From 012a8ff577f95211c6ffd3b77a94c34ebae009b6 Mon Sep 17 00:00:00 2001
From: Sean Hefty <sean.hefty@intel.com>
Date: Thu, 2 Jun 2011 09:01:33 -0700
Subject: IB/mlx4: Add support for XRC domains

Support creating and destroying XRC domains.  Any sharing of the XRCD
is managed above the low-level driver.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/hw/mlx4/main.c | 59 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 59 insertions(+)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fa643f4f4e28..23e45df9ae36 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -566,6 +566,57 @@ static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 	return 0;
 }
 
+static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata)
+{
+	struct mlx4_ib_xrcd *xrcd;
+	int err;
+
+	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
+	if (!xrcd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
+	if (err)
+		goto err1;
+
+	xrcd->pd = ib_alloc_pd(ibdev);
+	if (IS_ERR(xrcd->pd)) {
+		err = PTR_ERR(xrcd->pd);
+		goto err2;
+	}
+
+	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
+	if (IS_ERR(xrcd->cq)) {
+		err = PTR_ERR(xrcd->cq);
+		goto err3;
+	}
+
+	return &xrcd->ibxrcd;
+
+err3:
+	ib_dealloc_pd(xrcd->pd);
+err2:
+	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
+err1:
+	kfree(xrcd);
+	return ERR_PTR(err);
+}
+
+static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	ib_destroy_cq(to_mxrcd(xrcd)->cq);
+	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
+	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
+	kfree(xrcd);
+
+	return 0;
+}
+
 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 {
 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
@@ -1093,6 +1144,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
 	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
 
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
+		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
+		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
+		ibdev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
+			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+	}
+
 	spin_lock_init(&iboe->lock);
 
 	if (init_node_data(ibdev))
-- 
cgit v1.2.3


From 18abd5ea571608a7c726fc56e21d3e31f9febfd0 Mon Sep 17 00:00:00 2001
From: Sean Hefty <sean.hefty@intel.com>
Date: Thu, 2 Jun 2011 10:43:26 -0700
Subject: IB/mlx4: Add support for XRC SRQs

Allow the user to create XRC SRQs.  This patch is based on a patch
from Jack Morgenstrein <jackm@dev.mellanox.co.il>.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/hw/mlx4/main.c |  3 ++-
 drivers/infiniband/hw/mlx4/srq.c  | 13 +++++++++----
 drivers/net/mlx4/srq.c            | 20 +++++++++++---------
 include/linux/mlx4/device.h       |  4 ++--
 4 files changed, 24 insertions(+), 16 deletions(-)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 23e45df9ae36..42a538e5df36 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1095,7 +1095,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
-		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ);
 
 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
diff --git a/drivers/infiniband/hw/mlx4/srq.c b/drivers/infiniband/hw/mlx4/srq.c
index 4f7f7600d27c..39542f3703b8 100644
--- a/drivers/infiniband/hw/mlx4/srq.c
+++ b/drivers/infiniband/hw/mlx4/srq.c
@@ -76,14 +76,13 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 	struct mlx4_ib_srq *srq;
 	struct mlx4_wqe_srq_next_seg *next;
 	struct mlx4_wqe_data_seg *scatter;
+	u32 cqn;
+	u16 xrcdn;
 	int desc_size;
 	int buf_size;
 	int err;
 	int i;
 
-	if (init_attr->srq_type != IB_SRQT_BASIC)
-		return ERR_PTR(-ENOSYS);
-
 	/* Sanity check SRQ size before proceeding */
 	if (init_attr->attr.max_wr  >= dev->dev->caps.max_srq_wqes ||
 	    init_attr->attr.max_sge >  dev->dev->caps.max_srq_sge)
@@ -177,12 +176,18 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 		}
 	}
 
-	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt,
+	cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
+	xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
+		(u16) dev->dev->caps.reserved_xrcds;
+	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
 			     srq->db.dma, &srq->msrq);
 	if (err)
 		goto err_wrid;
 
 	srq->msrq.event = mlx4_ib_srq_event;
+	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
 
 	if (pd->uobject)
 		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
diff --git a/drivers/net/mlx4/srq.c b/drivers/net/mlx4/srq.c
index 3b07b80a0456..a20b141dbb5c 100644
--- a/drivers/net/mlx4/srq.c
+++ b/drivers/net/mlx4/srq.c
@@ -40,20 +40,20 @@
 struct mlx4_srq_context {
 	__be32			state_logsize_srqn;
 	u8			logstride;
-	u8			reserved1[3];
-	u8			pg_offset;
-	u8			reserved2[3];
-	u32			reserved3;
+	u8			reserved1;
+	__be16			xrcd;
+	__be32			pg_offset_cqn;
+	u32			reserved2;
 	u8			log_page_size;
-	u8			reserved4[2];
+	u8			reserved3[2];
 	u8			mtt_base_addr_h;
 	__be32			mtt_base_addr_l;
 	__be32			pd;
 	__be16			limit_watermark;
 	__be16			wqe_cnt;
-	u16			reserved5;
+	u16			reserved4;
 	__be16			wqe_counter;
-	u32			reserved6;
+	u32			reserved5;
 	__be64			db_rec_addr;
 };
 
@@ -109,8 +109,8 @@ static int mlx4_QUERY_SRQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox
 			    MLX4_CMD_TIME_CLASS_A);
 }
 
-int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
-		   u64 db_rec, struct mlx4_srq *srq)
+int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcd,
+		   struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq)
 {
 	struct mlx4_srq_table *srq_table = &mlx4_priv(dev)->srq_table;
 	struct mlx4_cmd_mailbox *mailbox;
@@ -148,6 +148,8 @@ int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
 	srq_context->state_logsize_srqn = cpu_to_be32((ilog2(srq->max) << 24) |
 						      srq->srqn);
 	srq_context->logstride          = srq->wqe_shift - 4;
+	srq_context->xrcd		= cpu_to_be16(xrcd);
+	srq_context->pg_offset_cqn	= cpu_to_be32(cqn & 0xffffff);
 	srq_context->log_page_size      = mtt->page_shift - MLX4_ICM_PAGE_SHIFT;
 
 	mtt_addr = mlx4_mtt_addr(dev, mtt);
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 6a3478d0b330..9ab0f6a80d60 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -543,8 +543,8 @@ void mlx4_qp_release_range(struct mlx4_dev *dev, int base_qpn, int cnt);
 int mlx4_qp_alloc(struct mlx4_dev *dev, int qpn, struct mlx4_qp *qp);
 void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp);
 
-int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
-		   u64 db_rec, struct mlx4_srq *srq);
+int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcdn,
+		   struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq);
 void mlx4_srq_free(struct mlx4_dev *dev, struct mlx4_srq *srq);
 int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark);
 int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark);
-- 
cgit v1.2.3


From 0a1405da9952a72dd587829a3321695adde7dca1 Mon Sep 17 00:00:00 2001
From: Sean Hefty <sean.hefty@intel.com>
Date: Thu, 2 Jun 2011 11:32:15 -0700
Subject: IB/mlx4: Add support for XRC QPs

Support the creation of XRC INI and TGT QPs.  To handle the case where
a CQ or PD is not provided, we allocate them internally with the xrcd.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/hw/mlx4/main.c    |   2 +
 drivers/infiniband/hw/mlx4/mlx4_ib.h |   1 +
 drivers/infiniband/hw/mlx4/qp.c      | 120 +++++++++++++++++++++++++----------
 drivers/net/mlx4/qp.c                |   3 +
 include/linux/mlx4/qp.h              |   3 +-
 5 files changed, 95 insertions(+), 34 deletions(-)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 42a538e5df36..aec76ad77872 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -128,6 +128,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
+		props->device_cap_flags |= IB_DEVICE_XRC;
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index ce150b0e2cc8..ed80345c99ae 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -145,6 +145,7 @@ struct mlx4_ib_qp {
 	struct mlx4_mtt		mtt;
 	int			buf_size;
 	struct mutex		mutex;
+	u16			xrcdn;
 	u32			flags;
 	u8			port;
 	u8			alt_port;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 3a91d9d8dc51..1bbe64103674 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -302,15 +302,14 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
 }
 
 static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
-		       int is_user, int has_srq, struct mlx4_ib_qp *qp)
+		       int is_user, int has_rq, struct mlx4_ib_qp *qp)
 {
 	/* Sanity check RQ size before proceeding */
 	if (cap->max_recv_wr  > dev->dev->caps.max_wqes  ||
 	    cap->max_recv_sge > dev->dev->caps.max_rq_sg)
 		return -EINVAL;
 
-	if (has_srq) {
-		/* QPs attached to an SRQ should have no RQ */
+	if (!has_rq) {
 		if (cap->max_recv_wr)
 			return -EINVAL;
 
@@ -463,6 +462,14 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev,
 	return 0;
 }
 
+static int qp_has_rq(struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
+		return 0;
+
+	return !attr->srq;
+}
+
 static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
 			    struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp)
@@ -479,7 +486,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
 		qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
 
-	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, !!init_attr->srq, qp);
+	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, qp_has_rq(init_attr), qp);
 	if (err)
 		goto err;
 
@@ -513,7 +520,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err_mtt;
 
-		if (!init_attr->srq) {
+		if (qp_has_rq(init_attr)) {
 			err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
 						  ucmd.db_addr, &qp->db);
 			if (err)
@@ -532,7 +539,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err;
 
-		if (!init_attr->srq) {
+		if (qp_has_rq(init_attr)) {
 			err = mlx4_db_alloc(dev->dev, &qp->db, 0);
 			if (err)
 				goto err;
@@ -575,6 +582,9 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	if (err)
 		goto err_qpn;
 
+	if (init_attr->qp_type == IB_QPT_XRC_TGT)
+		qp->mqp.qpn |= (1 << 23);
+
 	/*
 	 * Hardware wants QPN written in big-endian order (after
 	 * shifting) for send doorbell.  Precompute this value to save
@@ -592,9 +602,8 @@ err_qpn:
 
 err_wrid:
 	if (pd->uobject) {
-		if (!init_attr->srq)
-			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context),
-					      &qp->db);
+		if (qp_has_rq(init_attr))
+			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db);
 	} else {
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
@@ -610,7 +619,7 @@ err_buf:
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
 
 err_db:
-	if (!pd->uobject && !init_attr->srq)
+	if (!pd->uobject && qp_has_rq(init_attr))
 		mlx4_db_free(dev->dev, &qp->db);
 
 err:
@@ -671,6 +680,33 @@ static void del_gid_entries(struct mlx4_ib_qp *qp)
 	}
 }
 
+static struct mlx4_ib_pd *get_pd(struct mlx4_ib_qp *qp)
+{
+	if (qp->ibqp.qp_type == IB_QPT_XRC_TGT)
+		return to_mpd(to_mxrcd(qp->ibqp.xrcd)->pd);
+	else
+		return to_mpd(qp->ibqp.pd);
+}
+
+static void get_cqs(struct mlx4_ib_qp *qp,
+		    struct mlx4_ib_cq **send_cq, struct mlx4_ib_cq **recv_cq)
+{
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_XRC_TGT:
+		*send_cq = to_mcq(to_mxrcd(qp->ibqp.xrcd)->cq);
+		*recv_cq = *send_cq;
+		break;
+	case IB_QPT_XRC_INI:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = *send_cq;
+		break;
+	default:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = to_mcq(qp->ibqp.recv_cq);
+		break;
+	}
+}
+
 static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			      int is_user)
 {
@@ -682,8 +718,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			printk(KERN_WARNING "mlx4_ib: modify QP %06x to RESET failed.\n",
 			       qp->mqp.qpn);
 
-	send_cq = to_mcq(qp->ibqp.send_cq);
-	recv_cq = to_mcq(qp->ibqp.recv_cq);
+	get_cqs(qp, &send_cq, &recv_cq);
 
 	mlx4_ib_lock_cqs(send_cq, recv_cq);
 
@@ -706,7 +741,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 	mlx4_mtt_cleanup(dev->dev, &qp->mtt);
 
 	if (is_user) {
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context),
 					      &qp->db);
 		ib_umem_release(qp->umem);
@@ -714,7 +749,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_db_free(dev->dev, &qp->db);
 	}
 
@@ -725,10 +760,10 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
-	struct mlx4_ib_dev *dev = to_mdev(pd->device);
 	struct mlx4_ib_sqp *sqp;
 	struct mlx4_ib_qp *qp;
 	int err;
+	u16 xrcdn = 0;
 
 	/*
 	 * We only support LSO and multicast loopback blocking, and
@@ -739,10 +774,20 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 		return ERR_PTR(-EINVAL);
 
 	if (init_attr->create_flags &&
-	    (pd->uobject || init_attr->qp_type != IB_QPT_UD))
+	    (udata || init_attr->qp_type != IB_QPT_UD))
 		return ERR_PTR(-EINVAL);
 
 	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+		pd = to_mxrcd(init_attr->xrcd)->pd;
+		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+		init_attr->send_cq = to_mxrcd(init_attr->xrcd)->cq;
+		/* fall through */
+	case IB_QPT_XRC_INI:
+		if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+			return ERR_PTR(-ENOSYS);
+		init_attr->recv_cq = init_attr->send_cq;
+		/* fall through */
 	case IB_QPT_RC:
 	case IB_QPT_UC:
 	case IB_QPT_UD:
@@ -751,13 +796,14 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 		if (!qp)
 			return ERR_PTR(-ENOMEM);
 
-		err = create_qp_common(dev, pd, init_attr, udata, 0, qp);
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, 0, qp);
 		if (err) {
 			kfree(qp);
 			return ERR_PTR(err);
 		}
 
 		qp->ibqp.qp_num = qp->mqp.qpn;
+		qp->xrcdn = xrcdn;
 
 		break;
 	}
@@ -765,7 +811,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	case IB_QPT_GSI:
 	{
 		/* Userspace is not allowed to create special QPs: */
-		if (pd->uobject)
+		if (udata)
 			return ERR_PTR(-EINVAL);
 
 		sqp = kzalloc(sizeof *sqp, GFP_KERNEL);
@@ -774,8 +820,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 
 		qp = &sqp->qp;
 
-		err = create_qp_common(dev, pd, init_attr, udata,
-				       dev->dev->caps.sqp_start +
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
+				       to_mdev(pd->device)->dev->caps.sqp_start +
 				       (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
 				       init_attr->port_num - 1,
 				       qp);
@@ -801,11 +847,13 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
 {
 	struct mlx4_ib_dev *dev = to_mdev(qp->device);
 	struct mlx4_ib_qp *mqp = to_mqp(qp);
+	struct mlx4_ib_pd *pd;
 
 	if (is_qp0(dev, mqp))
 		mlx4_CLOSE_PORT(dev->dev, mqp->port);
 
-	destroy_qp_common(dev, mqp, !!qp->pd->uobject);
+	pd = get_pd(mqp);
+	destroy_qp_common(dev, mqp, !!pd->ibpd.uobject);
 
 	if (is_sqp(dev, mqp))
 		kfree(to_msqp(mqp));
@@ -821,6 +869,8 @@ static int to_mlx4_st(enum ib_qp_type type)
 	case IB_QPT_RC:		return MLX4_QP_ST_RC;
 	case IB_QPT_UC:		return MLX4_QP_ST_UC;
 	case IB_QPT_UD:		return MLX4_QP_ST_UD;
+	case IB_QPT_XRC_INI:
+	case IB_QPT_XRC_TGT:	return MLX4_QP_ST_XRC;
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:	return MLX4_QP_ST_MLX;
 	default:		return -1;
@@ -959,6 +1009,8 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 {
 	struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
+	struct mlx4_ib_pd *pd;
+	struct mlx4_ib_cq *send_cq, *recv_cq;
 	struct mlx4_qp_context *context;
 	enum mlx4_qp_optpar optpar = 0;
 	int sqd_event;
@@ -1014,8 +1066,10 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3;
 	context->sq_size_stride |= qp->sq.wqe_shift - 4;
 
-	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
 		context->sq_size_stride |= !!qp->sq_no_prefetch << 7;
+		context->xrcd = cpu_to_be32((u32) qp->xrcdn);
+	}
 
 	if (qp->ibqp.uobject)
 		context->usr_page = cpu_to_be32(to_mucontext(ibqp->uobject->context)->uar.index);
@@ -1079,8 +1133,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH;
 	}
 
-	context->pd	    = cpu_to_be32(to_mpd(ibqp->pd)->pdn);
-	context->params1    = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
+	pd = get_pd(qp);
+	get_cqs(qp, &send_cq, &recv_cq);
+	context->pd       = cpu_to_be32(pd->pdn);
+	context->cqn_send = cpu_to_be32(send_cq->mcq.cqn);
+	context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn);
+	context->params1  = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
 
 	/* Set "fast registration enabled" for all kernel QPs */
 	if (!qp->ibqp.uobject)
@@ -1106,8 +1164,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_SQ_PSN)
 		context->next_send_psn = cpu_to_be32(attr->sq_psn);
 
-	context->cqn_send = cpu_to_be32(to_mcq(ibqp->send_cq)->mcq.cqn);
-
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
 		if (attr->max_dest_rd_atomic)
 			context->params2 |=
@@ -1130,8 +1186,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_RQ_PSN)
 		context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
 
-	context->cqn_recv = cpu_to_be32(to_mcq(ibqp->recv_cq)->mcq.cqn);
-
 	if (attr_mask & IB_QP_QKEY) {
 		context->qkey = cpu_to_be32(attr->qkey);
 		optpar |= MLX4_QP_OPTPAR_Q_KEY;
@@ -1140,7 +1194,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (ibqp->srq)
 		context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn);
 
-	if (!ibqp->srq && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
 		context->db_rec_addr = cpu_to_be64(qp->db.dma);
 
 	if (cur_state == IB_QPS_INIT &&
@@ -1225,17 +1279,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	 * entries and reinitialize the QP.
 	 */
 	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
-		mlx4_ib_cq_clean(to_mcq(ibqp->recv_cq), qp->mqp.qpn,
+		mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
 				 ibqp->srq ? to_msrq(ibqp->srq): NULL);
-		if (ibqp->send_cq != ibqp->recv_cq)
-			mlx4_ib_cq_clean(to_mcq(ibqp->send_cq), qp->mqp.qpn, NULL);
+		if (send_cq != recv_cq)
+			mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
 
 		qp->rq.head = 0;
 		qp->rq.tail = 0;
 		qp->sq.head = 0;
 		qp->sq.tail = 0;
 		qp->sq_next_wqe = 0;
-		if (!ibqp->srq)
+		if (qp->rq.wqe_cnt)
 			*qp->db.db  = 0;
 	}
 
diff --git a/drivers/net/mlx4/qp.c b/drivers/net/mlx4/qp.c
index ec9350e5f21a..51c53898c35f 100644
--- a/drivers/net/mlx4/qp.c
+++ b/drivers/net/mlx4/qp.c
@@ -280,6 +280,9 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	 * We reserve 2 extra QPs per port for the special QPs.  The
 	 * block of special QPs must be aligned to a multiple of 8, so
 	 * round up.
+	 *
+	 * We also reserve the MSB of the 24-bit QP number to indicate
+	 * that a QP is an XRC QP.
 	 */
 	dev->caps.sqp_start =
 		ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 4001c8249dbb..48cc4cb97858 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -75,6 +75,7 @@ enum {
 	MLX4_QP_ST_UC				= 0x1,
 	MLX4_QP_ST_RD				= 0x2,
 	MLX4_QP_ST_UD				= 0x3,
+	MLX4_QP_ST_XRC				= 0x6,
 	MLX4_QP_ST_MLX				= 0x7
 };
 
@@ -137,7 +138,7 @@ struct mlx4_qp_context {
 	__be32			ssn;
 	__be32			params2;
 	__be32			rnr_nextrecvpsn;
-	__be32			srcd;
+	__be32			xrcd;
 	__be32			cqn_recv;
 	__be64			db_rec_addr;
 	__be32			qkey;
-- 
cgit v1.2.3


From 42849b2697c36abdafa6aef64186b15055392046 Mon Sep 17 00:00:00 2001
From: Sean Hefty <sean.hefty@intel.com>
Date: Thu, 11 Aug 2011 13:57:43 -0700
Subject: RDMA/uverbs: Export ib_open_qp() capability to user space

Allow processes that share the same XRC domain to open an existing
shareable QP.  This permits those processes to receive events on the
shared QP and transfer ownership, so that any process may modify the
QP.  The latter allows the creating process to exit, while a remaining
process can still transition it for path migration purposes.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/core/uverbs.h      |  1 +
 drivers/infiniband/core/uverbs_cmd.c  | 92 +++++++++++++++++++++++++++++++++++
 drivers/infiniband/core/uverbs_main.c |  3 +-
 drivers/infiniband/hw/mlx4/main.c     |  3 +-
 include/rdma/ib_user_verbs.h          | 14 +++++-
 5 files changed, 110 insertions(+), 3 deletions(-)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h
index 00ce032fcead..5bcb2afd3dcb 100644
--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h
@@ -195,6 +195,7 @@ IB_UVERBS_DECLARE_CMD(poll_cq);
 IB_UVERBS_DECLARE_CMD(req_notify_cq);
 IB_UVERBS_DECLARE_CMD(destroy_cq);
 IB_UVERBS_DECLARE_CMD(create_qp);
+IB_UVERBS_DECLARE_CMD(open_qp);
 IB_UVERBS_DECLARE_CMD(query_qp);
 IB_UVERBS_DECLARE_CMD(modify_qp);
 IB_UVERBS_DECLARE_CMD(destroy_qp);
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index c4c308cd2034..254f1649c734 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -1544,6 +1544,98 @@ err_put:
 	return ret;
 }
 
+ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
+			  const char __user *buf, int in_len, int out_len)
+{
+	struct ib_uverbs_open_qp        cmd;
+	struct ib_uverbs_create_qp_resp resp;
+	struct ib_udata                 udata;
+	struct ib_uqp_object           *obj;
+	struct ib_xrcd		       *xrcd;
+	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
+	struct ib_qp                   *qp;
+	struct ib_qp_open_attr          attr;
+	int ret;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	INIT_UDATA(&udata, buf + sizeof cmd,
+		   (unsigned long) cmd.response + sizeof resp,
+		   in_len - sizeof cmd, out_len - sizeof resp);
+
+	obj = kmalloc(sizeof *obj, GFP_KERNEL);
+	if (!obj)
+		return -ENOMEM;
+
+	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
+	down_write(&obj->uevent.uobject.mutex);
+
+	xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
+	if (!xrcd) {
+		ret = -EINVAL;
+		goto err_put;
+	}
+
+	attr.event_handler = ib_uverbs_qp_event_handler;
+	attr.qp_context    = file;
+	attr.qp_num        = cmd.qpn;
+	attr.qp_type       = cmd.qp_type;
+
+	obj->uevent.events_reported = 0;
+	INIT_LIST_HEAD(&obj->uevent.event_list);
+	INIT_LIST_HEAD(&obj->mcast_list);
+
+	qp = ib_open_qp(xrcd, &attr);
+	if (IS_ERR(qp)) {
+		ret = PTR_ERR(qp);
+		goto err_put;
+	}
+
+	qp->uobject = &obj->uevent.uobject;
+
+	obj->uevent.uobject.object = qp;
+	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
+	if (ret)
+		goto err_destroy;
+
+	memset(&resp, 0, sizeof resp);
+	resp.qpn       = qp->qp_num;
+	resp.qp_handle = obj->uevent.uobject.id;
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp)) {
+		ret = -EFAULT;
+		goto err_remove;
+	}
+
+	put_xrcd_read(xrcd_uobj);
+
+	mutex_lock(&file->mutex);
+	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
+	mutex_unlock(&file->mutex);
+
+	obj->uevent.uobject.live = 1;
+
+	up_write(&obj->uevent.uobject.mutex);
+
+	return in_len;
+
+err_remove:
+	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
+
+err_destroy:
+	ib_destroy_qp(qp);
+
+err_put:
+	put_xrcd_read(xrcd_uobj);
+	put_uobj_write(&obj->uevent.uobject);
+	return ret;
+}
+
 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
 			   const char __user *buf, int in_len,
 			   int out_len)
diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index 9c877e24eb60..485ea6c60d8b 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -110,7 +110,8 @@ static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
 	[IB_USER_VERBS_CMD_DESTROY_SRQ]		= ib_uverbs_destroy_srq,
 	[IB_USER_VERBS_CMD_OPEN_XRCD]		= ib_uverbs_open_xrcd,
 	[IB_USER_VERBS_CMD_CLOSE_XRCD]		= ib_uverbs_close_xrcd,
-	[IB_USER_VERBS_CMD_CREATE_XSRQ]		= ib_uverbs_create_xsrq
+	[IB_USER_VERBS_CMD_CREATE_XSRQ]		= ib_uverbs_create_xsrq,
+	[IB_USER_VERBS_CMD_OPEN_QP]		= ib_uverbs_open_qp
 };
 
 static void ib_uverbs_add_one(struct ib_device *device);
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index aec76ad77872..8a9ae720b56a 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1098,7 +1098,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
-		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ);
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
 
 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
diff --git a/include/rdma/ib_user_verbs.h b/include/rdma/ib_user_verbs.h
index 100abbbd6e6e..81aba3a73aa3 100644
--- a/include/rdma/ib_user_verbs.h
+++ b/include/rdma/ib_user_verbs.h
@@ -84,7 +84,8 @@ enum {
 	IB_USER_VERBS_CMD_POST_SRQ_RECV,
 	IB_USER_VERBS_CMD_OPEN_XRCD,
 	IB_USER_VERBS_CMD_CLOSE_XRCD,
-	IB_USER_VERBS_CMD_CREATE_XSRQ
+	IB_USER_VERBS_CMD_CREATE_XSRQ,
+	IB_USER_VERBS_CMD_OPEN_QP
 };
 
 /*
@@ -422,6 +423,17 @@ struct ib_uverbs_create_qp {
 	__u64 driver_data[0];
 };
 
+struct ib_uverbs_open_qp {
+	__u64 response;
+	__u64 user_handle;
+	__u32 pd_handle;
+	__u32 qpn;
+	__u8  qp_type;
+	__u8  reserved[7];
+	__u64 driver_data[0];
+};
+
+/* also used for open response */
 struct ib_uverbs_create_qp_resp {
 	__u32 qp_handle;
 	__u32 qpn;
-- 
cgit v1.2.3


From a5e12dff757b562bbecd6a2359fdc4c43d4d97de Mon Sep 17 00:00:00 2001
From: Marcel Apfelbaum <marcela@dev.mellanox.co.il>
Date: Mon, 3 Oct 2011 19:04:20 +0300
Subject: IB/mlx4: Configure extended active speeds

Set the extended active speeds based on the hardware configuration.

Signed-off-by: Marcel Apfelbaum <marcela@dev.mellanox.co.il>
Reviewed-by: Hal Rosenstock <hal@mellanox.com>

[ Move FDR-10 handling into ib_link_query_port().  - Roland ]

Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/hw/mlx4/main.c | 39 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 38 insertions(+), 1 deletion(-)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fa643f4f4e28..4e5b654e6375 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -181,8 +181,12 @@ mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 
 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 			      struct ib_port_attr *props,
+			      struct ib_smp *in_mad,
 			      struct ib_smp *out_mad)
 {
+	int ext_active_speed;
+	int err;
+
 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
 	props->lmc		= out_mad->data[34] & 0x7;
 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
@@ -203,6 +207,39 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 	props->max_vl_num	= out_mad->data[37] >> 4;
 	props->init_type_reply	= out_mad->data[41] >> 4;
 
+	/* Check if extended speeds (EDR/FDR/...) are supported */
+	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
+		ext_active_speed = out_mad->data[62] >> 4;
+
+		switch (ext_active_speed) {
+		case 1:
+			props->active_speed = 16; /* FDR */
+			break;
+		case 2:
+			props->active_speed = 32; /* EDR */
+			break;
+		}
+	}
+
+	/* If reported active speed is QDR, check if is FDR-10 */
+	if (props->active_speed == 4) {
+		if (to_mdev(ibdev)->dev->caps.ext_port_cap[port] &
+		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
+			init_query_mad(in_mad);
+			in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
+			in_mad->attr_mod = cpu_to_be32(port);
+
+			err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
+					   NULL, NULL, in_mad, out_mad);
+			if (err)
+				return err;
+
+			/* Checking LinkSpeedActive for FDR-10 */
+			if (out_mad->data[15] & 0x1)
+				props->active_speed = 8;
+		}
+	}
+
 	return 0;
 }
 
@@ -274,7 +311,7 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 		goto out;
 
 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
-		ib_link_query_port(ibdev, port, props, out_mad) :
+		ib_link_query_port(ibdev, port, props, in_mad, out_mad) :
 		eth_link_query_port(ibdev, port, props, out_mad);
 
 out:
-- 
cgit v1.2.3


From bcacb897569f0e7aab7643b22567d8de22ef9dfc Mon Sep 17 00:00:00 2001
From: Or Gerlitz <ogerlitz@mellanox.com>
Date: Mon, 10 Oct 2011 10:53:41 +0200
Subject: IB/mlx4: Enable 4K mtu for IBoE

The IBoE port MTU is derived from the corresponding Ethernet netdevice
MTU, which can support jumbo frames of 9K, and hence surely supports
the max IB mtu of 4K.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
---
 drivers/infiniband/hw/mlx4/main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/infiniband/hw/mlx4/main.c')

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fa643f4f4e28..e953cf9ffac6 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -227,7 +227,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 	props->pkey_tbl_len	= 1;
 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
-	props->max_mtu		= IB_MTU_2048;
+	props->max_mtu		= IB_MTU_4096;
 	props->subnet_timeout	= 0;
 	props->max_vl_num	= out_mad->data[37] >> 4;
 	props->init_type_reply	= 0;
-- 
cgit v1.2.3