From: Xi Wang <wangxi11@huawei.com>
Date: Tue, 10 Mar 2020 19:18:01 +0800
Subject: RDMA/hns: Optimize wqe buffer filling process for post send
Patch-mainline: v5.7-rc1
Git-commit: 00a59d30f3f953b4dbfffcae96e927e62533e6cf
References: jsc#SLE-14777

Encapsulates the wqe buffer process details for datagram seg, fast mr seg
and atomic seg.

Link: https://lore.kernel.org/r/1583839084-31579-3-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c |   63 ++++++++++++++---------------
 1 file changed, 32 insertions(+), 31 deletions(-)

--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -57,10 +57,10 @@ static void set_data_seg_v2(struct hns_r
 }
 
 static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
-			 struct hns_roce_wqe_frmr_seg *fseg,
-			 const struct ib_reg_wr *wr)
+			 void *wqe, const struct ib_reg_wr *wr)
 {
 	struct hns_roce_mr *mr = to_hr_mr(wr->mr);
+	struct hns_roce_wqe_frmr_seg *fseg = wqe;
 
 	/* use ib_access_flags */
 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_FRMR_WQE_BYTE_4_BIND_EN_S,
@@ -92,16 +92,26 @@ static void set_frmr_seg(struct hns_roce
 		     V2_RC_FRMR_WQE_BYTE_40_BLK_MODE_S, 0);
 }
 
-static void set_atomic_seg(struct hns_roce_wqe_atomic_seg *aseg,
-			   const struct ib_atomic_wr *wr)
+static void set_atomic_seg(const struct ib_send_wr *wr, void *wqe,
+			   struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
+			   int valid_num_sge)
 {
-	if (wr->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
-		aseg->fetchadd_swap_data = cpu_to_le64(wr->swap);
-		aseg->cmp_data  = cpu_to_le64(wr->compare_add);
+	struct hns_roce_wqe_atomic_seg *aseg;
+
+	set_data_seg_v2(wqe, wr->sg_list);
+	aseg = wqe + sizeof(struct hns_roce_v2_wqe_data_seg);
+
+	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+		aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
+		aseg->cmp_data = cpu_to_le64(atomic_wr(wr)->compare_add);
 	} else {
-		aseg->fetchadd_swap_data = cpu_to_le64(wr->compare_add);
+		aseg->fetchadd_swap_data =
+			cpu_to_le64(atomic_wr(wr)->compare_add);
 		aseg->cmp_data  = 0;
 	}
+
+	roce_set_field(rc_sq_wqe->byte_16, V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
+		       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
 }
 
 static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
@@ -154,11 +164,11 @@ static void set_extend_sge(struct hns_ro
 static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			     struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
 			     void *wqe, unsigned int *sge_ind,
-			     int valid_num_sge,
-			     const struct ib_send_wr **bad_wr)
+			     int valid_num_sge)
 {
 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
 	struct hns_roce_v2_wqe_data_seg *dseg = wqe;
+	struct ib_device *ibdev = &hr_dev->ib_dev;
 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
 	int j = 0;
 	int i;
@@ -166,15 +176,14 @@ static int set_rwqe_data_seg(struct ib_q
 	if (wr->send_flags & IB_SEND_INLINE && valid_num_sge) {
 		if (le32_to_cpu(rc_sq_wqe->msg_len) >
 		    hr_dev->caps.max_sq_inline) {
-			*bad_wr = wr;
-			dev_err(hr_dev->dev, "inline len(1-%d)=%d, illegal",
-				rc_sq_wqe->msg_len, hr_dev->caps.max_sq_inline);
+			ibdev_err(ibdev, "inline len(1-%d)=%d, illegal",
+				  rc_sq_wqe->msg_len,
+				  hr_dev->caps.max_sq_inline);
 			return -EINVAL;
 		}
 
 		if (wr->opcode == IB_WR_RDMA_READ) {
-			*bad_wr =  wr;
-			dev_err(hr_dev->dev, "Not support inline data!\n");
+			ibdev_err(ibdev, "Not support inline data!\n");
 			return -EINVAL;
 		}
 
@@ -285,7 +294,6 @@ static int hns_roce_v2_post_send(struct
 	struct hns_roce_v2_ud_send_wqe *ud_sq_wqe;
 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe;
 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
-	struct hns_roce_wqe_frmr_seg *fseg;
 	struct device *dev = hr_dev->dev;
 	unsigned int owner_bit;
 	unsigned int sge_idx;
@@ -547,8 +555,7 @@ static int hns_roce_v2_post_send(struct
 				break;
 			case IB_WR_REG_MR:
 				hr_op = HNS_ROCE_V2_WQE_OP_FAST_REG_PMR;
-				fseg = wqe;
-				set_frmr_seg(rc_sq_wqe, fseg, reg_wr(wr));
+				set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
 				break;
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 				hr_op = HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP;
@@ -582,23 +589,17 @@ static int hns_roce_v2_post_send(struct
 				       V2_RC_SEND_WQE_BYTE_4_OPCODE_S, hr_op);
 
 			if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
-			    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
-				struct hns_roce_v2_wqe_data_seg *dseg;
-
-				dseg = wqe;
-				set_data_seg_v2(dseg, wr->sg_list);
-				wqe += sizeof(struct hns_roce_v2_wqe_data_seg);
-				set_atomic_seg(wqe, atomic_wr(wr));
-				roce_set_field(rc_sq_wqe->byte_16,
-					       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
-					       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S,
+			    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+				set_atomic_seg(wr, wqe, rc_sq_wqe,
 					       valid_num_sge);
-			} else if (wr->opcode != IB_WR_REG_MR) {
+			else if (wr->opcode != IB_WR_REG_MR) {
 				ret = set_rwqe_data_seg(ibqp, wr, rc_sq_wqe,
 							wqe, &sge_idx,
-							valid_num_sge, bad_wr);
-				if (ret)
+							valid_num_sge);
+				if (ret) {
+					*bad_wr = wr;
 					goto out;
+				}
 			}
 		} else {
 			dev_err(dev, "Illegal qp_type(0x%x)\n", ibqp->qp_type);