[PATCH v2 for-next 0/7] RDMA/hns: Updates for kernel v5.8

[PATCH v2 for-next 0/7] RDMA/hns: Updates for kernel v5.8

[Weihang Li]

These are some miscellaneous changes of hns driver. #1 ~ #5 are cleanups
and #6 ~ #7 are small fixes.

Changes since v1:
- Fix comments from Leon about the judgment of return value in #2.
- Rewrite the commit message and remove the unused macro in #3.

Lang Cheng (4):
  RDMA/hns: Remove redundant hardware opcode definitions
  RDMA/hns: Remove support for HIP08_A
  RDMA/hns: Delete unnecessary memset when allocating VF resource
  RDMA/hns: Fix error during modify qp RTS2RTS

Weihang Li (2):
  RDMA/hns: Refactor hns_roce_v2_set_hem()
  RDMA/hns: Remove redundant parameters in set_rc_wqe()

Xi Wang (1):
  RDMA/hns: Fix the unneeded process when getting a general type of CQE
    error

 drivers/infiniband/hw/hns/hns_roce_device.h |   5 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 209 +++++++++++++++-------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  19 +--
 drivers/infiniband/hw/hns/hns_roce_qp.c     |  10 --
 4 files changed, 115 insertions(+), 128 deletions(-)

-- 
2.8.1

[PATCH v2 for-next 1/7] RDMA/hns: Remove redundant hardware opcode definitions

[Weihang Li]

From: Lang Cheng <chenglang@huawei.com>

HNS_ROCE_SQ_OPCODE_XXXs and HNS_ROCE_V2_WQE_OP_XXXs have same values, so
remove a set of redundant definitions. In addition, remove the suffix of
HNS_ROCE_V2_WQE_OP_BIND_MW_TYPE.

Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 26 +++++++++++++-------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 18 +-----------------
 2 files changed, 14 insertions(+), 30 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 2b0b676..35d46b7 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -3169,51 +3169,51 @@ static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
 		/* SQ corresponding to CQE */
 		switch (roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_OPCODE_M,
 				       V2_CQE_BYTE_4_OPCODE_S) & 0x1f) {
-		case HNS_ROCE_SQ_OPCODE_SEND:
+		case HNS_ROCE_V2_WQE_OP_SEND:
 			wc->opcode = IB_WC_SEND;
 			break;
-		case HNS_ROCE_SQ_OPCODE_SEND_WITH_INV:
+		case HNS_ROCE_V2_WQE_OP_SEND_WITH_INV:
 			wc->opcode = IB_WC_SEND;
 			break;
-		case HNS_ROCE_SQ_OPCODE_SEND_WITH_IMM:
+		case HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM:
 			wc->opcode = IB_WC_SEND;
 			wc->wc_flags |= IB_WC_WITH_IMM;
 			break;
-		case HNS_ROCE_SQ_OPCODE_RDMA_READ:
+		case HNS_ROCE_V2_WQE_OP_RDMA_READ:
 			wc->opcode = IB_WC_RDMA_READ;
 			wc->byte_len = le32_to_cpu(cqe->byte_cnt);
 			break;
-		case HNS_ROCE_SQ_OPCODE_RDMA_WRITE:
+		case HNS_ROCE_V2_WQE_OP_RDMA_WRITE:
 			wc->opcode = IB_WC_RDMA_WRITE;
 			break;
-		case HNS_ROCE_SQ_OPCODE_RDMA_WRITE_WITH_IMM:
+		case HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM:
 			wc->opcode = IB_WC_RDMA_WRITE;
 			wc->wc_flags |= IB_WC_WITH_IMM;
 			break;
-		case HNS_ROCE_SQ_OPCODE_LOCAL_INV:
+		case HNS_ROCE_V2_WQE_OP_LOCAL_INV:
 			wc->opcode = IB_WC_LOCAL_INV;
 			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
 			break;
-		case HNS_ROCE_SQ_OPCODE_ATOMIC_COMP_AND_SWAP:
+		case HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP:
 			wc->opcode = IB_WC_COMP_SWAP;
 			wc->byte_len  = 8;
 			break;
-		case HNS_ROCE_SQ_OPCODE_ATOMIC_FETCH_AND_ADD:
+		case HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD:
 			wc->opcode = IB_WC_FETCH_ADD;
 			wc->byte_len  = 8;
 			break;
-		case HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_COMP_AND_SWAP:
+		case HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP:
 			wc->opcode = IB_WC_MASKED_COMP_SWAP;
 			wc->byte_len  = 8;
 			break;
-		case HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_FETCH_AND_ADD:
+		case HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD:
 			wc->opcode = IB_WC_MASKED_FETCH_ADD;
 			wc->byte_len  = 8;
 			break;
-		case HNS_ROCE_SQ_OPCODE_FAST_REG_WR:
+		case HNS_ROCE_V2_WQE_OP_FAST_REG_PMR:
 			wc->opcode = IB_WC_REG_MR;
 			break;
-		case HNS_ROCE_SQ_OPCODE_BIND_MW:
+		case HNS_ROCE_V2_WQE_OP_BIND_MW:
 			wc->opcode = IB_WC_REG_MR;
 			break;
 		default:
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index e176b0a..53c26f3 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -179,27 +179,11 @@ enum {
 	HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD	= 0x9,
 	HNS_ROCE_V2_WQE_OP_FAST_REG_PMR			= 0xa,
 	HNS_ROCE_V2_WQE_OP_LOCAL_INV			= 0xb,
-	HNS_ROCE_V2_WQE_OP_BIND_MW_TYPE			= 0xc,
+	HNS_ROCE_V2_WQE_OP_BIND_MW			= 0xc,
 	HNS_ROCE_V2_WQE_OP_MASK				= 0x1f,
 };
 
 enum {
-	HNS_ROCE_SQ_OPCODE_SEND = 0x0,
-	HNS_ROCE_SQ_OPCODE_SEND_WITH_INV = 0x1,
-	HNS_ROCE_SQ_OPCODE_SEND_WITH_IMM = 0x2,
-	HNS_ROCE_SQ_OPCODE_RDMA_WRITE = 0x3,
-	HNS_ROCE_SQ_OPCODE_RDMA_WRITE_WITH_IMM = 0x4,
-	HNS_ROCE_SQ_OPCODE_RDMA_READ = 0x5,
-	HNS_ROCE_SQ_OPCODE_ATOMIC_COMP_AND_SWAP = 0x6,
-	HNS_ROCE_SQ_OPCODE_ATOMIC_FETCH_AND_ADD = 0x7,
-	HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_COMP_AND_SWAP = 0x8,
-	HNS_ROCE_SQ_OPCODE_ATOMIC_MASK_FETCH_AND_ADD = 0x9,
-	HNS_ROCE_SQ_OPCODE_FAST_REG_WR = 0xa,
-	HNS_ROCE_SQ_OPCODE_LOCAL_INV = 0xb,
-	HNS_ROCE_SQ_OPCODE_BIND_MW = 0xc,
-};
-
-enum {
 	/* rq operations */
 	HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM = 0x0,
 	HNS_ROCE_V2_OPCODE_SEND = 0x1,
-- 
2.8.1

[PATCH v2 for-next 2/7] RDMA/hns: Refactor hns_roce_v2_set_hem()

[Weihang Li]

The parts about preparing and sending mailbox to hardware is not strongly
related to other codes in hns_roce_v2_set_hem(), and can be encapsulated
into a separate function.

Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 45 +++++++++++++++++-------------
 1 file changed, 26 insertions(+), 19 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 35d46b7..516e246 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -3373,11 +3373,33 @@ static int get_op_for_set_hem(struct hns_roce_dev *hr_dev, u32 type,
 	return op + step_idx;
 }
 
+static int set_hem_to_hw(struct hns_roce_dev *hr_dev, int obj, u64 bt_ba,
+			 u32 hem_type, int step_idx)
+{
+	struct hns_roce_cmd_mailbox *mailbox;
+	int ret;
+	int op;
+
+	op = get_op_for_set_hem(hr_dev, hem_type, step_idx);
+	if (op < 0)
+		return 0;
+
+	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	ret = hns_roce_cmd_mbox(hr_dev, bt_ba, mailbox->dma, obj,
+				0, op, HNS_ROCE_CMD_TIMEOUT_MSECS);
+
+	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
+
+	return ret;
+}
+
 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 			       struct hns_roce_hem_table *table, int obj,
 			       int step_idx)
 {
-	struct hns_roce_cmd_mailbox *mailbox;
 	struct hns_roce_hem_iter iter;
 	struct hns_roce_hem_mhop mhop;
 	struct hns_roce_hem *hem;
@@ -3389,7 +3411,6 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 	u64 bt_ba = 0;
 	u32 chunk_ba_num;
 	u32 hop_num;
-	int op;
 
 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
 		return 0;
@@ -3411,14 +3432,6 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 		hem_idx = i;
 	}
 
-	op = get_op_for_set_hem(hr_dev, table->type, step_idx);
-	if (op == -EINVAL)
-		return 0;
-
-	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
-	if (IS_ERR(mailbox))
-		return PTR_ERR(mailbox);
-
 	if (table->type == HEM_TYPE_SCCC)
 		obj = mhop.l0_idx;
 
@@ -3427,11 +3440,8 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 		for (hns_roce_hem_first(hem, &iter);
 		     !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) {
 			bt_ba = hns_roce_hem_addr(&iter);
-
-			/* configure the ba, tag, and op */
-			ret = hns_roce_cmd_mbox(hr_dev, bt_ba, mailbox->dma,
-						obj, 0, op,
-						HNS_ROCE_CMD_TIMEOUT_MSECS);
+			ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type,
+					    step_idx);
 		}
 	} else {
 		if (step_idx == 0)
@@ -3439,12 +3449,9 @@ static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
 		else if (step_idx == 1 && hop_num == 2)
 			bt_ba = table->bt_l1_dma_addr[l1_idx];
 
-		/* configure the ba, tag, and op */
-		ret = hns_roce_cmd_mbox(hr_dev, bt_ba, mailbox->dma, obj,
-					0, op, HNS_ROCE_CMD_TIMEOUT_MSECS);
+		ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type, step_idx);
 	}
 
-	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
 	return ret;
 }
 
-- 
2.8.1

[PATCH v2 for-next 3/7] RDMA/hns: Remove support for HIP08_A

[Weihang Li]

From: Lang Cheng <chenglang@huawei.com>

HIP08_A is an temporary version and all features of it are supported by
HIP08_B. So remove the relevant code.

Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_device.h |  5 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 97 +++++++++++++----------------
 drivers/infiniband/hw/hns/hns_roce_qp.c     | 10 ---
 3 files changed, 47 insertions(+), 65 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index 6e64931..846954e 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -37,9 +37,8 @@
 
 #define DRV_NAME "hns_roce"
 
-/* hip08 is a pci device, it includes two version according pci version id */
-#define PCI_REVISION_ID_HIP08_A			0x20
-#define PCI_REVISION_ID_HIP08_B			0x21
+/* hip08 is a pci device */
+#define PCI_REVISION_ID_HIP08			0x21
 
 #define HNS_ROCE_HW_VER1	('h' << 24 | 'i' << 16 | '0' << 8 | '6')
 
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 516e246..ff738db 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -1744,27 +1744,25 @@ static void set_default_caps(struct hns_roce_dev *hr_dev)
 	caps->max_srq_wrs	= HNS_ROCE_V2_MAX_SRQ_WR;
 	caps->max_srq_sges	= HNS_ROCE_V2_MAX_SRQ_SGE;
 
-	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP08_B) {
-		caps->flags |= HNS_ROCE_CAP_FLAG_ATOMIC | HNS_ROCE_CAP_FLAG_MW |
-			       HNS_ROCE_CAP_FLAG_SRQ | HNS_ROCE_CAP_FLAG_FRMR |
-			       HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL;
-
-		caps->num_qpc_timer	  = HNS_ROCE_V2_MAX_QPC_TIMER_NUM;
-		caps->qpc_timer_entry_sz  = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
-		caps->qpc_timer_ba_pg_sz  = 0;
-		caps->qpc_timer_buf_pg_sz = 0;
-		caps->qpc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
-		caps->num_cqc_timer	  = HNS_ROCE_V2_MAX_CQC_TIMER_NUM;
-		caps->cqc_timer_entry_sz  = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
-		caps->cqc_timer_ba_pg_sz  = 0;
-		caps->cqc_timer_buf_pg_sz = 0;
-		caps->cqc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
-
-		caps->sccc_entry_sz	  = HNS_ROCE_V2_SCCC_ENTRY_SZ;
-		caps->sccc_ba_pg_sz	  = 0;
-		caps->sccc_buf_pg_sz	  = 0;
-		caps->sccc_hop_num	  = HNS_ROCE_SCCC_HOP_NUM;
-	}
+	caps->flags |= HNS_ROCE_CAP_FLAG_ATOMIC | HNS_ROCE_CAP_FLAG_MW |
+		       HNS_ROCE_CAP_FLAG_SRQ | HNS_ROCE_CAP_FLAG_FRMR |
+		       HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL;
+
+	caps->num_qpc_timer	  = HNS_ROCE_V2_MAX_QPC_TIMER_NUM;
+	caps->qpc_timer_entry_sz  = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
+	caps->qpc_timer_ba_pg_sz  = 0;
+	caps->qpc_timer_buf_pg_sz = 0;
+	caps->qpc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
+	caps->num_cqc_timer	  = HNS_ROCE_V2_MAX_CQC_TIMER_NUM;
+	caps->cqc_timer_entry_sz  = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
+	caps->cqc_timer_ba_pg_sz  = 0;
+	caps->cqc_timer_buf_pg_sz = 0;
+	caps->cqc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
+
+	caps->sccc_entry_sz	  = HNS_ROCE_V2_SCCC_ENTRY_SZ;
+	caps->sccc_ba_pg_sz	  = 0;
+	caps->sccc_buf_pg_sz	  = 0;
+	caps->sccc_hop_num	  = HNS_ROCE_SCCC_HOP_NUM;
 }
 
 static void calc_pg_sz(int obj_num, int obj_size, int hop_num, int ctx_bt_num,
@@ -1995,20 +1993,18 @@ static int hns_roce_query_pf_caps(struct hns_roce_dev *hr_dev)
 		   caps->srqc_bt_num, &caps->srqc_buf_pg_sz,
 		   &caps->srqc_ba_pg_sz, HEM_TYPE_SRQC);
 
-	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP08_B) {
-		caps->sccc_hop_num = ctx_hop_num;
-		caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
-		caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
+	caps->sccc_hop_num = ctx_hop_num;
+	caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
+	caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
 
-		calc_pg_sz(caps->num_qps, caps->sccc_entry_sz,
-			   caps->sccc_hop_num, caps->sccc_bt_num,
-			   &caps->sccc_buf_pg_sz, &caps->sccc_ba_pg_sz,
-			   HEM_TYPE_SCCC);
-		calc_pg_sz(caps->num_cqc_timer, caps->cqc_timer_entry_sz,
-			   caps->cqc_timer_hop_num, caps->cqc_timer_bt_num,
-			   &caps->cqc_timer_buf_pg_sz,
-			   &caps->cqc_timer_ba_pg_sz, HEM_TYPE_CQC_TIMER);
-	}
+	calc_pg_sz(caps->num_qps, caps->sccc_entry_sz,
+		   caps->sccc_hop_num, caps->sccc_bt_num,
+		   &caps->sccc_buf_pg_sz, &caps->sccc_ba_pg_sz,
+		   HEM_TYPE_SCCC);
+	calc_pg_sz(caps->num_cqc_timer, caps->cqc_timer_entry_sz,
+		   caps->cqc_timer_hop_num, caps->cqc_timer_bt_num,
+		   &caps->cqc_timer_buf_pg_sz,
+		   &caps->cqc_timer_ba_pg_sz, HEM_TYPE_CQC_TIMER);
 
 	calc_pg_sz(caps->num_cqe_segs, caps->mtt_entry_sz, caps->cqe_hop_num,
 		   1, &caps->cqe_buf_pg_sz, &caps->cqe_ba_pg_sz, HEM_TYPE_CQE);
@@ -2055,22 +2051,19 @@ static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
 		return ret;
 	}
 
-	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP08_B) {
-		ret = hns_roce_query_pf_timer_resource(hr_dev);
-		if (ret) {
-			dev_err(hr_dev->dev,
-				"Query pf timer resource fail, ret = %d.\n",
-				ret);
-			return ret;
-		}
+	ret = hns_roce_query_pf_timer_resource(hr_dev);
+	if (ret) {
+		dev_err(hr_dev->dev,
+			"failed to query pf timer resource, ret = %d.\n", ret);
+		return ret;
+	}
 
-		ret = hns_roce_set_vf_switch_param(hr_dev, 0);
-		if (ret) {
-			dev_err(hr_dev->dev,
-				"Set function switch param fail, ret = %d.\n",
-				ret);
-			return ret;
-		}
+	ret = hns_roce_set_vf_switch_param(hr_dev, 0);
+	if (ret) {
+		dev_err(hr_dev->dev,
+			"failed to set function switch param, ret = %d.\n",
+			ret);
+		return ret;
 	}
 
 	hr_dev->vendor_part_id = hr_dev->pci_dev->device;
@@ -2336,8 +2329,7 @@ static void hns_roce_v2_exit(struct hns_roce_dev *hr_dev)
 {
 	struct hns_roce_v2_priv *priv = hr_dev->priv;
 
-	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP08_B)
-		hns_roce_function_clear(hr_dev);
+	hns_roce_function_clear(hr_dev);
 
 	hns_roce_free_link_table(hr_dev, &priv->tpq);
 	hns_roce_free_link_table(hr_dev, &priv->tsq);
@@ -4231,12 +4223,13 @@ static int hns_roce_v2_set_path(struct ib_qp *ibqp,
 	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_HOP_LIMIT_M,
 		       V2_QPC_BYTE_24_HOP_LIMIT_S, 0);
 
-	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP08_B && is_udp)
+	if (is_udp)
 		roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
 			       V2_QPC_BYTE_24_TC_S, grh->traffic_class >> 2);
 	else
 		roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
 			       V2_QPC_BYTE_24_TC_S, grh->traffic_class);
+
 	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
 		       V2_QPC_BYTE_24_TC_S, 0);
 	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_FL_M,
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index a0a47bd..e94ca13 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -411,7 +411,6 @@ static int set_extend_sge_param(struct hns_roce_dev *hr_dev, u32 sq_wqe_cnt,
 				struct hns_roce_qp *hr_qp,
 				struct ib_qp_cap *cap)
 {
-	struct ib_device *ibdev = &hr_dev->ib_dev;
 	u32 cnt;
 
 	cnt = max(1U, cap->max_send_sge);
@@ -431,15 +430,6 @@ static int set_extend_sge_param(struct hns_roce_dev *hr_dev, u32 sq_wqe_cnt,
 	} else if (hr_qp->sq.max_gs > HNS_ROCE_SGE_IN_WQE) {
 		cnt = roundup_pow_of_two(sq_wqe_cnt *
 				     (hr_qp->sq.max_gs - HNS_ROCE_SGE_IN_WQE));
-
-		if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08_A) {
-			if (cnt > hr_dev->caps.max_extend_sg) {
-				ibdev_err(ibdev,
-					  "failed to check exSGE num, exSGE num = %d.\n",
-					  cnt);
-				return -EINVAL;
-			}
-		}
 	} else {
 		cnt = 0;
 	}
-- 
2.8.1

[PATCH v2 for-next 4/7] RDMA/hns: Remove redundant parameters in set_rc_wqe()

[Weihang Li]

There are some functions called by set_rc_wqe() use two parameters:
"void *wqe" and "struct hns_roce_v2_rc_send_wqe *rc_sq_wqe", but the first
one can be got from the second one. So remove the redundant wqe from
related functions.

Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 30 +++++++++++++++++-------------
 1 file changed, 17 insertions(+), 13 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index ff738db..8cda4a9 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -91,10 +91,11 @@ static u32 to_hr_opcode(u32 ib_opcode)
 }
 
 static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
-			 void *wqe, const struct ib_reg_wr *wr)
+			 const struct ib_reg_wr *wr)
 {
+	struct hns_roce_wqe_frmr_seg *fseg =
+		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
 	struct hns_roce_mr *mr = to_hr_mr(wr->mr);
-	struct hns_roce_wqe_frmr_seg *fseg = wqe;
 	u64 pbl_ba;
 
 	/* use ib_access_flags */
@@ -128,14 +129,16 @@ static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
 		     V2_RC_FRMR_WQE_BYTE_40_BLK_MODE_S, 0);
 }
 
-static void set_atomic_seg(const struct ib_send_wr *wr, void *wqe,
+static void set_atomic_seg(const struct ib_send_wr *wr,
 			   struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
 			   unsigned int valid_num_sge)
 {
-	struct hns_roce_wqe_atomic_seg *aseg;
+	struct hns_roce_v2_wqe_data_seg *dseg =
+		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
+	struct hns_roce_wqe_atomic_seg *aseg =
+		(void *)dseg + sizeof(struct hns_roce_v2_wqe_data_seg);
 
-	set_data_seg_v2(wqe, wr->sg_list);
-	aseg = wqe + sizeof(struct hns_roce_v2_wqe_data_seg);
+	set_data_seg_v2(dseg, wr->sg_list);
 
 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
 		aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
@@ -143,7 +146,7 @@ static void set_atomic_seg(const struct ib_send_wr *wr, void *wqe,
 	} else {
 		aseg->fetchadd_swap_data =
 			cpu_to_le64(atomic_wr(wr)->compare_add);
-		aseg->cmp_data  = 0;
+		aseg->cmp_data = 0;
 	}
 
 	roce_set_field(rc_sq_wqe->byte_16, V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
@@ -176,13 +179,15 @@ static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
 
 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,
+			     unsigned int *sge_ind,
 			     unsigned 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 hns_roce_v2_wqe_data_seg *dseg =
+		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
 	struct ib_device *ibdev = &hr_dev->ib_dev;
 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
+	void *wqe = dseg;
 	int j = 0;
 	int i;
 
@@ -438,7 +443,6 @@ static inline int set_rc_wqe(struct hns_roce_qp *qp,
 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_OWNER_S,
 		     owner_bit);
 
-	wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
 	switch (wr->opcode) {
 	case IB_WR_RDMA_READ:
 	case IB_WR_RDMA_WRITE:
@@ -451,7 +455,7 @@ static inline int set_rc_wqe(struct hns_roce_qp *qp,
 		rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
 		break;
 	case IB_WR_REG_MR:
-		set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
+		set_frmr_seg(rc_sq_wqe, reg_wr(wr));
 		break;
 	case IB_WR_ATOMIC_CMP_AND_SWP:
 	case IB_WR_ATOMIC_FETCH_AND_ADD:
@@ -468,10 +472,10 @@ static inline int set_rc_wqe(struct hns_roce_qp *qp,
 
 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
 	    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
-		set_atomic_seg(wr, wqe, rc_sq_wqe, valid_num_sge);
+		set_atomic_seg(wr, rc_sq_wqe, valid_num_sge);
 	else if (wr->opcode != IB_WR_REG_MR)
 		ret = set_rwqe_data_seg(&qp->ibqp, wr, rc_sq_wqe,
-					wqe, &curr_idx, valid_num_sge);
+					&curr_idx, valid_num_sge);
 
 	*sge_idx = curr_idx;
 
-- 
2.8.1

[PATCH v2 for-next 5/7] RDMA/hns: Delete unnecessary memset when allocating VF resource

[Weihang Li]

From: Lang Cheng <chenglang@huawei.com>

The hns_roce_cmq_setup_basic_desc() can clear the whole desc, so removes
these redundant memset operations.

Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 8cda4a9..9d64804 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -1514,8 +1514,6 @@ static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
 
 	req_a = (struct hns_roce_vf_res_a *)desc[0].data;
 	req_b = (struct hns_roce_vf_res_b *)desc[1].data;
-	memset(req_a, 0, sizeof(*req_a));
-	memset(req_b, 0, sizeof(*req_b));
 	for (i = 0; i < 2; i++) {
 		hns_roce_cmq_setup_basic_desc(&desc[i],
 					      HNS_ROCE_OPC_ALLOC_VF_RES, false);
-- 
2.8.1

[PATCH v2 for-next 6/7] RDMA/hns: Fix error during modify qp RTS2RTS

[Weihang Li]

From: Lang Cheng <chenglang@huawei.com>

One qp state migrations legal configuration was deleted mistakenly.

Fixes: 357f34294686 ("RDMA/hns: Simplify the state judgment code of qp")
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 9d64804..0c81100 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -4262,7 +4262,9 @@ static bool check_qp_state(enum ib_qp_state cur_state,
 		[IB_QPS_RTR] = { [IB_QPS_RESET] = true,
 				 [IB_QPS_RTS] = true,
 				 [IB_QPS_ERR] = true },
-		[IB_QPS_RTS] = { [IB_QPS_RESET] = true, [IB_QPS_ERR] = true },
+		[IB_QPS_RTS] = { [IB_QPS_RESET] = true,
+				 [IB_QPS_RTS] = true,
+				 [IB_QPS_ERR] = true },
 		[IB_QPS_SQD] = {},
 		[IB_QPS_SQE] = {},
 		[IB_QPS_ERR] = { [IB_QPS_RESET] = true, [IB_QPS_ERR] = true }
-- 
2.8.1

[PATCH v2 for-next 7/7] RDMA/hns: Fix the unneeded process when getting a general type of CQE error

[Weihang Li]

From: Xi Wang <wangxi11@huawei.com>

If the hns ROCEE reports a general error CQE (types not specified by the IB
General Specifications), it's no need to change the QP state to error, and
the driver should just skip it.

Fixes: 7c044adca272 ("RDMA/hns: Simplify the cqe code of poll cq")
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 9 +++++++++
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 1 +
 2 files changed, 10 insertions(+)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 0c81100..d51b332 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -3046,6 +3046,7 @@ static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
 		  IB_WC_RETRY_EXC_ERR },
 		{ HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR, IB_WC_RNR_RETRY_EXC_ERR },
 		{ HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR, IB_WC_REM_ABORT_ERR },
+		{ HNS_ROCE_CQE_V2_GENERAL_ERR, IB_WC_GENERAL_ERR}
 	};
 
 	u32 cqe_status = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_STATUS_M,
@@ -3068,6 +3069,14 @@ static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
 		       sizeof(*cqe), false);
 
 	/*
+	 * For hns ROCEE, GENERAL_ERR is an error type that is not defined in
+	 * the standard protocol, the driver must ignore it and needn't to set
+	 * the QP to an error state.
+	 */
+	if (cqe_status == HNS_ROCE_CQE_V2_GENERAL_ERR)
+		return;
+
+	/*
 	 * Hip08 hardware cannot flush the WQEs in SQ/RQ if the QP state gets
 	 * into errored mode. Hence, as a workaround to this hardware
 	 * limitation, driver needs to assist in flushing. But the flushing
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
index 53c26f3..1fb1c58 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.h
@@ -214,6 +214,7 @@ enum {
 	HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR		= 0x15,
 	HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR		= 0x16,
 	HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR		= 0x22,
+	HNS_ROCE_CQE_V2_GENERAL_ERR			= 0x23,
 
 	HNS_ROCE_V2_CQE_STATUS_MASK			= 0xff,
 };
-- 
2.8.1

Re: [PATCH v2 for-next 0/7] RDMA/hns: Updates for kernel v5.8

[Jason Gunthorpe]

On Tue, Jul 28, 2020 at 06:42:14PM +0800, Weihang Li wrote:
> These are some miscellaneous changes of hns driver. #1 ~ #5 are cleanups
> and #6 ~ #7 are small fixes.
> 
> Changes since v1:
> - Fix comments from Leon about the judgment of return value in #2.
> - Rewrite the commit message and remove the unused macro in #3.
> 
> Lang Cheng (4):
>   RDMA/hns: Remove redundant hardware opcode definitions
>   RDMA/hns: Remove support for HIP08_A
>   RDMA/hns: Delete unnecessary memset when allocating VF resource
>   RDMA/hns: Fix error during modify qp RTS2RTS
> 
> Weihang Li (2):
>   RDMA/hns: Refactor hns_roce_v2_set_hem()
>   RDMA/hns: Remove redundant parameters in set_rc_wqe()
> 
> Xi Wang (1):
>   RDMA/hns: Fix the unneeded process when getting a general type of CQE
>     error

Applied to for-next, thanks

Jason