RDMA/hns: Support cq record doorbell for the user space

This patch updates to support cq record doorbell for
the user space.

Signed-off-by: Yixian Liu <[email protected]>
Signed-off-by: Lijun Ou <[email protected]>
Signed-off-by: Wei Hu (Xavier) <[email protected]>
Signed-off-by: Shaobo Xu <[email protected]>
Signed-off-by: Doug Ledford <[email protected]>

Author: Yixian Liu

drivers/infiniband/hw/hns/hns_roce_cq.c

@@ -315,6 +315,7 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
 	struct device *dev = hr_dev->dev;
 	struct hns_roce_ib_create_cq ucmd;
+	struct hns_roce_ib_create_cq_resp resp;
 	struct hns_roce_cq *hr_cq = NULL;
 	struct hns_roce_uar *uar = NULL;
 	int vector = attr->comp_vector;
@@ -378,6 +379,16 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 		goto err_mtt;
 	}
 
+	if (context && (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
+	    (udata->outlen == sizeof(resp))) {
+		ret = hns_roce_db_map_user(to_hr_ucontext(context),
+					   ucmd.db_addr, &hr_cq->db);
+		if (ret) {
+			dev_err(dev, "cq record doorbell map failed!\n");
+			goto err_cqc;
+		}
+	}
+
 	/*
 	 * For the QP created by kernel space, tptr value should be initialized
 	 * to zero; For the QP created by user space, it will cause synchronous
@@ -393,14 +404,27 @@ struct ib_cq *hns_roce_ib_create_cq(struct ib_device *ib_dev,
 	hr_cq->cq_depth = cq_entries;
 
 	if (context) {
-		if (ib_copy_to_udata(udata, &hr_cq->cqn, sizeof(u64))) {
-			ret = -EFAULT;
-			goto err_cqc;
-		}
+		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
+					(udata->outlen == sizeof(resp))) {
+			hr_cq->db_en = 1;
+			resp.cqn = hr_cq->cqn;
+			resp.cap_flags |= HNS_ROCE_SUPPORT_CQ_RECORD_DB;
+			ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
+		} else
+			ret = ib_copy_to_udata(udata, &hr_cq->cqn, sizeof(u64));
+
+		if (ret)
+			goto err_dbmap;
 	}
 
 	return &hr_cq->ib_cq;
 
+err_dbmap:
+	if (context && (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
+	    (udata->outlen == sizeof(resp)))
+		hns_roce_db_unmap_user(to_hr_ucontext(context),
+				       &hr_cq->db);
+
 err_cqc:
 	hns_roce_free_cq(hr_dev, hr_cq);
 
@@ -430,12 +454,18 @@ int hns_roce_ib_destroy_cq(struct ib_cq *ib_cq)
 		hns_roce_free_cq(hr_dev, hr_cq);
 		hns_roce_mtt_cleanup(hr_dev, &hr_cq->hr_buf.hr_mtt);
 
-		if (ib_cq->uobject)
+		if (ib_cq->uobject) {
 			ib_umem_release(hr_cq->umem);
-		else
+
+			if (hr_cq->db_en == 1)
+				hns_roce_db_unmap_user(
+					to_hr_ucontext(ib_cq->uobject->context),
+					&hr_cq->db);
+		} else {
 			/* Free the buff of stored cq */
 			hns_roce_ib_free_cq_buf(hr_dev, &hr_cq->hr_buf,
 						ib_cq->cqe);
+		}
 
 		kfree(hr_cq);
 	}

drivers/infiniband/hw/hns/hns_roce_device.h

@@ -109,6 +109,10 @@ enum {
 	HNS_ROCE_SUPPORT_RQ_RECORD_DB = 1 << 0,
 };
 
+enum {
+	HNS_ROCE_SUPPORT_CQ_RECORD_DB = 1 << 0,
+};
+
 enum hns_roce_qp_state {
 	HNS_ROCE_QP_STATE_RST,
 	HNS_ROCE_QP_STATE_INIT,
@@ -381,6 +385,8 @@ struct hns_roce_cq_buf {
 struct hns_roce_cq {
 	struct ib_cq			ib_cq;
 	struct hns_roce_cq_buf		hr_buf;
+	struct hns_roce_db		db;
+	u8				db_en;
 	spinlock_t			lock;
 	struct ib_umem			*umem;
 	void (*comp)(struct hns_roce_cq *cq);

drivers/infiniband/hw/hns/hns_roce_hw_v2.c

@@ -1638,6 +1638,16 @@ static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
 	roce_set_field(cq_context->byte_40_cqe_ba, V2_CQC_BYTE_40_CQE_BA_M,
 		       V2_CQC_BYTE_40_CQE_BA_S, (dma_handle >> (32 + 3)));
 
+	if (hr_cq->db_en)
+		roce_set_bit(cq_context->byte_44_db_record,
+			     V2_CQC_BYTE_44_DB_RECORD_EN_S, 1);
+
+	roce_set_field(cq_context->byte_44_db_record,
+		       V2_CQC_BYTE_44_DB_RECORD_ADDR_M,
+		       V2_CQC_BYTE_44_DB_RECORD_ADDR_S,
+		       ((u32)hr_cq->db.dma) >> 1);
+	cq_context->db_record_addr = hr_cq->db.dma >> 32;
+
 	roce_set_field(cq_context->byte_56_cqe_period_maxcnt,
 		       V2_CQC_BYTE_56_CQ_MAX_CNT_M,
 		       V2_CQC_BYTE_56_CQ_MAX_CNT_S,

drivers/infiniband/hw/hns/hns_roce_hw_v2.h

@@ -299,6 +299,9 @@ struct hns_roce_v2_cq_context {
 
 #define	V2_CQC_BYTE_44_DB_RECORD_EN_S 0
 
+#define	V2_CQC_BYTE_44_DB_RECORD_ADDR_S 1
+#define	V2_CQC_BYTE_44_DB_RECORD_ADDR_M GENMASK(31, 1)
+
 #define	V2_CQC_BYTE_52_CQE_CNT_S 0
 #define	V2_CQC_BYTE_52_CQE_CNT_M GENMASK(23, 0)
 

include/uapi/rdma/hns-abi.h

@@ -38,6 +38,13 @@
 
 struct hns_roce_ib_create_cq {
 	__u64   buf_addr;
+	__u64	db_addr;
+};
+
+struct hns_roce_ib_create_cq_resp {
+	__u32	cqn;
+	__u32	reserved;
+	__u64	cap_flags;
 };
 
 struct hns_roce_ib_create_qp {