Re: [PATCH for-next v4 08/12] RDMA/erdma: Add verbs implementation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 3/18/22 8:24 PM, Wenpeng Liang wrote:
On 2022/3/14 14:47, Cheng Xu wrote:
The RDMA verbs implementation of erdma is divided into three files:
erdma_qp.c, erdma_cq.c, and erdma_verbs.c. Internal used functions and
datapath functions of QP/CQ are put in erdma_qp.c and erdma_cq.c, the reset
is in erdma_verbs.c.

reset -> rest.

Will fix.


<...>
+static int erdma_poll_one_cqe(struct erdma_cq *cq, struct erdma_cqe *cqe,
+			      struct ib_wc *wc)
+{
+	struct erdma_dev *dev = to_edev(cq->ibcq.device);
+	struct erdma_qp *qp;
+	struct erdma_kqp *kern_qp;
+	u64 *wqe_hdr;
+	u64 *id_table;
+	u32 qpn = be32_to_cpu(cqe->qpn);
+	u16 wqe_idx = be32_to_cpu(cqe->qe_idx);
+	u32 hdr = be32_to_cpu(cqe->hdr);
+	u16 depth;
+	u8 opcode, syndrome, qtype;
+
+	qp = find_qp_by_qpn(dev, qpn);
+	kern_qp = &qp->kern_qp;

The qp returned by find_qp_by_qpn may be null.


Thanks, this is a bug, will fix.

+
+	qtype = FIELD_GET(ERDMA_CQE_HDR_QTYPE_MASK, hdr);
+	syndrome = FIELD_GET(ERDMA_CQE_HDR_SYNDROME_MASK, hdr);
+	opcode = FIELD_GET(ERDMA_CQE_HDR_OPCODE_MASK, hdr);
+
+	if (qtype == ERDMA_CQE_QTYPE_SQ) {
+		id_table = kern_qp->swr_tbl;
+		depth = qp->attrs.sq_size;
+		wqe_hdr = (u64 *)get_sq_entry(qp, wqe_idx);

The pointer type returned by get_sq_entry is "void *",
which does not need to be cast.


Also will fix.

<...>
+int erdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	struct erdma_cq *cq = to_ecq(ibcq);
+	struct erdma_cqe *cqe;
+	unsigned long flags;
+	u32 owner;
+	u32 ci;
+	int i, ret;
+	int new = 0;
+	u32 hdr;
+
+	spin_lock_irqsave(&cq->kern_cq.lock, flags);
+
+	owner = cq->kern_cq.owner;
+	ci = cq->kern_cq.ci;
+
+	for (i = 0; i < num_entries; i++) {
+		cqe = &cq->kern_cq.qbuf[ci & (cq->depth - 1)];
+
+		hdr = be32_to_cpu(READ_ONCE(cqe->hdr));
+		if (FIELD_GET(ERDMA_CQE_HDR_OWNER_MASK, hdr) != owner)
+			break;
+
+		/* cqbuf should be ready when we poll*/

"poll*/" -> "poll */".

Will fix.


+		dma_rmb();
+		ret = erdma_poll_one_cqe(cq, cqe, wc);
+		ci++;
+		if ((ci & (cq->depth - 1)) == 0)
+			owner = !owner;
+		if (ret)

The return value of erdma_poll_one_cqe is always 0.

As you mentioned above, find_qp_by_qpn may return NULL, and then
erdma_poll_one_cqe should return a non-zero value. The code here is all
right, and I will fix the bug in erdma_poll_one_cqe.


+			continue;
+		wc++;
+		new++;
+	}
+	cq->kern_cq.owner = owner;
+	cq->kern_cq.ci = ci;
+
+	spin_unlock_irqrestore(&cq->kern_cq.lock, flags);
+	return new;
+}
<...>
+struct ib_qp *erdma_get_ibqp(struct ib_device *ibdev, int id)
+{
+	struct erdma_qp *qp = find_qp_by_qpn(to_edev(ibdev), id);
+
+	if (qp)
+		return &qp->ibqp;
+
+	return (struct ib_qp *)NULL;

Redundant cast?

Will fix.

<...>
+int erdma_modify_qp_internal(struct erdma_qp *qp, struct erdma_qp_attrs *attrs,
+			     enum erdma_qp_attr_mask mask)
+{
+	int drop_conn = 0, ret = 0;
+
+	if (!mask)
+		return 0;
+
+	if (!(mask & ERDMA_QP_ATTR_STATE))
+		return 0;
+
+	switch (qp->attrs.state) {
+	case ERDMA_QP_STATE_IDLE:
+	case ERDMA_QP_STATE_RTR:
+		if (attrs->state == ERDMA_QP_STATE_RTS) {
+			ret = erdma_modify_qp_state_to_rts(qp, attrs, mask);
+		} else if (attrs->state == ERDMA_QP_STATE_ERROR) {
+			qp->attrs.state = ERDMA_QP_STATE_ERROR;
+			if (qp->cep) {
+				erdma_cep_put(qp->cep);
+				qp->cep = NULL;
+			}
+			ret = erdma_modify_qp_state_to_stop(qp, attrs, mask);
+		}
+		break;
+	case ERDMA_QP_STATE_RTS:
+		if (attrs->state == ERDMA_QP_STATE_CLOSING) {
+			ret = erdma_modify_qp_state_to_stop(qp, attrs, mask);
+			drop_conn = 1;
+		} else if (attrs->state == ERDMA_QP_STATE_TERMINATE) {
+			qp->attrs.state = ERDMA_QP_STATE_TERMINATE;
+			ret = erdma_modify_qp_state_to_stop(qp, attrs, mask);
+			drop_conn = 1;
+		} else if (attrs->state == ERDMA_QP_STATE_ERROR) {
+			ret = erdma_modify_qp_state_to_stop(qp, attrs, mask);
+			qp->attrs.state = ERDMA_QP_STATE_ERROR;
+			drop_conn = 1;
+		}
+		break;
+	case ERDMA_QP_STATE_TERMINATE:
+		if (attrs->state == ERDMA_QP_STATE_ERROR)
+			qp->attrs.state = ERDMA_QP_STATE_ERROR;
+		break;
+	case ERDMA_QP_STATE_CLOSING:
+		if (attrs->state == ERDMA_QP_STATE_IDLE)
+			qp->attrs.state = ERDMA_QP_STATE_IDLE;
+		else if (attrs->state == ERDMA_QP_STATE_ERROR) {
+			ret = erdma_modify_qp_state_to_stop(qp, attrs, mask);
+			qp->attrs.state = ERDMA_QP_STATE_ERROR;
+		} else if (attrs->state != ERDMA_QP_STATE_CLOSING) {
+			return -ECONNABORTED;
+		}

If the conditional statement has a branch with curly braces,
then all branches use curly braces.


Yes, will fix.

+		break;
+	default:
+		break;
+	}
+
+	if (drop_conn)
+		erdma_qp_cm_drop(qp);

Can this conditional statement be placed in the ERDMA_QP_STATE_RTS branch
of the switch-case alternative branch?

The logic related to drop_conn is concentrated into one piece to improve
code aggregation.


Fine, it's better, and I will fix it.

+
+	return ret;
+}
<...>
+static int erdma_push_one_sqe(struct erdma_qp *qp, u16 *pi,
+			      const struct ib_send_wr *send_wr)
+{
+	struct erdma_write_sqe *write_sqe;
+	struct erdma_send_sqe *send_sqe;
+	struct erdma_readreq_sqe *read_sqe;
+	struct erdma_reg_mr_sqe *regmr_sge;
+	struct erdma_mr *mr;
+	struct ib_rdma_wr *rdma_wr;
+	struct ib_sge *sge;
+	u32 wqe_size, wqebb_cnt, hw_op;
+	int ret;
+	u32 flags = send_wr->send_flags;
+	u32 idx = *pi & (qp->attrs.sq_size - 1);
+	u64 *entry = (u64 *)get_sq_entry(qp, idx);

<...>

+		sge = (struct ib_sge *)get_sq_entry(qp, idx + 1);

<...>

+			inline_data = (u64 *)get_sq_entry(qp, idx + 1);

The pointer type returned by get_sq_entry is "void *",
which does not need to be cast.


Will fix.

<...>
+static int erdma_post_recv_one(struct ib_qp *ibqp,
+			       const struct ib_recv_wr *recv_wr,
+			       const struct ib_recv_wr **bad_recv_wr)

bad_recv_wr is a redundant input parameter.


Will fix.

+{
+	struct erdma_qp *qp = to_eqp(ibqp);
+	struct erdma_rqe *rqe;
+	unsigned int rq_pi;
+	u16 idx;
+
+	rq_pi = qp->kern_qp.rq_pi;
+	idx = rq_pi & (qp->attrs.rq_size - 1);
+	rqe = (struct erdma_rqe *)qp->kern_qp.rq_buf + idx;
+
+	rqe->qe_idx = rq_pi + 1;
+	rqe->qpn = QP_ID(qp);
+
+	if (recv_wr->num_sge == 0) {
+		rqe->length = 0;
+	} else if (recv_wr->num_sge == 1) {
+		rqe->stag = recv_wr->sg_list[0].lkey;
+		rqe->to = recv_wr->sg_list[0].addr;
+		rqe->length = recv_wr->sg_list[0].length;
+	} else {
+		return -EINVAL;
+	}
+
+	*(u64 *)qp->kern_qp.rq_db_info = *(u64 *)rqe;
+	writeq(*(u64 *)rqe, qp->kern_qp.hw_rq_db);
+
+	qp->kern_qp.rwr_tbl[idx] = recv_wr->wr_id;
+	qp->kern_qp.rq_pi = rq_pi + 1;
+
+	return 0;
+}
+
+int erdma_post_recv(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
+		    const struct ib_recv_wr **bad_recv_wr)
+{
+	struct erdma_qp *eqp = to_eqp(qp);
+	int ret = 0;
+	const struct ib_recv_wr *wr = recv_wr;
+	unsigned long flags;
+
+	if (!qp || !recv_wr)
+		return -EINVAL;
+
+	spin_lock_irqsave(&eqp->lock, flags);
+	while (wr) {
+		ret = erdma_post_recv_one(qp, wr, bad_recv_wr);
+		if (ret) {
+			*bad_recv_wr = wr;
+			break;
+		}
+		wr = wr->next;
+	}
+	spin_unlock_irqrestore(&eqp->lock, flags);
+	return ret;
+}

The "ret" does not need to be initialized to 0, it has been reassigned
before the function returns.


Will fix.

<...>
+static int erdma_create_stag(struct erdma_dev *dev, u32 *stag)
+{
+	int stag_idx;
+	u32 key = 0;
+
+	stag_idx = erdma_alloc_idx(&dev->res_cb[ERDMA_RES_TYPE_STAG_IDX]);
+	if (stag_idx < 0)
+		return stag_idx;
+
+	*stag = (stag_idx << 8) | (key & 0xFF);

The "key" is initialized to 0 and never assigned again.


Will fix it.

<...>
+int erdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
+		   int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr)
+{
+	struct erdma_qp *qp;
+	struct erdma_dev *dev;
+
+	if (ibqp && qp_attr && qp_init_attr) {
+		qp = to_eqp(ibqp);
+		dev = to_edev(ibqp->device);
+	} else
+		return -EINVAL;

If the conditional statement has a branch with curly braces,
then all branches use curly braces.


Will fix it.

Thanks,
Cheng Xu

Thanks,
Wenpeng



[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Photo]     [Yosemite News]     [Yosemite Photos]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux