[PATCH rdma-next 00/10] Prepare drivers to move QP allocation to ib_core

[PATCH rdma-next 00/10] Prepare drivers to move QP allocation to ib_core

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

Hi,

This series mainly changes mlx4, mlx5, and mthca drivers to future
change of QP allocation scheme.

The rdmavt driver will be sent separately.

Thanks

Leon Romanovsky (10):
  RDMA/mlx5: Embed GSI QP into general mlx5_ib QP
  RDMA/mlx5: Reuse existing fields in parent QP storage object
  RDMA/mlx5: Change GSI QP to have same creation flow like other QPs
  RDMA/mlx5: Delete not needed GSI QP signature QP type
  RDMA/mlx4: Embed GSI QP into general mlx4_ib QP
  RDMA/mlx4: Prepare QP allocation to remove from the driver
  RDMA/core: Align write and ioctl checks of QP types
  RDMA/drivers: Remove udata check from special QP
  RDMA/mthca: Combine special QP struct with mthca QP
  RDMA/i40iw: Remove intermediate pointer that points to the same struct

 drivers/infiniband/core/uverbs_cmd.c         |  17 +-
 drivers/infiniband/hw/hns/hns_roce_qp.c      |  57 ++--
 drivers/infiniband/hw/i40iw/i40iw_verbs.c    |   9 +-
 drivers/infiniband/hw/i40iw/i40iw_verbs.h    |   1 -
 drivers/infiniband/hw/mlx4/mlx4_ib.h         |  25 +-
 drivers/infiniband/hw/mlx4/qp.c              | 285 ++++++++-----------
 drivers/infiniband/hw/mlx5/gsi.c             | 138 +++------
 drivers/infiniband/hw/mlx5/main.c            |   6 +-
 drivers/infiniband/hw/mlx5/mlx5_ib.h         |  28 +-
 drivers/infiniband/hw/mlx5/qp.c              |  50 ++--
 drivers/infiniband/hw/mthca/mthca_dev.h      |   2 +-
 drivers/infiniband/hw/mthca/mthca_provider.c |  17 +-
 drivers/infiniband/hw/mthca/mthca_provider.h |  27 +-
 drivers/infiniband/hw/mthca/mthca_qp.c       |  75 +++--
 drivers/infiniband/hw/qedr/verbs.c           |   8 -
 drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c |   3 +-
 16 files changed, 329 insertions(+), 419 deletions(-)

--
2.26.2

[PATCH rdma-next 01/10] RDMA/mlx5: Embed GSI QP into general mlx5_ib QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

The GSI QPs have different create flow from the regular QPs, but it is
not really needed. Update the code to use mlx5_ib_qp as a storage class
for all outside of GSI calls.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/gsi.c     | 38 +++++++---------------------
 drivers/infiniband/hw/mlx5/mlx5_ib.h | 25 +++++++++++++++---
 drivers/infiniband/hw/mlx5/qp.c      |  2 +-
 3 files changed, 32 insertions(+), 33 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index d9f300f78a82..0b18558ba7b0 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -39,26 +39,6 @@ struct mlx5_ib_gsi_wr {
 	bool completed:1;
 };
 
-struct mlx5_ib_gsi_qp {
-	struct ib_qp ibqp;
-	struct ib_qp *rx_qp;
-	u8 port_num;
-	struct ib_qp_cap cap;
-	enum ib_sig_type sq_sig_type;
-	/* Serialize qp state modifications */
-	struct mutex mutex;
-	struct ib_cq *cq;
-	struct mlx5_ib_gsi_wr *outstanding_wrs;
-	u32 outstanding_pi, outstanding_ci;
-	int num_qps;
-	/* Protects access to the tx_qps. Post send operations synchronize
-	 * with tx_qp creation in setup_qp(). Also protects the
-	 * outstanding_wrs array and indices.
-	 */
-	spinlock_t lock;
-	struct ib_qp **tx_qps;
-};
-
 static struct mlx5_ib_gsi_qp *gsi_qp(struct ib_qp *qp)
 {
 	return container_of(qp, struct mlx5_ib_gsi_qp, ibqp);
@@ -116,6 +96,7 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 				    struct ib_qp_init_attr *init_attr)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_qp *mqp;
 	struct mlx5_ib_gsi_qp *gsi;
 	struct ib_qp_init_attr hw_init_attr = *init_attr;
 	const u8 port_num = init_attr->port_num;
@@ -130,10 +111,11 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 			num_qps = MLX5_MAX_PORTS;
 	}
 
-	gsi = kzalloc(sizeof(*gsi), GFP_KERNEL);
-	if (!gsi)
+	mqp = kzalloc(sizeof(struct mlx5_ib_qp), GFP_KERNEL);
+	if (!mqp)
 		return ERR_PTR(-ENOMEM);
 
+	gsi = &mqp->gsi;
 	gsi->tx_qps = kcalloc(num_qps, sizeof(*gsi->tx_qps), GFP_KERNEL);
 	if (!gsi->tx_qps) {
 		ret = -ENOMEM;
@@ -216,20 +198,18 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 err_free_tx:
 	kfree(gsi->tx_qps);
 err_free:
-	kfree(gsi);
+	kfree(mqp);
 	return ERR_PTR(ret);
 }
 
-int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp)
+int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp)
 {
-	struct mlx5_ib_dev *dev = to_mdev(qp->device);
-	struct mlx5_ib_gsi_qp *gsi = gsi_qp(qp);
+	struct mlx5_ib_dev *dev = to_mdev(mqp->ibqp.device);
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 	const int port_num = gsi->port_num;
 	int qp_index;
 	int ret;
 
-	mlx5_ib_dbg(dev, "destroying GSI QP\n");
-
 	mutex_lock(&dev->devr.mutex);
 	ret = mlx5_ib_destroy_qp(gsi->rx_qp, NULL);
 	if (ret) {
@@ -253,7 +233,7 @@ int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp)
 
 	kfree(gsi->outstanding_wrs);
 	kfree(gsi->tx_qps);
-	kfree(gsi);
+	kfree(mqp);
 
 	return 0;
 }
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 184abd5f493c..7e0eb815724b 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -384,6 +384,26 @@ struct mlx5_ib_dct {
 	u32                     *in;
 };
 
+struct mlx5_ib_gsi_qp {
+	struct ib_qp ibqp;
+	struct ib_qp *rx_qp;
+	u8 port_num;
+	struct ib_qp_cap cap;
+	enum ib_sig_type sq_sig_type;
+	/* Serialize qp state modifications */
+	struct mutex mutex;
+	struct ib_cq *cq;
+	struct mlx5_ib_gsi_wr *outstanding_wrs;
+	u32 outstanding_pi, outstanding_ci;
+	int num_qps;
+	/* Protects access to the tx_qps. Post send operations synchronize
+	 * with tx_qp creation in setup_qp(). Also protects the
+	 * outstanding_wrs array and indices.
+	 */
+	spinlock_t lock;
+	struct ib_qp **tx_qps;
+};
+
 struct mlx5_ib_qp {
 	struct ib_qp		ibqp;
 	union {
@@ -391,6 +411,7 @@ struct mlx5_ib_qp {
 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
 		struct mlx5_ib_rss_qp rss_qp;
 		struct mlx5_ib_dct dct;
+		struct mlx5_ib_gsi_qp gsi;
 	};
 	struct mlx5_frag_buf	buf;
 
@@ -693,8 +714,6 @@ struct mlx5_mr_cache {
 	unsigned long		last_add;
 };
 
-struct mlx5_ib_gsi_qp;
-
 struct mlx5_ib_port_resources {
 	struct mlx5_ib_resources *devr;
 	struct mlx5_ib_gsi_qp *gsi;
@@ -1322,7 +1341,7 @@ void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
 /* GSI QP helper functions */
 struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 				    struct ib_qp_init_attr *init_attr);
-int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp);
+int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
 			  int attr_mask);
 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index e8dd32d6a7ab..92dda0e50f5d 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -3037,7 +3037,7 @@ int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 	struct mlx5_ib_qp *mqp = to_mqp(qp);
 
 	if (unlikely(qp->qp_type == IB_QPT_GSI))
-		return mlx5_ib_gsi_destroy_qp(qp);
+		return mlx5_ib_destroy_gsi(mqp);
 
 	if (mqp->type == MLX5_IB_QPT_DCT)
 		return mlx5_ib_destroy_dct(mqp);
-- 
2.26.2

[PATCH rdma-next 02/10] RDMA/mlx5: Reuse existing fields in parent QP storage object

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

Remove duplication of mlx5_ib_qp and mlx5_ib_gsi_qp fields.
This change returns the memory footprint of mlx5_ib QP to be
as it was before embedding GSI QP.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/gsi.c     | 66 +++++++++++++---------------
 drivers/infiniband/hw/mlx5/main.c    |  6 +--
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  4 --
 3 files changed, 31 insertions(+), 45 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index 0b18558ba7b0..53c2b8a8c0fb 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -39,20 +39,16 @@ struct mlx5_ib_gsi_wr {
 	bool completed:1;
 };
 
-static struct mlx5_ib_gsi_qp *gsi_qp(struct ib_qp *qp)
-{
-	return container_of(qp, struct mlx5_ib_gsi_qp, ibqp);
-}
-
 static bool mlx5_ib_deth_sqpn_cap(struct mlx5_ib_dev *dev)
 {
 	return MLX5_CAP_GEN(dev->mdev, set_deth_sqpn);
 }
 
 /* Call with gsi->lock locked */
-static void generate_completions(struct mlx5_ib_gsi_qp *gsi)
+static void generate_completions(struct mlx5_ib_qp *mqp)
 {
-	struct ib_cq *gsi_cq = gsi->ibqp.send_cq;
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
+	struct ib_cq *gsi_cq = mqp->ibqp.send_cq;
 	struct mlx5_ib_gsi_wr *wr;
 	u32 index;
 
@@ -78,6 +74,7 @@ static void handle_single_completion(struct ib_cq *cq, struct ib_wc *wc)
 	struct mlx5_ib_gsi_qp *gsi = cq->cq_context;
 	struct mlx5_ib_gsi_wr *wr =
 		container_of(wc->wr_cqe, struct mlx5_ib_gsi_wr, cqe);
+	struct mlx5_ib_qp *mqp = container_of(gsi, struct mlx5_ib_qp, gsi);
 	u64 wr_id;
 	unsigned long flags;
 
@@ -86,9 +83,9 @@ static void handle_single_completion(struct ib_cq *cq, struct ib_wc *wc)
 	wr_id = wr->wc.wr_id;
 	wr->wc = *wc;
 	wr->wc.wr_id = wr_id;
-	wr->wc.qp = &gsi->ibqp;
+	wr->wc.qp = &mqp->ibqp;
 
-	generate_completions(gsi);
+	generate_completions(mqp);
 	spin_unlock_irqrestore(&gsi->lock, flags);
 }
 
@@ -130,7 +127,7 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 		goto err_free_tx;
 	}
 
-	mutex_init(&gsi->mutex);
+	mutex_init(&mqp->mutex);
 
 	mutex_lock(&dev->devr.mutex);
 
@@ -145,7 +142,7 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 
 	gsi->cap = init_attr->cap;
 	gsi->sq_sig_type = init_attr->sq_sig_type;
-	gsi->ibqp.qp_num = 1;
+	mqp->ibqp.qp_num = 1;
 	gsi->port_num = port_num;
 
 	gsi->cq = ib_alloc_cq(pd->device, gsi, init_attr->cap.max_send_wr, 0,
@@ -188,7 +185,7 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 
 	mutex_unlock(&dev->devr.mutex);
 
-	return &gsi->ibqp;
+	return &mqp->ibqp;
 
 err_destroy_cq:
 	ib_free_cq(gsi->cq);
@@ -362,56 +359,54 @@ static void setup_qp(struct mlx5_ib_gsi_qp *gsi, u16 qp_index)
 
 static void setup_qps(struct mlx5_ib_gsi_qp *gsi)
 {
+	struct mlx5_ib_dev *dev = to_mdev(gsi->rx_qp->device);
 	u16 qp_index;
 
+	mutex_lock(&dev->devr.mutex);
 	for (qp_index = 0; qp_index < gsi->num_qps; ++qp_index)
 		setup_qp(gsi, qp_index);
+	mutex_unlock(&dev->devr.mutex);
 }
 
 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
 			  int attr_mask)
 {
 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
-	struct mlx5_ib_gsi_qp *gsi = gsi_qp(qp);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 	int ret;
 
 	mlx5_ib_dbg(dev, "modifying GSI QP to state %d\n", attr->qp_state);
 
-	mutex_lock(&gsi->mutex);
 	ret = ib_modify_qp(gsi->rx_qp, attr, attr_mask);
 	if (ret) {
 		mlx5_ib_warn(dev, "unable to modify GSI rx QP: %d\n", ret);
-		goto unlock;
+		return ret;
 	}
 
 	if (to_mqp(gsi->rx_qp)->state == IB_QPS_RTS)
 		setup_qps(gsi);
-
-unlock:
-	mutex_unlock(&gsi->mutex);
-
-	return ret;
+	return 0;
 }
 
 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
 			 int qp_attr_mask,
 			 struct ib_qp_init_attr *qp_init_attr)
 {
-	struct mlx5_ib_gsi_qp *gsi = gsi_qp(qp);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 	int ret;
 
-	mutex_lock(&gsi->mutex);
 	ret = ib_query_qp(gsi->rx_qp, qp_attr, qp_attr_mask, qp_init_attr);
 	qp_init_attr->cap = gsi->cap;
-	mutex_unlock(&gsi->mutex);
-
 	return ret;
 }
 
 /* Call with gsi->lock locked */
-static int mlx5_ib_add_outstanding_wr(struct mlx5_ib_gsi_qp *gsi,
+static int mlx5_ib_add_outstanding_wr(struct mlx5_ib_qp *mqp,
 				      struct ib_ud_wr *wr, struct ib_wc *wc)
 {
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 	struct mlx5_ib_dev *dev = to_mdev(gsi->rx_qp->device);
 	struct mlx5_ib_gsi_wr *gsi_wr;
 
@@ -440,22 +435,21 @@ static int mlx5_ib_add_outstanding_wr(struct mlx5_ib_gsi_qp *gsi,
 }
 
 /* Call with gsi->lock locked */
-static int mlx5_ib_gsi_silent_drop(struct mlx5_ib_gsi_qp *gsi,
-				    struct ib_ud_wr *wr)
+static int mlx5_ib_gsi_silent_drop(struct mlx5_ib_qp *mqp, struct ib_ud_wr *wr)
 {
 	struct ib_wc wc = {
 		{ .wr_id = wr->wr.wr_id },
 		.status = IB_WC_SUCCESS,
 		.opcode = IB_WC_SEND,
-		.qp = &gsi->ibqp,
+		.qp = &mqp->ibqp,
 	};
 	int ret;
 
-	ret = mlx5_ib_add_outstanding_wr(gsi, wr, &wc);
+	ret = mlx5_ib_add_outstanding_wr(mqp, wr, &wc);
 	if (ret)
 		return ret;
 
-	generate_completions(gsi);
+	generate_completions(mqp);
 
 	return 0;
 }
@@ -482,7 +476,8 @@ static struct ib_qp *get_tx_qp(struct mlx5_ib_gsi_qp *gsi, struct ib_ud_wr *wr)
 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
 			  const struct ib_send_wr **bad_wr)
 {
-	struct mlx5_ib_gsi_qp *gsi = gsi_qp(qp);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 	struct ib_qp *tx_qp;
 	unsigned long flags;
 	int ret;
@@ -495,14 +490,14 @@ int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
 		spin_lock_irqsave(&gsi->lock, flags);
 		tx_qp = get_tx_qp(gsi, &cur_wr);
 		if (!tx_qp) {
-			ret = mlx5_ib_gsi_silent_drop(gsi, &cur_wr);
+			ret = mlx5_ib_gsi_silent_drop(mqp, &cur_wr);
 			if (ret)
 				goto err;
 			spin_unlock_irqrestore(&gsi->lock, flags);
 			continue;
 		}
 
-		ret = mlx5_ib_add_outstanding_wr(gsi, &cur_wr, NULL);
+		ret = mlx5_ib_add_outstanding_wr(mqp, &cur_wr, NULL);
 		if (ret)
 			goto err;
 
@@ -526,7 +521,8 @@ int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
 			  const struct ib_recv_wr **bad_wr)
 {
-	struct mlx5_ib_gsi_qp *gsi = gsi_qp(qp);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+	struct mlx5_ib_gsi_qp *gsi = &mqp->gsi;
 
 	return ib_post_recv(gsi->rx_qp, wr, bad_wr);
 }
@@ -536,7 +532,5 @@ void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi)
 	if (!gsi)
 		return;
 
-	mutex_lock(&gsi->mutex);
 	setup_qps(gsi);
-	mutex_unlock(&gsi->mutex);
 }
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index f3bece081ad0..8696ad50d8fb 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -2714,9 +2714,7 @@ static void pkey_change_handler(struct work_struct *work)
 		container_of(work, struct mlx5_ib_port_resources,
 			     pkey_change_work);
 
-	mutex_lock(&ports->devr->mutex);
 	mlx5_ib_gsi_pkey_change(ports->gsi);
-	mutex_unlock(&ports->devr->mutex);
 }
 
 static void mlx5_ib_handle_internal_error(struct mlx5_ib_dev *ibdev)
@@ -3142,11 +3140,9 @@ static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
 	atomic_inc(&devr->p0->usecnt);
 	atomic_set(&devr->s1->usecnt, 0);
 
-	for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) {
+	for (port = 0; port < ARRAY_SIZE(devr->ports); ++port)
 		INIT_WORK(&devr->ports[port].pkey_change_work,
 			  pkey_change_handler);
-		devr->ports[port].devr = devr;
-	}
 
 	return 0;
 
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 7e0eb815724b..0ad4824716a9 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -385,13 +385,10 @@ struct mlx5_ib_dct {
 };
 
 struct mlx5_ib_gsi_qp {
-	struct ib_qp ibqp;
 	struct ib_qp *rx_qp;
 	u8 port_num;
 	struct ib_qp_cap cap;
 	enum ib_sig_type sq_sig_type;
-	/* Serialize qp state modifications */
-	struct mutex mutex;
 	struct ib_cq *cq;
 	struct mlx5_ib_gsi_wr *outstanding_wrs;
 	u32 outstanding_pi, outstanding_ci;
@@ -715,7 +712,6 @@ struct mlx5_mr_cache {
 };
 
 struct mlx5_ib_port_resources {
-	struct mlx5_ib_resources *devr;
 	struct mlx5_ib_gsi_qp *gsi;
 	struct work_struct pkey_change_work;
 };
-- 
2.26.2

[PATCH rdma-next 03/10] RDMA/mlx5: Change GSI QP to have same creation flow like other QPs

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

There is no reason to have separate create flow for the GSI QP,
while general create_qp routine has all needed checks and ability to
allocate and free the proper struct mlx5_ib_qp.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/gsi.c     | 42 ++++++++++------------------
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  6 ++--
 drivers/infiniband/hw/mlx5/qp.c      | 36 +++++++++++++-----------
 3 files changed, 38 insertions(+), 46 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index 53c2b8a8c0fb..f5aa1167cb9c 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -89,14 +89,13 @@ static void handle_single_completion(struct ib_cq *cq, struct ib_wc *wc)
 	spin_unlock_irqrestore(&gsi->lock, flags);
 }
 
-struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
-				    struct ib_qp_init_attr *init_attr)
+int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
+		       struct ib_qp_init_attr *attr)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
-	struct mlx5_ib_qp *mqp;
 	struct mlx5_ib_gsi_qp *gsi;
-	struct ib_qp_init_attr hw_init_attr = *init_attr;
-	const u8 port_num = init_attr->port_num;
+	struct ib_qp_init_attr hw_init_attr = *attr;
+	const u8 port_num = attr->port_num;
 	int num_qps = 0;
 	int ret;
 
@@ -108,27 +107,19 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 			num_qps = MLX5_MAX_PORTS;
 	}
 
-	mqp = kzalloc(sizeof(struct mlx5_ib_qp), GFP_KERNEL);
-	if (!mqp)
-		return ERR_PTR(-ENOMEM);
-
 	gsi = &mqp->gsi;
 	gsi->tx_qps = kcalloc(num_qps, sizeof(*gsi->tx_qps), GFP_KERNEL);
-	if (!gsi->tx_qps) {
-		ret = -ENOMEM;
-		goto err_free;
-	}
+	if (!gsi->tx_qps)
+		return -ENOMEM;
 
-	gsi->outstanding_wrs = kcalloc(init_attr->cap.max_send_wr,
-				       sizeof(*gsi->outstanding_wrs),
-				       GFP_KERNEL);
+	gsi->outstanding_wrs =
+		kcalloc(attr->cap.max_send_wr, sizeof(*gsi->outstanding_wrs),
+			GFP_KERNEL);
 	if (!gsi->outstanding_wrs) {
 		ret = -ENOMEM;
 		goto err_free_tx;
 	}
 
-	mutex_init(&mqp->mutex);
-
 	mutex_lock(&dev->devr.mutex);
 
 	if (dev->devr.ports[port_num - 1].gsi) {
@@ -140,12 +131,11 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 	gsi->num_qps = num_qps;
 	spin_lock_init(&gsi->lock);
 
-	gsi->cap = init_attr->cap;
-	gsi->sq_sig_type = init_attr->sq_sig_type;
-	mqp->ibqp.qp_num = 1;
+	gsi->cap = attr->cap;
+	gsi->sq_sig_type = attr->sq_sig_type;
 	gsi->port_num = port_num;
 
-	gsi->cq = ib_alloc_cq(pd->device, gsi, init_attr->cap.max_send_wr, 0,
+	gsi->cq = ib_alloc_cq(pd->device, gsi, attr->cap.max_send_wr, 0,
 			      IB_POLL_SOFTIRQ);
 	if (IS_ERR(gsi->cq)) {
 		mlx5_ib_warn(dev, "unable to create send CQ for GSI QP. error %ld\n",
@@ -181,11 +171,11 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 	INIT_LIST_HEAD(&gsi->rx_qp->rdma_mrs);
 	INIT_LIST_HEAD(&gsi->rx_qp->sig_mrs);
 
-	dev->devr.ports[init_attr->port_num - 1].gsi = gsi;
+	dev->devr.ports[attr->port_num - 1].gsi = gsi;
 
 	mutex_unlock(&dev->devr.mutex);
 
-	return &mqp->ibqp;
+	return 0;
 
 err_destroy_cq:
 	ib_free_cq(gsi->cq);
@@ -194,9 +184,7 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 	kfree(gsi->outstanding_wrs);
 err_free_tx:
 	kfree(gsi->tx_qps);
-err_free:
-	kfree(mqp);
-	return ERR_PTR(ret);
+	return ret;
 }
 
 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp)
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 0ad4824716a9..1cea4940991f 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -1335,8 +1335,8 @@ void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
 
 /* GSI QP helper functions */
-struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
-				    struct ib_qp_init_attr *init_attr);
+int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
+		       struct ib_qp_init_attr *attr);
 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
 			  int attr_mask);
@@ -1375,7 +1375,7 @@ static inline void init_query_mad(struct ib_smp *mad)
 
 static inline int is_qp1(enum ib_qp_type qp_type)
 {
-	return qp_type == MLX5_IB_QPT_HW_GSI;
+	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
 }
 
 #define MLX5_MAX_UMR_SHIFT 16
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 92dda0e50f5d..7e9bf75c33e4 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2785,21 +2785,23 @@ static int create_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		goto out;
 	}
 
-	if (qp->type == MLX5_IB_QPT_DCT) {
+	switch (qp->type) {
+	case MLX5_IB_QPT_DCT:
 		err = create_dct(dev, pd, qp, params);
-		goto out;
-	}
-
-	if (qp->type == IB_QPT_XRC_TGT) {
+		break;
+	case IB_QPT_XRC_TGT:
 		err = create_xrc_tgt_qp(dev, qp, params);
-		goto out;
+		break;
+	case IB_QPT_GSI:
+		err = mlx5_ib_create_gsi(pd, qp, params->attr);
+		break;
+	default:
+		if (params->udata)
+			err = create_user_qp(dev, pd, qp, params);
+		else
+			err = create_kernel_qp(dev, pd, qp, params);
 	}
 
-	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);
@@ -2939,9 +2941,6 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	if (err)
 		return ERR_PTR(err);
 
-	if (attr->qp_type == IB_QPT_GSI)
-		return mlx5_ib_gsi_create_qp(pd, attr);
-
 	params.udata = udata;
 	params.uidx = MLX5_IB_DEFAULT_UIDX;
 	params.attr = attr;
@@ -3010,9 +3009,14 @@ struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attr,
 	return &qp->ibqp;
 
 destroy_qp:
-	if (qp->type == MLX5_IB_QPT_DCT) {
+	switch (qp->type) {
+	case MLX5_IB_QPT_DCT:
 		mlx5_ib_destroy_dct(qp);
-	} else {
+		break;
+	case IB_QPT_GSI:
+		mlx5_ib_destroy_gsi(qp);
+		break;
+	default:
 		/*
 		 * These lines below are temp solution till QP allocation
 		 * will be moved to be under IB/core responsiblity.
-- 
2.26.2

[PATCH rdma-next 04/10] RDMA/mlx5: Delete not needed GSI QP signature QP type

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

GSI QP doesn't need signature QP type because it is initialized
statically to zero, which is IB_SIGNAL_ALL_WR also wr->send_flags isn't
set too. This means that the GSI QP signature QP type can be removed.

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

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index f5aa1167cb9c..7fcad9135276 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -35,7 +35,6 @@
 struct mlx5_ib_gsi_wr {
 	struct ib_cqe cqe;
 	struct ib_wc wc;
-	int send_flags;
 	bool completed:1;
 };
 
@@ -59,10 +58,7 @@ static void generate_completions(struct mlx5_ib_qp *mqp)
 		if (!wr->completed)
 			break;
 
-		if (gsi->sq_sig_type == IB_SIGNAL_ALL_WR ||
-		    wr->send_flags & IB_SEND_SIGNALED)
-			WARN_ON_ONCE(mlx5_ib_generate_wc(gsi_cq, &wr->wc));
-
+		WARN_ON_ONCE(mlx5_ib_generate_wc(gsi_cq, &wr->wc));
 		wr->completed = false;
 	}
 
@@ -132,7 +128,6 @@ int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
 	spin_lock_init(&gsi->lock);
 
 	gsi->cap = attr->cap;
-	gsi->sq_sig_type = attr->sq_sig_type;
 	gsi->port_num = port_num;
 
 	gsi->cq = ib_alloc_cq(pd->device, gsi, attr->cap.max_send_wr, 0,
@@ -236,7 +231,6 @@ static struct ib_qp *create_gsi_ud_qp(struct mlx5_ib_gsi_qp *gsi)
 			.max_send_sge = gsi->cap.max_send_sge,
 			.max_inline_data = gsi->cap.max_inline_data,
 		},
-		.sq_sig_type = gsi->sq_sig_type,
 		.qp_type = IB_QPT_UD,
 		.create_flags = MLX5_IB_QP_CREATE_SQPN_QP1,
 	};
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 1cea4940991f..4921701b666a 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -388,7 +388,6 @@ struct mlx5_ib_gsi_qp {
 	struct ib_qp *rx_qp;
 	u8 port_num;
 	struct ib_qp_cap cap;
-	enum ib_sig_type sq_sig_type;
 	struct ib_cq *cq;
 	struct mlx5_ib_gsi_wr *outstanding_wrs;
 	u32 outstanding_pi, outstanding_ci;
-- 
2.26.2

[PATCH rdma-next 05/10] RDMA/mlx4: Embed GSI QP into general mlx4_ib QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

Refactor the storage struct of mlx4 GSI QP to be embedded
in mlx4_ib QP. This allows to remove internal memory allocation
of QP struct which is hidden inside the mlx4_ib_create_qp() flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx4/mlx4_ib.h |  25 +++-
 drivers/infiniband/hw/mlx4/qp.c      | 166 ++++++++++++---------------
 2 files changed, 100 insertions(+), 91 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 70636f70dd8c..2f9b1a6d01eb 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -299,6 +299,26 @@ struct mlx4_ib_rss {
 	u8			rss_key[MLX4_EN_RSS_KEY_SIZE];
 };
 
+enum {
+	/*
+	 * Largest possible UD header: send with GRH and immediate
+	 * data plus 18 bytes for an Ethernet header with VLAN/802.1Q
+	 * tag.  (LRH would only use 8 bytes, so Ethernet is the
+	 * biggest case)
+	 */
+	MLX4_IB_UD_HEADER_SIZE		= 82,
+	MLX4_IB_LSO_HEADER_SPARE	= 128,
+};
+
+struct mlx4_ib_sqp {
+	int pkey_index;
+	u32 qkey;
+	u32 send_psn;
+	struct ib_ud_header ud_header;
+	u8 header_buf[MLX4_IB_UD_HEADER_SIZE];
+	struct ib_qp *roce_v2_gsi;
+};
+
 struct mlx4_ib_qp {
 	union {
 		struct ib_qp	ibqp;
@@ -344,7 +364,10 @@ struct mlx4_ib_qp {
 	struct mlx4_wqn_range	*wqn_range;
 	/* Number of RSS QP parents that uses this WQ */
 	u32			rss_usecnt;
-	struct mlx4_ib_rss	*rss_ctx;
+	union {
+		struct mlx4_ib_rss *rss_ctx;
+		struct mlx4_ib_sqp *sqp;
+	};
 };
 
 struct mlx4_ib_srq {
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 0cb1a6e2fed4..f30acb563993 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -67,27 +67,6 @@ enum {
 	MLX4_IB_LINK_TYPE_ETH		= 1
 };
 
-enum {
-	/*
-	 * Largest possible UD header: send with GRH and immediate
-	 * data plus 18 bytes for an Ethernet header with VLAN/802.1Q
-	 * tag.  (LRH would only use 8 bytes, so Ethernet is the
-	 * biggest case)
-	 */
-	MLX4_IB_UD_HEADER_SIZE		= 82,
-	MLX4_IB_LSO_HEADER_SPARE	= 128,
-};
-
-struct mlx4_ib_sqp {
-	struct mlx4_ib_qp	qp;
-	int			pkey_index;
-	u32			qkey;
-	u32			send_psn;
-	struct ib_ud_header	ud_header;
-	u8			header_buf[MLX4_IB_UD_HEADER_SIZE];
-	struct ib_qp		*roce_v2_gsi;
-};
-
 enum {
 	MLX4_IB_MIN_SQ_STRIDE	= 6,
 	MLX4_IB_CACHE_LINE_SIZE	= 64,
@@ -123,11 +102,6 @@ enum mlx4_ib_source_type {
 	MLX4_IB_RWQ_SRC	= 1,
 };
 
-static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
-{
-	return container_of(mqp, struct mlx4_ib_sqp, qp);
-}
-
 static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 {
 	if (!mlx4_is_master(dev->dev))
@@ -993,7 +967,6 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 	struct mlx4_ib_dev *dev = to_mdev(pd->device);
 	int qpn;
 	int err;
-	struct mlx4_ib_sqp *sqp = NULL;
 	struct mlx4_ib_qp *qp;
 	struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context(
 		udata, struct mlx4_ib_ucontext, ibucontext);
@@ -1043,17 +1016,18 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 	}
 
 	if (!*caller_qp) {
+		qp = kzalloc(sizeof(struct mlx4_ib_qp), GFP_KERNEL);
+		if (!qp)
+			return -ENOMEM;
+
 		if (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI ||
 		    (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER |
 				MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) {
-			sqp = kzalloc(sizeof(struct mlx4_ib_sqp), GFP_KERNEL);
-			if (!sqp)
-				return -ENOMEM;
-			qp = &sqp->qp;
-		} else {
-			qp = kzalloc(sizeof(struct mlx4_ib_qp), GFP_KERNEL);
-			if (!qp)
+			qp->sqp = kzalloc(sizeof(struct mlx4_ib_sqp), GFP_KERNEL);
+			if (!qp->sqp) {
+				kfree(qp);
 				return -ENOMEM;
+			}
 		}
 		qp->pri.vid = 0xFFFF;
 		qp->alt.vid = 0xFFFF;
@@ -1291,9 +1265,10 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 		mlx4_db_free(dev->dev, &qp->db);
 
 err:
-	if (!sqp && !*caller_qp)
+	if (!*caller_qp) {
+		kfree(qp->sqp);
 		kfree(qp);
-	kfree(sqp);
+	}
 
 	return err;
 }
@@ -1652,7 +1627,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	if (!IS_ERR(ibqp) &&
 	    (init_attr->qp_type == IB_QPT_GSI) &&
 	    !(init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI)) {
-		struct mlx4_ib_sqp *sqp = to_msqp((to_mqp(ibqp)));
+		struct mlx4_ib_qp *qp = to_mqp(ibqp);
+		struct mlx4_ib_sqp *sqp = qp->sqp;
 		int is_eth = rdma_cap_eth_ah(&dev->ib_dev, init_attr->port_num);
 
 		if (is_eth &&
@@ -1664,8 +1640,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				pr_err("Failed to create GSI QP for RoCEv2 (%ld)\n", PTR_ERR(sqp->roce_v2_gsi));
 				sqp->roce_v2_gsi = NULL;
 			} else {
-				sqp = to_msqp(to_mqp(sqp->roce_v2_gsi));
-				sqp->qp.flags |= MLX4_IB_ROCE_V2_GSI_QP;
+				to_mqp(sqp->roce_v2_gsi)->flags |=
+					MLX4_IB_ROCE_V2_GSI_QP;
 			}
 
 			init_attr->create_flags &= ~MLX4_IB_QP_CREATE_ROCE_V2_GSI;
@@ -1699,9 +1675,8 @@ static int _mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 	}
 
 	if (is_sqp(dev, mqp))
-		kfree(to_msqp(mqp));
-	else
-		kfree(mqp);
+		kfree(mqp->sqp);
+	kfree(mqp);
 
 	return 0;
 }
@@ -1711,7 +1686,7 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 	struct mlx4_ib_qp *mqp = to_mqp(qp);
 
 	if (mqp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI) {
-		struct mlx4_ib_sqp *sqp = to_msqp(mqp);
+		struct mlx4_ib_sqp *sqp = mqp->sqp;
 
 		if (sqp->roce_v2_gsi)
 			ib_destroy_qp(sqp->roce_v2_gsi);
@@ -2573,7 +2548,7 @@ static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type,
 		qp->alt_port = attr->alt_port_num;
 
 	if (is_sqp(dev, qp))
-		store_sqp_attrs(to_msqp(qp), attr, attr_mask);
+		store_sqp_attrs(qp->sqp, attr, attr_mask);
 
 	/*
 	 * If we moved QP0 to RTR, bring the IB link up; if we moved
@@ -2850,7 +2825,7 @@ int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	ret = _mlx4_ib_modify_qp(ibqp, attr, attr_mask, udata);
 
 	if (mqp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI) {
-		struct mlx4_ib_sqp *sqp = to_msqp(mqp);
+		struct mlx4_ib_sqp *sqp = mqp->sqp;
 		int err = 0;
 
 		if (sqp->roce_v2_gsi)
@@ -2875,12 +2850,13 @@ static int vf_get_qp0_qkey(struct mlx4_dev *dev, int qpn, u32 *qkey)
 	return -EINVAL;
 }
 
-static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
+static int build_sriov_qp0_header(struct mlx4_ib_qp *qp,
 				  const struct ib_ud_wr *wr,
 				  void *wqe, unsigned *mlx_seg_len)
 {
-	struct mlx4_ib_dev *mdev = to_mdev(sqp->qp.ibqp.device);
-	struct ib_device *ib_dev = &mdev->ib_dev;
+	struct mlx4_ib_dev *mdev = to_mdev(qp->ibqp.device);
+	struct mlx4_ib_sqp *sqp = qp->sqp;
+	struct ib_device *ib_dev = qp->ibqp.device;
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
 	struct mlx4_ib_ah *ah = to_mah(wr->ah);
@@ -2902,12 +2878,12 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 
 	/* for proxy-qp0 sends, need to add in size of tunnel header */
 	/* for tunnel-qp0 sends, tunnel header is already in s/g list */
-	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER)
+	if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER)
 		send_size += sizeof (struct mlx4_ib_tunnel_header);
 
 	ib_ud_header_init(send_size, 1, 0, 0, 0, 0, 0, 0, &sqp->ud_header);
 
-	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) {
+	if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) {
 		sqp->ud_header.lrh.service_level =
 			be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
 		sqp->ud_header.lrh.destination_lid =
@@ -2924,26 +2900,26 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 
 	sqp->ud_header.lrh.virtual_lane    = 0;
 	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
-	err = ib_get_cached_pkey(ib_dev, sqp->qp.port, 0, &pkey);
+	err = ib_get_cached_pkey(ib_dev, qp->port, 0, &pkey);
 	if (err)
 		return err;
 	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
-	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
+	if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
 		sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
 	else
 		sqp->ud_header.bth.destination_qpn =
-			cpu_to_be32(mdev->dev->caps.spec_qps[sqp->qp.port - 1].qp0_tunnel);
+			cpu_to_be32(mdev->dev->caps.spec_qps[qp->port - 1].qp0_tunnel);
 
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
 	if (mlx4_is_master(mdev->dev)) {
-		if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
+		if (mlx4_get_parav_qkey(mdev->dev, qp->mqp.qpn, &qkey))
 			return -EINVAL;
 	} else {
-		if (vf_get_qp0_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
+		if (vf_get_qp0_qkey(mdev->dev, qp->mqp.qpn, &qkey))
 			return -EINVAL;
 	}
 	sqp->ud_header.deth.qkey = cpu_to_be32(qkey);
-	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.mqp.qpn);
+	sqp->ud_header.deth.source_qpn = cpu_to_be32(qp->mqp.qpn);
 
 	sqp->ud_header.bth.opcode        = IB_OPCODE_UD_SEND_ONLY;
 	sqp->ud_header.immediate_present = 0;
@@ -3027,10 +3003,11 @@ static int fill_gid_by_hw_index(struct mlx4_ib_dev *ibdev, u8 port_num,
 }
 
 #define MLX4_ROCEV2_QP1_SPORT 0xC000
-static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
+static int build_mlx_header(struct mlx4_ib_qp *qp, const struct ib_ud_wr *wr,
 			    void *wqe, unsigned *mlx_seg_len)
 {
-	struct ib_device *ib_dev = sqp->qp.ibqp.device;
+	struct mlx4_ib_sqp *sqp = qp->sqp;
+	struct ib_device *ib_dev = qp->ibqp.device;
 	struct mlx4_ib_dev *ibdev = to_mdev(ib_dev);
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
 	struct mlx4_wqe_ctrl_seg *ctrl = wqe;
@@ -3054,7 +3031,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 	for (i = 0; i < wr->wr.num_sge; ++i)
 		send_size += wr->wr.sg_list[i].length;
 
-	is_eth = rdma_port_get_link_layer(sqp->qp.ibqp.device, sqp->qp.port) == IB_LINK_LAYER_ETHERNET;
+	is_eth = rdma_port_get_link_layer(qp->ibqp.device, qp->port) == IB_LINK_LAYER_ETHERNET;
 	is_grh = mlx4_ib_ah_grh_present(ah);
 	if (is_eth) {
 		enum ib_gid_type gid_type;
@@ -3068,9 +3045,9 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 			if (err)
 				return err;
 		} else  {
-			err = fill_gid_by_hw_index(ibdev, sqp->qp.port,
-					    ah->av.ib.gid_index,
-					    &sgid, &gid_type);
+			err = fill_gid_by_hw_index(ibdev, qp->port,
+						   ah->av.ib.gid_index, &sgid,
+						   &gid_type);
 			if (!err) {
 				is_udp = gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP;
 				if (is_udp) {
@@ -3115,13 +3092,18 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 				 * indexes don't necessarily match the hw ones, so
 				 * we must use our own cache
 				 */
-				sqp->ud_header.grh.source_gid.global.subnet_prefix =
-					cpu_to_be64(atomic64_read(&(to_mdev(ib_dev)->sriov.
-								    demux[sqp->qp.port - 1].
-								    subnet_prefix)));
-				sqp->ud_header.grh.source_gid.global.interface_id =
-					to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
-						       guid_cache[ah->av.ib.gid_index];
+				sqp->ud_header.grh.source_gid.global
+					.subnet_prefix =
+					cpu_to_be64(atomic64_read(
+						&(to_mdev(ib_dev)
+							  ->sriov
+							  .demux[qp->port - 1]
+							  .subnet_prefix)));
+				sqp->ud_header.grh.source_gid.global
+					.interface_id =
+					to_mdev(ib_dev)
+						->sriov.demux[qp->port - 1]
+						.guid_cache[ah->av.ib.gid_index];
 			} else {
 				sqp->ud_header.grh.source_gid =
 					ah->ibah.sgid_attr->gid;
@@ -3153,10 +3135,13 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 	mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
 
 	if (!is_eth) {
-		mlx->flags |= cpu_to_be32((!sqp->qp.ibqp.qp_num ? MLX4_WQE_MLX_VL15 : 0) |
-					  (sqp->ud_header.lrh.destination_lid ==
-					   IB_LID_PERMISSIVE ? MLX4_WQE_MLX_SLR : 0) |
-					  (sqp->ud_header.lrh.service_level << 8));
+		mlx->flags |=
+			cpu_to_be32((!qp->ibqp.qp_num ? MLX4_WQE_MLX_VL15 : 0) |
+				    (sqp->ud_header.lrh.destination_lid ==
+						     IB_LID_PERMISSIVE ?
+					     MLX4_WQE_MLX_SLR :
+					     0) |
+				    (sqp->ud_header.lrh.service_level << 8));
 		if (ah->av.ib.port_pd & cpu_to_be32(0x80000000))
 			mlx->flags |= cpu_to_be32(0x1); /* force loopback */
 		mlx->rlid = sqp->ud_header.lrh.destination_lid;
@@ -3202,21 +3187,23 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 			sqp->ud_header.vlan.tag = cpu_to_be16(vlan | pcp);
 		}
 	} else {
-		sqp->ud_header.lrh.virtual_lane    = !sqp->qp.ibqp.qp_num ? 15 :
-							sl_to_vl(to_mdev(ib_dev),
-								 sqp->ud_header.lrh.service_level,
-								 sqp->qp.port);
-		if (sqp->qp.ibqp.qp_num && sqp->ud_header.lrh.virtual_lane == 15)
+		sqp->ud_header.lrh.virtual_lane =
+			!qp->ibqp.qp_num ?
+				15 :
+				sl_to_vl(to_mdev(ib_dev),
+					 sqp->ud_header.lrh.service_level,
+					 qp->port);
+		if (qp->ibqp.qp_num && sqp->ud_header.lrh.virtual_lane == 15)
 			return -EINVAL;
 		if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
 			sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
 	}
 	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
-	if (!sqp->qp.ibqp.qp_num)
-		err = ib_get_cached_pkey(ib_dev, sqp->qp.port, sqp->pkey_index,
+	if (!qp->ibqp.qp_num)
+		err = ib_get_cached_pkey(ib_dev, qp->port, sqp->pkey_index,
 					 &pkey);
 	else
-		err = ib_get_cached_pkey(ib_dev, sqp->qp.port, wr->pkey_index,
+		err = ib_get_cached_pkey(ib_dev, qp->port, wr->pkey_index,
 					 &pkey);
 	if (err)
 		return err;
@@ -3226,7 +3213,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, const struct ib_ud_wr *wr,
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
 	sqp->ud_header.deth.qkey = cpu_to_be32(wr->remote_qkey & 0x80000000 ?
 					       sqp->qkey : wr->remote_qkey);
-	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.ibqp.qp_num);
+	sqp->ud_header.deth.source_qpn = cpu_to_be32(qp->ibqp.qp_num);
 
 	header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);
 
@@ -3549,14 +3536,14 @@ static int _mlx4_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 
 	if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI) {
-		struct mlx4_ib_sqp *sqp = to_msqp(qp);
+		struct mlx4_ib_sqp *sqp = qp->sqp;
 
 		if (sqp->roce_v2_gsi) {
 			struct mlx4_ib_ah *ah = to_mah(ud_wr(wr)->ah);
 			enum ib_gid_type gid_type;
 			union ib_gid gid;
 
-			if (!fill_gid_by_hw_index(mdev, sqp->qp.port,
+			if (!fill_gid_by_hw_index(mdev, qp->port,
 					   ah->av.ib.gid_index,
 					   &gid, &gid_type))
 				qp = (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ?
@@ -3676,8 +3663,8 @@ static int _mlx4_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			break;
 
 		case MLX4_IB_QPT_TUN_SMI_OWNER:
-			err =  build_sriov_qp0_header(to_msqp(qp), ud_wr(wr),
-					ctrl, &seglen);
+			err = build_sriov_qp0_header(qp, ud_wr(wr), ctrl,
+						     &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
@@ -3713,8 +3700,8 @@ static int _mlx4_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			break;
 
 		case MLX4_IB_QPT_PROXY_SMI_OWNER:
-			err = build_sriov_qp0_header(to_msqp(qp), ud_wr(wr),
-					ctrl, &seglen);
+			err = build_sriov_qp0_header(qp, ud_wr(wr), ctrl,
+						     &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
@@ -3747,8 +3734,7 @@ static int _mlx4_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 
 		case MLX4_IB_QPT_SMI:
 		case MLX4_IB_QPT_GSI:
-			err = build_mlx_header(to_msqp(qp), ud_wr(wr), ctrl,
-					&seglen);
+			err = build_mlx_header(qp, ud_wr(wr), ctrl, &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
-- 
2.26.2

[PATCH rdma-next 06/10] RDMA/mlx4: Prepare QP allocation to remove from the driver

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

Since all mlx4 QP have same storage type, move the QP allocation to be
in one place. This change is preparation to removal of such allocation
from the driver.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx4/qp.c | 142 +++++++++++++-------------------
 1 file changed, 56 insertions(+), 86 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index f30acb563993..39c52c311dad 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -630,8 +630,6 @@ static int create_qp_rss(struct mlx4_ib_dev *dev,
 	if (err)
 		goto err_qpn;
 
-	mutex_init(&qp->mutex);
-
 	INIT_LIST_HEAD(&qp->gid_list);
 	INIT_LIST_HEAD(&qp->steering_rules);
 
@@ -670,80 +668,72 @@ static int create_qp_rss(struct mlx4_ib_dev *dev,
 	return err;
 }
 
-static struct ib_qp *_mlx4_ib_create_qp_rss(struct ib_pd *pd,
-					    struct ib_qp_init_attr *init_attr,
-					    struct ib_udata *udata)
+static int _mlx4_ib_create_qp_rss(struct ib_pd *pd, struct mlx4_ib_qp *qp,
+				  struct ib_qp_init_attr *init_attr,
+				  struct ib_udata *udata)
 {
-	struct mlx4_ib_qp *qp;
 	struct mlx4_ib_create_qp_rss ucmd = {};
 	size_t required_cmd_sz;
 	int err;
 
 	if (!udata) {
 		pr_debug("RSS QP with NULL udata\n");
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 	}
 
 	if (udata->outlen)
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 
 	required_cmd_sz = offsetof(typeof(ucmd), reserved1) +
 					sizeof(ucmd.reserved1);
 	if (udata->inlen < required_cmd_sz) {
 		pr_debug("invalid inlen\n");
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 	}
 
 	if (ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata->inlen))) {
 		pr_debug("copy failed\n");
-		return ERR_PTR(-EFAULT);
+		return -EFAULT;
 	}
 
 	if (memchr_inv(ucmd.reserved, 0, sizeof(ucmd.reserved)))
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 
 	if (ucmd.comp_mask || ucmd.reserved1)
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 
 	if (udata->inlen > sizeof(ucmd) &&
 	    !ib_is_udata_cleared(udata, sizeof(ucmd),
 				 udata->inlen - sizeof(ucmd))) {
 		pr_debug("inlen is not supported\n");
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 	}
 
 	if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
 		pr_debug("RSS QP with unsupported QP type %d\n",
 			 init_attr->qp_type);
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 	}
 
 	if (init_attr->create_flags) {
 		pr_debug("RSS QP doesn't support create flags\n");
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 	}
 
 	if (init_attr->send_cq || init_attr->cap.max_send_wr) {
 		pr_debug("RSS QP with unsupported send attributes\n");
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 	}
 
-	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
-	if (!qp)
-		return ERR_PTR(-ENOMEM);
-
 	qp->pri.vid = 0xFFFF;
 	qp->alt.vid = 0xFFFF;
 
 	err = create_qp_rss(to_mdev(pd->device), init_attr, &ucmd, qp);
-	if (err) {
-		kfree(qp);
-		return ERR_PTR(err);
-	}
+	if (err)
+		return err;
 
 	qp->ibqp.qp_num = qp->mqp.qpn;
-
-	return &qp->ibqp;
+	return 0;
 }
 
 /*
@@ -847,7 +837,6 @@ static int create_rq(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 
 	qp->mlx4_ib_qp_type = MLX4_IB_QPT_RAW_PACKET;
 
-	mutex_init(&qp->mutex);
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
 	INIT_LIST_HEAD(&qp->gid_list);
@@ -962,12 +951,11 @@ static int create_rq(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 
 static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 			    struct ib_udata *udata, int sqpn,
-			    struct mlx4_ib_qp **caller_qp)
+			    struct mlx4_ib_qp *qp, bool is_sqp)
 {
 	struct mlx4_ib_dev *dev = to_mdev(pd->device);
 	int qpn;
 	int err;
-	struct mlx4_ib_qp *qp;
 	struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context(
 		udata, struct mlx4_ib_ucontext, ibucontext);
 	enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type;
@@ -1015,28 +1003,20 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 		sqpn = qpn;
 	}
 
-	if (!*caller_qp) {
-		qp = kzalloc(sizeof(struct mlx4_ib_qp), GFP_KERNEL);
-		if (!qp)
-			return -ENOMEM;
-
+	if (is_sqp) {
 		if (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI ||
 		    (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER |
 				MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) {
 			qp->sqp = kzalloc(sizeof(struct mlx4_ib_sqp), GFP_KERNEL);
-			if (!qp->sqp) {
-				kfree(qp);
+			if (!qp->sqp)
 				return -ENOMEM;
-			}
 		}
 		qp->pri.vid = 0xFFFF;
 		qp->alt.vid = 0xFFFF;
-	} else
-		qp = *caller_qp;
+	}
 
 	qp->mlx4_ib_qp_type = qp_type;
 
-	mutex_init(&qp->mutex);
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
 	INIT_LIST_HEAD(&qp->gid_list);
@@ -1211,9 +1191,6 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 
 	qp->mqp.event = mlx4_ib_qp_event;
 
-	if (!*caller_qp)
-		*caller_qp = qp;
-
 	spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
 	mlx4_ib_lock_cqs(to_mcq(init_attr->send_cq),
 			 to_mcq(init_attr->recv_cq));
@@ -1265,11 +1242,7 @@ static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
 		mlx4_db_free(dev->dev, &qp->db);
 
 err:
-	if (!*caller_qp) {
-		kfree(qp->sqp);
-		kfree(qp);
-	}
-
+	kfree(qp->sqp);
 	return err;
 }
 
@@ -1502,17 +1475,16 @@ static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr)
 		return dev->dev->caps.spec_qps[attr->port_num - 1].qp1_proxy;
 }
 
-static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
-					struct ib_qp_init_attr *init_attr,
-					struct ib_udata *udata)
+static int _mlx4_ib_create_qp(struct ib_pd *pd, struct mlx4_ib_qp *qp,
+			      struct ib_qp_init_attr *init_attr,
+			      struct ib_udata *udata)
 {
-	struct mlx4_ib_qp *qp = NULL;
 	int err;
 	int sup_u_create_flags = MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK;
 	u16 xrcdn = 0;
 
 	if (init_attr->rwq_ind_tbl)
-		return _mlx4_ib_create_qp_rss(pd, init_attr, udata);
+		return _mlx4_ib_create_qp_rss(pd, qp, init_attr, udata);
 
 	/*
 	 * We only support LSO, vendor flag1, and multicast loopback blocking,
@@ -1524,16 +1496,16 @@ static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
 					MLX4_IB_SRIOV_SQP |
 					MLX4_IB_QP_NETIF |
 					MLX4_IB_QP_CREATE_ROCE_V2_GSI))
-		return ERR_PTR(-EINVAL);
+		return -EINVAL;
 
 	if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
 		if (init_attr->qp_type != IB_QPT_UD)
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 	}
 
 	if (init_attr->create_flags) {
 		if (udata && init_attr->create_flags & ~(sup_u_create_flags))
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 
 		if ((init_attr->create_flags & ~(MLX4_IB_SRIOV_SQP |
 						 MLX4_IB_QP_CREATE_ROCE_V2_GSI  |
@@ -1543,7 +1515,7 @@ static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
 		     init_attr->qp_type > IB_QPT_GSI) ||
 		    (init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI &&
 		     init_attr->qp_type != IB_QPT_GSI))
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 	}
 
 	switch (init_attr->qp_type) {
@@ -1554,31 +1526,22 @@ static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
 		fallthrough;
 	case IB_QPT_XRC_INI:
 		if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
-			return ERR_PTR(-ENOSYS);
+			return -ENOSYS;
 		init_attr->recv_cq = init_attr->send_cq;
 		fallthrough;
 	case IB_QPT_RC:
 	case IB_QPT_UC:
 	case IB_QPT_RAW_PACKET:
-		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
-		if (!qp)
-			return ERR_PTR(-ENOMEM);
+	case IB_QPT_UD:
 		qp->pri.vid = 0xFFFF;
 		qp->alt.vid = 0xFFFF;
-		fallthrough;
-	case IB_QPT_UD:
-	{
-		err = create_qp_common(pd, init_attr, udata, 0, &qp);
-		if (err) {
-			kfree(qp);
-			return ERR_PTR(err);
-		}
+		err = create_qp_common(pd, init_attr, udata, 0, qp, false);
+		if (err)
+			return err;
 
 		qp->ibqp.qp_num = qp->mqp.qpn;
 		qp->xrcdn = xrcdn;
-
 		break;
-	}
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:
 	{
@@ -1586,21 +1549,21 @@ static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
 
 		/* Userspace is not allowed to create special QPs: */
 		if (udata)
-			return ERR_PTR(-EINVAL);
+			return -EINVAL;
 		if (init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI) {
 			int res = mlx4_qp_reserve_range(to_mdev(pd->device)->dev,
 							1, 1, &sqpn, 0,
 							MLX4_RES_USAGE_DRIVER);
 
 			if (res)
-				return ERR_PTR(res);
+				return res;
 		} else {
 			sqpn = get_sqp_num(to_mdev(pd->device), init_attr);
 		}
 
-		err = create_qp_common(pd, init_attr, udata, sqpn, &qp);
+		err = create_qp_common(pd, init_attr, udata, sqpn, qp, true);
 		if (err)
-			return ERR_PTR(err);
+			return err;
 
 		qp->port	= init_attr->port_num;
 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 :
@@ -1609,25 +1572,32 @@ static struct ib_qp *_mlx4_ib_create_qp(struct ib_pd *pd,
 	}
 	default:
 		/* Don't support raw QPs */
-		return ERR_PTR(-EOPNOTSUPP);
+		return -EOPNOTSUPP;
 	}
-
-	return &qp->ibqp;
+	return 0;
 }
 
 struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata) {
 	struct ib_device *device = pd ? pd->device : init_attr->xrcd->device;
-	struct ib_qp *ibqp;
 	struct mlx4_ib_dev *dev = to_mdev(device);
+	struct mlx4_ib_qp *qp;
+	int ret;
 
-	ibqp = _mlx4_ib_create_qp(pd, init_attr, udata);
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&qp->mutex);
+	ret = _mlx4_ib_create_qp(pd, qp, init_attr, udata);
+	if (ret) {
+		kfree(qp);
+		return ERR_PTR(ret);
+	}
 
-	if (!IS_ERR(ibqp) &&
-	    (init_attr->qp_type == IB_QPT_GSI) &&
+	if (init_attr->qp_type == IB_QPT_GSI &&
 	    !(init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI)) {
-		struct mlx4_ib_qp *qp = to_mqp(ibqp);
 		struct mlx4_ib_sqp *sqp = qp->sqp;
 		int is_eth = rdma_cap_eth_ah(&dev->ib_dev, init_attr->port_num);
 
@@ -1647,7 +1617,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 			init_attr->create_flags &= ~MLX4_IB_QP_CREATE_ROCE_V2_GSI;
 		}
 	}
-	return ibqp;
+	return &qp->ibqp;
 }
 
 static int _mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
@@ -1674,8 +1644,7 @@ static int _mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 		destroy_qp_common(dev, mqp, MLX4_IB_QP_SRC, udata);
 	}
 
-	if (is_sqp(dev, mqp))
-		kfree(mqp->sqp);
+	kfree(mqp->sqp);
 	kfree(mqp);
 
 	return 0;
@@ -4156,6 +4125,7 @@ struct ib_wq *mlx4_ib_create_wq(struct ib_pd *pd,
 	if (!qp)
 		return ERR_PTR(-ENOMEM);
 
+	mutex_init(&qp->mutex);
 	qp->pri.vid = 0xFFFF;
 	qp->alt.vid = 0xFFFF;
 
-- 
2.26.2

[PATCH rdma-next 07/10] RDMA/core: Align write and ioctl checks of QP types

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

The ioctl flow checks that the user provides only a supported
list of QP types, while write flow didn't do it and relied on
the driver to check it. Align those flows to fail as early as
possible.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/core/uverbs_cmd.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 408a1a4b67f6..b4e383505eda 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -1275,8 +1275,21 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
 	bool has_sq = true;
 	struct ib_device *ib_dev;
 
-	if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
-		return -EPERM;
+	switch (cmd->qp_type) {
+	case IB_QPT_RAW_PACKET:
+		if (!capable(CAP_NET_RAW))
+			return -EPERM;
+		fallthrough;
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_XRC_INI:
+	case IB_QPT_XRC_TGT:
+	case IB_QPT_DRIVER:
+		break;
+	default:
+		return -EINVAL;
+	};
 
 	obj = (struct ib_uqp_object *)uobj_alloc(UVERBS_OBJECT_QP, attrs,
 						 &ib_dev);
-- 
2.26.2

[PATCH rdma-next 08/10] RDMA/drivers: Remove udata check from special QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

GSI QP can't be created from the user space, hence the udata check is
always true (udata == NULL). Remove that check and simplify the flow.

Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/hns/hns_roce_qp.c      | 57 +++++++-------------
 drivers/infiniband/hw/mlx4/qp.c              |  3 --
 drivers/infiniband/hw/mlx5/qp.c              | 12 -----
 drivers/infiniband/hw/mthca/mthca_provider.c |  4 --
 drivers/infiniband/hw/qedr/verbs.c           |  8 ---
 drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c |  3 +-
 6 files changed, 19 insertions(+), 68 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index 975281f03468..7aafbccb5787 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -1014,53 +1014,32 @@ struct ib_qp *hns_roce_create_qp(struct ib_pd *pd,
 	int ret;
 
 	switch (init_attr->qp_type) {
-	case IB_QPT_RC: {
-		hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
-		if (!hr_qp)
-			return ERR_PTR(-ENOMEM);
-
-		ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata,
-						hr_qp);
-		if (ret) {
-			ibdev_err(ibdev, "Create QP 0x%06lx failed(%d)\n",
-				  hr_qp->qpn, ret);
-			kfree(hr_qp);
-			return ERR_PTR(ret);
-		}
-
+	case IB_QPT_RC:
+	case IB_QPT_GSI:
 		break;
+	default:
+		ibdev_err(ibdev, "not support QP type %d\n",
+			  init_attr->qp_type);
+		return ERR_PTR(-EOPNOTSUPP);
 	}
-	case IB_QPT_GSI: {
-		/* Userspace is not allowed to create special QPs: */
-		if (udata) {
-			ibdev_err(ibdev, "not support usr space GSI\n");
-			return ERR_PTR(-EINVAL);
-		}
 
-		hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
-		if (!hr_qp)
-			return ERR_PTR(-ENOMEM);
+	hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
+	if (!hr_qp)
+		return ERR_PTR(-ENOMEM);
 
+	if (init_attr->qp_type == IB_QPT_GSI) {
 		hr_qp->port = init_attr->port_num - 1;
 		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
-
-		ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata,
-						hr_qp);
-		if (ret) {
-			ibdev_err(ibdev, "Create GSI QP failed!\n");
-			kfree(hr_qp);
-			return ERR_PTR(ret);
-		}
-
-		break;
-	}
-	default:{
-		ibdev_err(ibdev, "not support QP type %d\n",
-			  init_attr->qp_type);
-		return ERR_PTR(-EOPNOTSUPP);
-	}
 	}
 
+	ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata, hr_qp);
+	if (ret) {
+		ibdev_err(ibdev, "Create QP type 0x%x failed(%d)\n",
+			  init_attr->qp_type, ret);
+		ibdev_err(ibdev, "Create GSI QP failed!\n");
+		kfree(hr_qp);
+		return ERR_PTR(ret);
+	}
 	return &hr_qp->ibqp;
 }
 
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 39c52c311dad..7f0290112db7 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1547,9 +1547,6 @@ static int _mlx4_ib_create_qp(struct ib_pd *pd, struct mlx4_ib_qp *qp,
 	{
 		int sqpn;
 
-		/* Userspace is not allowed to create special QPs: */
-		if (udata)
-			return -EINVAL;
 		if (init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI) {
 			int res = mlx4_qp_reserve_range(to_mdev(pd->device)->dev,
 							1, 1, &sqpn, 0,
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 7e9bf75c33e4..22f678ba1a68 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2511,18 +2511,6 @@ static int check_valid_flow(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		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.
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 82ee252fe5aa..5dbddf8faf99 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -535,10 +535,6 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:
 	{
-		/* Don't allow userspace to create special QPs */
-		if (udata)
-			return ERR_PTR(-EINVAL);
-
 		qp = kzalloc(sizeof(struct mthca_sqp), GFP_KERNEL);
 		if (!qp)
 			return ERR_PTR(-ENOMEM);
diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index 208d89495ad6..3337d4b6a841 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -1213,14 +1213,6 @@ static int qedr_check_qp_attrs(struct ib_pd *ibpd, struct qedr_dev *dev,
 		return -EINVAL;
 	}
 
-	/* Unprivileged user space cannot create special QP */
-	if (udata && attrs->qp_type == IB_QPT_GSI) {
-		DP_ERR(dev,
-		       "create qp: userspace can't create special QPs of type=0x%x\n",
-		       attrs->qp_type);
-		return -EINVAL;
-	}
-
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c
index 8a385acf6f0c..428256c55065 100644
--- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c
+++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c
@@ -232,8 +232,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd,
 	switch (init_attr->qp_type) {
 	case IB_QPT_GSI:
 		if (init_attr->port_num == 0 ||
-		    init_attr->port_num > pd->device->phys_port_cnt ||
-		    udata) {
+		    init_attr->port_num > pd->device->phys_port_cnt) {
 			dev_warn(&dev->pdev->dev, "invalid queuepair attrs\n");
 			ret = -EINVAL;
 			goto err_qp;
-- 
2.26.2

[PATCH rdma-next 09/10] RDMA/mthca: Combine special QP struct with mthca QP

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

As preparation for the removal of QP allocation logic, we need to ensure
that ib_core allocates the right amount of memory before a call to the
driver create_qp(). It requires from driver to have the same structs for
all types of QPs.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mthca/mthca_dev.h      |  2 +-
 drivers/infiniband/hw/mthca/mthca_provider.c | 13 +++-
 drivers/infiniband/hw/mthca/mthca_provider.h | 27 +++----
 drivers/infiniband/hw/mthca/mthca_qp.c       | 75 ++++++++++----------
 4 files changed, 59 insertions(+), 58 deletions(-)

diff --git a/drivers/infiniband/hw/mthca/mthca_dev.h b/drivers/infiniband/hw/mthca/mthca_dev.h
index 7550e9d03dec..9dbbf4d16796 100644
--- a/drivers/infiniband/hw/mthca/mthca_dev.h
+++ b/drivers/infiniband/hw/mthca/mthca_dev.h
@@ -548,7 +548,7 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
 		    struct ib_qp_cap *cap,
 		    int qpn,
 		    int port,
-		    struct mthca_sqp *sqp,
+		    struct mthca_qp *qp,
 		    struct ib_udata *udata);
 void mthca_free_qp(struct mthca_dev *dev, struct mthca_qp *qp);
 int mthca_create_ah(struct mthca_dev *dev,
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 5dbddf8faf99..31b558ff8218 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -535,9 +535,14 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:
 	{
-		qp = kzalloc(sizeof(struct mthca_sqp), GFP_KERNEL);
+		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
 		if (!qp)
 			return ERR_PTR(-ENOMEM);
+		qp->sqp = kzalloc(sizeof(struct mthca_sqp), GFP_KERNEL);
+		if (!qp->sqp) {
+			kfree(qp);
+			return ERR_PTR(-ENOMEM);
+		}
 
 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
 
@@ -546,7 +551,7 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
 				      to_mcq(init_attr->recv_cq),
 				      init_attr->sq_sig_type, &init_attr->cap,
 				      qp->ibqp.qp_num, init_attr->port_num,
-				      to_msqp(qp), udata);
+				      qp, udata);
 		break;
 	}
 	default:
@@ -555,6 +560,7 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
 	}
 
 	if (err) {
+		kfree(qp->sqp);
 		kfree(qp);
 		return ERR_PTR(err);
 	}
@@ -587,7 +593,8 @@ static int mthca_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
 				    to_mqp(qp)->rq.db_index);
 	}
 	mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
-	kfree(qp);
+	kfree(to_mqp(qp)->sqp);
+	kfree(to_mqp(qp));
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.h b/drivers/infiniband/hw/mthca/mthca_provider.h
index 84c64bff0d92..8a77483bb33c 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.h
+++ b/drivers/infiniband/hw/mthca/mthca_provider.h
@@ -240,6 +240,16 @@ struct mthca_wq {
 	__be32    *db;
 };
 
+struct mthca_sqp {
+	int             pkey_index;
+	u32             qkey;
+	u32             send_psn;
+	struct ib_ud_header ud_header;
+	int             header_buf_size;
+	void           *header_buf;
+	dma_addr_t      header_dma;
+};
+
 struct mthca_qp {
 	struct ib_qp           ibqp;
 	int                    refcount;
@@ -265,17 +275,7 @@ struct mthca_qp {
 
 	wait_queue_head_t      wait;
 	struct mutex	       mutex;
-};
-
-struct mthca_sqp {
-	struct mthca_qp qp;
-	int             pkey_index;
-	u32             qkey;
-	u32             send_psn;
-	struct ib_ud_header ud_header;
-	int             header_buf_size;
-	void           *header_buf;
-	dma_addr_t      header_dma;
+	struct mthca_sqp *sqp;
 };
 
 static inline struct mthca_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
@@ -313,9 +313,4 @@ static inline struct mthca_qp *to_mqp(struct ib_qp *ibqp)
 	return container_of(ibqp, struct mthca_qp, ibqp);
 }
 
-static inline struct mthca_sqp *to_msqp(struct mthca_qp *qp)
-{
-	return container_of(qp, struct mthca_sqp, qp);
-}
-
 #endif /* MTHCA_PROVIDER_H */
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index c6e95d0d760a..08a2a7afafd3 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -809,7 +809,7 @@ static int __mthca_modify_qp(struct ib_qp *ibqp,
 		qp->alt_port = attr->alt_port_num;
 
 	if (is_sqp(dev, qp))
-		store_attrs(to_msqp(qp), attr, attr_mask);
+		store_attrs(qp->sqp, attr, attr_mask);
 
 	/*
 	 * If we moved QP0 to RTR, bring the IB link up; if we moved
@@ -1368,39 +1368,40 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
 		    struct ib_qp_cap *cap,
 		    int qpn,
 		    int port,
-		    struct mthca_sqp *sqp,
+		    struct mthca_qp *qp,
 		    struct ib_udata *udata)
 {
 	u32 mqpn = qpn * 2 + dev->qp_table.sqp_start + port - 1;
 	int err;
 
-	sqp->qp.transport = MLX;
-	err = mthca_set_qp_size(dev, cap, pd, &sqp->qp);
+	qp->transport = MLX;
+	err = mthca_set_qp_size(dev, cap, pd, qp);
 	if (err)
 		return err;
 
-	sqp->header_buf_size = sqp->qp.sq.max * MTHCA_UD_HEADER_SIZE;
-	sqp->header_buf = dma_alloc_coherent(&dev->pdev->dev, sqp->header_buf_size,
-					     &sqp->header_dma, GFP_KERNEL);
-	if (!sqp->header_buf)
+	qp->sqp->header_buf_size = qp->sq.max * MTHCA_UD_HEADER_SIZE;
+	qp->sqp->header_buf =
+		dma_alloc_coherent(&dev->pdev->dev, qp->sqp->header_buf_size,
+				   &qp->sqp->header_dma, GFP_KERNEL);
+	if (!qp->sqp->header_buf)
 		return -ENOMEM;
 
 	spin_lock_irq(&dev->qp_table.lock);
 	if (mthca_array_get(&dev->qp_table.qp, mqpn))
 		err = -EBUSY;
 	else
-		mthca_array_set(&dev->qp_table.qp, mqpn, sqp);
+		mthca_array_set(&dev->qp_table.qp, mqpn, qp->sqp);
 	spin_unlock_irq(&dev->qp_table.lock);
 
 	if (err)
 		goto err_out;
 
-	sqp->qp.port      = port;
-	sqp->qp.qpn       = mqpn;
-	sqp->qp.transport = MLX;
+	qp->port      = port;
+	qp->qpn       = mqpn;
+	qp->transport = MLX;
 
 	err = mthca_alloc_qp_common(dev, pd, send_cq, recv_cq,
-				    send_policy, &sqp->qp, udata);
+				    send_policy, qp, udata);
 	if (err)
 		goto err_out_free;
 
@@ -1421,10 +1422,9 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
 
 	mthca_unlock_cqs(send_cq, recv_cq);
 
- err_out:
-	dma_free_coherent(&dev->pdev->dev, sqp->header_buf_size,
-			  sqp->header_buf, sqp->header_dma);
-
+err_out:
+	dma_free_coherent(&dev->pdev->dev, qp->sqp->header_buf_size,
+			  qp->sqp->header_buf, qp->sqp->header_dma);
 	return err;
 }
 
@@ -1487,20 +1487,19 @@ void mthca_free_qp(struct mthca_dev *dev,
 
 	if (is_sqp(dev, qp)) {
 		atomic_dec(&(to_mpd(qp->ibqp.pd)->sqp_count));
-		dma_free_coherent(&dev->pdev->dev,
-				  to_msqp(qp)->header_buf_size,
-				  to_msqp(qp)->header_buf,
-				  to_msqp(qp)->header_dma);
+		dma_free_coherent(&dev->pdev->dev, qp->sqp->header_buf_size,
+				  qp->sqp->header_buf, qp->sqp->header_dma);
 	} else
 		mthca_free(&dev->qp_table.alloc, qp->qpn);
 }
 
 /* Create UD header for an MLX send and build a data segment for it */
-static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
-			    int ind, const struct ib_ud_wr *wr,
+static int build_mlx_header(struct mthca_dev *dev, struct mthca_qp *qp, int ind,
+			    const struct ib_ud_wr *wr,
 			    struct mthca_mlx_seg *mlx,
 			    struct mthca_data_seg *data)
 {
+	struct mthca_sqp *sqp = qp->sqp;
 	int header_size;
 	int err;
 	u16 pkey;
@@ -1513,7 +1512,7 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	if (err)
 		return err;
 	mlx->flags &= ~cpu_to_be32(MTHCA_NEXT_SOLICIT | 1);
-	mlx->flags |= cpu_to_be32((!sqp->qp.ibqp.qp_num ? MTHCA_MLX_VL15 : 0) |
+	mlx->flags |= cpu_to_be32((!qp->ibqp.qp_num ? MTHCA_MLX_VL15 : 0) |
 				  (sqp->ud_header.lrh.destination_lid ==
 				   IB_LID_PERMISSIVE ? MTHCA_MLX_SLR : 0) |
 				  (sqp->ud_header.lrh.service_level << 8));
@@ -1534,29 +1533,29 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 		return -EINVAL;
 	}
 
-	sqp->ud_header.lrh.virtual_lane    = !sqp->qp.ibqp.qp_num ? 15 : 0;
+	sqp->ud_header.lrh.virtual_lane    = !qp->ibqp.qp_num ? 15 : 0;
 	if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
 		sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
 	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
-	if (!sqp->qp.ibqp.qp_num)
-		ib_get_cached_pkey(&dev->ib_dev, sqp->qp.port,
-				   sqp->pkey_index, &pkey);
+	if (!qp->ibqp.qp_num)
+		ib_get_cached_pkey(&dev->ib_dev, qp->port, sqp->pkey_index,
+				   &pkey);
 	else
-		ib_get_cached_pkey(&dev->ib_dev, sqp->qp.port,
-				   wr->pkey_index, &pkey);
+		ib_get_cached_pkey(&dev->ib_dev, qp->port, wr->pkey_index,
+				   &pkey);
 	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
 	sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
 	sqp->ud_header.deth.qkey = cpu_to_be32(wr->remote_qkey & 0x80000000 ?
 					       sqp->qkey : wr->remote_qkey);
-	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.ibqp.qp_num);
+	sqp->ud_header.deth.source_qpn = cpu_to_be32(qp->ibqp.qp_num);
 
 	header_size = ib_ud_header_pack(&sqp->ud_header,
 					sqp->header_buf +
 					ind * MTHCA_UD_HEADER_SIZE);
 
 	data->byte_count = cpu_to_be32(header_size);
-	data->lkey       = cpu_to_be32(to_mpd(sqp->qp.ibqp.pd)->ntmr.ibmr.lkey);
+	data->lkey       = cpu_to_be32(to_mpd(qp->ibqp.pd)->ntmr.ibmr.lkey);
 	data->addr       = cpu_to_be64(sqp->header_dma +
 				       ind * MTHCA_UD_HEADER_SIZE);
 
@@ -1735,9 +1734,9 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			break;
 
 		case MLX:
-			err = build_mlx_header(dev, to_msqp(qp), ind, ud_wr(wr),
-					       wqe - sizeof (struct mthca_next_seg),
-					       wqe);
+			err = build_mlx_header(
+				dev, qp, ind, ud_wr(wr),
+				wqe - sizeof(struct mthca_next_seg), wqe);
 			if (err) {
 				*bad_wr = wr;
 				goto out;
@@ -2065,9 +2064,9 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			break;
 
 		case MLX:
-			err = build_mlx_header(dev, to_msqp(qp), ind, ud_wr(wr),
-					       wqe - sizeof (struct mthca_next_seg),
-					       wqe);
+			err = build_mlx_header(
+				dev, qp, ind, ud_wr(wr),
+				wqe - sizeof(struct mthca_next_seg), wqe);
 			if (err) {
 				*bad_wr = wr;
 				goto out;
-- 
2.26.2

[PATCH rdma-next 10/10] RDMA/i40iw: Remove intermediate pointer that points to the same struct

[Leon Romanovsky]

From: Leon Romanovsky <leonro@nvidia.com>

There is no real need to have an intermediate pointer for the same
struct, remove it, and use struct directly.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/i40iw/i40iw_verbs.c | 9 +++------
 drivers/infiniband/hw/i40iw/i40iw_verbs.h | 1 -
 2 files changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.c b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
index 4511e175c498..f5d54cb46f78 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_verbs.c
+++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
@@ -380,7 +380,7 @@ void i40iw_free_qp_resources(struct i40iw_device *iwdev,
 	i40iw_free_dma_mem(iwdev->sc_dev.hw, &iwqp->kqp.dma_mem);
 	kfree(iwqp->kqp.wrid_mem);
 	iwqp->kqp.wrid_mem = NULL;
-	kfree(iwqp->allocated_buffer);
+	kfree(iwqp);
 }
 
 /**
@@ -525,7 +525,6 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
 	struct i40iw_create_qp_req req;
 	struct i40iw_create_qp_resp uresp;
 	u32 qp_num = 0;
-	void *mem;
 	enum i40iw_status_code ret;
 	int err_code;
 	int sq_size;
@@ -567,12 +566,10 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
 	init_info.qp_uk_init_info.max_rq_frag_cnt = init_attr->cap.max_recv_sge;
 	init_info.qp_uk_init_info.max_inline_data = init_attr->cap.max_inline_data;
 
-	mem = kzalloc(sizeof(*iwqp), GFP_KERNEL);
-	if (!mem)
+	iwqp = kzalloc(sizeof(*iwqp), GFP_KERNEL);
+	if (!iwqp)
 		return ERR_PTR(-ENOMEM);
 
-	iwqp = (struct i40iw_qp *)mem;
-	iwqp->allocated_buffer = mem;
 	qp = &iwqp->sc_qp;
 	qp->back_qp = (void *)iwqp;
 	qp->push_idx = I40IW_INVALID_PUSH_PAGE_INDEX;
diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.h b/drivers/infiniband/hw/i40iw/i40iw_verbs.h
index 331bc21cbcc7..1bd790575bd4 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_verbs.h
+++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.h
@@ -138,7 +138,6 @@ struct i40iw_qp {
 	struct i40iw_pd *iwpd;
 	struct i40iw_qp_host_ctx_info ctx_info;
 	struct i40iwarp_offload_info iwarp_info;
-	void *allocated_buffer;
 	atomic_t refcount;
 	struct iw_cm_id *cm_id;
 	void *cm_node;
-- 
2.26.2

RE: [PATCH rdma-next 10/10] RDMA/i40iw: Remove intermediate pointer that points to the same struct

[Saleem, Shiraz]

> Subject: [PATCH rdma-next 10/10] RDMA/i40iw: Remove intermediate pointer that
> points to the same struct
> 
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> There is no real need to have an intermediate pointer for the same struct, remove it,
> and use struct directly.
> 
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> ---

Looks good. Thanks!

Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>

Re: [PATCH rdma-next 00/10] Prepare drivers to move QP allocation to ib_core

[Jason Gunthorpe]

On Thu, Sep 10, 2020 at 05:00:36PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> Hi,
> 
> This series mainly changes mlx4, mlx5, and mthca drivers to future
> change of QP allocation scheme.
> 
> The rdmavt driver will be sent separately.
> 
> Thanks
> 
> Leon Romanovsky (10):
>   RDMA/mlx5: Embed GSI QP into general mlx5_ib QP
>   RDMA/mlx5: Reuse existing fields in parent QP storage object
>   RDMA/mlx5: Change GSI QP to have same creation flow like other QPs
>   RDMA/mlx5: Delete not needed GSI QP signature QP type
>   RDMA/mlx4: Embed GSI QP into general mlx4_ib QP
>   RDMA/mlx4: Prepare QP allocation to remove from the driver
>   RDMA/core: Align write and ioctl checks of QP types
>   RDMA/drivers: Remove udata check from special QP
>   RDMA/mthca: Combine special QP struct with mthca QP
>   RDMA/i40iw: Remove intermediate pointer that points to the same struct

This all seems fine, can you re-send it with the fixes? i40iw also
needs a trivial rebase

Thanks,
Jason

Re: [PATCH rdma-next 04/10] RDMA/mlx5: Delete not needed GSI QP signature QP type

[Jason Gunthorpe]

On Thu, Sep 10, 2020 at 05:00:40PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> GSI QP doesn't need signature QP type because it is initialized
> statically to zero, which is IB_SIGNAL_ALL_WR also wr->send_flags isn't
> set too. This means that the GSI QP signature QP type can be removed.

sig here is "signal" not "signature"

Jason

Re: [PATCH rdma-next 07/10] RDMA/core: Align write and ioctl checks of QP types

[Jason Gunthorpe]

On Thu, Sep 10, 2020 at 05:00:43PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> The ioctl flow checks that the user provides only a supported
> list of QP types, while write flow didn't do it and relied on
> the driver to check it. Align those flows to fail as early as
> possible.
> 
> Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
>  drivers/infiniband/core/uverbs_cmd.c | 17 +++++++++++++++--
>  1 file changed, 15 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
> index 408a1a4b67f6..b4e383505eda 100644
> +++ b/drivers/infiniband/core/uverbs_cmd.c
> @@ -1275,8 +1275,21 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
>  	bool has_sq = true;
>  	struct ib_device *ib_dev;
>  
> -	if (cmd->qp_type == IB_QPT_RAW_PACKET && !capable(CAP_NET_RAW))
> -		return -EPERM;
> +	switch (cmd->qp_type) {
> +	case IB_QPT_RAW_PACKET:
> +		if (!capable(CAP_NET_RAW))
> +			return -EPERM;
> +		fallthrough;

I think the consensus is this fallthrough should be break

Jason

Re: [PATCH rdma-next 08/10] RDMA/drivers: Remove udata check from special QP

[Jason Gunthorpe]

On Thu, Sep 10, 2020 at 05:00:44PM +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> GSI QP can't be created from the user space, hence the udata check is
> always true (udata == NULL). Remove that check and simplify the flow.

Most of the cases are 'always false'

Jason