From d2ae16d576086bd3e539f4e94f95df4e37ea2419 Mon Sep 17 00:00:00 2001
From: Roland Dreier <rolandd@cisco.com>
Date: Wed, 16 Apr 2008 21:01:07 -0700
Subject: IB/mlx4: Endianness annotations

Trivial fixes to stamp_send_wqe().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/hw/mlx4/qp.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

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

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 958e205b6d7c..ac965ab28845 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -122,7 +122,7 @@ static void *get_send_wqe(struct mlx4_ib_qp *qp, int n)
  */
 static void stamp_send_wqe(struct mlx4_ib_qp *qp, int n, int size)
 {
-	u32 *wqe;
+	__be32 *wqe;
 	int i;
 	int s;
 	int ind;
@@ -143,7 +143,7 @@ static void stamp_send_wqe(struct mlx4_ib_qp *qp, int n, int size)
 		buf = get_send_wqe(qp, n & (qp->sq.wqe_cnt - 1));
 		for (i = 64; i < s; i += 64) {
 			wqe = buf + i;
-			*wqe = 0xffffffff;
+			*wqe = cpu_to_be32(0xffffffff);
 		}
 	}
 }
-- 
cgit v1.2.3


From 8ff095ec4bce7be943beff3b330562e2f0e42167 Mon Sep 17 00:00:00 2001
From: Eli Cohen <eli@dev.mellanox.co.il>
Date: Wed, 16 Apr 2008 21:01:10 -0700
Subject: IB/mlx4: Add IPoIB checksum offload support

ConnectX devices support checksum generation and verification of TCP
and UDP packets for UD IPoIB messages.  This patch checks if the HCA
supports this and sets the IB_DEVICE_UD_IP_CSUM capability flag if it
does.  It implements support for handling the IB_SEND_IP_CSUM send
flag and setting the csum_ok field in receive work completions.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Ali Ayub <ali@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/hw/mlx4/cq.c   | 16 ++++++++++++++++
 drivers/infiniband/hw/mlx4/main.c |  2 ++
 drivers/infiniband/hw/mlx4/qp.c   |  3 +++
 drivers/net/mlx4/fw.c             |  4 ++++
 include/linux/mlx4/cq.h           | 14 ++++++++++++--
 include/linux/mlx4/qp.h           | 10 ++++++----
 6 files changed, 43 insertions(+), 6 deletions(-)

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

diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c
index 7360bbafbe84..d2e32b03e2f7 100644
--- a/drivers/infiniband/hw/mlx4/cq.c
+++ b/drivers/infiniband/hw/mlx4/cq.c
@@ -297,6 +297,20 @@ static void mlx4_ib_handle_error_cqe(struct mlx4_err_cqe *cqe,
 	wc->vendor_err = cqe->vendor_err_syndrome;
 }
 
+static int mlx4_ib_ipoib_csum_ok(__be32 status, __be16 checksum)
+{
+	return ((status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4	|
+				      MLX4_CQE_IPOIB_STATUS_IPV4F	|
+				      MLX4_CQE_IPOIB_STATUS_IPV4OPT	|
+				      MLX4_CQE_IPOIB_STATUS_IPV6	|
+				      MLX4_CQE_IPOIB_STATUS_IPOK)) ==
+		cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4	|
+			    MLX4_CQE_IPOIB_STATUS_IPOK))		&&
+		(status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_UDP	|
+				      MLX4_CQE_IPOIB_STATUS_TCP))	&&
+		checksum == cpu_to_be16(0xffff);
+}
+
 static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
 			    struct mlx4_ib_qp **cur_qp,
 			    struct ib_wc *wc)
@@ -434,6 +448,8 @@ static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
 		wc->dlid_path_bits = (g_mlpath_rqpn >> 24) & 0x7f;
 		wc->wc_flags	  |= g_mlpath_rqpn & 0x80000000 ? IB_WC_GRH : 0;
 		wc->pkey_index     = be32_to_cpu(cqe->immed_rss_invalid) & 0x7f;
+		wc->csum_ok	   = mlx4_ib_ipoib_csum_ok(cqe->ipoib_status,
+							   cqe->checksum);
 	}
 
 	return 0;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index d5512011999c..6ea4746c2e9b 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -99,6 +99,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
+		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index ac965ab28845..31b2b5b230bd 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1436,6 +1436,9 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			 cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE) : 0) |
 			(wr->send_flags & IB_SEND_SOLICITED ?
 			 cpu_to_be32(MLX4_WQE_CTRL_SOLICITED) : 0) |
+			((wr->send_flags & IB_SEND_IP_CSUM) ?
+			 cpu_to_be32(MLX4_WQE_CTRL_IP_CSUM |
+				     MLX4_WQE_CTRL_TCP_UDP_CSUM) : 0) |
 			qp->sq_signal_bits;
 
 		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
diff --git a/drivers/net/mlx4/fw.c b/drivers/net/mlx4/fw.c
index 61dc4951d6b0..f494c3e8bce3 100644
--- a/drivers/net/mlx4/fw.c
+++ b/drivers/net/mlx4/fw.c
@@ -696,6 +696,10 @@ int mlx4_INIT_HCA(struct mlx4_dev *dev, struct mlx4_init_hca_param *param)
 	/* Check port for UD address vector: */
 	*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1);
 
+	/* Enable IPoIB checksumming if we can: */
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
+		*(inbox + INIT_HCA_FLAGS_OFFSET / 4) |= cpu_to_be32(1 << 3);
+
 	/* QPC/EEC/CQC/EQC/RDMARC attributes */
 
 	MLX4_PUT(inbox, param->qpc_base,      INIT_HCA_QPC_BASE_OFFSET);
diff --git a/include/linux/mlx4/cq.h b/include/linux/mlx4/cq.h
index 0181e0a57cbf..1243ebace561 100644
--- a/include/linux/mlx4/cq.h
+++ b/include/linux/mlx4/cq.h
@@ -45,11 +45,11 @@ struct mlx4_cqe {
 	u8			sl;
 	u8			reserved1;
 	__be16			rlid;
-	u32			reserved2;
+	__be32			ipoib_status;
 	__be32			byte_cnt;
 	__be16			wqe_index;
 	__be16			checksum;
-	u8			reserved3[3];
+	u8			reserved2[3];
 	u8			owner_sr_opcode;
 };
 
@@ -85,6 +85,16 @@ enum {
 	MLX4_CQE_SYNDROME_REMOTE_ABORTED_ERR		= 0x22,
 };
 
+enum {
+	MLX4_CQE_IPOIB_STATUS_IPV4			= 1 << 22,
+	MLX4_CQE_IPOIB_STATUS_IPV4F			= 1 << 23,
+	MLX4_CQE_IPOIB_STATUS_IPV6			= 1 << 24,
+	MLX4_CQE_IPOIB_STATUS_IPV4OPT			= 1 << 25,
+	MLX4_CQE_IPOIB_STATUS_TCP			= 1 << 26,
+	MLX4_CQE_IPOIB_STATUS_UDP			= 1 << 27,
+	MLX4_CQE_IPOIB_STATUS_IPOK			= 1 << 28,
+};
+
 static inline void mlx4_cq_arm(struct mlx4_cq *cq, u32 cmd,
 			       void __iomem *uar_page,
 			       spinlock_t *doorbell_lock)
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 09a2230923f2..31f9eb3ccbb3 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -158,10 +158,12 @@ struct mlx4_qp_context {
 #define MLX4_FW_VER_WQE_CTRL_NEC mlx4_fw_ver(2, 2, 232)
 
 enum {
-	MLX4_WQE_CTRL_NEC	= 1 << 29,
-	MLX4_WQE_CTRL_FENCE	= 1 << 6,
-	MLX4_WQE_CTRL_CQ_UPDATE	= 3 << 2,
-	MLX4_WQE_CTRL_SOLICITED	= 1 << 1,
+	MLX4_WQE_CTRL_NEC		= 1 << 29,
+	MLX4_WQE_CTRL_FENCE		= 1 << 6,
+	MLX4_WQE_CTRL_CQ_UPDATE		= 3 << 2,
+	MLX4_WQE_CTRL_SOLICITED		= 1 << 1,
+	MLX4_WQE_CTRL_IP_CSUM		= 1 << 4,
+	MLX4_WQE_CTRL_TCP_UDP_CSUM	= 1 << 5,
 };
 
 struct mlx4_wqe_ctrl_seg {
-- 
cgit v1.2.3


From b846f25aa2a353355aec5202fe4dbdc6674dfc64 Mon Sep 17 00:00:00 2001
From: Eli Cohen <eli@dev.mellanox.co.il>
Date: Wed, 16 Apr 2008 21:09:27 -0700
Subject: IB/core: Add creation flags to struct ib_qp_init_attr

Add a create_flags member to struct ib_qp_init_attr that will allow a
kernel verbs consumer to create a pass special flags when creating a QP.
Add a flag value for telling low-level drivers that a QP will be used
for IPoIB UD LSO.  The create_flags member will also be useful for XRC
and ehca low-latency QP support.

Since no create_flags handling is implemented yet, add code to all
low-level drivers to return -EINVAL if create_flags is non-zero.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/core/uverbs_cmd.c         | 1 +
 drivers/infiniband/hw/amso1100/c2_provider.c | 3 +++
 drivers/infiniband/hw/ehca/ehca_qp.c         | 3 +++
 drivers/infiniband/hw/ipath/ipath_qp.c       | 5 +++++
 drivers/infiniband/hw/mlx4/qp.c              | 3 +++
 drivers/infiniband/hw/mthca/mthca_provider.c | 3 +++
 drivers/infiniband/hw/nes/nes_verbs.c        | 3 +++
 include/rdma/ib_verbs.h                      | 5 +++++
 8 files changed, 26 insertions(+)

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

diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 495c803fb11d..9e98cec6230f 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -1065,6 +1065,7 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 	attr.srq           = srq;
 	attr.sq_sig_type   = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
 	attr.qp_type       = cmd.qp_type;
+	attr.create_flags  = 0;
 
 	attr.cap.max_send_wr     = cmd.max_send_wr;
 	attr.cap.max_recv_wr     = cmd.max_recv_wr;
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c
index 21fb46d39319..e10d27a6e145 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -245,6 +245,9 @@ static struct ib_qp *c2_create_qp(struct ib_pd *pd,
 
 	pr_debug("%s:%u\n", __func__, __LINE__);
 
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
 	switch (init_attr->qp_type) {
 	case IB_QPT_RC:
 		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c
index a9fd419855cb..3eb14a52cbf2 100644
--- a/drivers/infiniband/hw/ehca/ehca_qp.c
+++ b/drivers/infiniband/hw/ehca/ehca_qp.c
@@ -421,6 +421,9 @@ static struct ehca_qp *internal_create_qp(
 	u32 swqe_size = 0, rwqe_size = 0, ib_qp_num;
 	unsigned long flags;
 
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
 	memset(&parms, 0, sizeof(parms));
 	qp_type = init_attr->qp_type;
 
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 6a4a5e3b78ba..812b42c500e1 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -747,6 +747,11 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 	size_t sz;
 	struct ib_qp *ret;
 
+	if (init_attr->create_flags) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
 	if (init_attr->cap.max_send_sge > ib_ipath_max_sges ||
 	    init_attr->cap.max_send_wr > ib_ipath_max_qp_wrs) {
 		ret = ERR_PTR(-EINVAL);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 31b2b5b230bd..320c25fa74b1 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -673,6 +673,9 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	struct mlx4_ib_qp *qp;
 	int err;
 
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
 	switch (init_attr->qp_type) {
 	case IB_QPT_RC:
 	case IB_QPT_UC:
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index ee9bc1456226..81b257e18bb6 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -540,6 +540,9 @@ static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
 	struct mthca_qp *qp;
 	int err;
 
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
 	switch (init_attr->qp_type) {
 	case IB_QPT_RC:
 	case IB_QPT_UC:
diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c
index 46b3b8ebcc14..7c27420c2240 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.c
+++ b/drivers/infiniband/hw/nes/nes_verbs.c
@@ -1252,6 +1252,9 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 	u8 rq_encoded_size;
 	/* int counter; */
 
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
 	atomic_inc(&qps_created);
 	switch (init_attr->qp_type) {
 		case IB_QPT_RC:
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 40ff51244d19..c3299be41c6f 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -495,6 +495,10 @@ enum ib_qp_type {
 	IB_QPT_RAW_ETY
 };
 
+enum ib_qp_create_flags {
+	IB_QP_CREATE_IPOIB_UD_LSO	= 1 << 0,
+};
+
 struct ib_qp_init_attr {
 	void                  (*event_handler)(struct ib_event *, void *);
 	void		       *qp_context;
@@ -504,6 +508,7 @@ struct ib_qp_init_attr {
 	struct ib_qp_cap	cap;
 	enum ib_sig_type	sq_sig_type;
 	enum ib_qp_type		qp_type;
+	enum ib_qp_create_flags	create_flags;
 	u8			port_num; /* special QP types only */
 };
 
-- 
cgit v1.2.3


From b832be1e4007f4a54954ec68bd865ff05d6babca Mon Sep 17 00:00:00 2001
From: Eli Cohen <eli@dev.mellanox.co.il>
Date: Wed, 16 Apr 2008 21:09:27 -0700
Subject: IB/mlx4: Add IPoIB LSO support

Add TSO support to the mlx4_ib driver.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/hw/mlx4/cq.c      |  3 ++
 drivers/infiniband/hw/mlx4/main.c    |  2 +
 drivers/infiniband/hw/mlx4/mlx4_ib.h |  5 +++
 drivers/infiniband/hw/mlx4/qp.c      | 72 +++++++++++++++++++++++++++++++-----
 drivers/net/mlx4/fw.c                |  9 +++++
 drivers/net/mlx4/fw.h                |  1 +
 drivers/net/mlx4/main.c              |  1 +
 include/linux/mlx4/device.h          |  1 +
 include/linux/mlx4/qp.h              |  5 +++
 9 files changed, 90 insertions(+), 9 deletions(-)

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

diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c
index d2e32b03e2f7..7d70af7952b0 100644
--- a/drivers/infiniband/hw/mlx4/cq.c
+++ b/drivers/infiniband/hw/mlx4/cq.c
@@ -420,6 +420,9 @@ static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
 		case MLX4_OPCODE_BIND_MW:
 			wc->opcode    = IB_WC_BIND_MW;
 			break;
+		case MLX4_OPCODE_LSO:
+			wc->opcode    = IB_WC_LSO;
+			break;
 		}
 	} else {
 		wc->byte_len = be32_to_cpu(cqe->byte_cnt);
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 6ea4746c2e9b..e9330a0d6c03 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -101,6 +101,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
+	if (dev->dev->caps.max_gso_sz)
+		props->device_cap_flags |= IB_DEVICE_UD_TSO;
 
 	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 3726e451a327..3f8bd0a37b96 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -110,6 +110,10 @@ struct mlx4_ib_wq {
 	unsigned		tail;
 };
 
+enum mlx4_ib_qp_flags {
+	MLX4_IB_QP_LSO		= 1 << 0
+};
+
 struct mlx4_ib_qp {
 	struct ib_qp		ibqp;
 	struct mlx4_qp		mqp;
@@ -129,6 +133,7 @@ struct mlx4_ib_qp {
 	struct mlx4_mtt		mtt;
 	int			buf_size;
 	struct mutex		mutex;
+	u32			flags;
 	u8			port;
 	u8			alt_port;
 	u8			atomic_rd_en;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 320c25fa74b1..2ba243084089 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -71,6 +71,7 @@ enum {
 
 static const __be32 mlx4_ib_opcode[] = {
 	[IB_WR_SEND]			= __constant_cpu_to_be32(MLX4_OPCODE_SEND),
+	[IB_WR_LSO]			= __constant_cpu_to_be32(MLX4_OPCODE_LSO),
 	[IB_WR_SEND_WITH_IMM]		= __constant_cpu_to_be32(MLX4_OPCODE_SEND_IMM),
 	[IB_WR_RDMA_WRITE]		= __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
 	[IB_WR_RDMA_WRITE_WITH_IMM]	= __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
@@ -242,7 +243,7 @@ static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type)
 	}
 }
 
-static int send_wqe_overhead(enum ib_qp_type type)
+static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
 {
 	/*
 	 * UD WQEs must have a datagram segment.
@@ -253,7 +254,8 @@ static int send_wqe_overhead(enum ib_qp_type type)
 	switch (type) {
 	case IB_QPT_UD:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
-			sizeof (struct mlx4_wqe_datagram_seg);
+			sizeof (struct mlx4_wqe_datagram_seg) +
+			((flags & MLX4_IB_QP_LSO) ? 64 : 0);
 	case IB_QPT_UC:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
 			sizeof (struct mlx4_wqe_raddr_seg);
@@ -315,7 +317,7 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 	/* Sanity check SQ size before proceeding */
 	if (cap->max_send_wr	 > dev->dev->caps.max_wqes  ||
 	    cap->max_send_sge	 > dev->dev->caps.max_sq_sg ||
-	    cap->max_inline_data + send_wqe_overhead(type) +
+	    cap->max_inline_data + send_wqe_overhead(type, qp->flags) +
 	    sizeof (struct mlx4_wqe_inline_seg) > dev->dev->caps.max_sq_desc_sz)
 		return -EINVAL;
 
@@ -329,7 +331,7 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 
 	s = max(cap->max_send_sge * sizeof (struct mlx4_wqe_data_seg),
 		cap->max_inline_data + sizeof (struct mlx4_wqe_inline_seg)) +
-		send_wqe_overhead(type);
+		send_wqe_overhead(type, qp->flags);
 
 	/*
 	 * Hermon supports shrinking WQEs, such that a single work
@@ -394,7 +396,8 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 	}
 
 	qp->sq.max_gs = ((qp->sq_max_wqes_per_wr << qp->sq.wqe_shift) -
-			 send_wqe_overhead(type)) / sizeof (struct mlx4_wqe_data_seg);
+			 send_wqe_overhead(type, qp->flags)) /
+		sizeof (struct mlx4_wqe_data_seg);
 
 	qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
 		(qp->sq.wqe_cnt << qp->sq.wqe_shift);
@@ -503,6 +506,9 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	} else {
 		qp->sq_no_prefetch = 0;
 
+		if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)
+			qp->flags |= MLX4_IB_QP_LSO;
+
 		err = set_kernel_sq_size(dev, &init_attr->cap, init_attr->qp_type, qp);
 		if (err)
 			goto err;
@@ -673,7 +679,11 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	struct mlx4_ib_qp *qp;
 	int err;
 
-	if (init_attr->create_flags)
+	/* We only support LSO, and only for kernel UD QPs. */
+	if (init_attr->create_flags & ~IB_QP_CREATE_IPOIB_UD_LSO)
+		return ERR_PTR(-EINVAL);
+	if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO &&
+	    (pd->uobject || init_attr->qp_type != IB_QPT_UD))
 		return ERR_PTR(-EINVAL);
 
 	switch (init_attr->qp_type) {
@@ -879,10 +889,15 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		}
 	}
 
-	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI ||
-	    ibqp->qp_type == IB_QPT_UD)
+	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI)
 		context->mtu_msgmax = (IB_MTU_4096 << 5) | 11;
-	else if (attr_mask & IB_QP_PATH_MTU) {
+	else if (ibqp->qp_type == IB_QPT_UD) {
+		if (qp->flags & MLX4_IB_QP_LSO)
+			context->mtu_msgmax = (IB_MTU_4096 << 5) |
+					      ilog2(dev->dev->caps.max_gso_sz);
+		else
+			context->mtu_msgmax = (IB_MTU_4096 << 5) | 11;
+	} else if (attr_mask & IB_QP_PATH_MTU) {
 		if (attr->path_mtu < IB_MTU_256 || attr->path_mtu > IB_MTU_4096) {
 			printk(KERN_ERR "path MTU (%u) is invalid\n",
 			       attr->path_mtu);
@@ -1399,6 +1414,34 @@ static void __set_data_seg(struct mlx4_wqe_data_seg *dseg, struct ib_sge *sg)
 	dseg->addr       = cpu_to_be64(sg->addr);
 }
 
+static int build_lso_seg(struct mlx4_lso_seg *wqe, struct ib_send_wr *wr,
+			 struct mlx4_ib_qp *qp, unsigned *lso_seg_len)
+{
+	unsigned halign = ALIGN(sizeof *wqe + wr->wr.ud.hlen, 16);
+
+	/*
+	 * This is a temporary limitation and will be removed in
+	 * a forthcoming FW release:
+	 */
+	if (unlikely(halign > 64))
+		return -EINVAL;
+
+	if (unlikely(!(qp->flags & MLX4_IB_QP_LSO) &&
+		     wr->num_sge > qp->sq.max_gs - (halign >> 4)))
+		return -EINVAL;
+
+	memcpy(wqe->header, wr->wr.ud.header, wr->wr.ud.hlen);
+
+	/* make sure LSO header is written before overwriting stamping */
+	wmb();
+
+	wqe->mss_hdr_size = cpu_to_be32((wr->wr.ud.mss - wr->wr.ud.hlen) << 16 |
+					wr->wr.ud.hlen);
+
+	*lso_seg_len = halign;
+	return 0;
+}
+
 int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		      struct ib_send_wr **bad_wr)
 {
@@ -1412,6 +1455,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	unsigned ind;
 	int uninitialized_var(stamp);
 	int uninitialized_var(size);
+	unsigned seglen;
 	int i;
 
 	spin_lock_irqsave(&qp->sq.lock, flags);
@@ -1490,6 +1534,16 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			set_datagram_seg(wqe, wr);
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
+
+			if (wr->opcode == IB_WR_LSO) {
+				err = build_lso_seg(wqe, wr, qp, &seglen);
+				if (unlikely(err)) {
+					*bad_wr = wr;
+					goto out;
+				}
+				wqe  += seglen;
+				size += seglen / 16;
+			}
 			break;
 
 		case IB_QPT_SMI:
diff --git a/drivers/net/mlx4/fw.c b/drivers/net/mlx4/fw.c
index f494c3e8bce3..d82f2751d2c7 100644
--- a/drivers/net/mlx4/fw.c
+++ b/drivers/net/mlx4/fw.c
@@ -133,6 +133,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 #define QUERY_DEV_CAP_MAX_AV_OFFSET		0x27
 #define QUERY_DEV_CAP_MAX_REQ_QP_OFFSET		0x29
 #define QUERY_DEV_CAP_MAX_RES_QP_OFFSET		0x2b
+#define QUERY_DEV_CAP_MAX_GSO_OFFSET		0x2d
 #define QUERY_DEV_CAP_MAX_RDMA_OFFSET		0x2f
 #define QUERY_DEV_CAP_RSZ_SRQ_OFFSET		0x33
 #define QUERY_DEV_CAP_ACK_DELAY_OFFSET		0x35
@@ -215,6 +216,13 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev_cap->max_requester_per_qp = 1 << (field & 0x3f);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RES_QP_OFFSET);
 	dev_cap->max_responder_per_qp = 1 << (field & 0x3f);
+	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_GSO_OFFSET);
+	field &= 0x1f;
+	if (!field)
+		dev_cap->max_gso_sz = 0;
+	else
+		dev_cap->max_gso_sz = 1 << field;
+
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_RDMA_OFFSET);
 	dev_cap->max_rdma_global = 1 << (field & 0x3f);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_ACK_DELAY_OFFSET);
@@ -377,6 +385,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 		 dev_cap->max_sq_desc_sz, dev_cap->max_sq_sg);
 	mlx4_dbg(dev, "Max RQ desc size: %d, max RQ S/G: %d\n",
 		 dev_cap->max_rq_desc_sz, dev_cap->max_rq_sg);
+	mlx4_dbg(dev, "Max GSO size: %d\n", dev_cap->max_gso_sz);
 
 	dump_dev_cap_flags(dev, dev_cap->flags);
 
diff --git a/drivers/net/mlx4/fw.h b/drivers/net/mlx4/fw.h
index e16dec890413..306cb9b0242d 100644
--- a/drivers/net/mlx4/fw.h
+++ b/drivers/net/mlx4/fw.h
@@ -96,6 +96,7 @@ struct mlx4_dev_cap {
 	u8  bmme_flags;
 	u32 reserved_lkey;
 	u64 max_icm_sz;
+	int max_gso_sz;
 };
 
 struct mlx4_adapter {
diff --git a/drivers/net/mlx4/main.c b/drivers/net/mlx4/main.c
index 08bfc130a33e..7cfbe75114d1 100644
--- a/drivers/net/mlx4/main.c
+++ b/drivers/net/mlx4/main.c
@@ -159,6 +159,7 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
 	dev->caps.flags		     = dev_cap->flags;
 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
+	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
 
 	return 0;
 }
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 6cdf813cd478..ff7df1a2222f 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -186,6 +186,7 @@ struct mlx4_caps {
 	u32			flags;
 	u16			stat_rate_support;
 	u8			port_width_cap[MLX4_MAX_PORTS + 1];
+	int			max_gso_sz;
 };
 
 struct mlx4_buf_list {
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 31f9eb3ccbb3..a5e43febee4f 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -219,6 +219,11 @@ struct mlx4_wqe_datagram_seg {
 	__be32			reservd[2];
 };
 
+struct mlx4_lso_seg {
+	__be32			mss_hdr_size;
+	__be32			header[0];
+};
+
 struct mlx4_wqe_bind_seg {
 	__be32			flags1;
 	__be32			flags2;
-- 
cgit v1.2.3


From f438000f7a31fad7cfd27f33ad324a250f4cd2df Mon Sep 17 00:00:00 2001
From: Roland Dreier <rolandd@cisco.com>
Date: Wed, 16 Apr 2008 21:09:28 -0700
Subject: IB/mlx4: Micro-optimize mlx4_ib_post_send()

Rather than have build_mlx_header() return a negative value on failure
and the length of the segments it builds on success, add a pointer
parameter to return the length and return 0 on success.  This matches
the calling convention used for build_lso_seg() and generates slightly
smaller code -- eg, on 64-bit x86:

add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-22 (-22)
function                                     old     new   delta
mlx4_ib_post_send                           2023    2001     -22

Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/hw/mlx4/qp.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

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

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 2ba243084089..f5210c17e312 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1200,7 +1200,7 @@ out:
 }
 
 static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
-			    void *wqe)
+			    void *wqe, unsigned *mlx_seg_len)
 {
 	struct ib_device *ib_dev = &to_mdev(sqp->qp.ibqp.device)->ib_dev;
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
@@ -1321,7 +1321,9 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 		i = 2;
 	}
 
-	return ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + header_size, 16);
+	*mlx_seg_len =
+		ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + header_size, 16);
+	return 0;
 }
 
 static int mlx4_wq_overflow(struct mlx4_ib_wq *wq, int nreq, struct ib_cq *ib_cq)
@@ -1548,15 +1550,13 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 
 		case IB_QPT_SMI:
 		case IB_QPT_GSI:
-			err = build_mlx_header(to_msqp(qp), wr, ctrl);
-			if (err < 0) {
+			err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen);
+			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
 			}
-			wqe  += err;
-			size += err / 16;
-
-			err = 0;
+			wqe  += seglen;
+			size += seglen / 16;
 			break;
 
 		default:
-- 
cgit v1.2.3


From 0f39cf3d54e67a705773fd0ec56ca3dcd3e9272f Mon Sep 17 00:00:00 2001
From: Roland Dreier <rolandd@cisco.com>
Date: Wed, 16 Apr 2008 21:09:32 -0700
Subject: IB/core: Add support for "send with invalidate" work requests

Add a new IB_WR_SEND_WITH_INV send opcode that can be used to mark a
"send with invalidate" work request as defined in the iWARP verbs and
the InfiniBand base memory management extensions.  Also put "imm_data"
and a new "invalidate_rkey" member in a new "ex" union in struct
ib_send_wr. The invalidate_rkey member can be used to pass in an
R_Key/STag to be invalidated.  Add this new union to struct
ib_uverbs_send_wr.  Add code to copy the invalidate_rkey field in
ib_uverbs_post_send().

Fix up low-level drivers to deal with the change to struct ib_send_wr,
and just remove the imm_data initialization from net/sunrpc/xprtrdma/,
since that code never does any send with immediate operations.

Also, move the existing IB_DEVICE_SEND_W_INV flag to a new bit, since
the iWARP drivers currently in the tree set the bit.  The amso1100
driver at least will silently fail to honor the IB_SEND_INVALIDATE bit
if passed in as part of userspace send requests (since it does not
implement kernel bypass work request queueing).  Remove the flag from
all existing drivers that set it until we know which ones are OK.

The values chosen for the new flag is not consecutive to avoid clashing
with flags defined in the XRC patches, which are not merged yet but
which are already in use and are likely to be merged soon.

This resurrects a patch sent long ago by Mikkel Hagen <mhagen@iol.unh.edu>.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/core/uverbs_cmd.c        | 13 +++++++++++--
 drivers/infiniband/hw/amso1100/c2_rnic.c    |  2 +-
 drivers/infiniband/hw/cxgb3/iwch_provider.c |  3 +--
 drivers/infiniband/hw/cxgb3/iwch_qp.c       |  4 ++--
 drivers/infiniband/hw/ehca/ehca_reqs.c      |  2 +-
 drivers/infiniband/hw/ipath/ipath_rc.c      |  8 ++++----
 drivers/infiniband/hw/ipath/ipath_ruc.c     |  4 ++--
 drivers/infiniband/hw/ipath/ipath_uc.c      |  8 ++++----
 drivers/infiniband/hw/ipath/ipath_ud.c      |  4 ++--
 drivers/infiniband/hw/mlx4/qp.c             |  4 ++--
 drivers/infiniband/hw/mthca/mthca_qp.c      |  6 +++---
 drivers/infiniband/hw/nes/nes_hw.c          |  2 +-
 include/rdma/ib_user_verbs.h                |  5 ++++-
 include/rdma/ib_verbs.h                     | 11 ++++++++---
 net/sunrpc/xprtrdma/verbs.c                 |  1 -
 15 files changed, 46 insertions(+), 31 deletions(-)

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

diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 9e98cec6230f..2c3bff5fe867 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -1463,7 +1463,6 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 		next->num_sge    = user_wr->num_sge;
 		next->opcode     = user_wr->opcode;
 		next->send_flags = user_wr->send_flags;
-		next->imm_data   = (__be32 __force) user_wr->imm_data;
 
 		if (is_ud) {
 			next->wr.ud.ah = idr_read_ah(user_wr->wr.ud.ah,
@@ -1476,14 +1475,24 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 			next->wr.ud.remote_qkey = user_wr->wr.ud.remote_qkey;
 		} else {
 			switch (next->opcode) {
-			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
+				next->ex.imm_data =
+					(__be32 __force) user_wr->ex.imm_data;
+			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_READ:
 				next->wr.rdma.remote_addr =
 					user_wr->wr.rdma.remote_addr;
 				next->wr.rdma.rkey        =
 					user_wr->wr.rdma.rkey;
 				break;
+			case IB_WR_SEND_WITH_IMM:
+				next->ex.imm_data =
+					(__be32 __force) user_wr->ex.imm_data;
+				break;
+			case IB_WR_SEND_WITH_INV:
+				next->ex.invalidate_rkey =
+					user_wr->ex.invalidate_rkey;
+				break;
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 			case IB_WR_ATOMIC_FETCH_AND_ADD:
 				next->wr.atomic.remote_addr =
diff --git a/drivers/infiniband/hw/amso1100/c2_rnic.c b/drivers/infiniband/hw/amso1100/c2_rnic.c
index 7a625524e0c5..b1441aeb60c2 100644
--- a/drivers/infiniband/hw/amso1100/c2_rnic.c
+++ b/drivers/infiniband/hw/amso1100/c2_rnic.c
@@ -455,7 +455,7 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 	     IB_DEVICE_CURR_QP_STATE_MOD |
 	     IB_DEVICE_SYS_IMAGE_GUID |
 	     IB_DEVICE_ZERO_STAG |
-	     IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW);
+	     IB_DEVICE_MEM_WINDOW);
 
 	/* Allocate the qptr_array */
 	c2dev->qptr_array = vmalloc(C2_MAX_CQS * sizeof(void *));
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c
index 50e1f2a16e0c..ca7265443c05 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c
@@ -1109,8 +1109,7 @@ int iwch_register_device(struct iwch_dev *dev)
 	memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
 	dev->ibdev.owner = THIS_MODULE;
 	dev->device_cap_flags =
-	    (IB_DEVICE_ZERO_STAG |
-	     IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW);
+	    (IB_DEVICE_ZERO_STAG | IB_DEVICE_MEM_WINDOW);
 
 	dev->ibdev.uverbs_cmd_mask =
 	    (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c
index bc5d9b0813e5..8891c3b0a3d5 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_qp.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c
@@ -72,7 +72,7 @@ static int iwch_build_rdma_send(union t3_wr *wqe, struct ib_send_wr *wr,
 	wqe->send.reserved[2] = 0;
 	if (wr->opcode == IB_WR_SEND_WITH_IMM) {
 		plen = 4;
-		wqe->send.sgl[0].stag = wr->imm_data;
+		wqe->send.sgl[0].stag = wr->ex.imm_data;
 		wqe->send.sgl[0].len = __constant_cpu_to_be32(0);
 		wqe->send.num_sgle = __constant_cpu_to_be32(0);
 		*flit_cnt = 5;
@@ -112,7 +112,7 @@ static int iwch_build_rdma_write(union t3_wr *wqe, struct ib_send_wr *wr,
 
 	if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
 		plen = 4;
-		wqe->write.sgl[0].stag = wr->imm_data;
+		wqe->write.sgl[0].stag = wr->ex.imm_data;
 		wqe->write.sgl[0].len = __constant_cpu_to_be32(0);
 		wqe->write.num_sgle = __constant_cpu_to_be32(0);
 		*flit_cnt = 6;
diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
index 2ce8cffb8664..a20bbf466188 100644
--- a/drivers/infiniband/hw/ehca/ehca_reqs.c
+++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
@@ -188,7 +188,7 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 	if (send_wr->opcode == IB_WR_SEND_WITH_IMM ||
 	    send_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
 		/* this might not work as long as HW does not support it */
-		wqe_p->immediate_data = be32_to_cpu(send_wr->imm_data);
+		wqe_p->immediate_data = be32_to_cpu(send_wr->ex.imm_data);
 		wqe_p->wr_flag |= WQE_WRFLAG_IMM_DATA_PRESENT;
 	}
 
diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c
index 467981905bbe..c405dfba5531 100644
--- a/drivers/infiniband/hw/ipath/ipath_rc.c
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c
@@ -308,7 +308,7 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 			else {
 				qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE);
 				/* Immediate data comes after the BTH */
-				ohdr->u.imm_data = wqe->wr.imm_data;
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
 				hwords += 1;
 			}
 			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
@@ -346,7 +346,7 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 				qp->s_state =
 					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
 				/* Immediate data comes after RETH */
-				ohdr->u.rc.imm_data = wqe->wr.imm_data;
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
 				hwords += 1;
 				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 					bth0 |= 1 << 23;
@@ -490,7 +490,7 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		else {
 			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
 			/* Immediate data comes after the BTH */
-			ohdr->u.imm_data = wqe->wr.imm_data;
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
 			hwords += 1;
 		}
 		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
@@ -526,7 +526,7 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		else {
 			qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
 			/* Immediate data comes after the BTH */
-			ohdr->u.imm_data = wqe->wr.imm_data;
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
 			hwords += 1;
 			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 				bth0 |= 1 << 23;
diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c
index bcaa2914e341..8ac5c1d82ccd 100644
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c
@@ -310,7 +310,7 @@ again:
 	switch (wqe->wr.opcode) {
 	case IB_WR_SEND_WITH_IMM:
 		wc.wc_flags = IB_WC_WITH_IMM;
-		wc.imm_data = wqe->wr.imm_data;
+		wc.imm_data = wqe->wr.ex.imm_data;
 		/* FALLTHROUGH */
 	case IB_WR_SEND:
 		if (!ipath_get_rwqe(qp, 0)) {
@@ -339,7 +339,7 @@ again:
 			goto err;
 		}
 		wc.wc_flags = IB_WC_WITH_IMM;
-		wc.imm_data = wqe->wr.imm_data;
+		wc.imm_data = wqe->wr.ex.imm_data;
 		if (!ipath_get_rwqe(qp, 1))
 			goto rnr_nak;
 		/* FALLTHROUGH */
diff --git a/drivers/infiniband/hw/ipath/ipath_uc.c b/drivers/infiniband/hw/ipath/ipath_uc.c
index 2dd8de20d221..bfe8926b5514 100644
--- a/drivers/infiniband/hw/ipath/ipath_uc.c
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c
@@ -94,7 +94,7 @@ int ipath_make_uc_req(struct ipath_qp *qp)
 				qp->s_state =
 					OP(SEND_ONLY_WITH_IMMEDIATE);
 				/* Immediate data comes after the BTH */
-				ohdr->u.imm_data = wqe->wr.imm_data;
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
 				hwords += 1;
 			}
 			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
@@ -123,7 +123,7 @@ int ipath_make_uc_req(struct ipath_qp *qp)
 				qp->s_state =
 					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
 				/* Immediate data comes after the RETH */
-				ohdr->u.rc.imm_data = wqe->wr.imm_data;
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
 				hwords += 1;
 				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 					bth0 |= 1 << 23;
@@ -152,7 +152,7 @@ int ipath_make_uc_req(struct ipath_qp *qp)
 		else {
 			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
 			/* Immediate data comes after the BTH */
-			ohdr->u.imm_data = wqe->wr.imm_data;
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
 			hwords += 1;
 		}
 		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
@@ -177,7 +177,7 @@ int ipath_make_uc_req(struct ipath_qp *qp)
 			qp->s_state =
 				OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
 			/* Immediate data comes after the BTH */
-			ohdr->u.imm_data = wqe->wr.imm_data;
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
 			hwords += 1;
 			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
 				bth0 |= 1 << 23;
diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c
index 918f52070653..8b6a261c89e3 100644
--- a/drivers/infiniband/hw/ipath/ipath_ud.c
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c
@@ -95,7 +95,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 
 	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
 		wc.wc_flags = IB_WC_WITH_IMM;
-		wc.imm_data = swqe->wr.imm_data;
+		wc.imm_data = swqe->wr.ex.imm_data;
 	} else {
 		wc.wc_flags = 0;
 		wc.imm_data = 0;
@@ -327,7 +327,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 	}
 	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
 		qp->s_hdrwords++;
-		ohdr->u.ud.imm_data = wqe->wr.imm_data;
+		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
 		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
 	} else
 		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index f5210c17e312..38e651a67589 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1249,7 +1249,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 	case IB_WR_SEND_WITH_IMM:
 		sqp->ud_header.bth.opcode	 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
 		sqp->ud_header.immediate_present = 1;
-		sqp->ud_header.immediate_data    = wr->imm_data;
+		sqp->ud_header.immediate_data    = wr->ex.imm_data;
 		break;
 	default:
 		return -EINVAL;
@@ -1492,7 +1492,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 
 		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
 		    wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
-			ctrl->imm = wr->imm_data;
+			ctrl->imm = wr->ex.imm_data;
 		else
 			ctrl->imm = 0;
 
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index 8433897624bc..b3fd6b05d79d 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1532,7 +1532,7 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	case IB_WR_SEND_WITH_IMM:
 		sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
 		sqp->ud_header.immediate_present = 1;
-		sqp->ud_header.immediate_data = wr->imm_data;
+		sqp->ud_header.immediate_data = wr->ex.imm_data;
 		break;
 	default:
 		return -EINVAL;
@@ -1679,7 +1679,7 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			cpu_to_be32(1);
 		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
 		    wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
-			((struct mthca_next_seg *) wqe)->imm = wr->imm_data;
+			((struct mthca_next_seg *) wqe)->imm = wr->ex.imm_data;
 
 		wqe += sizeof (struct mthca_next_seg);
 		size = sizeof (struct mthca_next_seg) / 16;
@@ -2020,7 +2020,7 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			cpu_to_be32(1);
 		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
 		    wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
-			((struct mthca_next_seg *) wqe)->imm = wr->imm_data;
+			((struct mthca_next_seg *) wqe)->imm = wr->ex.imm_data;
 
 		wqe += sizeof (struct mthca_next_seg);
 		size = sizeof (struct mthca_next_seg) / 16;
diff --git a/drivers/infiniband/hw/nes/nes_hw.c b/drivers/infiniband/hw/nes/nes_hw.c
index 134189d77ed3..aa53aab91bf8 100644
--- a/drivers/infiniband/hw/nes/nes_hw.c
+++ b/drivers/infiniband/hw/nes/nes_hw.c
@@ -393,7 +393,7 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 	nesadapter->base_pd = 1;
 
 	nesadapter->device_cap_flags =
-			IB_DEVICE_ZERO_STAG | IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW;
+		IB_DEVICE_ZERO_STAG | IB_DEVICE_MEM_WINDOW;
 
 	nesadapter->allocated_qps = (unsigned long *)&(((unsigned char *)nesadapter)
 			[(sizeof(struct nes_adapter)+(sizeof(unsigned long)-1))&(~(sizeof(unsigned long)-1))]);
diff --git a/include/rdma/ib_user_verbs.h b/include/rdma/ib_user_verbs.h
index 64a721fcbc1c..8d65bf0a625b 100644
--- a/include/rdma/ib_user_verbs.h
+++ b/include/rdma/ib_user_verbs.h
@@ -533,7 +533,10 @@ struct ib_uverbs_send_wr {
 	__u32 num_sge;
 	__u32 opcode;
 	__u32 send_flags;
-	__u32 imm_data;
+	union {
+		__u32 imm_data;
+		__u32 invalidate_rkey;
+	} ex;
 	union {
 		struct {
 			__u64 remote_addr;
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 66928e9cab19..c48f6af5ef9a 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -94,7 +94,7 @@ enum ib_device_cap_flags {
 	IB_DEVICE_SRQ_RESIZE		= (1<<13),
 	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
 	IB_DEVICE_ZERO_STAG		= (1<<15),
-	IB_DEVICE_SEND_W_INV		= (1<<16),
+	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
 	IB_DEVICE_MEM_WINDOW		= (1<<17),
 	/*
 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
@@ -105,6 +105,7 @@ enum ib_device_cap_flags {
 	 */
 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
 	IB_DEVICE_UD_TSO		= (1<<19),
+	IB_DEVICE_SEND_W_INV		= (1<<21),
 };
 
 enum ib_atomic_cap {
@@ -625,7 +626,8 @@ enum ib_wr_opcode {
 	IB_WR_RDMA_READ,
 	IB_WR_ATOMIC_CMP_AND_SWP,
 	IB_WR_ATOMIC_FETCH_AND_ADD,
-	IB_WR_LSO
+	IB_WR_LSO,
+	IB_WR_SEND_WITH_INV,
 };
 
 enum ib_send_flags {
@@ -649,7 +651,10 @@ struct ib_send_wr {
 	int			num_sge;
 	enum ib_wr_opcode	opcode;
 	int			send_flags;
-	__be32			imm_data;
+	union {
+		__be32		imm_data;
+		u32		invalidate_rkey;
+	} ex;
 	union {
 		struct {
 			u64	remote_addr;
diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index ffbf22a1d2ca..8ea283ecc522 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -1573,7 +1573,6 @@ rpcrdma_ep_post(struct rpcrdma_ia *ia,
 	send_wr.sg_list = req->rl_send_iov;
 	send_wr.num_sge = req->rl_niovs;
 	send_wr.opcode = IB_WR_SEND;
-	send_wr.imm_data = 0;
 	if (send_wr.num_sge == 4)	/* no need to sync any pad (constant) */
 		ib_dma_sync_single_for_device(ia->ri_id->device,
 			req->rl_send_iov[3].addr, req->rl_send_iov[3].length,
-- 
cgit v1.2.3


From 0df67030958bad07c3f82f392e3ffb785c56e48d Mon Sep 17 00:00:00 2001
From: Dotan Barak <dotanb@dev.mellanox.co.il>
Date: Wed, 16 Apr 2008 21:09:34 -0700
Subject: IB/mlx4: Update QP state if query QP succeeds

If the QP was moved to another state (such as SQE) by the hardware,
then after this change the user won't have to set the IBV_QP_CUR_STATE
mask in order to execute modify QP in order to recover from this state.

Signed-off-by: Dotan Barak <dotanb@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
---
 drivers/infiniband/hw/mlx4/qp.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

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

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 38e651a67589..b75efae7e449 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1785,7 +1785,9 @@ int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
 	struct mlx4_qp_context context;
 	int mlx4_state;
-	int err;
+	int err = 0;
+
+	mutex_lock(&qp->mutex);
 
 	if (qp->state == IB_QPS_RESET) {
 		qp_attr->qp_state = IB_QPS_RESET;
@@ -1793,12 +1795,15 @@ int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr
 	}
 
 	err = mlx4_qp_query(dev->dev, &qp->mqp, &context);
-	if (err)
-		return -EINVAL;
+	if (err) {
+		err = -EINVAL;
+		goto out;
+	}
 
 	mlx4_state = be32_to_cpu(context.flags) >> 28;
 
-	qp_attr->qp_state	     = to_ib_qp_state(mlx4_state);
+	qp->state		     = to_ib_qp_state(mlx4_state);
+	qp_attr->qp_state	     = qp->state;
 	qp_attr->path_mtu	     = context.mtu_msgmax >> 5;
 	qp_attr->path_mig_state	     =
 		to_ib_mig_state((be32_to_cpu(context.flags) >> 11) & 0x3);
@@ -1857,6 +1862,8 @@ done:
 
 	qp_init_attr->cap	     = qp_attr->cap;
 
-	return 0;
+out:
+	mutex_unlock(&qp->mutex);
+	return err;
 }
 
-- 
cgit v1.2.3