RAW QP unsupported
command flags to the function that checks all such flags.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 3807e1687cb2..8daa8bc6b9c7 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1652,13 +1652,6 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		return -EOPNOTSUPP;
 	}
 
-	if (ucmd->flags & ~(MLX5_QP_FLAG_TUNNEL_OFFLOADS |
-			   MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC |
-			   MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC)) {
-		mlx5_ib_dbg(dev, "invalid flags\n");
-		return -EOPNOTSUPP;
-	}
-
 	if (ucmd->rx_hash_fields_mask & MLX5_RX_HASH_INNER &&
 	    !(ucmd->flags & MLX5_QP_FLAG_TUNNEL_OFFLOADS)) {
 		mlx5_ib_dbg(dev, "Tunnel offloads must be set for inner RSS\n");
@@ -2687,11 +2680,20 @@ static int process_vendor_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_BFREG_INDEX, true, qp);
 	process_vendor_flag(dev, &flags, MLX5_QP_FLAG_UAR_PAGE_INDEX, true, qp);
 
+	cond = qp->flags_en & ~(MLX5_QP_FLAG_TUNNEL_OFFLOADS |
+				MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_UC |
+				MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC);
+	if (attr->rwq_ind_tbl && cond) {
+		mlx5_ib_dbg(dev, "RSS RAW QP has unsupported flags 0x%X\n",
+			    cond);
+		return -EINVAL;
+	}
+
 	if (flags)
 		mlx5_ib_dbg(dev, "udata has unsupported flags 0x%X\n", flags);
 
 	return (flags) ? -EINVAL : 0;
-}
+	}
 
 static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
 				bool cond, struct mlx5_ib_qp *qp)
-- 
2.25.3

[PATCH rdma-next v1 32/36] RDMA/mlx5: Handle udate outlen checks in one place

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Place in one function all udata size checks. This will allow
us move ib_copy_to_udata() in general place and ensure that
it will be performed after call to the FW.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 48 ++++++++++++++++++++-------------
 1 file changed, 30 insertions(+), 18 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 8daa8bc6b9c7..0d06706e6ce1 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1613,6 +1613,7 @@ static void destroy_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *q
 struct mlx5_create_qp_params {
 	struct ib_udata *udata;
 	size_t inlen;
+	size_t outlen;
 	void *ucmd;
 	u8 is_rss_raw : 1;
 	struct ib_qp_init_attr *attr;
@@ -1638,15 +1639,9 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	void *hfso;
 	u32 selected_fields = 0;
 	u32 outer_l4;
-	size_t min_resp_len;
 	u32 tdn = mucontext->tdn;
 	u8 lb_flag = 0;
 
-	min_resp_len =
-		offsetof(typeof(resp), bfreg_index) + sizeof(resp.bfreg_index);
-	if (udata->outlen < min_resp_len)
-		return -EINVAL;
-
 	if (ucmd->comp_mask) {
 		mlx5_ib_dbg(dev, "invalid comp mask\n");
 		return -EOPNOTSUPP;
@@ -2780,26 +2775,43 @@ static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	return (create_flags) ? -EINVAL : 0;
 }
 
-static size_t process_udata_size(struct ib_qp_init_attr *attr,
-				 struct ib_udata *udata)
+static int process_udata_size(struct mlx5_ib_dev *dev,
+			      struct mlx5_create_qp_params *params)
 {
 	size_t ucmd = sizeof(struct mlx5_ib_create_qp);
+	struct ib_qp_init_attr *attr = params->attr;
+	struct ib_udata *udata = params->udata;
+	size_t outlen = udata->outlen;
 	size_t inlen = udata->inlen;
 
-	if (attr->qp_type == IB_QPT_DRIVER)
-		return (inlen < ucmd) ? 0 : ucmd;
+	params->outlen = min(outlen, sizeof(struct mlx5_ib_create_qp_resp));
+	if (attr->qp_type == IB_QPT_DRIVER) {
+		params->inlen = (inlen < ucmd) ? 0 : ucmd;
+		goto out;
+	}
 
-	if (!attr->rwq_ind_tbl)
-		return ucmd;
+	if (!params->is_rss_raw) {
+		params->inlen = ucmd;
+		goto out;
+	}
 
+	/* RSS RAW QP */
 	if (inlen < offsetofend(struct mlx5_ib_create_qp_rss, flags))
-		return 0;
+		return -EINVAL;
+
+	if (outlen < offsetofend(struct mlx5_ib_create_qp_resp, bfreg_index))
+		return -EINVAL;
 
 	ucmd = sizeof(struct mlx5_ib_create_qp_rss);
 	if (inlen > ucmd && !ib_is_udata_cleared(udata, ucmd, inlen - ucmd))
-		return 0;
+		return -EINVAL;
+
+	params->inlen = min(ucmd, inlen);
+out:
+	if (!params->inlen)
+		mlx5_ib_dbg(dev, "udata is too small or not cleared\n");
 
-	return min(ucmd, inlen);
+	return (params->inlen) ? 0 : -EINVAL;
 }
 
 static int create_raw_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
@@ -2883,9 +2895,9 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	params.is_rss_raw = !!attr->rwq_ind_tbl;
 
 	if (udata) {
-		params.inlen = process_udata_size(attr, udata);
-		if (!params.inlen)
-			return ERR_PTR(-EINVAL);
+		err = process_udata_size(dev, &params);
+		if (err)
+			return ERR_PTR(err);
 
 		params.ucmd = kzalloc(params.inlen, GFP_KERNEL);
 		if (!params.ucmd)
-- 
2.25.3

[PATCH rdma-next v1 33/36] RDMA/mlx5: Copy response to the user in one place

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Update all the places in create QP flows to copy response
to the user in one place.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 113 +++++++++++++++-----------------
 1 file changed, 52 insertions(+), 61 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 0d06706e6ce1..9ca742189281 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1015,17 +1015,8 @@ static int _create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		goto err_free;
 	}
 
-	err = ib_copy_to_udata(udata, resp, min(udata->outlen, sizeof(*resp)));
-	if (err) {
-		mlx5_ib_dbg(dev, "copy failed\n");
-		goto err_unmap;
-	}
-
 	return 0;
 
-err_unmap:
-	mlx5_ib_db_unmap_user(context, &qp->db);
-
 err_free:
 	kvfree(*in);
 
@@ -1551,14 +1542,8 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 
 	qp->trans_qp.base.mqp.qpn = qp->sq.wqe_cnt ? sq->base.mqp.qpn :
 						     rq->base.mqp.qpn;
-	err = ib_copy_to_udata(udata, resp, min(udata->outlen, sizeof(*resp)));
-	if (err)
-		goto err_destroy_tir;
-
 	return 0;
 
-err_destroy_tir:
-	destroy_raw_packet_qp_tir(dev, rq, qp->flags_en, pd);
 err_destroy_rq:
 	destroy_raw_packet_qp_rq(dev, rq);
 err_destroy_sq:
@@ -1618,6 +1603,7 @@ struct mlx5_create_qp_params {
 	u8 is_rss_raw : 1;
 	struct ib_qp_init_attr *attr;
 	u32 uidx;
+	struct mlx5_ib_create_qp_resp resp;
 };
 
 static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
@@ -1629,7 +1615,6 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	struct ib_udata *udata = params->udata;
 	struct mlx5_ib_ucontext *mucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
-	struct mlx5_ib_create_qp_resp resp = {};
 	int inlen;
 	int outlen;
 	int err;
@@ -1662,12 +1647,6 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (qp->flags_en & MLX5_QP_FLAG_TIR_ALLOW_SELF_LB_MC)
 		lb_flag |= MLX5_TIRC_SELF_LB_BLOCK_BLOCK_MULTICAST;
 
-	err = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));
-	if (err) {
-		mlx5_ib_dbg(dev, "copy failed\n");
-		return -EINVAL;
-	}
-
 	inlen = MLX5_ST_SZ_BYTES(create_tir_in);
 	outlen = MLX5_ST_SZ_BYTES(create_tir_out);
 	in = kvzalloc(inlen + outlen, GFP_KERNEL);
@@ -1803,34 +1782,30 @@ static int create_rss_raw_qp_tir(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		goto err;
 
 	if (mucontext->devx_uid) {
-		resp.comp_mask |= MLX5_IB_CREATE_QP_RESP_MASK_TIRN;
-		resp.tirn = qp->rss_qp.tirn;
+		params->resp.comp_mask |= MLX5_IB_CREATE_QP_RESP_MASK_TIRN;
+		params->resp.tirn = qp->rss_qp.tirn;
 		if (MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, sw_owner)) {
-			resp.tir_icm_addr =
+			params->resp.tir_icm_addr =
 				MLX5_GET(create_tir_out, out, icm_address_31_0);
-			resp.tir_icm_addr |= (u64)MLX5_GET(create_tir_out, out,
-							   icm_address_39_32)
-					     << 32;
-			resp.tir_icm_addr |= (u64)MLX5_GET(create_tir_out, out,
-							   icm_address_63_40)
-					     << 40;
-			resp.comp_mask |=
+			params->resp.tir_icm_addr |=
+				(u64)MLX5_GET(create_tir_out, out,
+					      icm_address_39_32)
+				<< 32;
+			params->resp.tir_icm_addr |=
+				(u64)MLX5_GET(create_tir_out, out,
+					      icm_address_63_40)
+				<< 40;
+			params->resp.comp_mask |=
 				MLX5_IB_CREATE_QP_RESP_MASK_TIR_ICM_ADDR;
 		}
 	}
 
-	err = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));
-	if (err)
-		goto err_copy;
-
 	kvfree(in);
 	/* qpn is reserved for that QP */
 	qp->trans_qp.base.mqp.qpn = 0;
 	qp->is_rss = true;
 	return 0;
 
-err_copy:
-	mlx5_cmd_destroy_tir(dev->mdev, qp->rss_qp.tirn, mucontext->devx_uid);
 err:
 	kvfree(in);
 	return err;
@@ -1995,7 +1970,6 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
 	struct mlx5_core_dev *mdev = dev->mdev;
-	struct mlx5_ib_create_qp_resp resp = {};
 	struct mlx5_ib_cq *send_cq;
 	struct mlx5_ib_cq *recv_cq;
 	unsigned long flags;
@@ -2038,8 +2012,8 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 	if (ucmd->sq_wqe_count > (1 << MLX5_CAP_GEN(mdev, log_max_qp_sz)))
 		return -EINVAL;
 
-	err = _create_user_qp(dev, pd, qp, udata, init_attr, &in, &resp, &inlen,
-			      base, ucmd);
+	err = _create_user_qp(dev, pd, qp, udata, init_attr, &in, &params->resp,
+			      &inlen, base, ucmd);
 	if (err)
 		return err;
 
@@ -2139,7 +2113,7 @@ static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		qp->raw_packet_qp.sq.ubuffer.buf_addr = ucmd->sq_buf_addr;
 		raw_packet_qp_copy_info(qp, &qp->raw_packet_qp);
 		err = create_raw_packet_qp(dev, qp, in, inlen, pd, udata,
-					   &resp);
+					   &params->resp);
 	} else
 		err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
 
@@ -2865,6 +2839,25 @@ static int get_qp_uidx(struct mlx5_ib_qp *qp,
 	return get_qp_user_index(ucontext, ucmd, sizeof(*ucmd), &params->uidx);
 }
 
+static int mlx5_ib_destroy_dct(struct mlx5_ib_qp *mqp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(mqp->ibqp.device);
+
+	if (mqp->state == IB_QPS_RTR) {
+		int err;
+
+		err = mlx5_core_destroy_dct(dev, &mqp->dct.mdct);
+		if (err) {
+			mlx5_ib_warn(dev, "failed to destroy DCT %d\n", err);
+			return err;
+		}
+	}
+
+	kfree(mqp->dct.in);
+	kfree(mqp);
+	return 0;
+}
+
 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 				struct ib_udata *udata)
 {
@@ -2955,6 +2948,7 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	}
 
 	kfree(params.ucmd);
+	params.ucmd = NULL;
 
 	if (is_qp0(attr->qp_type))
 		qp->ibqp.qp_num = 0;
@@ -2965,8 +2959,24 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 
 	qp->trans_qp.xrcdn = xrcdn;
 
+	if (udata)
+		/*
+		 * It is safe to copy response for all user create QP flows,
+		 * including MLX5_IB_QPT_DCT, which doesn't need it.
+		 * In that case, resp will be filled with zeros.
+		 */
+		err = ib_copy_to_udata(udata, &params.resp, params.outlen);
+	if (err)
+		goto destroy_qp;
+
 	return &qp->ibqp;
 
+destroy_qp:
+	if (qp->type == MLX5_IB_QPT_DCT)
+		mlx5_ib_destroy_dct(qp);
+	else
+		destroy_qp_common(dev, qp, udata);
+	qp = NULL;
 free_qp:
 	kfree(qp);
 free_ucmd:
@@ -2974,25 +2984,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	return ERR_PTR(err);
 }
 
-static int mlx5_ib_destroy_dct(struct mlx5_ib_qp *mqp)
-{
-	struct mlx5_ib_dev *dev = to_mdev(mqp->ibqp.device);
-
-	if (mqp->state == IB_QPS_RTR) {
-		int err;
-
-		err = mlx5_core_destroy_dct(dev, &mqp->dct.mdct);
-		if (err) {
-			mlx5_ib_warn(dev, "failed to destroy DCT %d\n", err);
-			return err;
-		}
-	}
-
-	kfree(mqp->dct.in);
-	kfree(mqp);
-	return 0;
-}
-
 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 {
 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
-- 
2.25.3

[PATCH rdma-next v1 34/36] RDMA/mlx5: Remove redundant destroy QP call

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

After major refactoring in create QP flow, it is no needed to call
to destroy QP in XRC_TGT flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 9ca742189281..d7983a951e8d 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1887,7 +1887,6 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			     struct mlx5_create_qp_params *params)
 {
 	struct ib_qp_init_attr *attr = params->attr;
-	struct ib_udata *udata = params->udata;
 	u32 uidx = params->uidx;
 	struct mlx5_ib_resources *devr = &dev->devr;
 	int inlen = MLX5_ST_SZ_BYTES(create_qp_in);
@@ -1944,10 +1943,8 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	base = &qp->trans_qp.base;
 	err = mlx5_core_create_qp(dev, &base->mqp, in, inlen);
 	kvfree(in);
-	if (err) {
-		destroy_qp(dev, qp, base, udata);
+	if (err)
 		return err;
-	}
 
 	base->container_mibqp = qp;
 	base->mqp.event = mlx5_ib_qp_event;
-- 
2.25.3

[PATCH rdma-next v1 35/36] RDMA/mlx5: Consolidate into special function all create QP calls

[Leon Romanovsky]

From: Leon Romanovsky <leonro@mellanox.com>

Finish separation to blocks of mlx5_ib_create_qp() functions,
so all internal create QP implementation are located in one place.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 85 +++++++++++++++++++--------------
 1 file changed, 49 insertions(+), 36 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index d7983a951e8d..18c0a25da47a 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1953,6 +1953,7 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	list_add_tail(&qp->qps_list, &dev->qp_list);
 	spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
 
+	qp->trans_qp.xrcdn = to_mxrcd(attr->xrcd)->xrcdn;
 	return 0;
 }
 
@@ -2785,14 +2786,54 @@ static int process_udata_size(struct mlx5_ib_dev *dev,
 	return (params->inlen) ? 0 : -EINVAL;
 }
 
-static int create_raw_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
-			 struct mlx5_ib_qp *qp,
-			 struct mlx5_create_qp_params *params)
+static int create_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+		     struct mlx5_ib_qp *qp,
+		     struct mlx5_create_qp_params *params)
 {
-	if (params->is_rss_raw)
-		return create_rss_raw_qp_tir(dev, pd, qp, params);
+	int err;
+
+	if (params->is_rss_raw) {
+		err = create_rss_raw_qp_tir(dev, pd, qp, params);
+		goto out;
+	}
+
+	if (qp->type == MLX5_IB_QPT_DCT) {
+		err = create_dct(pd, qp, params);
+		goto out;
+	}
+
+	if (qp->type == IB_QPT_XRC_TGT) {
+		err = create_xrc_tgt_qp(dev, qp, params);
+		goto out;
+	}
+
+	if (params->udata)
+		err = create_user_qp(dev, pd, qp, params);
+	else
+		err = create_kernel_qp(dev, pd, qp, params);
+
+out:
+	if (err) {
+		mlx5_ib_err(dev, "Create QP type %d failed\n", qp->type);
+		return err;
+	}
+
+	if (is_qp0(qp->type))
+		qp->ibqp.qp_num = 0;
+	else if (is_qp1(qp->type))
+		qp->ibqp.qp_num = 1;
+	else
+		qp->ibqp.qp_num = qp->trans_qp.base.mqp.qpn;
+
+	mlx5_ib_dbg(dev,
+		"QP type %d, ib qpn 0x%X, mlx qpn 0x%x, rcqn 0x%x, scqn 0x%x\n",
+		qp->type, qp->ibqp.qp_num, qp->trans_qp.base.mqp.qpn,
+		params->attr->recv_cq ? to_mcq(params->attr->recv_cq)->mcq.cqn :
+					-1,
+		params->attr->send_cq ? to_mcq(params->attr->send_cq)->mcq.cqn :
+					-1);
 
-	return create_user_qp(dev, pd, qp, params);
+	return 0;
 }
 
 static int check_qp_attr(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
@@ -2862,7 +2903,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	struct mlx5_ib_dev *dev;
 	struct mlx5_ib_qp *qp;
 	enum ib_qp_type type;
-	u16 xrcdn = 0;
 	int err;
 
 	dev = pd ? to_mdev(pd->device) :
@@ -2922,40 +2962,13 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	if (err)
 		goto free_qp;
 
-	switch (qp->type) {
-	case IB_QPT_RAW_PACKET:
-		err = create_raw_qp(dev, pd, qp, &params);
-		break;
-	case MLX5_IB_QPT_DCT:
-		err = create_dct(pd, qp, &params);
-		break;
-	case IB_QPT_XRC_TGT:
-		xrcdn = to_mxrcd(attr->xrcd)->xrcdn;
-		err = create_xrc_tgt_qp(dev, qp, &params);
-		break;
-	default:
-		if (udata)
-			err = create_user_qp(dev, pd, qp, &params);
-		else
-			err = create_kernel_qp(dev, pd, qp, &params);
-	}
-	if (err) {
-		mlx5_ib_err(dev, "create_qp failed %d\n", err);
+	err = create_qp(dev, pd, qp, &params);
+	if (err)
 		goto free_qp;
-	}
 
 	kfree(params.ucmd);
 	params.ucmd = NULL;
 
-	if (is_qp0(attr->qp_type))
-		qp->ibqp.qp_num = 0;
-	else if (is_qp1(attr->qp_type))
-		qp->ibqp.qp_num = 1;
-	else
-		qp->ibqp.qp_num = qp->trans_qp.base.mqp.qpn;
-
-	qp->trans_qp.xrcdn = xrcdn;
-
 	if (udata)
 		/*
 		 * It is safe to copy response for all user create QP flows,
-- 
2.25.3

[PATCH rdma-next v1 36/36] RDMA/mlx5: Verify that QP is created with RQ or SQ

[Leon Romanovsky]

From: Aharon Landau <aharonl@mellanox.com>

RAW packet QP and underlay QP must be created with either
RQ or SQ, check that.

Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Aharon Landau <aharonl@mellanox.com>
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 18c0a25da47a..14f4f0982e4e 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1482,6 +1482,8 @@ static int create_raw_packet_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	u16 uid = to_mpd(pd)->uid;
 	u32 out[MLX5_ST_SZ_DW(create_tir_out)] = {};
 
+	if (!qp->sq.wqe_cnt && !qp->rq.wqe_cnt)
+		return -EINVAL;
 	if (qp->sq.wqe_cnt) {
 		err = create_raw_packet_qp_tis(dev, qp, sq, tdn, pd);
 		if (err)
-- 
2.25.3

Re: [PATCH rdma-next v1 00/36] Refactor mlx5_ib_create_qp

[Jason Gunthorpe]

On Mon, Apr 27, 2020 at 06:46:00PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@mellanox.com>
> 
> Changelog
> v1: * Combined both series to easy apply
>     * Rebsed on top of rdma-next + mlx5-next
>     * Fixed create_qp flags check
> v0: https://lore.kernel.org/lkml/20200420151105.282848-1-leon@kernel.org
>     https://lore.kernel.org/lkml/20200423190303.12856-1-leon@kernel.org
> 
> 
> Hi,
> 
> This is first part of series which tries to return some sanity
> to mlx5_ib_create_qp() function. Such refactoring is required
> to make extension of that function with less worries of breaking
> driver.
> 
> Extra goal of such refactoring is to ensure that QP is allocated
> at the beginning of function and released at the end. It will allow
> us to move QP allocation to be under IB/core responsibility.
> 
> It is based on previously sent [1] "[PATCH mlx5-next 00/24] Mass
> conversion to light mlx5 command interface"
> 
> Thanks
> 
> [1]
> https://lore.kernel.org/linux-rdma/20200420114136.264924-1-leon@kernel.org
> 
> Aharon Landau (1):
>   RDMA/mlx5: Verify that QP is created with RQ or SQ
> 
> Leon Romanovsky (35):
>   RDMA/mlx5: Organize QP types checks in one place
>   RDMA/mlx5: Delete impossible GSI port check
>   RDMA/mlx5: Perform check if QP creation flow is valid
>   RDMA/mlx5: Prepare QP allocation for future removal
>   RDMA/mlx5: Avoid setting redundant NULL for XRC QPs
>   RDMA/mlx5: Set QP subtype immediately when it is known
>   RDMA/mlx5: Separate create QP flows to be based on type
>   RDMA/mlx5: Split scatter CQE configuration for DCT QP
>   RDMA/mlx5: Update all DRIVER QP places to use QP subtype
>   RDMA/mlx5: Move DRIVER QP flags check into separate function
>   RDMA/mlx5: Remove second copy from user for non RSS RAW QPs
>   RDMA/mlx5: Initial separation of RAW_PACKET QP from common flow
>   RDMA/mlx5: Delete create QP flags obfuscation
>   RDMA/mlx5: Process create QP flags in one place
>   RDMA/mlx5: Use flags_en mechanism to mark QP created with WQE
>     signature
>   RDMA/mlx5: Change scatter CQE flag to be set like other vendor flags
>   RDMA/mlx5: Return all configured create flags through query QP
>   RDMA/mlx5: Process all vendor flags in one place

I grabbed the 'part 1' of this into for-next

Thanks,
Jason

Re: [PATCH rdma-next v1 00/36] Refactor mlx5_ib_create_qp

[Jason Gunthorpe]

On Mon, Apr 27, 2020 at 06:46:00PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@mellanox.com>
> 
> Changelog
> v1: * Combined both series to easy apply
>     * Rebsed on top of rdma-next + mlx5-next
>     * Fixed create_qp flags check
> v0: https://lore.kernel.org/lkml/20200420151105.282848-1-leon@kernel.org
>     https://lore.kernel.org/lkml/20200423190303.12856-1-leon@kernel.org
> 
> 
> Hi,
> 
> This is first part of series which tries to return some sanity
> to mlx5_ib_create_qp() function. Such refactoring is required
> to make extension of that function with less worries of breaking
> driver.
> 
> Extra goal of such refactoring is to ensure that QP is allocated
> at the beginning of function and released at the end. It will allow
> us to move QP allocation to be under IB/core responsibility.
> 
> It is based on previously sent [1] "[PATCH mlx5-next 00/24] Mass
> conversion to light mlx5 command interface"
> 
> Thanks
> 
> [1]
> https://lore.kernel.org/linux-rdma/20200420114136.264924-1-leon@kernel.org
> 
> Aharon Landau (1):
>   RDMA/mlx5: Verify that QP is created with RQ or SQ
> 
> Leon Romanovsky (35):
>   RDMA/mlx5: Organize QP types checks in one place
>   RDMA/mlx5: Delete impossible GSI port check
>   RDMA/mlx5: Perform check if QP creation flow is valid
>   RDMA/mlx5: Prepare QP allocation for future removal
>   RDMA/mlx5: Avoid setting redundant NULL for XRC QPs
>   RDMA/mlx5: Set QP subtype immediately when it is known
>   RDMA/mlx5: Separate create QP flows to be based on type
>   RDMA/mlx5: Split scatter CQE configuration for DCT QP
>   RDMA/mlx5: Update all DRIVER QP places to use QP subtype
>   RDMA/mlx5: Move DRIVER QP flags check into separate function
>   RDMA/mlx5: Remove second copy from user for non RSS RAW QPs
>   RDMA/mlx5: Initial separation of RAW_PACKET QP from common flow
>   RDMA/mlx5: Delete create QP flags obfuscation
>   RDMA/mlx5: Process create QP flags in one place
>   RDMA/mlx5: Use flags_en mechanism to mark QP created with WQE
>     signature
>   RDMA/mlx5: Change scatter CQE flag to be set like other vendor flags
>   RDMA/mlx5: Return all configured create flags through query QP
>   RDMA/mlx5: Process all vendor flags in one place
>   RDMA/mlx5: Delete unsupported QP types
>   RDMA/mlx5: Store QP type in the vendor QP structure
>   RDMA/mlx5: Promote RSS RAW QP attribute check in higher level
>   RDMA/mlx5: Combine copy of create QP command in RSS RAW QP
>   RDMA/mlx5: Remove second user copy in create_user_qp
>   RDMA/mlx5: Rely on existence of udata to separate kernel/user flows
>   RDMA/mlx5: Delete impossible inlen check
>   RDMA/mlx5: Globally parse DEVX UID
>   RDMA/mlx5: Separate XRC_TGT QP creation from common flow
>   RDMA/mlx5: Separate to user/kernel create QP flows
>   RDMA/mlx5: Reduce amount of duplication in QP destroy
>   RDMA/mlx5: Group all create QP parameters to simplify in-kernel
>     interfaces
>   RDMA/mlx5: Promote RSS RAW QP flags check to higher level
>   RDMA/mlx5: Handle udate outlen checks in one place
>   RDMA/mlx5: Copy response to the user in one place
>   RDMA/mlx5: Remove redundant destroy QP call
>   RDMA/mlx5: Consolidate into special function all create QP calls

Part II applies too thanks

Jason