[ofa-general] [PATCH 10 of 17] mlx4: add LSO support

Eli Cohen eli at mellanox.co.il
Tue Sep 11 08:54:43 PDT 2007


Add LSO support to mlx4

Signed-off-by: Eli Cohen <eli at mellnaox.co.il>

---

Index: ofa_1_3_dev_kernel/drivers/net/mlx4/fw.c
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/net/mlx4/fw.c	2007-09-11 21:15:26.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/net/mlx4/fw.c	2007-09-11 21:15:28.000000000 +0300
@@ -133,6 +133,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *
 #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_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_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);
 
Index: ofa_1_3_dev_kernel/drivers/net/mlx4/fw.h
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/net/mlx4/fw.h	2007-09-11 21:14:34.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/net/mlx4/fw.h	2007-09-11 21:15:28.000000000 +0300
@@ -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 {
Index: ofa_1_3_dev_kernel/drivers/net/mlx4/main.c
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/net/mlx4/main.c	2007-09-11 21:15:18.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/net/mlx4/main.c	2007-09-11 21:15:28.000000000 +0300
@@ -159,6 +159,7 @@ static int __devinit mlx4_dev_cap(struct
 	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;
 }
Index: ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/main.c
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/infiniband/hw/mlx4/main.c	2007-09-11 21:15:26.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/main.c	2007-09-11 21:15:28.000000000 +0300
@@ -102,6 +102,8 @@ static int mlx4_ib_query_device(struct i
 		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_IP_CSUM;
+	if (dev->dev->caps.max_gso_sz)
+		props->device_cap_flags |= IB_DEVICE_TCP_GSO;
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
@@ -630,6 +632,8 @@ static void *mlx4_ib_add(struct mlx4_dev
 
 	if (ibdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 		ibdev->ib_dev.flags |= IB_DEVICE_IP_CSUM;
+	if (ibdev->dev->caps.max_gso_sz)
+		ibdev->ib_dev.flags |= IB_DEVICE_TCP_GSO;
 
 	if (init_node_data(ibdev))
 		goto err_map;
Index: ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/qp.c
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/infiniband/hw/mlx4/qp.c	2007-09-11 21:15:27.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/qp.c	2007-09-11 21:15:28.000000000 +0300
@@ -69,6 +69,7 @@ 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),
@@ -748,9 +749,11 @@ static int __mlx4_ib_modify_qp(struct ib
 		}
 	}
 
-	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 (ibqp->qp_type == IB_QPT_UD)
+		context->mtu_msgmax = (IB_MTU_4096 << 5) |
+			ilog2(dev->dev->caps.max_gso_sz);
 	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",
@@ -1240,6 +1243,29 @@ static void set_data_seg(struct mlx4_wqe
 	dseg->byte_count = cpu_to_be32(sg->length);
 }
 
+static int build_lso_seg(struct mlx4_lso_seg *wqe, struct ib_send_wr *wr,
+			 struct mlx4_ib_qp *qp, int *lso_seg_len)
+{
+	int halign;
+
+	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);
+
+	halign = ALIGN(wr->wr.ud.hlen, 16);
+
+	if (unlikely(wr->num_sge > qp->sq.max_gs - (halign >> 4)))
+		return -EINVAL;
+
+	*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)
 {
@@ -1331,6 +1357,19 @@ int mlx4_ib_post_send(struct ib_qp *ibqp
 			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) {
+				int hlen;
+
+				err = build_lso_seg(wqe, wr, qp, &hlen);
+				if (err) {
+					*bad_wr = wr;
+					goto out;
+				}
+				wqe += hlen;
+				size += hlen >> 4;
+			}
+
 			break;
 
 		case IB_QPT_SMI:
Index: ofa_1_3_dev_kernel/include/linux/mlx4/device.h
===================================================================
--- ofa_1_3_dev_kernel.orig/include/linux/mlx4/device.h	2007-09-11 21:15:11.000000000 +0300
+++ ofa_1_3_dev_kernel/include/linux/mlx4/device.h	2007-09-11 21:15:28.000000000 +0300
@@ -181,6 +181,7 @@ 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 {
Index: ofa_1_3_dev_kernel/include/linux/mlx4/qp.h
===================================================================
--- ofa_1_3_dev_kernel.orig/include/linux/mlx4/qp.h	2007-09-11 21:15:26.000000000 +0300
+++ ofa_1_3_dev_kernel/include/linux/mlx4/qp.h	2007-09-11 21:15:28.000000000 +0300
@@ -215,6 +215,11 @@ 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;
Index: ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/cq.c
===================================================================
--- ofa_1_3_dev_kernel.orig/drivers/infiniband/hw/mlx4/cq.c	2007-09-11 21:15:26.000000000 +0300
+++ ofa_1_3_dev_kernel/drivers/infiniband/hw/mlx4/cq.c	2007-09-11 21:15:28.000000000 +0300
@@ -403,6 +403,9 @@ static int mlx4_ib_poll_one(struct mlx4_
 		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);




More information about the general mailing list