Enables the fastpath verb ib_post_recv.
Signed-off-by: Eddie Wai <eddie....@broadcom.com>
Signed-off-by: Devesh Sharma <devesh.sha...@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.ko...@broadcom.com>
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapa...@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xav...@broadcom.com>
---
drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.c | 100 ++++++++++++++++++
drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.h | 8 ++
drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.c | 133 ++++++++++++++++++++++++
drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.h | 2 +
drivers/infiniband/hw/bnxtre/bnxt_re_main.c | 2 +
5 files changed, 245 insertions(+)
diff --git a/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.c
b/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.c
index 419efe2..67188ce 100644
--- a/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.c
+++ b/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.c
@@ -1107,6 +1107,37 @@ void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
return NULL;
}
+u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
+{
+ struct bnxt_qplib_q *rq = &qp->rq;
+
+ return HWQ_CMP(rq->hwq.prod, &rq->hwq);
+}
+
+dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32
index)
+{
+ return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
+}
+
+void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
+ struct bnxt_qplib_sge *sge)
+{
+ struct bnxt_qplib_q *rq = &qp->rq;
+ u32 sw_prod;
+
+ memset(sge, 0, sizeof(*sge));
+
+ if (qp->rq_hdr_buf) {
+ sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
+ sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
+ sw_prod * qp->rq_hdr_buf_size);
+ sge->lkey = 0xFFFFFFFF;
+ sge->size = qp->rq_hdr_buf_size;
+ return qp->rq_hdr_buf + sw_prod * sge->size;
+ }
+ return NULL;
+}
+
void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
{
struct bnxt_qplib_q *sq = &qp->sq;
@@ -1355,6 +1386,75 @@ int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
return rc;
}
+void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp)
+{
+ struct bnxt_qplib_q *rq = &qp->rq;
+ struct dbr_dbr db_msg = { 0 };
+ u32 sw_prod;
+
+ sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
+ db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) &
+ DBR_DBR_INDEX_MASK);
+ db_msg.type_xid =
+ cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) |
+ DBR_DBR_TYPE_RQ);
+
+ /* Flush the writes to HW Rx WQE before the ringing Rx DB */
+ wmb();
+ __iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64));
+}
+
+int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
+ struct bnxt_qplib_swqe *wqe)
+{
+ struct bnxt_qplib_q *rq = &qp->rq;
+ struct rq_wqe *rqe, **rqe_ptr;
+ struct sq_sge *hw_sge;
+ u32 sw_prod;
+ int i, rc = 0;
+
+ if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
+ dev_err(&rq->hwq.pdev->dev,
+ "QPLIB: FP: QP (0x%x) is in the 0x%x state",
+ qp->id, qp->state);
+ rc = -EINVAL;
+ goto done;
+ }
+ if (HWQ_CMP((rq->hwq.prod + 1), &rq->hwq) ==
+ HWQ_CMP(rq->hwq.cons, &rq->hwq)) {
+ dev_err(&rq->hwq.pdev->dev,
+ "QPLIB: FP: QP (0x%x) RQ is full!", qp->id);
+ rc = -EINVAL;
+ goto done;
+ }
+ sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
+ rq->swq[sw_prod].wr_id = wqe->wr_id;
+
+ rqe_ptr = (struct rq_wqe **)rq->hwq.pbl_ptr;
+ rqe = &rqe_ptr[RQE_PG(sw_prod)][RQE_IDX(sw_prod)];
+
+ memset(rqe, 0, BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
+
+ /* Calculate wqe_size16 and data_len */
+ for (i = 0, hw_sge = (struct sq_sge *)rqe->data;
+ i < wqe->num_sge; i++, hw_sge++) {
+ hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
+ hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
+ hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
+ }
+ rqe->wqe_type = wqe->type;
+ rqe->flags = wqe->flags;
+ rqe->wqe_size = wqe->num_sge +
+ ((offsetof(typeof(*rqe), data) + 15) >> 4);
+
+ /* Supply the rqe->wr_id index to the wr_id_tbl for now */
+ rqe->wr_id[0] = cpu_to_le32(sw_prod);
+
+ rq->hwq.prod++;
+done:
+ return rc;
+}
+
/* CQ */
/* Spinlock must be held */
diff --git a/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.h
b/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.h
index 7fe98db..d9f2611 100644
--- a/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.h
+++ b/drivers/infiniband/hw/bnxtre/bnxt_qplib_fp.h
@@ -395,9 +395,17 @@ int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct
bnxt_qplib_qp *qp);
int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp
*qp);
void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
struct bnxt_qplib_sge *sge);
+void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
+ struct bnxt_qplib_sge *sge);
+u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp);
+dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp,
+ u32 index);
void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp);
int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
struct bnxt_qplib_swqe *wqe);
+void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp);
+int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
+ struct bnxt_qplib_swqe *wqe);
int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);
int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq
*cq);
diff --git a/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.c
b/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.c
index 540f2f2..5d6c89c 100644
--- a/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.c
+++ b/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.c
@@ -1644,6 +1644,61 @@ static int bnxt_re_build_qp1_send_v2(struct bnxt_re_qp
*qp,
return rc;
}
+/* For the MAD layer, it only provides the recv SGE the size of
+ * ib_grh + MAD datagram. No Ethernet headers, Ethertype, BTH, DETH,
+ * nor RoCE iCRC. The Cu+ solution must provide buffer for the entire
+ * receive packet (334 bytes) with no VLAN and then copy the GRH
+ * and the MAD datagram out to the provided SGE.
+ */
+static int bnxt_re_build_qp1_shadow_qp_recv(struct bnxt_re_qp *qp,
+ struct ib_recv_wr *wr,
+ struct bnxt_qplib_swqe *wqe,
+ int payload_size)
+{
+ struct bnxt_qplib_sge ref, sge;
+ int rc = 0;
+ u32 rq_prod_index;
+ struct bnxt_re_sqp_entries *sqp_entry;
+
+ rq_prod_index = bnxt_qplib_get_rq_prod_index(&qp->qplib_qp);
+
+ if (bnxt_qplib_get_qp1_rq_buf(&qp->qplib_qp, &sge)) {
+ /* Create 1 SGE to receive the entire
+ * ethernet packet
+ */
+ /* Save the reference from ULP */
+ ref.addr = wqe->sg_list[0].addr;
+ ref.lkey = wqe->sg_list[0].lkey;
+ ref.size = wqe->sg_list[0].size;
+
+ sqp_entry = &qp->rdev->sqp_tbl[rq_prod_index];
+
+ /* SGE 1 */
+ wqe->sg_list[0].addr = sge.addr;
+ wqe->sg_list[0].lkey = sge.lkey;
+ wqe->sg_list[0].size = BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2;
+ sge.size -= wqe->sg_list[0].size;
+ if (sge.size < 0) {
+ dev_err(rdev_to_dev(qp->rdev),
+ "QP1 rq buffer is empty!");
+ rc = -ENOMEM;
+ goto done;
+ }
+
+ sqp_entry->sge.addr = ref.addr;
+ sqp_entry->sge.lkey = ref.lkey;
+ sqp_entry->sge.size = ref.size;
+ /* Store the wrid for reporting completion */
+ sqp_entry->wrid = wqe->wr_id;
+ /* change the wqe->wrid to table index */
+ wqe->wr_id = rq_prod_index;
+ }
+ return 0;
+done:
+
+ return rc;
+}
+
static int is_ud_qp(struct bnxt_re_qp *qp)
{
return qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD;
@@ -1988,6 +2043,84 @@ int bnxt_re_post_send(struct ib_qp *ib_qp, struct
ib_send_wr *wr,
return rc;
}
+int bnxt_re_post_recv_shadow_qp(struct bnxt_re_dev *rdev,
+ struct bnxt_re_qp *qp,
+ struct ib_recv_wr *wr)
+{
+ struct bnxt_qplib_swqe wqe;
+ int rc = 0, payload_sz = 0;
+
+ memset(&wqe, 0, sizeof(wqe));
+ while (wr) {
+ /* House keeping */
+ memset(&wqe, 0, sizeof(wqe));
+
+ /* Common */
+ wqe.num_sge = wr->num_sge;
+ if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
+ dev_err(rdev_to_dev(rdev),
+ "Limit exceeded for Receive SGEs");
+ rc = -EINVAL;
+ goto bad;
+ }
+ payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe.sg_list,
+ wr->num_sge);
+ wqe.wr_id = wr->wr_id;
+ wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
+
+ if (!rc)
+ rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
+bad:
+ if (rc)
+ break;
+
+ wr = wr->next;
+ }
+ bnxt_qplib_post_recv_db(&qp->qplib_qp);
+ return rc;
+}
+
+int bnxt_re_post_recv(struct ib_qp *ib_qp, struct ib_recv_wr *wr,
+ struct ib_recv_wr **bad_wr)
+{
+ struct bnxt_re_qp *qp = to_bnxt_re(ib_qp, struct bnxt_re_qp, ib_qp);
+ struct bnxt_qplib_swqe wqe;
+ int rc = 0, payload_sz = 0;
+
+ while (wr) {
+ /* House keeping */
+ memset(&wqe, 0, sizeof(wqe));
+
+ /* Common */
+ wqe.num_sge = wr->num_sge;
+ if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
+ dev_err(rdev_to_dev(qp->rdev),
+ "Limit exceeded for Receive SGEs");
+ rc = -EINVAL;
+ goto bad;
+ }
+
+ payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe.sg_list,
+ wr->num_sge);
+ wqe.wr_id = wr->wr_id;
+ wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
+
+ if (ib_qp->qp_type == IB_QPT_GSI)
+ rc = bnxt_re_build_qp1_shadow_qp_recv(qp, wr, &wqe,
+ payload_sz);
+ if (!rc)
+ rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
+bad:
+ if (rc) {
+ *bad_wr = wr;
+ break;
+ }
+ wr = wr->next;
+ }
+ bnxt_qplib_post_recv_db(&qp->qplib_qp);
+ return rc;
+}
+
/* Completion Queues */
int bnxt_re_destroy_cq(struct ib_cq *ib_cq)
{
diff --git a/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.h
b/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.h
index becdcdc..9f3dd49 100644
--- a/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.h
+++ b/drivers/infiniband/hw/bnxtre/bnxt_re_ib_verbs.h
@@ -164,6 +164,8 @@ int bnxt_re_query_qp(struct ib_qp *qp, struct ib_qp_attr
*qp_attr,
int bnxt_re_destroy_qp(struct ib_qp *qp);
int bnxt_re_post_send(struct ib_qp *qp, struct ib_send_wr *send_wr,
struct ib_send_wr **bad_send_wr);
+int bnxt_re_post_recv(struct ib_qp *qp, struct ib_recv_wr *recv_wr,
+ struct ib_recv_wr **bad_recv_wr);
struct ib_cq *bnxt_re_create_cq(struct ib_device *ibdev,
const struct ib_cq_init_attr *attr,
struct ib_ucontext *context,
diff --git a/drivers/infiniband/hw/bnxtre/bnxt_re_main.c
b/drivers/infiniband/hw/bnxtre/bnxt_re_main.c
index 14d1147..73dfadd 100644
--- a/drivers/infiniband/hw/bnxtre/bnxt_re_main.c
+++ b/drivers/infiniband/hw/bnxtre/bnxt_re_main.c
@@ -452,6 +452,8 @@ static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
ibdev->destroy_qp = bnxt_re_destroy_qp;
ibdev->post_send = bnxt_re_post_send;
+ ibdev->post_recv = bnxt_re_post_recv;
+
ibdev->create_cq = bnxt_re_create_cq;
ibdev->destroy_cq = bnxt_re_destroy_cq;
ibdev->req_notify_cq = bnxt_re_req_notify_cq;
--
2.5.5
